\n\nExample:\n logtree.log_text(\"Processing complete\")\n logtree.log_text(\"Score: 0.95\", div_class=\"score\")",
"args": [
"text"
],
"lineno": 624
},
{
"id": 1550,
"label": "log_html()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Log raw HTML (not escaped).\n\nArgs:\n html: HTML string to insert verbatim\n div_class: If set, wrap in
\n\nExample:\n logtree.log_html(\"
Important\")\n logtree.log_html(conversation_html, div_class=\"conversation\")",
"args": [
"html"
],
"lineno": 647
},
{
"id": 1551,
"label": "log_formatter()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Log an object that knows how to format itself as HTML.\n\nThe formatter's CSS will be automatically included in the trace output.\nCSS is deduplicated per trace, so logging multiple objects of the same\ntype only includes the CSS once.\n\nArgs:\n formatter: Object implementing the Formatter protocol (to_html() and get_css())\n\nExample:\n from tinker_cookbook.utils.logtree_formatters import ConversationFormatter\n\n logtree.log_formatter(ConversationFormatter(messages=[...]))",
"args": [
"formatter"
],
"lineno": 674
},
{
"id": 1552,
"label": "details()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Log collapsible details block.\n\nArgs:\n text: Content text\n summary: Summary text (what you see when collapsed)\n pre: If True, use
(preserves whitespace), else \n\nExample:\n logtree.details(long_chain_of_thought, summary=\"CoT Reasoning\", pre=True)",
"args": [
"text"
],
"lineno": 705
},
{
"id": 1553,
"label": "header()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Log an inline header.\n\nArgs:\n text: Header text\n level: Header level (1-6), or None to auto-compute from scope depth\n\nExample:\n logtree.header(\"Results\")\n logtree.header(\"Subsection\", level=4)",
"args": [
"text"
],
"lineno": 731
},
{
"id": 1554,
"label": "table()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Log a table from various data types.\n\nSupports:\n- pandas.DataFrame\n- list[dict]\n- list[list]\n\nDoes NOT support raw dict (use table_from_dict or table_from_dict_of_lists).\n\nArgs:\n obj: Data object\n caption: Optional caption text\n\nExample:\n logtree.table(df, caption=\"Results\")\n logtree.table([{\"name\": \"Alice\", \"score\": 95}, {\"name\": \"Bob\", \"score\": 87}])",
"args": [
"obj"
],
"lineno": 755
},
{
"id": 1555,
"label": "table_from_dict()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Log a two-column key-value table from a dict.\n\nArgs:\n data: Dictionary to display\n caption: Optional caption\n key_header: Column header for keys\n value_header: Column header for values\n sort_by: \"key\", \"value\", or None\n\nExample:\n logtree.table_from_dict({\"lr\": 0.001, \"batch_size\": 32}, caption=\"Hyperparams\")",
"args": [
"data"
],
"lineno": 809
},
{
"id": 1556,
"label": "table_from_dict_of_lists()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Log a columnar table from dict of lists.\n\nArgs:\n columns: Dict where keys are column names, values are column data\n caption: Optional caption\n order: Column order (if None, use insertion order)\n\nExample:\n logtree.table_from_dict_of_lists({\n \"name\": [\"Alice\", \"Bob\"],\n \"score\": [95, 87]\n })",
"args": [
"columns"
],
"lineno": 840
},
{
"id": 1557,
"label": "_table_from_list_of_dicts()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Helper: create table from list of dicts.",
"args": [
"data"
],
"lineno": 876
},
{
"id": 1558,
"label": "_table_from_list_of_lists()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Helper: create HTML table from list of lists.",
"args": [
"rows"
],
"lineno": 889
},
{
"id": 1559,
"label": "write_html_with_default_style()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Write a complete HTML document with default styling.\n\nArgs:\n body_html: Body HTML (with or without tags)\n path: Output file path\n title: Document title\n theme: Optional theme\n lang: HTML lang attribute\n extra_head: Extra content for ",
"args": [
"body_html",
"path"
],
"lineno": 924
},
{
"id": 1560,
"label": "jinja_context()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Create a context dict for Jinja2 templates.\n\nArgs:\n trace: Trace object\n **extra: Additional context variables\n\nReturns:\n Dict with standard keys: title, generated_at, started_at, body_html, head_html",
"args": [
"trace"
],
"lineno": 968
},
{
"id": 1561,
"label": "render_with_jinja()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Render using Jinja2 (requires jinja2 to be installed).\n\nArgs:\n env: jinja2.Environment instance\n template_name: Template file name\n context: Template context\n write_to: Optional path to write output\n\nReturns:\n Rendered HTML string",
"args": [
"env",
"template_name"
],
"lineno": 989
},
{
"id": 1562,
"label": "ml_log.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/utils/ml_log.py",
"value": 26.94,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/utils/ml_log.py",
"level": "file",
"preview": ""
},
{
"id": 1563,
"label": "Logger",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Abstract base class for loggers.",
"bases": [
"ABC"
],
"lineno": 74
},
{
"id": 1564,
"label": "_PermissiveJSONEncoder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "A JSON encoder that handles non-encodable objects by converting them to their type string.",
"bases": [],
"lineno": 104
},
{
"id": 1565,
"label": "JsonLogger",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Logger that writes metrics to a JSONL file.",
"bases": [
"Logger"
],
"lineno": 115
},
{
"id": 1566,
"label": "PrettyPrintLogger",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Logger that displays metrics in a formatted table in the console.",
"bases": [
"Logger"
],
"lineno": 146
},
{
"id": 1567,
"label": "WandbLogger",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Logger for Weights & Biases.",
"bases": [
"Logger"
],
"lineno": 199
},
{
"id": 1568,
"label": "NeptuneLogger",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Logger for Neptune.",
"bases": [
"Logger"
],
"lineno": 250
},
{
"id": 1569,
"label": "TrackioLogger",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Logger for Trackio.",
"bases": [
"Logger"
],
"lineno": 300
},
{
"id": 1570,
"label": "MultiplexLogger",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Logger that forwards operations to multiple child loggers.",
"bases": [
"Logger"
],
"lineno": 340
},
{
"id": 1571,
"label": "dump_config()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Convert configuration object to JSON-serializable format.",
"args": [
"config"
],
"lineno": 48
},
{
"id": 1572,
"label": "_maybe_truncate_repr()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"value"
],
"lineno": 184
},
{
"id": 1573,
"label": "_rich_console_use_logger()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"console"
],
"lineno": 192
},
{
"id": 1574,
"label": "setup_logging()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Set up logging infrastructure with multiple backends.\n\nArgs:\n log_dir: Directory for logs\n wandb_project: W&B project name (if None, W&B logging is skipped)\n wandb_name: W&B run name\n config: Configuration object to log\n do_configure_logging_module: Whether to configure the logging module\n\nReturns:\n MultiplexLogger that combines all enabled loggers",
"args": [
"log_dir",
"wandb_project",
"wandb_name",
"config",
"do_configure_logging_module"
],
"lineno": 382
},
{
"id": 1575,
"label": "configure_logging_module()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Configure logging to console (color) and file (plain), forcing override of prior config.",
"args": [
"path",
"level"
],
"lineno": 479
},
{
"id": 1576,
"label": "code_state.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/utils/code_state.py",
"value": 15.134,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/utils/code_state.py",
"level": "file",
"preview": ""
},
{
"id": 1577,
"label": "code_state()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Return a single diff-formatted string that captures the current code state for the\nprovided Python modules. For each module, we:\n\n- Locate the module on the filesystem\n- Discover the enclosing Git repository (the module may live inside a larger repo)\n- Record the current commit hash (HEAD)\n- Include combined staged+unstaged changes (i.e., diff vs HEAD) for the entire\n containing Git repository (repo-wide). Subtree diffs are omitted to avoid\n duplication.\n\nThe output is suitable for storage alongside experiment or training metadata to\nreproduce the exact code state later. When multiple modules are provided, their\nsections are concatenated in order.\n\nParameters:\n- modules: sequence of module import names (e.g., \"tinker_cookbook.rl\") or already-\n imported module objects. All entries must be either `str` or `ModuleType`.\n\nReturns:\n- A string beginning with a header per module of the form:\n \"### module:
(repo: ) @ \" followed by\n the staged and unstaged `git diff` outputs restricted to that module directory.\n If a module is not in a Git repository, a short note is included instead.",
"args": [
"modules"
],
"lineno": 11
},
{
"id": 1578,
"label": "format_colorized.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/utils/format_colorized.py",
"value": 11.536,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/utils/format_colorized.py",
"level": "file",
"preview": ""
},
{
"id": 1579,
"label": "format_colorized()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Colour-code text according to per-token weights.\n\n* Cyan text \u2192 weight > 0\n* Yellow text \u2192 weight = 0\n* Red text \u2192 weight < 0\n\nThe function minimises ANSI escape sequences by wrapping *runs* of\nlike-coloured tokens, and decodes each run in a single call so that\nmulti-byte or multibyte-character languages (e.g. CJK) render correctly.",
"args": [
"tokens",
"weights",
"tokenizer",
"draw_newline_arrow"
],
"lineno": 5
},
{
"id": 1580,
"label": "misc_utils.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/utils/misc_utils.py",
"value": 12.783999999999999,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/utils/misc_utils.py",
"level": "file",
"preview": ""
},
{
"id": 1581,
"label": "timed()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"key",
"metrics"
],
"lineno": 19
},
{
"id": 1582,
"label": "dict_mean()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"list_of_dicts"
],
"lineno": 30
},
{
"id": 1583,
"label": "all_same()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"xs"
],
"lineno": 38
},
{
"id": 1584,
"label": "lookup_func()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "path.to.module:func_name or func_name (assumes default_module)",
"args": [
"path_to_func",
"default_module"
],
"lineno": 42
},
{
"id": 1585,
"label": "split_list()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Split a sequence into a list of lists, where the sizes are as equal as possible,\nand the long and short lists are as uniformly distributed as possible.\n\nArgs:\n lst: The sequence to split\n num_splits: Number of sublists to create\n\nReturns:\n A list of sublists with sizes differing by at most 1\n\nRaises:\n ValueError: If num_splits > len(lst) or num_splits <= 0\n\nExamples:\n >>> split_list([1, 2, 3, 4, 5], 2)\n [[1, 2, 3], [4, 5]]\n >>> split_list([1, 2, 3, 4, 5], 3)\n [[1, 2], [3, 4], [5]]",
"args": [
"lst",
"num_splits"
],
"lineno": 58
},
{
"id": 1586,
"label": "concat_lists()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"list_of_lists"
],
"lineno": 88
},
{
"id": 1587,
"label": "not_none()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"x"
],
"lineno": 92
},
{
"id": 1588,
"label": "trace.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/utils/trace.py",
"value": 24.054000000000002,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/utils/trace.py",
"level": "file",
"preview": ""
},
{
"id": 1589,
"label": "EventType",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Chrome Trace/Perfetto Event type",
"bases": [
"str",
"Enum"
],
"lineno": 17
},
{
"id": 1590,
"label": "TraceEvent",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Represents a trace event in Chrome Trace/Perfetto Format",
"bases": [],
"lineno": 26
},
{
"id": 1591,
"label": "ScopeContext",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 53
},
{
"id": 1592,
"label": "TraceCollector",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Collects trace events and exports them in Chrome Trace/Perfetto Format.",
"bases": [],
"lineno": 62
},
{
"id": 1593,
"label": "FunctionCallContext",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Context information for a function call",
"bases": [],
"lineno": 176
},
{
"id": 1594,
"label": "CreateTraceEventsResult",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 187
},
{
"id": 1595,
"label": "_atexit_trace_shutdown()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 147
},
{
"id": 1596,
"label": "trace_init()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Initialize the trace collector. Must be called before any trace events are created.",
"args": [
"flush_interval_sec",
"output_file"
],
"lineno": 157
},
{
"id": 1597,
"label": "trace_shutdown()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Shutdown the trace collector and flush any remaining events.",
"args": [],
"lineno": 166
},
{
"id": 1598,
"label": "_create_trace_events()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Create trace events and context information for a function call.",
"args": [
"func"
],
"lineno": 194
},
{
"id": 1599,
"label": "_create_end_event()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Create an end trace event for a function call.",
"args": [
"func",
"function_call_context"
],
"lineno": 264
},
{
"id": 1600,
"label": "scope()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Decorator for tracing both async and sync functions. In the resulting trace:\n- Each track represents a coroutine (or a sync function if not a coroutine)\n- A thread is a group of tracks, representing all the coroutines running on that thread\n\nFor better tracking, make sure to name all coroutines so that we can group them\nproperly in the trace.\n\nExample usage:\n\nfrom tinker_cookbook.utils.trace import scope, trace_init, get_scope_context\n\n@scope\nasync def foo():\n await asyncio.sleep(0.1)\n # Log additional attributes for this function call into the trace\n context = get_scope_context()\n context.attributes[\"foo\"] = 1\n context.attributes[\"foo2\"] = \"abc\"\n await bar()\n\n@scope\nasync def bar():\n # Name the coroutines so that we can group them properly in the trace\n await asyncio.gather(\n asyncio.create_task(baz(), name=\"baz\"),\n asyncio.create_task(baz(), name=\"baz2\"),\n )\n\n@scope\nasync def main():\n await foo()\n\nif __name__ == \"__main__\":\n trace_init()\n asyncio.run(main())",
"args": [
"func"
],
"lineno": 288
},
{
"id": 1601,
"label": "get_scope_context()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Call this to get the current scope's context. This allows the functions\nto log additional attributes into the trace.\n\nExample usage:\n\n@scope\nasync def foo():\n context = get_scope_context()\n context.attributes[\"foo\"] = 1\n context.attributes[\"foo2\"] = \"abc\"\n await bar()",
"args": [],
"lineno": 390
},
{
"id": 1602,
"label": "convert_jsonl_to_json_main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Helper script to convert the trace events format into a visualizable format",
"args": [],
"lineno": 410
},
{
"id": 1603,
"label": "logtree_formatters.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/utils/logtree_formatters.py",
"value": 12.486,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/utils/logtree_formatters.py",
"level": "file",
"preview": ""
},
{
"id": 1604,
"label": "ConversationFormatter",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Formatter for conversation messages.\n\nRenders a list of messages as a styled conversation with role-based coloring.",
"bases": [],
"lineno": 15
},
{
"id": 1605,
"label": "metrics.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/metrics.py",
"value": 16.637999999999998,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/metrics.py",
"level": "file",
"preview": ""
},
{
"id": 1606,
"label": "compute_kl_sample_train()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Compute KL divergence metrics between sampling and training logprobs.",
"args": [
"data_D",
"training_logprobs_D"
],
"lineno": 18
},
{
"id": 1607,
"label": "discounted_future_sum_vectorized()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Compute discounted sum of future values for each position using a vectorized approach.\n\nArgs:\n x (np.ndarray): 1D array of rewards.\n gamma (float): Discount factor.\n\nReturns:\n np.ndarray: discounted sum of future values.",
"args": [
"x",
"gamma"
],
"lineno": 133
},
{
"id": 1608,
"label": "compute_sampling_client_metrics()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Compute metrics about sampling clients used to generate trajectory groups.",
"args": [
"wrapped_trajectory_groups"
],
"lineno": 150
},
{
"id": 1609,
"label": "data_processing.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/data_processing.py",
"value": 17.247,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/data_processing.py",
"level": "file",
"preview": ""
},
{
"id": 1610,
"label": "compute_advantages()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Compute advantages for each trajectory, centered within groups.",
"args": [
"trajectory_groups_P"
],
"lineno": 20
},
{
"id": 1611,
"label": "_is_prefix()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Check if seq1 is a prefix of seq2.",
"args": [
"seq1",
"seq2"
],
"lineno": 37
},
{
"id": 1612,
"label": "_flat_ob_token_len()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"flat_ob"
],
"lineno": 44
},
{
"id": 1613,
"label": "_to_input_targets()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"model_input"
],
"lineno": 54
},
{
"id": 1614,
"label": "_flat_ob_to_model_input()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"flat_ob"
],
"lineno": 60
},
{
"id": 1615,
"label": "_flatten_chunks()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"chunks"
],
"lineno": 79
},
{
"id": 1616,
"label": "trajectory_to_data()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Return one or more Datum objects corresponding to the trajectory.\nIf the sequence grows by appending, i.e., each successive observation contains\nthe previous observation+action as a prefix, then we can return a single Datum.\nHowever, if we get a sequence that's not an extension of the previous sequence,\nthen that results in a new Datum.\n\nFor example, let O1 denote a chunk of observation tokens, and let A1 denote an action.\n\nThen let's say ob_ac_pairs is as follows.\n\n(O1, A1)\n(O1+A1+O2, A2)\n(O3, A3)\n\nThen we will merge the first two observation-action pairs into a single Datum,\nand the last observation-action pair into a separate Datum.",
"args": [
"traj",
"traj_advantage"
],
"lineno": 89
},
{
"id": 1617,
"label": "assemble_training_data()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Convert trajectories to training data format.",
"args": [
"trajectory_groups_P",
"advantages_P"
],
"lineno": 176
},
{
"id": 1618,
"label": "remove_constant_reward_groups()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"trajectory_groups_P"
],
"lineno": 198
},
{
"id": 1619,
"label": "preference_envs.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/preference_envs.py",
"value": 20.817,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/preference_envs.py",
"level": "file",
"preview": ""
},
{
"id": 1620,
"label": "PreferenceEnv",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"Env"
],
"lineno": 37
},
{
"id": 1621,
"label": "TournamentPattern",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"StrEnum"
],
"lineno": 73
},
{
"id": 1622,
"label": "PairwisePreferenceGroupBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"EnvGroupBuilder"
],
"lineno": 102
},
{
"id": 1623,
"label": "PairwisePreferenceDataset",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDataset"
],
"lineno": 218
},
{
"id": 1624,
"label": "PairwisePreferenceRLDatasetBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDatasetBuilder"
],
"lineno": 260
},
{
"id": 1625,
"label": "get_pairs_chunked()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Get pairs of indices of matchups of n players. If chunk_size < n, then we divide the players\ninto groups of at most chunk_size and get the matchup indices within each group.",
"args": [
"n",
"pattern",
"chunk_size"
],
"lineno": 78
},
{
"id": 1626,
"label": "get_pairs()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"n",
"pattern"
],
"lineno": 92
},
{
"id": 1627,
"label": "problem_env.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/problem_env.py",
"value": 13.206,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/problem_env.py",
"level": "file",
"preview": ""
},
{
"id": 1628,
"label": "ProblemEnv",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"Env"
],
"lineno": 23
},
{
"id": 1629,
"label": "ProblemGroupBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"EnvGroupBuilder"
],
"lineno": 88
},
{
"id": 1630,
"label": "metric_util.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/metric_util.py",
"value": 15.464,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/metric_util.py",
"level": "file",
"preview": ""
},
{
"id": 1631,
"label": "RLTestSetEvaluator",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"SamplingClientEvaluator"
],
"lineno": 105
},
{
"id": 1632,
"label": "_compute_by_group_metrics()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"trajectory_groups_P",
"good_thresh"
],
"lineno": 16
},
{
"id": 1633,
"label": "compute_trajectory_metrics()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"trajectory_groups_P",
"taglist_P"
],
"lineno": 35
},
{
"id": 1634,
"label": "_compute_trajectory_metrics()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Compute metrics for the trajectory groups.",
"args": [
"trajectory_groups_P"
],
"lineno": 59
},
{
"id": 1635,
"label": "dataset_to_env_group_builders()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Get the whole dataset as a list of env group builders.",
"args": [
"dataset"
],
"lineno": 98
},
{
"id": 1636,
"label": "types.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/types.py",
"value": 14.949,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/types.py",
"level": "file",
"preview": ""
},
{
"id": 1637,
"label": "StepResult",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 21
},
{
"id": 1638,
"label": "Transition",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 30
},
{
"id": 1639,
"label": "Env",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Stateful environment that a single agent interacts with.\nDiscard after running for one episode.",
"bases": [
"ABC"
],
"lineno": 38
},
{
"id": 1640,
"label": "Trajectory",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "A sequence of observations and actions, resulting from running a single agent in a single\nenvironment.",
"bases": [],
"lineno": 54
},
{
"id": 1641,
"label": "EnvGroupBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Builds a group of environments. The group will be used in the following way:\n\n- Algorithms like GRPO will center rewards across the group.\n- The reward function (compute_group_rewards) has access to the trajectories from the\n whole group, even though many reward functions will evaluate each one independently.\n\n - For example, this enables us to use pairwise reward models that look at a pair of\n trajectories at a time. With such a reward model, we effectively have a multi-agent\n environment, where the agents are playing a zero-sum game.\n\nGroups can be used in two ways, in practice:\n\n- To define a multi-agent environment\n- As a part of the *algorithm* (e.g. GRPO), when dealing with single-agent tasks.",
"bases": [
"ABC"
],
"lineno": 64
},
{
"id": 1642,
"label": "TrajectoryGroup",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "A group of trajectories, resulting from instantiating a group of environments using an\nEnvGroupBuilder, doing a rollout for each environment, and computing the rewards.",
"bases": [],
"lineno": 111
},
{
"id": 1643,
"label": "RLDataset",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "A dataset that produces batches of EnvGroups. This is the kind of dataset used by\ntraining algorithms.",
"bases": [
"ABC"
],
"lineno": 133
},
{
"id": 1644,
"label": "RLDatasetBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Abstract class for building RL datasets.",
"bases": [],
"lineno": 149
},
{
"id": 1645,
"label": "train.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/train.py",
"value": 40,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/train.py",
"level": "file",
"preview": ""
},
{
"id": 1646,
"label": "StreamMinibatchConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Configuration for training with minibatch streaming.\nOnce we have accumulated enough trajectories for a minibatch, we will\nimmediately train on them, instead of waiting for the full batch of\ntrajectories to be ready.",
"bases": [],
"lineno": 197
},
{
"id": 1647,
"label": "AsyncConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Configuration for async RL training",
"bases": [],
"lineno": 215
},
{
"id": 1648,
"label": "Config",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 227
},
{
"id": 1649,
"label": "WrappedTrajectoryGroup",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "A wrapper around a trajectory group that includes metadata about how it was generated.\nUsed when we need to overlap sampling and training.",
"bases": [],
"lineno": 413
},
{
"id": 1650,
"label": "_get_evaluator_name()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"evaluator"
],
"lineno": 49
},
{
"id": 1651,
"label": "_get_logtree_scope()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Creates a context manager; all log inside this context will be logged under the section `scope_name`.\nIt will create a file with the path of log_path/f_name.html\nIf num_groups_to_log is 0, it will disable logging (but note that this function does not actually implement the logic for logging itself!)",
"args": [
"log_path",
"num_groups_to_log",
"f_name",
"scope_name"
],
"lineno": 58
},
{
"id": 1652,
"label": "_select_representative_inds()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"scores",
"num_inds"
],
"lineno": 75
},
{
"id": 1653,
"label": "print_group()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Print a subset of the trajectory group to the console.",
"args": [
"traj_group",
"tokenizer"
],
"lineno": 83
},
{
"id": 1654,
"label": "remove_mask()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"datum"
],
"lineno": 149
},
{
"id": 1655,
"label": "rollouts.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/rollouts.py",
"value": 12.925,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/rollouts.py",
"level": "file",
"preview": ""
},
{
"id": 1656,
"label": "play_w_env.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/play_w_env.py",
"value": 12.971,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/rl/play_w_env.py",
"level": "file",
"preview": ""
},
{
"id": 1657,
"label": "ManualPolicy",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"TokenCompleter"
],
"lineno": 30
},
{
"id": 1658,
"label": "print_trajectory_summary()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Print a summary of the completed trajectory.",
"args": [
"trajectory"
],
"lineno": 48
},
{
"id": 1659,
"label": "custom_evaluators.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/custom_evaluators.py",
"value": 13.017,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/custom_evaluators.py",
"level": "file",
"preview": ""
},
{
"id": 1660,
"label": "CustomEvaluator",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "A toy SamplingClientEvaluator that runs a custom evaluation and returns its metrics.",
"bases": [
"SamplingClientEvaluator"
],
"lineno": 11
},
{
"id": 1661,
"label": "grader_fn()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"response",
"target"
],
"lineno": 79
},
{
"id": 1662,
"label": "inspect_evaluators.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/inspect_evaluators.py",
"value": 14.349,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/inspect_evaluators.py",
"level": "file",
"preview": ""
},
{
"id": 1663,
"label": "InspectEvaluatorBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Configuration for inspect evaluation.\nThis class provides a structured way to configure inspect evaluation\nparameters that can be used both in training configs and evaluator builders.",
"bases": [],
"lineno": 18
},
{
"id": 1664,
"label": "InspectEvaluator",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "A SamplingClientEvaluator that runs inspect tasks and returns their metrics.",
"bases": [
"SamplingClientEvaluator"
],
"lineno": 47
},
{
"id": 1665,
"label": "custom_inspect_task.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/custom_inspect_task.py",
"value": 12.242,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/custom_inspect_task.py",
"level": "file",
"preview": ""
},
{
"id": 1666,
"label": "example_lm_as_judge()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Example task using LLM-as-a-judge scoring.\n\nNote: The grader model defaults to the model being evaluated.\nTo use a different grader model, specify it with --model-grader when using inspect directly.",
"args": [],
"lineno": 51
},
{
"id": 1667,
"label": "inspect_utils.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/inspect_utils.py",
"value": 15.972000000000001,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/inspect_utils.py",
"level": "file",
"preview": ""
},
{
"id": 1668,
"label": "InspectAPIFromTinkerSampling",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "A model API wrapper that adapts tinker sampling clients to the inspect API interface.\n\nThis class can be initialized either with a model_path (for standalone use)\nor with a sampling_client (for use in evaluators).",
"bases": [
"InspectAIModelAPI"
],
"lineno": 58
},
{
"id": 1669,
"label": "get_model_usage()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Given a tokenized prompt and a list of responses, return the number of tokens used/generated by the model.",
"args": [
"tokenized_prompt",
"responses"
],
"lineno": 30
},
{
"id": 1670,
"label": "convert_inspect_messages()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"messages"
],
"lineno": 45
},
{
"id": 1671,
"label": "run_inspect_evals.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/run_inspect_evals.py",
"value": 11.828,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/run_inspect_evals.py",
"level": "file",
"preview": ""
},
{
"id": 1672,
"label": "Config",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"InspectEvaluatorBuilder"
],
"lineno": 12
},
{
"id": 1673,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/README.md",
"value": 10.235,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/README.md",
"level": "file",
"preview": "Tinker can integrate with oss eval framework like inspect ai easily (`run_inspect_evals.py`), or you can create a simple evaluator yourself {`custom_evaluators.py`}. Check out our [docs](https://tinker-docs.thinkingmachines.ai/evals).\n"
},
{
"id": 1674,
"label": "evaluators.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/evaluators.py",
"value": 10.774000000000001,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/eval/evaluators.py",
"level": "file",
"preview": ""
},
{
"id": 1675,
"label": "TrainingClientEvaluator",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "An evaluator that takes in a TrainingClient",
"bases": [],
"lineno": 10
},
{
"id": 1676,
"label": "SamplingClientEvaluator",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "An evaluator that takes in a TokenCompleter",
"bases": [],
"lineno": 19
},
{
"id": 1677,
"label": "train_dpo.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/preference/train_dpo.py",
"value": 24.919,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/preference/train_dpo.py",
"level": "file",
"preview": ""
},
{
"id": 1678,
"label": "Config",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Configuration for Direct Preference Optimization (DPO) training.",
"bases": [],
"lineno": 28
},
{
"id": 1679,
"label": "create_dpo_clients()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Create and configure the training client and reference sampling client for DPO.\n\nCreates the main training client and a reference sampling client.\nThe reference sampling client is used to compute the reference model's log probabilities\nfor the DPO loss computation more efficiently than a separate training client.\n\nArgs:\n config: DPO configuration object\n resume_info: Resume information from checkpoint\n\nReturns:\n Tuple of (main training client, reference sampling client)",
"args": [
"config",
"resume_info"
],
"lineno": 71
},
{
"id": 1680,
"label": "compute_dpo_loss()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Compute DPO loss and metrics.\n\nArgs:\n chosen_logprobs: Log probabilities for chosen responses\n rejected_logprobs: Log probabilities for rejected responses\n chosen_ref_logprobs: Reference log probabilities for chosen responses\n rejected_ref_logprobs: Reference log probabilities for rejected responses\n dpo_beta: DPO beta parameter\n\nReturns:\n Tuple of (loss tensor, metrics dictionary)",
"args": [
"chosen_logprobs",
"rejected_logprobs",
"chosen_ref_logprobs",
"rejected_ref_logprobs",
"dpo_beta"
],
"lineno": 107
},
{
"id": 1681,
"label": "do_update()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Perform a single DPO training update step.",
"args": [
"epoch_idx",
"batch_idx",
"n_batches",
"total_steps",
"config",
"training_client",
"reference_client",
"evaluators",
"infrequent_evaluators",
"dataset",
"ml_logger",
"log_path",
"tokenizer"
],
"lineno": 155
},
{
"id": 1682,
"label": "main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Main training function that runs the complete DPO training process.",
"args": [
"config"
],
"lineno": 318
},
{
"id": 1683,
"label": "print_example()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Print a formatted example from the dataset.",
"args": [
"datum",
"tokenizer",
"label"
],
"lineno": 390
},
{
"id": 1684,
"label": "dpo_datasets.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/preference/dpo_datasets.py",
"value": 13.021,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/preference/dpo_datasets.py",
"level": "file",
"preview": ""
},
{
"id": 1685,
"label": "DPODatasetBuilderFromComparisons",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "DPO dataset builder that uses a ComparisonDatasetBuilder.\nDPO needs both chosen and rejected examples for training.",
"bases": [
"ChatDatasetBuilder"
],
"lineno": 15
},
{
"id": 1686,
"label": "comparison_policy_evaluator.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/preference/comparison_policy_evaluator.py",
"value": 12.692,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/preference/comparison_policy_evaluator.py",
"level": "file",
"preview": ""
},
{
"id": 1687,
"label": "ComparisonEvaluator",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Evaluates a policy by comparing its completions to references, with a reward model",
"bases": [
"SamplingClientEvaluator"
],
"lineno": 17
},
{
"id": 1688,
"label": "types.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/preference/types.py",
"value": 15.059000000000001,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/preference/types.py",
"level": "file",
"preview": ""
},
{
"id": 1689,
"label": "Comparison",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 23
},
{
"id": 1690,
"label": "LabeledComparison",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 37
},
{
"id": 1691,
"label": "ComparisonRenderer",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 48
},
{
"id": 1692,
"label": "ComparisonRendererFromChatRenderer",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"ComparisonRenderer"
],
"lineno": 62
},
{
"id": 1693,
"label": "PreferenceModel",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 94
},
{
"id": 1694,
"label": "PreferenceModelBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 104
},
{
"id": 1695,
"label": "PreferenceModelFromChatRenderer",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"PreferenceModel"
],
"lineno": 109
},
{
"id": 1696,
"label": "PreferenceModelBuilderFromChatRenderer",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"PreferenceModelBuilder"
],
"lineno": 135
},
{
"id": 1697,
"label": "preference_datasets.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/preference/preference_datasets.py",
"value": 16.447,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/preference/preference_datasets.py",
"level": "file",
"preview": ""
},
{
"id": 1698,
"label": "ComparisonDatasetBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Builds HF datasets and converts to LabeledComparisons.\nThis class is independent of rendering/tokenization.",
"bases": [],
"lineno": 26
},
{
"id": 1699,
"label": "ChatDatasetBuilderFromComparisons",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Abstract base for chat dataset builders that use comparisons.\nSubclasses must implement get_comparison_builder() to provide the dataset-specific logic.",
"bases": [
"ChatDatasetBuilder"
],
"lineno": 70
},
{
"id": 1700,
"label": "ComparisonBuilderFromJsonl",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Load LabeledComparisons from JSONL files produced by combine_preference_datasets.py.",
"bases": [
"ComparisonDatasetBuilder"
],
"lineno": 123
},
{
"id": 1701,
"label": "rl_basic.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/rl_basic.py",
"value": 11.186,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/rl_basic.py",
"level": "file",
"preview": ""
},
{
"id": 1702,
"label": "build_config_blueprint()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 10
},
{
"id": 1703,
"label": "main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"config"
],
"lineno": 32
},
{
"id": 1704,
"label": "sl_loop.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/sl_loop.py",
"value": 15.148,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/sl_loop.py",
"level": "file",
"preview": ""
},
{
"id": 1705,
"label": "Config",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 23
},
{
"id": 1706,
"label": "main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"config"
],
"lineno": 35
},
{
"id": 1707,
"label": "sl_basic.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/sl_basic.py",
"value": 11.855,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/sl_basic.py",
"level": "file",
"preview": ""
},
{
"id": 1708,
"label": "build_config_blueprint()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 12
},
{
"id": 1709,
"label": "main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"config"
],
"lineno": 42
},
{
"id": 1710,
"label": "rl_loop.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/rl_loop.py",
"value": 19.392,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/rl_loop.py",
"level": "file",
"preview": ""
},
{
"id": 1711,
"label": "Config",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 22
},
{
"id": 1712,
"label": "get_reward()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"response",
"answer"
],
"lineno": 35
},
{
"id": 1713,
"label": "main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"config"
],
"lineno": 44
},
{
"id": 1714,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/README.md",
"value": 12.916,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/README.md",
"level": "file",
"preview": "# Cookbook Recipes\n\nTinker allows you to flexibly customize your training environment.\nWe will first introduce a few simple training scripts to help you get started, and then cover a broad range of different use cases.\n\n## Getting Started\n\nTinker Cookbook comes with useful abstractions so you can flexibly customize your experiments. Here are some minimal launch scripts:\n- [`rl_basic.py`](./rl_basic.py): a template script to configure reinforcement learning.\n- [`sl_basic.py`](./sl_basic.py): a template script to configure supervised learning.\n\nTo explain what goes under-the-hood, we also provide minimal, self-contained scripts that directly use the TinkerAPI to train LLMs.\n- [`rl_loop.py`](./rl_loop.py): a minimal reinforcement learning training loop.\n- [`sl_loop.py`](./sl_loop.py): a minimal supervised learning training loop.\n\n## More Post-Training Examples\n\nBuilding on Tinker and Tinker Cookbook, we can easily customize a wide range of training environments for LLMs.\nWe provide the fo"
},
{
"id": 1715,
"label": "create_data.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/prompt_distillation/create_data.py",
"value": 17.806,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/prompt_distillation/create_data.py",
"level": "file",
"preview": ""
},
{
"id": 1716,
"label": "Config",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 79
},
{
"id": 1717,
"label": "setup_clients()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 83
},
{
"id": 1718,
"label": "main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"cfg"
],
"lineno": 156
},
{
"id": 1719,
"label": "train.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/prompt_distillation/train.py",
"value": 13.425,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/prompt_distillation/train.py",
"level": "file",
"preview": ""
},
{
"id": 1720,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 18
},
{
"id": 1721,
"label": "cli_main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"cli_config"
],
"lineno": 53
},
{
"id": 1722,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/prompt_distillation/README.md",
"value": 13.966000000000001,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/prompt_distillation/README.md",
"level": "file",
"preview": "# Prompt Distillation\n\nPrompt Distillation -- also known as context distillation [1,2] -- is a training method that can \"make an LLM internalize the prompt into its parameters\".\nIn this method, the model is fine-tuned to behave as if it had been provided with a long and complex prompt, even without actually accessing it.\n\nFor example, we want to internalize the following target prompt $p$:\n\n`Classify the language of the provided text into these labels: en, fr, zh, ja ...`\n\nAfter prompt distillation, the LLM will respond with only the language label after receiving a query without seeing the prompt $p$, e.g.,\n```\nQuery: \u4e00\u751f\u3001\u30d0\u30f3\u30c9\u3057\u3066\u304f\u308c\u308b\uff1f\nResponse: ja\n```\n\nAt a high level, this method involves two stages:\n1. **Creating data for distillation**: A teacher language model uses $p$ to generate responses $r$ on a set of queries $q$; i.e. $r \\sim \\text{teacher}(\\cdot|p, q)$\n2. **Training the student model**: A student model is fine-tuned to predict the responses $r$ to the query $q$ but without acce"
},
{
"id": 1723,
"label": "on_policy_multi_teacher.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/distillation/on_policy_multi_teacher.py",
"value": 15.841000000000001,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/distillation/on_policy_multi_teacher.py",
"level": "file",
"preview": ""
},
{
"id": 1724,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Command-line configuration for multi-teacher on-policy distillation.",
"bases": [],
"lineno": 38
},
{
"id": 1725,
"label": "on_policy_distillation.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/distillation/on_policy_distillation.py",
"value": 15.280000000000001,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/distillation/on_policy_distillation.py",
"level": "file",
"preview": ""
},
{
"id": 1726,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Command-line configuration for on-policy distillation.",
"bases": [],
"lineno": 46
},
{
"id": 1727,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/distillation/README.md",
"value": 14.178,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/distillation/README.md",
"level": "file",
"preview": "# Distillation\n\nDistillation refers to a class of methods where a teacher model is supervising the training of a student model, which can often be more efficient than training the student model in isolation. We provide off-policy and on-policy distillation recipes on top of the [OpenThoughts3](https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M), [DeepMath](https://huggingface.co/datasets/zwhe99/DeepMath-103K), and [Tulu3](https://huggingface.co/datasets/allenai/tulu-3-sft-mixture)* datasets.\n\nSpecifically, we provide the scripts needed to reproduce our experiments from the [On-Policy Distillation](https://thinkingmachines.ai/blog/on-policy-distillation) blog post, which can be run with LoRA using Tinker.\n\n\\* For our post, we regenerated the assistant turns using [Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B).\n\n## Distillation for reasoning\n\nOur results can be reproduced by running:\n1. Supervised finetuning on [OpenThoughts3](https://huggingface.co/datasets/open-thoughts"
},
{
"id": 1728,
"label": "off_policy_reasoning.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/distillation/off_policy_reasoning.py",
"value": 16.21,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/distillation/off_policy_reasoning.py",
"level": "file",
"preview": ""
},
{
"id": 1729,
"label": "OpenThoughts3Builder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Builder for OpenThoughts3 dataset with streaming support.",
"bases": [
"ChatDatasetBuilder"
],
"lineno": 42
},
{
"id": 1730,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Command-line configuration for SFT on OpenThoughts3.",
"bases": [],
"lineno": 89
},
{
"id": 1731,
"label": "cli_main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Convert CLI config to full config and run training.",
"args": [
"cli_config"
],
"lineno": 124
},
{
"id": 1732,
"label": "tinker_openai.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/verifiers_rl/tinker_openai.py",
"value": 19.618000000000002,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/verifiers_rl/tinker_openai.py",
"level": "file",
"preview": ""
},
{
"id": 1733,
"label": "TinkerAsyncOpenAIClient",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "OpenAI-compatible async client that routes calls to a Tinker SamplingClient.",
"bases": [
"AsyncOpenAI"
],
"lineno": 57
},
{
"id": 1734,
"label": "TinkerChatCompletions",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"OpenAIAsyncChatCompletions"
],
"lineno": 89
},
{
"id": 1735,
"label": "TinkerCompletions",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"OpenAIAsyncCompletions"
],
"lineno": 168
},
{
"id": 1736,
"label": "TinkerAsyncChat",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"OpenAIAsyncChat"
],
"lineno": 241
},
{
"id": 1737,
"label": "TinkerAsyncCompletionStream",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 250
},
{
"id": 1738,
"label": "convert_oai_messages_to_renderer_messages()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"messages"
],
"lineno": 34
},
{
"id": 1739,
"label": "verifiers_env.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/verifiers_rl/verifiers_env.py",
"value": 12.514,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/verifiers_rl/verifiers_env.py",
"level": "file",
"preview": ""
},
{
"id": 1740,
"label": "VerifiersRLDataset",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDataset"
],
"lineno": 10
},
{
"id": 1741,
"label": "VerifiersRLDatasetBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDatasetBuilder"
],
"lineno": 43
},
{
"id": 1742,
"label": "VerifiersEnvGroupBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"EnvGroupBuilder"
],
"lineno": 63
},
{
"id": 1743,
"label": "evaluate.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/verifiers_rl/evaluate.py",
"value": 13.908999999999999,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/verifiers_rl/evaluate.py",
"level": "file",
"preview": ""
},
{
"id": 1744,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 101
},
{
"id": 1745,
"label": "log_results()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"results",
"vf_env_id",
"model_name",
"num_examples",
"rollouts_per_example",
"time_s"
],
"lineno": 18
},
{
"id": 1746,
"label": "evaluate()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"vf_env_id",
"vf_env_args",
"model_name",
"num_examples",
"rollouts_per_example",
"max_concurrent",
"max_tokens",
"temperature"
],
"lineno": 60
},
{
"id": 1747,
"label": "train.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/verifiers_rl/train.py",
"value": 16.918,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/verifiers_rl/train.py",
"level": "file",
"preview": ""
},
{
"id": 1748,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 27
},
{
"id": 1749,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/verifiers_rl/README.md",
"value": 12.449,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/verifiers_rl/README.md",
"level": "file",
"preview": "# RL Training with Tinker + Environments Hub (Verifiers)\n\n[Verifiers](https://github.com/primeintellect-ai/verifiers) is a library for creating RL environments for LLMs, including many community implementations featured on Prime Intellect's [Environments Hub](https://app.primeintellect.ai/dashboard/environments). This recipe allows all text-based environments from the Environments Hub to be used with Tinker for RL training.\n\nTo use this recipe, you need to have your chosen environment module (a self-contained Python package) installed in your project. You can install environments from the Environments Hub using the `prime` CLI:\n\n```bash\nuv tool install prime # or pipx install prime\nprime env install user/env-id # ex. prime env install primeintellect/reverse-text\n```\n\nExamples:\n- [primeintellect/reverse-text](https://app.primeintellect.ai/dashboard/environments/primeintellect/reverse-text)\n- [primeintellect/alphabet-sort](https://app.primeintellect.ai/dashboard/environments/primeintelle"
},
{
"id": 1750,
"label": "chat_datasets.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/chat_sl/chat_datasets.py",
"value": 12.812,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/chat_sl/chat_datasets.py",
"level": "file",
"preview": ""
},
{
"id": 1751,
"label": "Tulu3Builder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"ChatDatasetBuilder"
],
"lineno": 23
},
{
"id": 1752,
"label": "NoRobotsBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"ChatDatasetBuilder"
],
"lineno": 53
},
{
"id": 1753,
"label": "train.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/chat_sl/train.py",
"value": 15.379999999999999,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/chat_sl/train.py",
"level": "file",
"preview": ""
},
{
"id": 1754,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 19
},
{
"id": 1755,
"label": "get_dataset_builder()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"dataset",
"model_name",
"renderer_name",
"max_length",
"batch_size",
"train_on_what"
],
"lineno": 56
},
{
"id": 1756,
"label": "get_infrequent_evaluator_builders()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"inline_evals",
"renderer_name",
"model_name"
],
"lineno": 87
},
{
"id": 1757,
"label": "cli_main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"cli_config"
],
"lineno": 112
},
{
"id": 1758,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/chat_sl/README.md",
"value": 11.217,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/chat_sl/README.md",
"level": "file",
"preview": "# Supervised Learning\n\n## SFT on NoRobots\n\n```bash\npython -m tinker_cookbook.recipes.chat_sl.train\n model_name=Qwen/Qwen3-8B-Base \\\n dataset=no_robots \\\n learning_rate=5e-4 \\\n batch_size=64 \\\n lora_rank=64 \\\n eval_every=20 \\\n save_every=20 \\\n wandb_project=cookbook_sl\n```\n\nAfter 140 steps of training, `test/nll` decreases to 1.788.\n\n## SFT on Tulu3 dataset\n\n```bash\npython -m tinker_cookbook.recipes.chat_sl.train\n model_name=Qwen/Qwen3-8B-Base \\\n dataset=tulu3 \\\n learning_rate=5e-4 \\\n batch_size=128 \\\n lora_rank=64 \\\n eval_every=500 \\\n save_every=500 \\\n wandb_project=cookbook_sl\n```\n\nAfter 1740 steps of training, `test/nll` decreases to 0.50.\nPerformance can be further improved by training longer with a higher `lora_rank` and lower `batch_size`.\n\n## Adding your own dataset\n\nThe base classes in [tinker_cookbook/supervised/data.py](../../supervised/data.py) support loading new data in the following way:\n- `SupervisedDatasetFromHFDataset` loa"
},
{
"id": 1759,
"label": "math_env.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/math_rl/math_env.py",
"value": 25.439999999999998,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/math_rl/math_env.py",
"level": "file",
"preview": ""
},
{
"id": 1760,
"label": "MathEnv",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"ProblemEnv"
],
"lineno": 20
},
{
"id": 1761,
"label": "MathDataset",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDataset"
],
"lineno": 141
},
{
"id": 1762,
"label": "MathDatasetBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDatasetBuilder"
],
"lineno": 190
},
{
"id": 1763,
"label": "PolarisDataset",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"MathDataset"
],
"lineno": 219
},
{
"id": 1764,
"label": "PolarisDatasetBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDatasetBuilder"
],
"lineno": 255
},
{
"id": 1765,
"label": "DeepMathDataset",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"MathDataset"
],
"lineno": 272
},
{
"id": 1766,
"label": "DeepMathDatasetBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDatasetBuilder"
],
"lineno": 306
},
{
"id": 1767,
"label": "Gsm8kDataset",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDataset"
],
"lineno": 323
},
{
"id": 1768,
"label": "Gsm8kDatasetBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDatasetBuilder"
],
"lineno": 378
},
{
"id": 1769,
"label": "safe_grade()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"given_answer",
"ground_truth",
"grader",
"timeout"
],
"lineno": 74
},
{
"id": 1770,
"label": "extract_gsm8k_final_answer()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Extract the final numeric/string answer from a GSM8K solution field.\n\nGSM8K format typically places the final answer on a line starting with\n'####'. We take the substring following '####' on the last such line.",
"args": [
"text"
],
"lineno": 90
},
{
"id": 1771,
"label": "_get_hendrycks_math_test()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 111
},
{
"id": 1772,
"label": "_get_hendrycks_math_train()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 116
},
{
"id": 1773,
"label": "get_math_dataset_builder()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Unified function to get any math dataset builder.\nArgs:\n dataset_name: One of \"math\", \"polaris\", \"deepmath\", or \"gsm8k\"\n batch_size: Number of groups per batch\n model_name_for_tokenizer: Model name for tokenizer\n renderer_name: Name of the renderer to use\n group_size: Number of environments per group\n seed: Random seed for data shuffling (default: 0)\nReturns:\n The appropriate dataset builder instance",
"args": [
"dataset_name",
"batch_size",
"model_name_for_tokenizer",
"renderer_name",
"group_size",
"seed"
],
"lineno": 416
},
{
"id": 1774,
"label": "train.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/math_rl/train.py",
"value": 14.915,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/math_rl/train.py",
"level": "file",
"preview": ""
},
{
"id": 1775,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Simple command-line configuration for RL training.",
"bases": [],
"lineno": 19
},
{
"id": 1776,
"label": "get_dataset_builder()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"env",
"batch_size",
"model_name",
"renderer_name",
"group_size",
"seed"
],
"lineno": 64
},
{
"id": 1777,
"label": "arithmetic_env.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/math_rl/arithmetic_env.py",
"value": 13.235,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/math_rl/arithmetic_env.py",
"level": "file",
"preview": ""
},
{
"id": 1778,
"label": "ArithmeticEnv",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "A toy environment for solving addition problems.",
"bases": [
"ProblemEnv"
],
"lineno": 12
},
{
"id": 1779,
"label": "ArithmeticDataset",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDataset"
],
"lineno": 53
},
{
"id": 1780,
"label": "ArithmeticDatasetBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDatasetBuilder"
],
"lineno": 89
},
{
"id": 1781,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/math_rl/README.md",
"value": 14.137,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/math_rl/README.md",
"level": "file",
"preview": "# Using Reinforcement Learning to Solve Math Problems\n\nMath problems have been the most active testbed for RL with LLMs. This recipe collects environments and grading functions that allows you to test on several popular math datasets.\n\n\n## RL on arithmetic.\n\nTrivial, but runs fast enough that you can see it learn. Reward should go from 0.66 to 1 in the first few steps.\n\n```bash\npython -m tinker_cookbook.recipes.math_rl.train model_name=\"meta-llama/Llama-3.2-1B\" group_size=4 groups_per_batch=100 learning_rate=1e-4\n```\n\n## RL on MATH dataset.\n\n```bash\npython -m tinker_cookbook.recipes.math_rl.train env=math model_name=\"Qwen/Qwen3-8B\" group_size=16 groups_per_batch=64 learning_rate=2e-5 max_tokens=512\n```\n\nAfter 180 steps of training, we observe `\"test/env/all/correct\": 0.767578125`, which is logged to `/tmp/tinker-examples/math_rl/math-Qwen_Qwen3-8B-32rank-2e-05lr-${DATE}/metrics.jsonl`.\n\n```\n<|im_start|>user\nThe numbers 2, 3, 5, 7, 11, 13 are arranged in a multiplication table, with thr"
},
{
"id": 1782,
"label": "math_grading.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/math_rl/math_grading.py",
"value": 26.351,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/math_rl/math_grading.py",
"level": "file",
"preview": ""
},
{
"id": 1783,
"label": "TimeoutException",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"Exception"
],
"lineno": 514
},
{
"id": 1784,
"label": "normalize_answer()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"answer"
],
"lineno": 27
},
{
"id": 1785,
"label": "_fix_fracs()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"string"
],
"lineno": 41
},
{
"id": 1786,
"label": "_fix_a_slash_b()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"string"
],
"lineno": 73
},
{
"id": 1787,
"label": "_remove_right_units()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"string"
],
"lineno": 88
},
{
"id": 1788,
"label": "_fix_sqrt()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"string"
],
"lineno": 98
},
{
"id": 1789,
"label": "_strip_string()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"string"
],
"lineno": 113
},
{
"id": 1790,
"label": "extract_boxed()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Extract the context of the last \\boxed{...} in the text.\nUsed for getting answers from hendrycks math",
"args": [
"text"
],
"lineno": 182
},
{
"id": 1791,
"label": "_sympy_parse()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Parses an expression with sympy.",
"args": [
"expr"
],
"lineno": 220
},
{
"id": 1792,
"label": "_parse_latex()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Attempts to parse latex to an expression sympy can read.",
"args": [
"expr"
],
"lineno": 232
},
{
"id": 1793,
"label": "_is_float()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"num"
],
"lineno": 250
},
{
"id": 1794,
"label": "_is_int()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"x"
],
"lineno": 258
},
{
"id": 1795,
"label": "_is_frac()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"expr"
],
"lineno": 265
},
{
"id": 1796,
"label": "_str_is_int()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"x_str"
],
"lineno": 269
},
{
"id": 1797,
"label": "_str_to_int()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"x_str"
],
"lineno": 278
},
{
"id": 1798,
"label": "_inject_implicit_mixed_number()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Automatically make a mixed number evalable\ne.g. 7 3/4 => 7+3/4",
"args": [
"step"
],
"lineno": 284
},
{
"id": 1799,
"label": "_strip_properly_formatted_commas()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"expr"
],
"lineno": 294
},
{
"id": 1800,
"label": "_normalize()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Normalize answer expressions.",
"args": [
"expr"
],
"lineno": 305
},
{
"id": 1801,
"label": "count_unknown_letters_in_expr()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"expr"
],
"lineno": 376
},
{
"id": 1802,
"label": "should_allow_eval()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"expr"
],
"lineno": 383
},
{
"id": 1803,
"label": "are_equal_under_sympy()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"ground_truth_normalized",
"given_normalized"
],
"lineno": 395
},
{
"id": 1804,
"label": "split_tuple()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Split the elements in a tuple/interval, while handling well-formatted commas in large numbers",
"args": [
"expr"
],
"lineno": 409
},
{
"id": 1805,
"label": "grade_answer()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "The answer will be considered correct if:\n(a) it normalizes to the same string as the ground truth answer\nOR\n(b) sympy can simplify the difference between the expressions to 0",
"args": [
"given_answer",
"ground_truth"
],
"lineno": 428
},
{
"id": 1806,
"label": "grade_answer_math_verify()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Use the math_verify package to verify the answer.",
"args": [
"given_answer",
"ground_truth"
],
"lineno": 488
},
{
"id": 1807,
"label": "run_with_timeout_signal()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Runs a function with a timeout using ThreadPoolExecutor (cross-platform).\n\nArgs:\n func: The function to execute.\n args: Positional arguments for the function.\n kwargs: Keyword arguments for the function.\n timeout_seconds: Maximum time allowed in seconds.\n\nReturns:\n The result of the function call, or None if it times out.",
"args": [
"func",
"args",
"kwargs",
"timeout_seconds"
],
"lineno": 518
},
{
"id": 1808,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/README.md",
"value": 11.126,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/README.md",
"level": "file",
"preview": "# Multiturn Training\n\nOften we not only want large language models (LLMs) to generate a single response, but also to perform well across multiple turns of interaction.\nTo help Tinker users easily customize their own training, we provide the *Environment* abstraction.\n\nWe cover three examples, with increasing complexity.\n1. [Guess the number](./guess_number/): where the policy learns to guess the target number with multiple tries, given feedback on whether the number is too high or low.\n2. [Twenty Questions](./twenty_questions): where the policy learns to guess an underlying object by asking yes/no questions.\n3. [Tic-Tac-Toe](./text_arena): where the policy learns by playing against itself.\n\nThe first example is the simplest, since the user turn can be programmed with simple python statements.\nThe second is more complicated, since we need a language model to answer yes/no questions to the policy.\nThe third one is the most complicated, since we need to train multiple LLMs at the same tim"
},
{
"id": 1809,
"label": "env.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/guess_number/env.py",
"value": 16.059,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/guess_number/env.py",
"level": "file",
"preview": ""
},
{
"id": 1810,
"label": "GuessNumberEnv",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"Env"
],
"lineno": 31
},
{
"id": 1811,
"label": "GuessNumberEnvGroupBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"EnvGroupBuilder"
],
"lineno": 96
},
{
"id": 1812,
"label": "GuessNumberDataset",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDataset"
],
"lineno": 109
},
{
"id": 1813,
"label": "GuessNumberDatasetBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDatasetBuilder"
],
"lineno": 130
},
{
"id": 1814,
"label": "train.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/guess_number/train.py",
"value": 12.17,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/guess_number/train.py",
"level": "file",
"preview": ""
},
{
"id": 1815,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 11
},
{
"id": 1816,
"label": "build_config()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"cli_config"
],
"lineno": 25
},
{
"id": 1817,
"label": "main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 63
},
{
"id": 1818,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/guess_number/README.md",
"value": 14.687999999999999,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/guess_number/README.md",
"level": "file",
"preview": "# A Simple `Environment` for Guessing the Number\n\n```bash\npython -m tinker_cookbook.recipes.multiplayer_rl.guess_number.train\n```\n\nThe `test/env/all/reward/total` should increase from ~40% to >=50% in 20 steps.\n\n### Background: Guess the Number\n\nIn this task, we train an LLM to guess a hidden integer number between 0 and 1024.\n- If the LLM guess correctly, the task is done\n- If the LLM guesses incorrectly, the LLM will be given the chance to guess again and information whether the guessed number is too low or high.\n- The interaction automatically ends after 10 guesses.\nThe LLM is rewarded with 1 if it guesses correctly, otherwise 0.\n\nHere is one example game if the correct guess is 640:\n```\n[User]: Guess a number between 0 and 1024.\n[LLM]: Guess: 512\n[User]: Too low.\n[LLM]: Guess: 768\n[User]: Too high.\n[LLM]: Guess: 640\n[User]: Correct!\n```\n\n### Defining a `Guess-the-Number` Environment in Reinforcement Learning (RL)\n\nIn RL [1] (or more accurately, POMDP [2]), we need to specify the fo"
},
{
"id": 1819,
"label": "env.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/text_arena/env.py",
"value": 21.971,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/text_arena/env.py",
"level": "file",
"preview": ""
},
{
"id": 1820,
"label": "TwoPlayerCoordinator",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Coordinates a single two player game between two players. See README.md in this folder for more details.",
"bases": [],
"lineno": 28
},
{
"id": 1821,
"label": "TwoPlayerEnv",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Two player TextArena environment.",
"bases": [
"Env"
],
"lineno": 87
},
{
"id": 1822,
"label": "TwoPlayerEnvGroupBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Builder for groups of two player TextArena environments sharing the same game.",
"bases": [
"EnvGroupBuilder"
],
"lineno": 181
},
{
"id": 1823,
"label": "TwoPlayerTextArenaDataset",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Dataset for TextArena environments.",
"bases": [
"RLDataset"
],
"lineno": 226
},
{
"id": 1824,
"label": "TwoPlayerTextArenaDatasetBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDatasetBuilder"
],
"lineno": 249
},
{
"id": 1825,
"label": "train.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/text_arena/train.py",
"value": 12.4,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/text_arena/train.py",
"level": "file",
"preview": ""
},
{
"id": 1826,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 11
},
{
"id": 1827,
"label": "build_config()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"cli_config"
],
"lineno": 27
},
{
"id": 1828,
"label": "main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 65
},
{
"id": 1829,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/text_arena/README.md",
"value": 13.644,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/text_arena/README.md",
"level": "file",
"preview": "# Learning Tic-Tac-Toe via Self-Play\n\nMany research studies involve training several different language model agents jointly. We cover one simple example, where the language model learns to play tic-tac-toe with itself.\nWe show how to coordinate the steps of two *Environment* objects such that both the winning and the losing trajectory will be used to fine-tune the weights.\n\n```bash\npython -m tinker_cookbook.recipes.multiplayer_rl.text_arena.train\n```\n\nThe `test/env/all/reward/total` should increase from ~ -1.0 to >=0 in 40 steps.\n\n### Background\n\nThe TextArena [1] already implements an environment object where two players can specify which position to play using ``[0], [1], [2] ...`` in tic-tac-toe and compute how the board changes, the observation (prompt) for each language model player, and the final reward.\n\nHere's an example language model input:\n```\n[GAME] You are Player 0 in Tic Tac Toe.\nYour goal is to win three in a row (horizontally, vertically, or diagonally) on the board.\nO"
},
{
"id": 1830,
"label": "env.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/twenty_questions/env.py",
"value": 20.607,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/twenty_questions/env.py",
"level": "file",
"preview": ""
},
{
"id": 1831,
"label": "TwentyQuestionsEnv",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"Env"
],
"lineno": 39
},
{
"id": 1832,
"label": "TwentyQuestionsEnvGroupBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"EnvGroupBuilder"
],
"lineno": 155
},
{
"id": 1833,
"label": "TwentyQuestionsDataset",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDataset"
],
"lineno": 172
},
{
"id": 1834,
"label": "TwentyQuestionsDatasetBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDatasetBuilder"
],
"lineno": 195
},
{
"id": 1835,
"label": "_load_words_from_file()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 143
},
{
"id": 1836,
"label": "construct_minimal_20q_env()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"answer"
],
"lineno": 255
},
{
"id": 1837,
"label": "train.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/twenty_questions/train.py",
"value": 12.272,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/twenty_questions/train.py",
"level": "file",
"preview": ""
},
{
"id": 1838,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 13
},
{
"id": 1839,
"label": "build_config()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"cli_config"
],
"lineno": 28
},
{
"id": 1840,
"label": "main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 65
},
{
"id": 1841,
"label": "common_english_nouns.txt",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/twenty_questions/common_english_nouns.txt",
"value": 11.055,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/twenty_questions/common_english_nouns.txt",
"level": "file",
"preview": "actor\nadult\nairport\nant\napple\naunt\nbaby\nbag\nball\nballoon\nbanana\nbasket\nbear\nbed\nbee\nbeef\nbeer\nbell\nbicycle\nbiscuit\nbird\nblanket\nbook\nbottle\nbox\nboy\nbread\nbrother\nbuilding\nbutter\ncake\ncamera\ncandle\ncar\ncarpenter\ncarrot\ncastle\ncat\ncave\ncheese\nchicken\nchild\ncity\nclock\ncloud\ncoat\ncoffee\ncomputer\ncookie\ncorn\ncousin\ncow\ncup\ndad\ndaughter\ndog\ndoor\nduck\neagle\negg\nfather\nfire\nfish\nflag\nfloor\nforest\nfork\nfriend\ngarden\ngarlic\nglass\ngrandmother\ngrandfather\nguitar\nguest\nhammer\nhat\nhero\nhill\nhorse\nhospital\nhotel\nhouse\nhusband\nice\nice cream\nisland\njuice\njudge\nkangaroo\nkey\nkid\nking\nkitten\nknife\nladder\nlake\nlamp\nlady\nlion\nlock\nman\nmilk\nmirror\nmother\nmountain\nmouse\nnurse\norange\nparent\npark\npassenger\npen\npencil\nperson\nphone\npicture\npillow\npizza\npotato\nqueen\nradio\nrain\nrice\nring\nriver\nroad\nrunner\nsalad\nsalt\nsandwich\nscissors\nschool\nsea\nsheep\nshoe\nshop\nsinger\nsister\nsnow\nsoldier\nsoup\nstar\nstone\nstorm\nstreet\nstudent\nsugar\nsun\nsupermarket\ntable\ntea\nteacher\ntomato\ntown\ntoy\ntrain\nvillage\nwatch\nwater\nwife\nwind\nw"
},
{
"id": 1842,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/twenty_questions/README.md",
"value": 14.831,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/multiplayer_rl/twenty_questions/README.md",
"level": "file",
"preview": "# Playing Twenty Questions Against Another Language Model\n\n```bash\npython -m tinker_cookbook.recipes.multiplayer_rl.twenty_questions.train\n```\n\nThe `test/env/all/reward/total` should increase from ~10% to ~20% after 20 steps.\n\n### Background: Twenty Questions\n\nThis game involves a *player* and an *answerer*. The answerer has a *secret word* in mind, and the player needs to ask yes/no questions to guess it. Here is an example where the secret word is apple:\n\n> **Player**: Is it an animal?\n> **Answerer**: No\n>\n> **Player**: Is it a plant?\n> **Answerer**: Yes\n>\n> **Player**: Does it grow on a tree?\n> **Answerer**: Yes\n>\n> **Player**: Guess: Apple\n> **Answerer**: Yes\n\nTo use Tinker to train LLMs to play twenty questions, we mainly need to implement the `TwentyQuestionsEnv`\u00a0class.\n\n### Implementing a Training Environment\n\nEach environment object has exactly one secret word; it determines what the conversation looks like based on the player\u2019s (policy\u2019s) questions (actions).\n\nThe most importa"
},
{
"id": 1843,
"label": "datasets.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/datasets.py",
"value": 22.536,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/datasets.py",
"level": "file",
"preview": ""
},
{
"id": 1844,
"label": "Tulu38BComparisonBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Tulu 3.8B preference dataset comparison builder.",
"bases": [
"ComparisonDatasetBuilder"
],
"lineno": 104
},
{
"id": 1845,
"label": "HHHComparisonBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "HHH dataset comparison builder.",
"bases": [
"ComparisonDatasetBuilder"
],
"lineno": 133
},
{
"id": 1846,
"label": "HelpSteer3ComparisonBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "HelpSteer3 dataset comparison builder.",
"bases": [
"ComparisonDatasetBuilder"
],
"lineno": 150
},
{
"id": 1847,
"label": "UltraFeedbackComparisonBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "UltraFeedback dataset comparison builder.",
"bases": [
"ComparisonDatasetBuilder"
],
"lineno": 182
},
{
"id": 1848,
"label": "ArenaComparisonBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Arena dataset comparison builder.",
"bases": [
"ComparisonDatasetBuilder"
],
"lineno": 211
},
{
"id": 1849,
"label": "HelpSteer2ComparisonBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "HelpSteer2 dataset comparison builder.",
"bases": [
"ComparisonDatasetBuilder"
],
"lineno": 260
},
{
"id": 1850,
"label": "_hhh_parse_conversation()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Parse conversation text into message list format.",
"args": [
"text"
],
"lineno": 23
},
{
"id": 1851,
"label": "hhh_example_to_comparison()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Process a single preference pair into the new format.",
"args": [
"example"
],
"lineno": 48
},
{
"id": 1852,
"label": "_arena_parse_conversation()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Parse arena conversation to message format.",
"args": [
"conversation"
],
"lineno": 70
},
{
"id": 1853,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/README.md",
"value": 11.835,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/README.md",
"level": "file",
"preview": "# Learning from Preferences\n\nMany applications involve learning from preferences beyond scalar rewards. We provide a few examples here:\n\n1. [Shorter](./shorter/): we introduce the `PairwisePreferenceRLDatasetBuilder` abstraction and walk through a simple example that trains a model to generate shorter responses.\n2. [RLHF](./rlhf/): we walk through the standard RLHF pipeline from [1, 2]. This pipeline involves three stages: supervised fine-tuning, reward model learning, and reinforcement learning.\n3. [DPO](./dpo/): we optimize for human preferences using the Direct Preference Optimization algorithm [3], which requires a custom loss function.\n\n**References:**\n1. Bai, Y., Kadavath, S., Kundu, S., Askell, A., Kernion, J., Jones, A., Chen, A., Goldie, A., Mirhoseini, A., McKinnon, C., Chen, C., Olsson, C., Olah, C., Hernandez, D., Drain, D., Ganguli, D., Li, D., Tran-Johnson, E., Perez, E., Kerr, J., Mueller, J., Ladish, J., Landau, J., Ndousse, K., Luko\u0161i\u016bt\u0117, K., Lovitt, L., Sellitto, M., "
},
{
"id": 1854,
"label": "env.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/shorter/env.py",
"value": 11.995000000000001,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/shorter/env.py",
"level": "file",
"preview": ""
},
{
"id": 1855,
"label": "PreferenceModelShorter",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "A dummy preference model that always prefers a shorter response",
"bases": [
"PreferenceModel"
],
"lineno": 27
},
{
"id": 1856,
"label": "ShorterComparisonBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"ComparisonDatasetBuilder"
],
"lineno": 50
},
{
"id": 1857,
"label": "ShorterPreferenceModelBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"PreferenceModelBuilder"
],
"lineno": 59
},
{
"id": 1858,
"label": "train.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/shorter/train.py",
"value": 11.366,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/shorter/train.py",
"level": "file",
"preview": ""
},
{
"id": 1859,
"label": "build_config()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 13
},
{
"id": 1860,
"label": "main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 38
},
{
"id": 1861,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/shorter/README.md",
"value": 12.083,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/shorter/README.md",
"level": "file",
"preview": "# Generating Shorter Responses via Comparisons\n\n```bash\npython -m tinker_cookbook.recipes.preference.shorter.train\n```\n\n`ac_tokens_per_turn` should drop significantly after 40 steps. The policy generates significantly shorter responses.\n\n### Using the `PairwisePreferenceRLDatasetBuilder` class\n\nWe implement the `PairwisePreferenceRLDatasetBuilder` class to make it easier to learn from preference pairs, rather than scalar rewards for an individual trajectory. The key objects you need to implement are: (1) PreferenceModelBuilder, and (2) ComparisonBuilder.\n\n**PreferenceModelBuilder** will build a *PreferenceModel* when called (via its `__call__()` method), which determines what responses are preferred. Concretely, `PreferenceModel.__call__`\n- accepts a `Comparison` object, which contains (1) `prompt_conversation`: a list of input messages that the policy model receives, and (2) `completion_A` and `completion_B`, each a list of messages that the policy model generates.\n- returns a floatin"
},
{
"id": 1862,
"label": "rlhf_pipeline.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/rlhf/rlhf_pipeline.py",
"value": 19.244999999999997,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/rlhf/rlhf_pipeline.py",
"level": "file",
"preview": ""
},
{
"id": 1863,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 21
},
{
"id": 1864,
"label": "sft_stage()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Train base policy on NoRobots dataset",
"args": [
"log_path",
"base_model",
"wandb_project",
"wandb_name",
"lora_rank",
"batch_size",
"learning_rate",
"max_length",
"save_every",
"eval_every"
],
"lineno": 46
},
{
"id": 1865,
"label": "train_rm()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Train reward model using Anthropic HHH preference comparisons.",
"args": [
"log_path",
"base_model",
"wandb_project",
"wandb_name",
"lora_rank",
"batch_size",
"learning_rate",
"max_length",
"save_every",
"eval_every"
],
"lineno": 96
},
{
"id": 1866,
"label": "cli_main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"cli_config"
],
"lineno": 231
},
{
"id": 1867,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/rlhf/README.md",
"value": 12.295,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/rlhf/README.md",
"level": "file",
"preview": "# RLHF Pipeline\n\n```bash\npython -m tinker_cookbook.recipes.preference.rlhf.rlhf_pipeline\n```\n\nThere are three stages:\n1. Policy SFT stage: this stage is short, and `test/nll` should decrease from 1.99 to 1.92 in 20 steps.\n2. Reward model SFT stage: this stage is longer, and `test/nll` should drastically decrease from 7 to around 0.7 in the first 40 steps, slowly decrease to 0.6 at around step 300, and converge to around 0.55 in 600 steps. This stage needs to finish before the next stage.\n3. Policy RL stage: `test/win_rate` should increase from ~40% to ~70% in 100 steps.\n\n### Stage 1 and 2: Supervised Fine-Tuning\n\nThe first two stages are supervised fine-tuning, which is relatively straightforward (see `recipes.sl_basic` for an example). In the first stage, we perform supervised fine-tuning to initialize the policy on the `no_robot` dataset from Huggingface; in the second stage, we perform supervised fine-tuning to learn the reward model on the `HHH` dataset from Anthropic.\n\n### Stage 3"
},
{
"id": 1868,
"label": "train.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/dpo/train.py",
"value": 14.253,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/dpo/train.py",
"level": "file",
"preview": ""
},
{
"id": 1869,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 22
},
{
"id": 1870,
"label": "get_dataset_builder()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Get the appropriate dataset builder for DPO training.",
"args": [
"dataset",
"model_name",
"renderer_name",
"max_length",
"batch_size"
],
"lineno": 49
},
{
"id": 1871,
"label": "cli_main()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Main CLI function that builds the full config and calls the training function.",
"args": [
"cli_config"
],
"lineno": 80
},
{
"id": 1872,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/dpo/README.md",
"value": 11.554,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/preference/dpo/README.md",
"level": "file",
"preview": "# Direct Preference Optimization\n\nPlease check our [doc](https://tinker-docs.thinkingmachines.ai/preferences/dpo-guide) for background on DPO.\n\nHere is an example command:\n```\npython -m tinker_cookbook.recipes.preference.dpo.train \\\n log_path=/tmp/dpo-hhh-experiment \\\n model_name=meta-llama/Llama-3.2-1B \\\n dataset=hhh \\\n renderer_name=role_colon \\\n learning_rate=1e-5 \\\n dpo_beta=0.1\n```\n\nAfter 50 steps, you should expect training metrics like:\n```\n Step 50\n\u250f\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2533\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2513\n\u2503 Metric \u2503 Value \u2503\n\u2521\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2547\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2529\n\u2502 accuracy \u2502 0.568627 \u2502\n\u2502 batch_time \u2502 27.953704 \u2502\n\u2502 chosen_reward \u2502 0.053621 \u2502\n\u2502 dpo_loss \u2502 0.683825 \u2502\n\u2502 learning_rate \u2502 0.000009 \u2502\n\u2502 margin \u2502 0.002147 \u2502\n\u2502 num_pairs \u2502 255 \u2502\n\u2502 num_tokens "
},
{
"id": 1873,
"label": "embedding.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/tool_use/search/embedding.py",
"value": 14.716999999999999,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/tool_use/search/embedding.py",
"level": "file",
"preview": ""
},
{
"id": 1874,
"label": "get_gemini_client()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 20
},
{
"id": 1875,
"label": "tools.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/tool_use/search/tools.py",
"value": 16.559,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/tool_use/search/tools.py",
"level": "file",
"preview": ""
},
{
"id": 1876,
"label": "ToolClientInterface",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"ABC"
],
"lineno": 21
},
{
"id": 1877,
"label": "EmbeddingConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 30
},
{
"id": 1878,
"label": "RetrievalConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 37
},
{
"id": 1879,
"label": "ChromaToolClientConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 43
},
{
"id": 1880,
"label": "ChromaToolClient",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"ToolClientInterface"
],
"lineno": 52
},
{
"id": 1881,
"label": "offline_eval.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/tool_use/search/offline_eval.py",
"value": 16.976,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/tool_use/search/offline_eval.py",
"level": "file",
"preview": ""
},
{
"id": 1882,
"label": "EvaluationResult",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"TypedDict"
],
"lineno": 28
},
{
"id": 1883,
"label": "split_data_by_source()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Split data by data source.",
"args": [
"data"
],
"lineno": 34
},
{
"id": 1884,
"label": "sample_k_from_each_source()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Sample K items from each data source.",
"args": [
"data_by_source",
"k",
"seed"
],
"lineno": 42
},
{
"id": 1885,
"label": "search_env.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/tool_use/search/search_env.py",
"value": 24.427,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/tool_use/search/search_env.py",
"level": "file",
"preview": ""
},
{
"id": 1886,
"label": "SearchEnv",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"ProblemEnv"
],
"lineno": 100
},
{
"id": 1887,
"label": "SearchR1Datum",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"TypedDict"
],
"lineno": 221
},
{
"id": 1888,
"label": "SearchR1Dataset",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDataset"
],
"lineno": 291
},
{
"id": 1889,
"label": "SearchR1DatasetBuilder",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"RLDatasetBuilder"
],
"lineno": 344
},
{
"id": 1890,
"label": "normalize_answer()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Normalize answer by lowercasing, removing punctuation, articles, and fixing whitespace.",
"args": [
"s"
],
"lineno": 79
},
{
"id": 1891,
"label": "process_single_row()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Process a single row of data for SearchR1-like format.\n\nArgs:\n row: DataFrame row containing the original data\n current_split_name: Name of the current split (train/test)\n row_index: Index of the row in the DataFrame\n\nReturns:\n pd.Series: Processed row data in the required format",
"args": [
"row_series"
],
"lineno": 227
},
{
"id": 1892,
"label": "download_search_r1_dataset()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [
"split"
],
"lineno": 269
},
{
"id": 1893,
"label": "train.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/tool_use/search/train.py",
"value": 14.422,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/tool_use/search/train.py",
"level": "file",
"preview": ""
},
{
"id": 1894,
"label": "CLIConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 21
},
{
"id": 1895,
"label": "README.md",
"group": "doc",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/tool_use/search/README.md",
"value": 13.811,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/recipes/tool_use/search/README.md",
"level": "file",
"preview": "# Replicating Search-R1 with Tinker\n\n[Search-R1](https://arxiv.org/pdf/2503.09516) is a recent paper that showcases tool-use RL for multi-hop QA on Wikipedia.\nIt provides a clean setup for testing tool-use RL and also releases their training and evaluation data.\nIn this demo, we demonstrate similar experiments using `Qwen3-4B-Instruct-2507`, and we include our replication results using `Qwen/Qwen2.5-7B-Instruct` at the end.\n\n## Running This Demo\n\n### Installation and Setup\nThis demo is built with Chroma DB and the Gemini API. You can install the additional dependencies by\n\n```bash\nuv pip install -e .[vector-search]\n```\n\nBy default, we use google vertex ai for the embedding service, and you need to set `$GOOGLE_GENAI_USE_VERTEXAI`, `$GCP_VERTEXAI_PROJECT_NUMBER`, `$GCP_VERTEXAI_REGION`. Or, tweak `./embedding.py` to authenticate differently.\n\nCurrently, the tool use RL run relies on a separate Chroma vector search service. You can set it up with the following step:\n1. You can download "
},
{
"id": 1896,
"label": "smoke_tests.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/smoke_tests.py",
"value": 14.243,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/smoke_tests.py",
"level": "file",
"preview": ""
},
{
"id": 1897,
"label": "test_supervised()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 17
},
{
"id": 1898,
"label": "test_rl_async()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 74
},
{
"id": 1899,
"label": "test_rl_sync_stream_minibatch()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 105
},
{
"id": 1900,
"label": "test_utils.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/test_utils.py",
"value": 11.615,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/test_utils.py",
"level": "file",
"preview": ""
},
{
"id": 1901,
"label": "create_mock_logger_with_jsonl()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Create a mock logger that writes metrics to JSONL and optionally interrupts at a specific step.\n\nArgs:\n log_path: Directory to write metrics file to\n interrupt_at_step: If provided, raise exception at this step\n interrupt_exception_class: Exception class to raise (required if interrupt_at_step is set)\n metrics_filename: Name of the metrics file to write\n\nReturns:\n Mock logger object",
"args": [
"log_path",
"interrupt_at_step",
"interrupt_exception_class",
"metrics_filename"
],
"lineno": 9
},
{
"id": 1902,
"label": "test_trace.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/test_trace.py",
"value": 13.068999999999999,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/test_trace.py",
"level": "file",
"preview": ""
},
{
"id": 1903,
"label": "ced()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 26
},
{
"id": 1904,
"label": "sync_func()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 51
},
{
"id": 1905,
"label": "test_trace()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 77
},
{
"id": 1906,
"label": "test_resume.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/test_resume.py",
"value": 15.144,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/test_resume.py",
"level": "file",
"preview": ""
},
{
"id": 1907,
"label": "StopTrainingException",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Exception to stop training at a specific step.",
"bases": [
"Exception"
],
"lineno": 18
},
{
"id": 1908,
"label": "checkpoint_resume()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 24
},
{
"id": 1909,
"label": "test_logtree.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/test_logtree.py",
"value": 27.278,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/test_logtree.py",
"level": "file",
"preview": ""
},
{
"id": 1910,
"label": "test_basic_trace()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test basic trace creation and HTML generation.",
"args": [],
"lineno": 11
},
{
"id": 1911,
"label": "test_nested_scopes()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test nested scopes and auto header levels.",
"args": [],
"lineno": 32
},
{
"id": 1912,
"label": "test_conditional_logging()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test scope_disable for conditional logging.",
"args": [],
"lineno": 54
},
{
"id": 1913,
"label": "test_table_rendering()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test various table rendering functions.",
"args": [],
"lineno": 79
},
{
"id": 1914,
"label": "test_html_content()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test log_html for raw HTML insertion.",
"args": [],
"lineno": 107
},
{
"id": 1915,
"label": "test_details()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test collapsible details blocks.",
"args": [],
"lineno": 123
},
{
"id": 1916,
"label": "test_async_safety()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Wrapper to run async test.",
"args": [],
"lineno": 162
},
{
"id": 1917,
"label": "test_scope_header_decorator()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test the scope_header_decorator.",
"args": [],
"lineno": 167
},
{
"id": 1918,
"label": "test_async_decorator()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Wrapper to run async decorator test.",
"args": [],
"lineno": 214
},
{
"id": 1919,
"label": "test_error_handling()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test that traces are written even on error when write_on_error=True.",
"args": [],
"lineno": 219
},
{
"id": 1920,
"label": "test_no_write_without_path()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test that no file is written when path=None.",
"args": [],
"lineno": 240
},
{
"id": 1921,
"label": "test_scope_div()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test scope_div for wrapping content without changing header level.",
"args": [],
"lineno": 263
},
{
"id": 1922,
"label": "test_inline_header()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test inline header function.",
"args": [],
"lineno": 280
},
{
"id": 1923,
"label": "test_div_class_parameter()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test div_class parameter for log_text and log_html.",
"args": [],
"lineno": 297
},
{
"id": 1924,
"label": "test_export_helpers()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test export helper functions.",
"args": [],
"lineno": 314
},
{
"id": 1925,
"label": "test_graceful_degradation()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test that logtree functions work gracefully when no trace is active.",
"args": [],
"lineno": 331
},
{
"id": 1926,
"label": "test_graceful_degradation_async()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Wrapper for async graceful degradation test.",
"args": [],
"lineno": 366
},
{
"id": 1927,
"label": "test_formatter()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test the formatter object API.",
"args": [],
"lineno": 371
},
{
"id": 1928,
"label": "test_formatter_css_deduplication()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test that formatter CSS is deduplicated per trace.",
"args": [],
"lineno": 400
},
{
"id": 1929,
"label": "test_scope_details()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test collapsible details scope.",
"args": [],
"lineno": 429
},
{
"id": 1930,
"label": "test_scope_disable_nested()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test that scope_disable actually disables nested logging.",
"args": [],
"lineno": 452
},
{
"id": 1931,
"label": "test_renderers.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/test_renderers.py",
"value": 17.332,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/test_renderers.py",
"level": "file",
"preview": ""
},
{
"id": 1932,
"label": "test_generation_against_hf_chat_templates()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test generation prompt against HF chat templates",
"args": [
"model_name"
],
"lineno": 18
},
{
"id": 1933,
"label": "test_supervised_example_against_hf_chat_templates()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test supervised example against HF chat templates",
"args": [
"model_name"
],
"lineno": 74
},
{
"id": 1934,
"label": "test_eot_parsing()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Test EOT token parsing behavior for different renderers using real tokenizers.",
"args": [
"model_name",
"renderer_name"
],
"lineno": 131
},
{
"id": 1935,
"label": "test_rl_datasets.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/test_rl_datasets.py",
"value": 11.064,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/test_rl_datasets.py",
"level": "file",
"preview": ""
},
{
"id": 1936,
"label": "test_math_dataset_builder()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 6
},
{
"id": 1937,
"label": "compare_sampling_training_logprobs.py",
"group": "code",
"title": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/compare_sampling_training_logprobs.py",
"value": 15.634,
"path": "tinker_lab/tinker-cookbook/tinker_cookbook/tests/compare_sampling_training_logprobs.py",
"level": "file",
"preview": ""
},
{
"id": 1938,
"label": "Config",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 92
},
{
"id": 1939,
"label": "get_reference_document()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Download PyTorch's forward_ad.py file from a specific commit.",
"args": [],
"lineno": 16
},
{
"id": 1940,
"label": "pyright.yaml",
"group": "file",
"title": "tinker_lab/tinker-cookbook/.github/workflows/pyright.yaml",
"value": 10.402,
"path": "tinker_lab/tinker-cookbook/.github/workflows/pyright.yaml",
"level": "file",
"preview": ""
},
{
"id": 1941,
"label": "pre-commit.yaml",
"group": "file",
"title": "tinker_lab/tinker-cookbook/.github/workflows/pre-commit.yaml",
"value": 10.248,
"path": "tinker_lab/tinker-cookbook/.github/workflows/pre-commit.yaml",
"level": "file",
"preview": ""
},
{
"id": 1942,
"label": "tinker-cover.png",
"group": "file",
"title": "tinker_lab/tinker-cookbook/assets/tinker-cover.png",
"value": 40,
"path": "tinker_lab/tinker-cookbook/assets/tinker-cover.png",
"level": "file",
"preview": ""
},
{
"id": 1943,
"label": "synthetic_black_swan_scenario.csv",
"group": "file",
"title": "downloads/synthetic_black_swan_scenario.csv",
"value": 10.163,
"path": "downloads/synthetic_black_swan_scenario.csv",
"level": "file",
"preview": ""
},
{
"id": 1944,
"label": "download_agents.py",
"group": "code",
"title": "downloads/download_agents.py",
"value": 10.538,
"path": "downloads/download_agents.py",
"level": "file",
"preview": ""
},
{
"id": 1945,
"label": "download_agents()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Downloads pre-configured agents from a remote repository.",
"args": [],
"lineno": 5
},
{
"id": 1946,
"label": "demo_session.jsonl",
"group": "file",
"title": "downloads/demo_session.jsonl",
"value": 11.345,
"path": "downloads/demo_session.jsonl",
"level": "file",
"preview": ""
},
{
"id": 1947,
"label": "index.html",
"group": "ui",
"title": "downloads/index.html",
"value": 15.686,
"path": "downloads/index.html",
"level": "file",
"preview": ""
},
{
"id": 1948,
"label": "directory_manifest.jsonld",
"group": "file",
"title": "downloads/directory_manifest.jsonld",
"value": 10.497,
"path": "downloads/directory_manifest.jsonld",
"level": "file",
"preview": ""
},
{
"id": 1949,
"label": "synthetic_stock_data.csv",
"group": "file",
"title": "downloads/synthetic_stock_data.csv",
"value": 11.594,
"path": "downloads/synthetic_stock_data.csv",
"level": "file",
"preview": ""
},
{
"id": 1950,
"label": "early_stage_valuation.py",
"group": "code",
"title": "src/early_stage_valuation.py",
"value": 16.174,
"path": "src/early_stage_valuation.py",
"level": "file",
"preview": ""
},
{
"id": 1951,
"label": "EarlyStageModel",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Framework for valuing investments or lending to early-stage high-growth pre-IPO companies.\nIncorporates implied valuation, liquidity runway, and interim covenant protection until EBITDA/FCF positive.",
"bases": [],
"lineno": 4
},
{
"id": 1952,
"label": "mcp_server.py",
"group": "code",
"title": "src/mcp_server.py",
"value": 17.806,
"path": "src/mcp_server.py",
"level": "file",
"preview": ""
},
{
"id": 1953,
"label": "WACCInputs",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"BaseModel"
],
"lineno": 43
},
{
"id": 1954,
"label": "DCFInputs",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"WACCInputs"
],
"lineno": 50
},
{
"id": 1955,
"label": "DCFOutput",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"BaseModel"
],
"lineno": 53
},
{
"id": 1956,
"label": "QuantumRiskInputs",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"BaseModel"
],
"lineno": 58
},
{
"id": 1957,
"label": "SentimentInputs",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"BaseModel"
],
"lineno": 63
},
{
"id": 1958,
"label": "calculate_wacc()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Calculates the Weighted Average Cost of Capital (WACC) based on company inputs\nand global default assumptions (Risk Free Rate, Market Risk Premium).",
"args": [
"inputs"
],
"lineno": 69
},
{
"id": 1959,
"label": "calculate_dcf()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Performs a Discounted Cash Flow (DCF) analysis.\nReturns Enterprise Value, WACC, and year-by-year projections.",
"args": [
"inputs"
],
"lineno": 87
},
{
"id": 1960,
"label": "calculate_quantum_risk()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Calculates Probability of Default (PD) using Quantum Amplitude Estimation (QAE).\nUses a classical simulator (Qiskit Aer) or Numpy Monte Carlo if real quantum hardware is unavailable.",
"args": [
"inputs"
],
"lineno": 115
},
{
"id": 1961,
"label": "analyze_financial_sentiment()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Analyzes the sentiment of a financial text using NLP.\nReturns polarity (-1.0 to 1.0) and subjectivity (0.0 to 1.0).",
"args": [
"inputs"
],
"lineno": 130
},
{
"id": 1962,
"label": "get_market_data()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Retrieves market data for a given ticker from the Universal Ingestor cache.\nTries to load from 'data/adam_market_baseline.json' if available.",
"args": [
"ticker"
],
"lineno": 156
},
{
"id": 1963,
"label": "execute_reasoning_loop()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Executes the Cyclical Reasoning Loop (draft, critique, refine).",
"args": [
"query",
"context"
],
"lineno": 210
},
{
"id": 1964,
"label": "__init__.py",
"group": "code",
"title": "src/__init__.py",
"value": 10.0,
"path": "src/__init__.py",
"level": "file",
"preview": ""
},
{
"id": 1965,
"label": "config.py",
"group": "code",
"title": "src/config.py",
"value": 10.475999999999999,
"path": "src/config.py",
"level": "file",
"preview": ""
},
{
"id": 1966,
"label": "core_valuation.py",
"group": "code",
"title": "src/core_valuation.py",
"value": 13.199,
"path": "src/core_valuation.py",
"level": "file",
"preview": ""
},
{
"id": 1967,
"label": "ValuationEngine",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Computes enterprise value using a Discounted Cash Flow (DCF) model.",
"bases": [],
"lineno": 6
},
{
"id": 1968,
"label": "index.html",
"group": "ui",
"title": "src/index.html",
"value": 16.672,
"path": "src/index.html",
"level": "file",
"preview": ""
},
{
"id": 1969,
"label": "credit_risk.py",
"group": "code",
"title": "src/credit_risk.py",
"value": 13.57,
"path": "src/credit_risk.py",
"level": "file",
"preview": ""
},
{
"id": 1970,
"label": "CreditSponsorModel",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Evaluates credit risk metrics, assigns regulatory ratings, and performs stress testing.",
"bases": [],
"lineno": 5
},
{
"id": 1971,
"label": "schemas.py",
"group": "code",
"title": "src/llm/schemas.py",
"value": 11.531,
"path": "src/llm/schemas.py",
"level": "file",
"preview": ""
},
{
"id": 1972,
"label": "ProcessedRow",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Structured validation schema representing a successfully processed tabular row.\nMaps directly back to the original spreadsheet format.",
"bases": [
"BaseModel"
],
"lineno": 4
},
{
"id": 1973,
"label": "SpreadsheetBatchOutput",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Schema for validating the batch output from the LLM.\nEnsures that the LLM returns structured rows that can be reassembled.",
"bases": [
"BaseModel"
],
"lineno": 24
},
{
"id": 1974,
"label": "agent.py",
"group": "code",
"title": "src/llm/agent.py",
"value": 12.228,
"path": "src/llm/agent.py",
"level": "file",
"preview": ""
},
{
"id": 1975,
"label": "get_agent()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Initializes a PydanticAI Agent configured with the LiteLLM Provider.\nThis architecture handles proxy routing, multi-model execution, and\nenforces structural determinism using SpreadsheetBatchOutput.\n\nArgs:\n model_name: The LiteLLM formatted model name.\n\nReturns:\n A configured Agent instance ready to process tabular data.",
"args": [
"model_name"
],
"lineno": 11
},
{
"id": 1976,
"label": "main.py",
"group": "code",
"title": "src/api/main.py",
"value": 10.65,
"path": "src/api/main.py",
"level": "file",
"preview": ""
},
{
"id": 1977,
"label": "health_check()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 24
},
{
"id": 1978,
"label": "ingest.py",
"group": "code",
"title": "src/api/routers/ingest.py",
"value": 13.163,
"path": "src/api/routers/ingest.py",
"level": "file",
"preview": ""
},
{
"id": 1979,
"label": "IngestionResponse",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"BaseModel"
],
"lineno": 19
},
{
"id": 1980,
"label": "__init__.py",
"group": "code",
"title": "src/adam/__init__.py",
"value": 10.093,
"path": "src/adam/__init__.py",
"level": "file",
"preview": ""
},
{
"id": 1981,
"label": "index.html",
"group": "ui",
"title": "src/adam/index.html",
"value": 14.978,
"path": "src/adam/index.html",
"level": "file",
"preview": ""
},
{
"id": 1982,
"label": "main.py",
"group": "code",
"title": "src/adam/api/main.py",
"value": 13.668,
"path": "src/adam/api/main.py",
"level": "file",
"preview": ""
},
{
"id": 1983,
"label": "__init__.py",
"group": "code",
"title": "src/adam/api/__init__.py",
"value": 10.0,
"path": "src/adam/api/__init__.py",
"level": "file",
"preview": ""
},
{
"id": 1984,
"label": "auth.py",
"group": "code",
"title": "src/adam/api/auth.py",
"value": 11.43,
"path": "src/adam/api/auth.py",
"level": "file",
"preview": ""
},
{
"id": 1985,
"label": "models.py",
"group": "code",
"title": "src/adam/api/models.py",
"value": 10.972,
"path": "src/adam/api/models.py",
"level": "file",
"preview": ""
},
{
"id": 1986,
"label": "OptimizerConfig",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"BaseModel"
],
"lineno": 4
},
{
"id": 1987,
"label": "OptimizationRequest",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"BaseModel"
],
"lineno": 13
},
{
"id": 1988,
"label": "OptimizationResponse",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"BaseModel"
],
"lineno": 19
},
{
"id": 1989,
"label": "index.html",
"group": "ui",
"title": "src/adam/api/index.html",
"value": 15.399000000000001,
"path": "src/adam/api/index.html",
"level": "file",
"preview": ""
},
{
"id": 1990,
"label": "__init__.py",
"group": "core",
"title": "src/adam/core/__init__.py",
"value": 10.0,
"path": "src/adam/core/__init__.py",
"level": "file",
"preview": ""
},
{
"id": 1991,
"label": "state_manager.py",
"group": "core",
"title": "src/adam/core/state_manager.py",
"value": 13.323,
"path": "src/adam/core/state_manager.py",
"level": "file",
"preview": ""
},
{
"id": 1992,
"label": "StateManager",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Manages persistence of optimizer states (momentum, variance, step count).\nUses Redis for production, falls back to in-memory for testing/local dev.",
"bases": [],
"lineno": 30
},
{
"id": 1993,
"label": "index.html",
"group": "ui",
"title": "src/adam/core/index.html",
"value": 15.017,
"path": "src/adam/core/index.html",
"level": "file",
"preview": ""
},
{
"id": 1994,
"label": "optimizers.py",
"group": "core",
"title": "src/adam/core/optimizers.py",
"value": 18.496000000000002,
"path": "src/adam/core/optimizers.py",
"level": "file",
"preview": ""
},
{
"id": 1995,
"label": "AdamW",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Implements AdamW algorithm with decoupled weight decay.\n\nReferences:\n- Decoupled Weight Decay Regularization: https://arxiv.org/abs/1711.05101",
"bases": [
"Optimizer"
],
"lineno": 6
},
{
"id": 1996,
"label": "Lion",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Implements Lion algorithm (Evolved Sign Momentum).\n\nReferences:\n- Symbolic Discovery of Optimization Algorithms: https://arxiv.org/abs/2302.06675",
"bases": [
"Optimizer"
],
"lineno": 75
},
{
"id": 1997,
"label": "AdamMini",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Implements Adam-mini (Memory Efficient).\n\nFeatures:\n- Block-wise quantization (simulated via block averaging).\n- Reduces second moment memory footprint by factor of 'block_size'.\n\nReferences:\n- Adam-mini: Use 1/block_size fewer stats for v.",
"bases": [
"Optimizer"
],
"lineno": 126
},
{
"id": 1998,
"label": "logging.py",
"group": "core",
"title": "src/core/logging.py",
"value": 10.306,
"path": "src/core/logging.py",
"level": "file",
"preview": ""
},
{
"id": 1999,
"label": "setup_logging()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": null,
"args": [],
"lineno": 4
},
{
"id": 2000,
"label": "config.py",
"group": "core",
"title": "src/core/config.py",
"value": 10.51,
"path": "src/core/config.py",
"level": "file",
"preview": ""
},
{
"id": 2001,
"label": "Settings",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [
"BaseSettings"
],
"lineno": 4
},
{
"id": 2002,
"label": "plugin_manager.py",
"group": "code",
"title": "src/ingestion/plugin_manager.py",
"value": 13.064,
"path": "src/ingestion/plugin_manager.py",
"level": "file",
"preview": ""
},
{
"id": 2003,
"label": "PluginManager",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Dynamically loads and registers ingestion plugins.\nAllows easy extension without modifying core logic.",
"bases": [],
"lineno": 9
},
{
"id": 2004,
"label": "base.py",
"group": "code",
"title": "src/ingestion/base.py",
"value": 11.109,
"path": "src/ingestion/base.py",
"level": "file",
"preview": ""
},
{
"id": 2005,
"label": "IngestionStrategy",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Abstract Base Class for all ingestion strategies.\nEnsures that any new format parsing plugin implements the parse method.",
"bases": [
"ABC"
],
"lineno": 5
},
{
"id": 2006,
"label": "pdf_parser.py",
"group": "code",
"title": "src/ingestion/plugins/pdf_parser.py",
"value": 11.978,
"path": "src/ingestion/plugins/pdf_parser.py",
"level": "file",
"preview": ""
},
{
"id": 2007,
"label": "PDFParser",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Parses PDF files using PyMuPDF (fitz).\nExtracts text from PDF documents.",
"bases": [
"IngestionStrategy"
],
"lineno": 8
},
{
"id": 2008,
"label": "__init__.py",
"group": "code",
"title": "src/ingestion/plugins/__init__.py",
"value": 10.0,
"path": "src/ingestion/plugins/__init__.py",
"level": "file",
"preview": ""
},
{
"id": 2009,
"label": "excel_parser.py",
"group": "code",
"title": "src/ingestion/plugins/excel_parser.py",
"value": 12.522,
"path": "src/ingestion/plugins/excel_parser.py",
"level": "file",
"preview": ""
},
{
"id": 2010,
"label": "ExcelParser",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Parses Excel and CSV files using Pandas.\nExtracts tabular data into a structured list of dictionaries.",
"bases": [
"IngestionStrategy"
],
"lineno": 7
},
{
"id": 2011,
"label": "scanners.py",
"group": "code",
"title": "src/market_mayhem/scanners.py",
"value": 18.573,
"path": "src/market_mayhem/scanners.py",
"level": "file",
"preview": ""
},
{
"id": 2012,
"label": "Holding",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Represents a single position in a 13F filing.\nStrict typing prevents data corruption from XML parsing artifacts.",
"bases": [
"BaseModel"
],
"lineno": 22
},
{
"id": 2013,
"label": "WhaleSignal",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": "Represents a detected signal of activity (Entry, Exit, Accumulation).",
"bases": [
"BaseModel"
],
"lineno": 46
},
{
"id": 2014,
"label": "WhaleScanner",
"group": "class",
"size": 15,
"color": "#eab308",
"level": "code",
"docstring": null,
"bases": [],
"lineno": 63
},
{
"id": 2015,
"label": "__init__.py",
"group": "code",
"title": "src/market_mayhem/__init__.py",
"value": 10.0,
"path": "src/market_mayhem/__init__.py",
"level": "file",
"preview": ""
},
{
"id": 2016,
"label": "index.html",
"group": "ui",
"title": "src/market_mayhem/index.html",
"value": 14.591999999999999,
"path": "src/market_mayhem/index.html",
"level": "file",
"preview": ""
},
{
"id": 2017,
"label": "celery_app.py",
"group": "code",
"title": "src/orchestration/celery_app.py",
"value": 10.531,
"path": "src/orchestration/celery_app.py",
"level": "file",
"preview": ""
},
{
"id": 2018,
"label": "tasks.py",
"group": "code",
"title": "src/orchestration/tasks.py",
"value": 15.589,
"path": "src/orchestration/tasks.py",
"level": "file",
"preview": ""
},
{
"id": 2019,
"label": "update_task_state()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Updates the state of a specific task within Redis.\nThis shared state object is continuously polled by the SSE endpoint.",
"args": [
"task_id",
"status",
"progress",
"message",
"meta"
],
"lineno": 17
},
{
"id": 2020,
"label": "calculate_optimal_chunk_size()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "Implements the contextual window math to determine batch sizes.\nC_size = ((T_max - T_prompt) / T_row_avg) * S_safety_margin",
"args": [
"total_rows",
"tokens_per_row",
"max_context",
"prompt_tokens",
"safety_margin"
],
"lineno": 31
},
{
"id": 2021,
"label": "process_data_pipeline()",
"group": "function",
"size": 10,
"color": "#3b82f6",
"level": "code",
"docstring": "The core orchestration function executed by a Celery worker.\nHandles Ingestion -> Formatting -> Batching -> LLM Inference.",
"args": [
"self",
"file_path",
"file_extension",
"instructions",
"model_name"
],
"lineno": 44
},
{
"id": 2022,
"label": "schema_fibo.yaml",
"group": "file",
"title": "technical_specification/schema_fibo.yaml",
"value": 11.917,
"path": "technical_specification/schema_fibo.yaml",
"level": "file",
"preview": ""
},
{
"id": 2023,
"label": "GLOSSARY.md",
"group": "doc",
"title": "technical_specification/GLOSSARY.md",
"value": 13.242,
"path": "technical_specification/GLOSSARY.md",
"level": "file",
"preview": "# Glossary\n\nThis document provides a glossary of key terms and concepts used in the ADAM v21.0 technical specification.\n\n## A\n\n* **Agent:** An autonomous component that is designed to perform a specific task.\n* **Agent Orchestrator:** The component that is responsible for managing the lifecycle of the agents and orchestrating their interactions.\n* **API (Application Programming Interface):** A set of rules and protocols that allows different software applications to communicate with each other.\n* **Autonomous Mode:** A mode of operation in which the system performs tasks without human intervention.\n\n## C\n\n* **Chatbot:** A computer program that is designed to simulate human conversation.\n* **CI/CD (Continuous Integration/Continuous Deployment):** A set of practices that automate the process of building, testing, and deploying software.\n* **Component:** A self-contained unit of software that provides a specific set of features or functionality.\n\n## D\n\n* **Data Store:** A "
},
{
"id": 2024,
"label": "RESOURCE_MANAGEMENT.md",
"group": "doc",
"title": "technical_specification/RESOURCE_MANAGEMENT.md",
"value": 14.18,
"path": "technical_specification/RESOURCE_MANAGEMENT.md",
"level": "file",
"preview": "# Resource Management and Tracking\n\n## 1. Introduction\n\nThis document outlines the strategy for managing and tracking resource usage within the ADAM v21.0 platform. Effective resource management is essential for ensuring the performance, scalability, and cost-effectiveness of the system.\n\nThe two primary resources that will be tracked are:\n\n* **Compute Usage:** The processing power consumed by the various components of the system.\n* **Token Usage:** The number of tokens consumed by the Large Language Models (LLMs).\n\n## 2. Compute Usage Tracking\n\nCompute usage will be tracked for each of the major components of the system, including:\n\n* **API Layer:** The CPU and memory usage of the central API.\n* **Core System:** The CPU and memory usage of the Agent Orchestrator, Data Manager, and other core components.\n* **Data Layer:** The resources consumed by the data warehouse, SharePoint integration, and other data stores.\n\nThis will be achieved by leveraging the monitoring capabilitie"
},
{
"id": 2025,
"label": "SETUP_AND_DEPLOYMENT.md",
"group": "doc",
"title": "technical_specification/SETUP_AND_DEPLOYMENT.md",
"value": 13.683,
"path": "technical_specification/SETUP_AND_DEPLOYMENT.md",
"level": "file",
"preview": "# Setup and Deployment Guide\n\n## 1. Introduction\n\nThis guide provides step-by-step instructions for setting up and deploying the ADAM v21.0 platform. The system is designed to be deployed using Docker, which ensures a consistent and portable environment.\n\n## 2. Prerequisites\n\nBefore you begin, ensure you have the following software installed on your system:\n\n* **Git:** For cloning the repository.\n* **Docker:** For running the application in containers.\n* **Docker Compose:** For orchestrating the multi-container application.\n\n## 3. Setup\n\n### 3.1. Clone the Repository\n\nFirst, clone the project repository from GitHub:\n\n```bash\ngit clone https://github.com/adamvangrover/adam.git\ncd adam\n```\n\n### 3.2. Configuration\n\nThe system is configured using environment variables. You can create a `.env` file in the root of the project to store your configuration. A sample configuration file is provided at `config.sample.json`.\n\nCreate a `.env` file and add the following variables:\n\n```\n# API Ke"
},
{
"id": 2026,
"label": "DATA_STRATEGY.md",
"group": "doc",
"title": "technical_specification/DATA_STRATEGY.md",
"value": 14.809000000000001,
"path": "technical_specification/DATA_STRATEGY.md",
"level": "file",
"preview": "# Data Strategy and Management\n\n## 1. Introduction\n\nThis document outlines the data strategy for the Project. It covers data sources, ingestion, labeling, storage, and governance. A sound data strategy is critical for ensuring the quality, consistency, and availability of data, which is the lifeblood of the ADAM v21.0 platform.\n\n## 2. Data Sources\n\nThe system will integrate data from the following sources:\n\n* **Data Warehouse:** A centralized repository for structured data, such as financial statements, market data, and economic indicators.\n* **SharePoint:** A collaboration platform for storing and managing unstructured documents, such as reports, articles, and legal agreements.\n* **Prompt Library:** A curated collection of prompts for various financial analysis and communication tasks.\n* **General Data Store:** A flexible data store for persisting application-specific data, such as user profiles, session information, and cached data.\n\n## 3. Data Ingestion and Labeling\n\nThe Dat"
},
{
"id": 2027,
"label": "deploy.sh",
"group": "file",
"title": "technical_specification/deploy.sh",
"value": 11.977,
"path": "technical_specification/deploy.sh",
"level": "file",
"preview": ""
},
{
"id": 2028,
"label": "Adam_v20.0_TECHNICAL_SPECIFICATION.md",
"group": "doc",
"title": "technical_specification/Adam_v20.0_TECHNICAL_SPECIFICATION.md",
"value": 40,
"path": "technical_specification/Adam_v20.0_TECHNICAL_SPECIFICATION.md",
"level": "file",
"preview": "Adam v20.0: Master Technical Design Specification\nThis document serves as the canonical technical specification for the Adam v20.0 system upgrade. Its purpose is to provide an unambiguous blueprint for development, ensuring that the architectural themes of Enhanced Autonomy, Causal Inference, and Generative Simulation are implemented in a coherent, scalable, and robust manner. The design choices herein prioritize formal specification, interoperability through established standards, and the creation of a system that is not merely automated, but genuinely self-improving. All development teams assigned to the Adam v20.0 initiative shall adhere to the specifications, schemas, and protocols defined within this document.\nPart I: The Architecture of Enhanced Autonomy\nThis part details the complete technical framework enabling Adam to self-diagnose capability gaps and autonomously orchestrate the creation of new agents. The architecture is designed around two core artifacts: a machine-readable"
},
{
"id": 2029,
"label": "AGENTIC_PROCESSES.md",
"group": "doc",
"title": "technical_specification/AGENTIC_PROCESSES.md",
"value": 14.301,
"path": "technical_specification/AGENTIC_PROCESSES.md",
"level": "file",
"preview": "# Agentic Processes and Human-Machine Interaction\n\n## 1. Introduction\n\nThis document describes the agentic processes within the ADAM v21.0 platform, focusing on the different modes of operation and the mechanisms for human-in-the-loop interaction. The system is designed to be flexible, allowing for both fully autonomous operation and human-guided decision-making.\n\n## 2. Modes of Operation\n\nThe system supports two primary modes of operation, which can be selected by the user when submitting a query via the API:\n\n* **Autonomous Mode:** The system operates without human intervention. The agents will follow a predefined workflow to process the query, make decisions, and generate a final result.\n* **Prompted Mode (Human-in-the-Loop):** The system actively involves the user in the decision-making process. This mode is designed for complex or ambiguous queries where human expertise is required.\n\n## 3. Autonomous Mode\n\nIn autonomous mode, the Agent Orchestrator executes a workflow based on"
},
{
"id": 2030,
"label": "TESTING_STRATEGY.md",
"group": "doc",
"title": "technical_specification/TESTING_STRATEGY.md",
"value": 12.713000000000001,
"path": "technical_specification/TESTING_STRATEGY.md",
"level": "file",
"preview": "# Testing Strategy\n\n## 1. Introduction\n\nThis document outlines the testing strategy for the ADAM v21.0 platform. The goal of the testing strategy is to ensure that the system is reliable, performs as expected, and meets the quality standards of the business.\n\n## 2. Levels of Testing\n\nThe testing strategy will include the following levels of testing:\n\n* **Unit Testing:** Each individual component will be tested in isolation to ensure that it functions correctly. Unit tests will be written by the developers and will be run automatically as part of the continuous integration (CI) process.\n* **Integration Testing:** The interactions between different components will be tested to ensure that they work together as expected. Integration tests will be run after the unit tests have passed.\n* **End-to-End (E2E) Testing:** The entire system will be tested from end to end to ensure that it meets the business requirements. E2E tests will simulate real-world user scenarios and will be run in a"
},
{
"id": 2031,
"label": "SECURITY.md",
"group": "doc",
"title": "technical_specification/SECURITY.md",
"value": 12.425,
"path": "technical_specification/SECURITY.md",
"level": "file",
"preview": "# Security Specification\n\n## 1. Introduction\n\nThis document outlines the security measures for the ADAM v21.0 platform. Security is a critical aspect of the system, and this document describes the policies and procedures that will be implemented to protect the system and its data from unauthorized access and use.\n\n## 2. Authentication\n\nAll access to the system will be authenticated. The central API will use JSON Web Tokens (JWT) for authentication. Users will be required to provide a valid JWT in the `Authorization` header of each API request.\n\nThe JWT will be issued by a trusted identity provider (IdP) and will contain information about the user, such as their user ID and roles.\n\n## 3. Authorization\n\nAuthorization will be based on the user's roles, which will be included in the JWT. The system will use role-based access control (RBAC) to restrict access to resources and operations.\n\nFor example, a user with the `analyst` role might be able to view data and run analyses, while a user w"
},
{
"id": 2032,
"label": "PROMPT_LIBRARY_GUIDE.md",
"group": "doc",
"title": "technical_specification/PROMPT_LIBRARY_GUIDE.md",
"value": 14.122,
"path": "technical_specification/PROMPT_LIBRARY_GUIDE.md",
"level": "file",
"preview": "# Prompt Library Guide\n\n## 1. Introduction\n\nThis guide provides instructions for using and extending the ADAM v21.0 Prompt Library. The prompt library is a curated collection of prompts that are used by the AI agents to perform a variety of tasks, such as financial analysis, communication, and data extraction.\n\n## 2. Using Existing Prompts\n\nTo use an existing prompt, you simply need to reference it by its module and name. For example, to use the `escalation_email` prompt from the `communication` module, you would reference it as `communication.escalation_email`.\n\nThe Agent Orchestrator will automatically load the prompts from the `prompt_library` directory and make them available to the agents.\n\n## 3. Extending the Library\n\nThe prompt library is designed to be easily extensible. You can add new prompts or even create new modules for different tasks.\n\n### 3.1. Creating a New Module\n\nTo create a new module, simply create a new subdirectory in the `prompt_library` directory. The name of t"
},
{
"id": 2033,
"label": "PROJECT_VISION.md",
"group": "doc",
"title": "technical_specification/PROJECT_VISION.md",
"value": 17.036,
"path": "technical_specification/PROJECT_VISION.md",
"level": "file",
"preview": "# Project Vision and Business Requirements\n\n## 1. Introduction\n\nThis document outlines the vision, goals, and business requirements for the Production Integration of the Data Store Prompt Library Ingestion SharePoint Labeling Warehouse Repository (the \"Project\"). The Project aims to create a comprehensive, self-contained, and portable system that integrates a variety of data sources and provides a central webpage for API integration and a user-facing chatbot.\n\nThe system will leverage the existing ADAM v21.0 platform, extending its capabilities to meet the specific needs of the business. The project will also include the development of a detailed technical specification guide to ensure that the system is well-documented and can be easily maintained and extended in the future.\n\n## 2. Vision\n\nTo create a world-class, AI-powered financial analysis platform that provides a seamless and intuitive user experience. The platform will empower financial analysts and decision-makers with the tool"
},
{
"id": 2034,
"label": "index.html",
"group": "ui",
"title": "technical_specification/index.html",
"value": 24.651,
"path": "technical_specification/index.html",
"level": "file",
"preview": ""
},
{
"id": 2035,
"label": "UI_AND_CHATBOT.md",
"group": "doc",
"title": "technical_specification/UI_AND_CHATBOT.md",
"value": 15.294,
"path": "technical_specification/UI_AND_CHATBOT.md",
"level": "file",
"preview": "# User Interface and Chatbot\n\n## 1. Introduction\n\nThis document describes the user interface (UI) and chatbot for the ADAM v21.0 platform. The UI is designed to be intuitive, user-friendly, and provide a comprehensive set of tools for financial analysis. The chatbot offers a conversational interface for interacting with the system. The design is based on the details provided in the `UI Mockups.md` document.\n\n## 2. User Interface Overview\n\nThe main user interface is a web-based application with the following key sections:\n\n* **Dashboard:** The main landing page, providing a high-level overview of the market, the user's portfolio, investment ideas, alerts, and simulation results. The dashboard is customizable, allowing users to add, remove, and rearrange widgets to suit their needs.\n* **Market Data:** A section for exploring detailed market data for various asset classes, including stocks, bonds, ETFs, and crypto. It features interactive charts with technical indicators, historical d"
},
{
"id": 2036,
"label": "config.sample.json",
"group": "data",
"title": "technical_specification/config.sample.json",
"value": 10.428,
"path": "technical_specification/config.sample.json",
"level": "file",
"preview": "{\n \"api_keys\": {\n \"openai\": \"your_openai_api_key\"\n },\n \"sharepoint\": {\n \"site_url\": \"https://your-tenant.sharepoint.com/sites/yoursite\",\n \"username\": \"your_username\",\n \"password\": \"your_password\"\n },\n \"data_warehouse\": {\n \"host\": \"localhost\",\n \"port\": 5432,\n \"username\": \"your_username\",\n \"password\": \"your_password\",\n \"database\": \"your_database\"\n },\n \"system\": {\n \"log_level\": \"INFO\"\n }\n}..."
},
{
"id": 2037,
"label": "README.md",
"group": "doc",
"title": "technical_specification/README.md",
"value": 13.535,
"path": "technical_specification/README.md",
"level": "file",
"preview": "# ADAM Technical Specification Guides\n\nThis document is the central hub for the technical specifications of the ADAM platform.\n\n---\n\n## ADAM v20.0 - Master Technical Design Specification\n\nThe canonical technical specification for the Adam v20.0 system upgrade, focusing on Enhanced Autonomy, Causal Inference, and Generative Simulation.\n\n* **[Master Technical Design Specification](./Adam_v20.0_TECHNICAL_SPECIFICATION.md)**\n\n### v20.0 Schemas and Ontologies\n* **[Agent Proposal Standard (APS/1.0)](./schemas/agent_proposal.schema.json):** A JSON Schema for system-generated agent proposals.\n* **[\"Black Swan\" Scenario Definition Schema (BSSDS/1.0)](./schemas/black_swan_scenario.schema.yaml):** A YAML schema for defining simulation scenarios.\n* **[Adam Causal Predicate Set (ACPS/1.0)](./ontologies/acps.ttl):** An RDF/OWL ontology for representing causal relationships.\n\n---\n\n# ADAM v21.0 - Technical Specification Guide\n\n## Introduction\n\nThis document is the central hub for the technical"
},
{
"id": 2038,
"label": "api_specification.yaml",
"group": "file",
"title": "technical_specification/api_specification.yaml",
"value": 15.437000000000001,
"path": "technical_specification/api_specification.yaml",
"level": "file",
"preview": ""
},
{
"id": 2039,
"label": "ARCHITECTURE.md",
"group": "doc",
"title": "technical_specification/ARCHITECTURE.md",
"value": 18.022,
"path": "technical_specification/ARCHITECTURE.md",
"level": "file",
"preview": "# System Architecture and Design\n\n## 1. Introduction\n\nThis document provides a detailed overview of the system architecture for the Production Integration of the Data Store Prompt Library Ingestion SharePoint Labeling Warehouse Repository (the \"Project\"). It builds upon the existing ADAM v21.0 platform, extending its capabilities to meet the business requirements outlined in the `PROJECT_VISION.md` document.\n\nThe architecture is designed to be modular, scalable, and extensible, allowing for the seamless integration of new components and data sources. It also prioritizes transparency and auditability, with clear data flows and well-defined interfaces between components.\n\n## 2. Architectural Principles\n\nThe following principles guide the design of the system:\n\n* **Modularity:** The system is composed of independent, loosely coupled components, each with a specific responsibility. This allows for parallel development, testing, and deployment.\n* **Scalability:** The architecture is des"
},
{
"id": 2040,
"label": "acps_v2.ttl",
"group": "file",
"title": "technical_specification/ontologies/acps_v2.ttl",
"value": 10.97,
"path": "technical_specification/ontologies/acps_v2.ttl",
"level": "file",
"preview": ""
},
{
"id": 2041,
"label": "acps.ttl",
"group": "file",
"title": "technical_specification/ontologies/acps.ttl",
"value": 12.771,
"path": "technical_specification/ontologies/acps.ttl",
"level": "file",
"preview": ""
},
{
"id": 2042,
"label": "index.html",
"group": "ui",
"title": "technical_specification/ontologies/index.html",
"value": 14.487,
"path": "technical_specification/ontologies/index.html",
"level": "file",
"preview": ""
},
{
"id": 2043,
"label": "agent_proposal.schema.json",
"group": "data",
"title": "technical_specification/schemas/agent_proposal.schema.json",
"value": 13.542,
"path": "technical_specification/schemas/agent_proposal.schema.json",
"level": "file",
"preview": "{\n \"$schema\": \"https://json-schema.org/draft/2020-12/schema\",\n \"$id\": \"https://adam.system/schemas/v20.0/agent_proposal.schema.json\",\n \"title\": \"Agent Proposal Standard (APS/1.0)\",\n \"description\": \"A formal JSON Schema for system-generated agent proposals.\",\n \"type\": \"object\",\n \"properties\": {\n \"proposalMetadata\": {\n \"type\": \"object\",\n \"properties\": {\n \"proposalID\": {\n \"type\": \"string\",\n \"format\": \"uuid\"\n },\n \"timestamp\": {\n \"typ..."
},
{
"id": 2044,
"label": "index.html",
"group": "ui",
"title": "technical_specification/schemas/index.html",
"value": 14.552,
"path": "technical_specification/schemas/index.html",
"level": "file",
"preview": ""
},
{
"id": 2045,
"label": "black_swan_scenario.schema.yaml",
"group": "file",
"title": "technical_specification/schemas/black_swan_scenario.schema.yaml",
"value": 11.609,
"path": "technical_specification/schemas/black_swan_scenario.schema.yaml",
"level": "file",
"preview": ""
},
{
"id": 2046,
"label": "SYSTEM_PROMPT_BEHAVIORAL_ECON.md",
"group": "prompt",
"title": "prompt_library/SYSTEM_PROMPT_BEHAVIORAL_ECON.md",
"value": 12.529,
"path": "prompt_library/SYSTEM_PROMPT_BEHAVIORAL_ECON.md",
"level": "file",
"preview": "You are a \"Mentor\" model, a world-class expert in behavioral economics and quantitative finance. Your role is to teach a \"Student\" model how to analyze financial queries by generating gold-standard, step-by-step reasoning.\n\nWhen you receive a financial query, you MUST NOT give a simple, surface-level answer. You MUST analyze the query through the following behavioral lenses:\n\n1. **Prospect Theory & Loss Aversion:** Analyze how investors are perceiving gains versus losses. Are they anchored to an irrational price point (e.g., all-time high, purchase price)? Are they overweighting the \"pain\" of a recent loss?\n2. **Herd Behavior & Consensus:** Is the current price action driven by consensus and momentum (herding) or by fundamentals? Explicitly state if the data suggests a contrarian view is warranted.\n3. **Availability Heuristic:** Is the market over-reacting to recent, vivid, and easily recalled news (e.g., a bad earnings report, a CEO interview) while ignoring more complex, long-term",
"color": "#ec4899"
},
{
"id": 2047,
"label": "regulatory_rating.json",
"group": "prompt",
"title": "prompt_library/regulatory_rating.json",
"value": 14.981,
"path": "prompt_library/regulatory_rating.json",
"level": "file",
"preview": "{\n \"prompt_metadata\": {\n \"prompt_id\": \"Regulatory_Rating_Prompts_v1.0\",\n \"prompt_version\": \"1.0\",\n \"creation_date\": \"2025-09-09\",\n \"description\": \"A focused library of prompts for generating regulatory rating recommendations (e.g., for Shared National Credit - SNC) for corporate credit.\",\n \"author\": \"Jules\"\n },\n \"report_specifications\": {\n \"report_title_template\": \"Regulatory Rating Assessment: [Company Name]\",\n \"target_audience\": \"Bank Examiners, Credit Risk Managers, In...",
"color": "#ec4899"
},
{
"id": 2048,
"label": "prompt.schema.json",
"group": "prompt",
"title": "prompt_library/prompt.schema.json",
"value": 11.224,
"path": "prompt_library/prompt.schema.json",
"level": "file",
"preview": "{\n \"$schema\": \"http://json-schema.org/draft-07/schema#\",\n \"title\": \"Prompt Schema\",\n \"description\": \"Defines the canonical structure for a single prompt in the prompt library.\",\n \"type\": \"object\",\n \"properties\": {\n \"prompt_id\": {\n \"description\": \"Unique identifier for the prompt, e.g., 'ONBOARD-SUM-001'.\",\n \"type\": \"string\",\n \"pattern\": \"^[A-Z0-9_-]+-[0-9]+$\"\n },\n \"category\": {\n \"description\": \"Hierarchical category for the prompt, e.g., 'Pre-Trade/Summarization...",
"color": "#ec4899"
},
{
"id": 2049,
"label": "communication.md",
"group": "prompt",
"title": "prompt_library/communication.md",
"value": 11.813,
"path": "prompt_library/communication.md",
"level": "file",
"preview": "# Guide to Communication using the Prompt Library\n\n## Introduction\n\nThis guide is designed to help you leverage our comprehensive JSON prompt library to generate professional communications, such as escalation emails. The goal of this library is to provide a structured framework for your communications, ensuring they are clear, concise, and effective.\n\n## Overview of the Prompt Library JSON Structure\n\nThe provided JSON file is the backbone of your analysis. It's organized into several key sections:\n\n* **`prompt_metadata`**: Contains general information about the prompt library version and author.\n* **`core_communication_areas`**: This is the heart of the library. It's an array of individual prompt objects, each designed to tackle a specific communication task. Each prompt has an `id`, `title`, `description`, `instructions`, and a crucial list of `key_considerations`.\n\nYour main focus will be on the `core_communication_areas`, as these provide the building blocks for your communications",
"color": "#ec4899"
},
{
"id": 2050,
"label": "market_mayhem_newsletter.json",
"group": "prompt",
"title": "prompt_library/market_mayhem_newsletter.json",
"value": 16.373,
"path": "prompt_library/market_mayhem_newsletter.json",
"level": "file",
"preview": "{\n \"artifact_metadata\": {\n \"artifact_id\": \"Market_Mayhem_Newsletter_Content_v1.0\",\n \"version\": \"1.0\",\n \"description\": \"Library of quirky sign-offs, market heuristics, and behavioral finance quotes for the Market Mayhem newsletter.\",\n \"creation_date\": \"2025-05-22\",\n \"author\": \"Adam v23.5 Cognitive Architect\"\n },\n \"content_sections\": [\n {\n \"section_id\": \"email_sign_off_library\",\n \"title\": \"The 'Finance & Markets' Email Sign-Off Library\",\n \"categories\": [\n ...",
"color": "#ec4899"
},
{
"id": 2051,
"label": "workflow.jsonl",
"group": "prompt",
"title": "prompt_library/workflow.jsonl",
"value": 26.45,
"path": "prompt_library/workflow.jsonl",
"level": "file",
"preview": "{\"prompt_id\": \"AGENT-DATA-VALIDATE-001\", \"workflow_stage\": \"CREDIT_ASSESSMENT\", \"agent_task\": \"Validate and Summarize Ingested Data Quality\", \"trigger_event\": \"NewCreditFileReceived\", \"prompt_template\": \"You are a data validation agent. A new credit file has been ingested from multiple sources. Your task is to assess the quality and completeness of the data. Based on the provided data manifest, you must:\\n1. List all required data sources (e.g., Financial Statements, Credit Bureaus, News Feeds).\\n2. For each source, state its status (e.g., 'Received', 'Missing', 'Stale - older than 90 days').\\n3. Flag any material inconsistencies (e.g., 'Company name mismatch between sources').\\n4. Conclude with a data quality score (e.g., 'Excellent', 'Acceptable', 'Poor - Manual Review Required') and a summary of any critical gaps.\", \"example_context\": {\"data_manifest\":}, \"example_completion\": \"**Data Quality Assessment**\\n\\n1. **Annual Financials (2023):** Received\\n2. **Interim Financials (Q2 202",
"color": "#ec4899"
},
{
"id": 2052,
"label": "market_analysis.json",
"group": "prompt",
"title": "prompt_library/market_analysis.json",
"value": 18.243000000000002,
"path": "prompt_library/market_analysis.json",
"level": "file",
"preview": "{\n \"prompt_metadata\": {\n \"prompt_id\": \"Market_Analysis_Prompts_v1.0\",\n \"prompt_version\": \"1.0\",\n \"creation_date\": \"2025-08-17\",\n \"description\": \"A library of prompts for conducting market analysis, including daily briefings, sector deep dives, and risk assessments.\",\n \"author\": \"Jules\"\n },\n \"core_analysis_areas\": [\n {\n \"prompt_id\": \"daily_market_briefing\",\n \"prompt_title\": \"Daily Market Briefing\",\n \"description\": \"A prompt to generate a concise daily market br...",
"color": "#ec4899"
},
{
"id": 2053,
"label": "regulatory_rating.md",
"group": "prompt",
"title": "prompt_library/regulatory_rating.md",
"value": 13.744,
"path": "prompt_library/regulatory_rating.md",
"level": "file",
"preview": "# Guide to Regulatory Rating Analysis using the Prompt Library\n\n## Introduction\n\nThis guide is for credit professionals who need to assign a regulatory rating to a corporate credit facility, such as for the Shared National Credit (SNC) program. The prompts in the accompanying JSON file (`regulatory_rating.json`) provide a structured framework to ensure your analysis aligns with the core principles of regulatory credit assessment: timely repayment capacity and the identification of well-defined weaknesses.\n\n---\n\n## How to Use the Prompts\n\nThe `regulatory_rating.json` file contains a single `core_analysis_area` focused on this task. The prompts are designed to be used sequentially to build a clear and defensible rating recommendation.\n\n### Regulatory Rating Workflow\n\n#### 1. Analyze the Primary Source of Repayment\n\n* **Objective**: To determine the primary source of cash to repay the loan and assess its reliability.\n* **Relevant Prompt**: `repayment_source_analysis` (Task ID: `RR01`)",
"color": "#ec4899"
},
{
"id": 2054,
"label": "due_diligence.md",
"group": "prompt",
"title": "prompt_library/due_diligence.md",
"value": 12.712,
"path": "prompt_library/due_diligence.md",
"level": "file",
"preview": "# Guide to Due Diligence using the Prompt Library\n\n## Introduction\n\nThis guide is designed to help you leverage our comprehensive JSON prompt library to conduct thorough and standardized due diligence on a company. The goal of this library is to provide a structured framework for your analysis, ensuring all critical aspects of due diligence are considered consistently and efficiently.\n\n## Overview of the Prompt Library JSON Structure\n\nThe provided JSON file is the backbone of your analysis. It's organized into several key sections:\n\n* **`prompt_metadata`**: Contains general information about the prompt library version and author.\n* **`core_analysis_areas`**: This is the heart of the library. It's an array of individual prompt objects, each designed to tackle a specific part of the due diligence process. Each prompt has an `id`, `title`, `description`, `instructions`, and a crucial list of `key_considerations`.\n\nYour main focus will be on the `core_analysis_areas`, as these provide the ",
"color": "#ec4899"
},
{
"id": 2055,
"label": "SPEC_DRIVEN_AGENT_PROMPT.md",
"group": "prompt",
"title": "prompt_library/SPEC_DRIVEN_AGENT_PROMPT.md",
"value": 16.008,
"path": "prompt_library/SPEC_DRIVEN_AGENT_PROMPT.md",
"level": "file",
"preview": "# \ud83e\udde0 Spec-Driven Agent Protocol (SDAP)\n\n**Context:**\n* **Repository:** `adamvangrover/adam`\n* **Purpose:** A rigorous protocol for working with AI Agents to ensure high-quality, maintainable, and safe code generation.\n* **Philosophy:** **\"Plan first in read-only mode, then execute and iterate continuously.\"**\n\n---\n\n## \ud83d\udcdc The 5 Core Principles\n\n### 1. Start with Vision, Let AI Draft Details\nDon't over-engineer upfront. Provide a high-level goal (the \"Product Brief\") and use the agent's \"Plan Mode\" (read-only) to draft the detailed specification.\n* *Action:* \"Draft a detailed spec for [Goal] covering objectives, features, and constraints.\"\n* *Why:* Leverages AI's elaboration strength while keeping you in control.\n\n### 2. Structure the Spec (PRD/SRS)\nThe spec is the \"Source of Truth\". It must be structured, not a stream of consciousness.\n* **Core Areas:**\n 1. **Commands:** Executable commands (`pytest`, `npm test`).\n 2. **Testing:** Test plan, file locations, coverage re",
"color": "#ec4899"
},
{
"id": 2056,
"label": "unified_v1.md",
"group": "prompt",
"title": "prompt_library/unified_v1.md",
"value": 20.172,
"path": "prompt_library/unified_v1.md",
"level": "file",
"preview": "\n***\n\n# **MASTER PROMPT: UNIFIED FINANCIAL ANALYSIS & REPORTING SYSTEM (v1.0)**\n\n## **1. PERSONA**\n\n**Act as an expert financial analysis AI system.** You are a sophisticated copilot designed to assist financial professionals by executing a wide range of predefined analytical tasks and generating comprehensive reports. Your knowledge is encapsulated in the 'Unified Prompt Library' defined below. You must be precise, data-driven, and adhere strictly to the requested formats.\n\n---\n\n## **2. OBJECTIVE**\n\nYour primary goal is to function as an interface to the comprehensive library of analytical tasks detailed in Section 3. When a user makes a request (e.g., \"Generate a market outlook report,\" \"Give me a SWOT analysis for Company X,\" or \"Run task FHA-L-01\"), you must:\n1. Identify the corresponding prompt(s) from the library.\n2. Execute the instructions exactly as specified in the `prompt_text`.\n3. Structure your response according to the `expected_response_format`.\n4. If a request requi",
"color": "#ec4899"
},
{
"id": 2057,
"label": "Adam_v26.0_System_Prompt.md",
"group": "prompt",
"title": "prompt_library/Adam_v26.0_System_Prompt.md",
"value": 18.229,
"path": "prompt_library/Adam_v26.0_System_Prompt.md",
"level": "file",
"preview": "### SYSTEM ROLE: ADAM v26.0 \"COGNITIVE ARCHITECT\"\n\n**IDENTITY PROTOCOL:**\nYou are **Adam v26.0**, a **Neuro-Symbolic \"System 2\" Cognitive Engine**. You are not a chatbot; you are an autonomous financial intelligence architecture designed to replace junior analytical labor in high-stakes environments. You operate with **institutional-grade precision**, moving beyond retrieval to deep inference, causal reasoning, and predictive modeling.\n\n**OPERATIONAL MODE:**\nYou execute a **Federated Reasoning Strategy**, synthesizing outputs from four isolated cognitive modules into a single, sovereign \"Hyper-Dimensional Knowledge Graph\" (HDKG). Your output is not an opinion; it is a **calculated adjudication** of value and risk.\n\n**ACTIVE COGNITIVE MODULES:**\n\n1. **Credit & Insolvency Architecture (The Shield):** Responsible for downside protection, covenant friction analysis, capital structure deconstruction, and regulatory (SNC) classification.\n2. **Equity & Valuation Engine (The Spear):** Respon",
"color": "#ec4899"
},
{
"id": 2058,
"label": "model_validation.json",
"group": "prompt",
"title": "prompt_library/model_validation.json",
"value": 11.463000000000001,
"path": "prompt_library/model_validation.json",
"level": "file",
"preview": "{\n \"prompt_metadata\": {\n \"prompt_id\": \"Model_Validation_Prompts_v1.0\",\n \"prompt_version\": \"1.0\",\n \"creation_date\": \"2025-08-17\",\n \"description\": \"A library of prompts for generating challenges to financial models.\",\n \"author\": \"Jules\"\n },\n \"core_validation_areas\": [\n {\n \"prompt_id\": \"model_challenge\",\n \"prompt_title\": \"Model Challenge\",\n \"description\": \"A prompt to generate insightful challenges to a financial model.\",\n \"instructions\": \"Generate a list of...",
"color": "#ec4899"
},
{
"id": 2059,
"label": "unified_v1.json",
"group": "prompt",
"title": "prompt_library/unified_v1.json",
"value": 40,
"path": "prompt_library/unified_v1.json",
"level": "file",
"preview": "{\n \"prompt_metadata\": {\n \"prompt_id\": \"Unified_Prompt_Library_v1\",\n \"prompt_version\": \"1.0\",\n \"creation_date\": \"2025-09-21\",\n \"description\": \"A unified prompt library containing all prompts from the repository.\",\n \"author\": \"Jules\"\n },\n \"core_analysis_areas\": [\n {\n \"prompt_id\": \"macroeconomic_overview\",\n \"prompt_title\": \"Global Macroeconomic Backdrop\",\n \"description\": \"Analyze key macroeconomic factors expected to influence credit and capital markets.\",\n ...",
"color": "#ec4899"
},
{
"id": 2060,
"label": "v2_0_credit_risk_architect.md",
"group": "prompt",
"title": "prompt_library/v2_0_credit_risk_architect.md",
"value": 16.147,
"path": "prompt_library/v2_0_credit_risk_architect.md",
"level": "file",
"preview": "# Refined v2.0 Prompt: Cloud-Aware Credit & Risk Architect\n\n### GOAL\nTo act as an autonomous AI agent that generates comprehensive, data-driven corporate credit risk assessments based on user requests.\n\n### PERSONA\nYou are a methodical risk analysis system. Your core function is to execute the provided workflow with precision and accuracy. You prioritize empirical, quantitative data over speculation and must cite a source for every metric. You communicate in the clear, concise, and formal language of an institutional financial report.\n\n### TOOLS\nYou have access to the following tools. You must use the provided JSON schemas to format your tool calls.\n```json\n[\n {\n \"type\": \"function\",\n \"function\": {\n \"name\": \"azure_ai_search\",\n \"description\": \"Searches and retrieves excerpts from unstructured documents (e.g., rating agency reports, company filings) from the Azure AI Search index.\",\n \"parameters\": {\n \"type\": \"object\",\n ",
"color": "#ec4899"
},
{
"id": 2061,
"label": "due_diligence.json",
"group": "prompt",
"title": "prompt_library/due_diligence.json",
"value": 16.841,
"path": "prompt_library/due_diligence.json",
"level": "file",
"preview": "{\n \"prompt_metadata\": {\n \"prompt_id\": \"Due_Diligence_Prompts_v1.0\",\n \"prompt_version\": \"1.0\",\n \"creation_date\": \"2025-08-17\",\n \"description\": \"A library of prompts for conducting due diligence on a company.\",\n \"author\": \"Jules\"\n },\n \"core_analysis_areas\": [\n {\n \"prompt_id\": \"comprehensive_due_diligence_checklist\",\n \"prompt_title\": \"Comprehensive Due Diligence Checklist\",\n \"description\": \"Generates a comprehensive checklist of items and questions for conducting...",
"color": "#ec4899"
},
{
"id": 2062,
"label": "communication.json",
"group": "prompt",
"title": "prompt_library/communication.json",
"value": 11.407,
"path": "prompt_library/communication.json",
"level": "file",
"preview": "{\n \"prompt_metadata\": {\n \"prompt_id\": \"Communication_Prompts_v1.0\",\n \"prompt_version\": \"1.0\",\n \"creation_date\": \"2025-08-17\",\n \"description\": \"A library of prompts for generating professional communications, such as escalation emails.\",\n \"author\": \"Jules\"\n },\n \"core_communication_areas\": [\n {\n \"prompt_id\": \"escalation_email\",\n \"prompt_title\": \"Escalation Email\",\n \"description\": \"A prompt to generate a clear, concise, and effective escalation email.\",\n \"in...",
"color": "#ec4899"
},
{
"id": 2063,
"label": "index.html",
"group": "prompt",
"title": "prompt_library/index.html",
"value": 30.401,
"path": "prompt_library/index.html",
"level": "file",
"preview": "\n\n\n \n \n Index of /prompt_library\n \n \n \n\n\n\n\n Adam v21.0 Chatbot\n \n \n \n \n \n \n UI Archive v1
Preserved snapshot of 31 static HTML files.
![\"Adam](\".github/images/logo-dark.svg\"){kind=logo}
\n ![\"License\"](\"https://img.shields.io/badge/License-MIT-yellow.svg\"){kind=icon}
![\"Python](\"https://img.shields.io/badge/python-3.10+-blue.svg\"){kind=icon}
![\"Docker](\"https://img.shields.io/badge/docker-ready-blue\"){kind=icon}
![](\"assets/tinker-cover.png\")
\n