Functions — github.com/OpenPipe/ART

hub / github.com/OpenPipe/ART / functions

Functions1,502 in github.com/OpenPipe/ART

⨍ Functions 1,502◇ Types & classes 261↳ Endpoints 14

↓ 1 callersMethod_create_identity_lora

(self, lora_path: str)

src/art/megatron/service.py:95

↓ 1 callersMethod_deep_merge_dicts

( base: dict[str, Any], updates: dict[str, Any] )

src/art/model.py:361

↓ 1 callersMethod_default_num_workers

(self)

src/art/tinker/server.py:271

↓ 1 callersMethod_define_wandb_step_metrics

(self, keys: Iterable[str])

src/art/model.py:564

↓ 1 callersMethod_emit_deltas

Emit deltas for any new content since last emission.

src/art/tinker/cookbook_v/renderers/kimi_k2.py:208

↓ 1 callersMethod_encode_multipart_content

( self, content: list[ContentPart] )

src/art/tinker/cookbook_v/renderers/kimi_k2.py:508

↓ 1 callersMethod_ensure_identity_lora

(self, lora_path: str)

src/art/megatron/service.py:119

↓ 1 callersMethod_ensure_megatron_running

Lazily start Megatron training process if not running.

src/art/megatron/service.py:159

↓ 1 callersFunction_ensure_tool_message

(item: MessageOrChoice)

tests/unit/test_trajectory_parquet.py:72

↓ 1 callersMethod_env_enabled

(self, env_name: str)

src/art/tinker_native/backend.py:132

↓ 1 callersFunction_estimate_anthropic_cost

( token_counts: _AnthropicTokenUsage | None, pricing: TokenPricing, )

src/art/api_costs.py:229

↓ 1 callersFunction_estimate_openai_cost

( token_counts: _OpenAITokenUsage | None, pricing: TokenPricing, )

src/art/api_costs.py:201

↓ 1 callersFunction_estimate_provider_cost

( provider_name: str, response: Any, pricing: TokenPricing, )

src/art/api_costs.py:262

↓ 1 callersMethod_eval_stage

(self)

src/art/pipeline_trainer/trainer.py:548

↓ 1 callersMethod_experimental_fork_checkpoint

Fork a checkpoint from another TinkerNative model to initialize this model. Loads the source model's training checkpoint into the destination

src/art/tinker_native/backend.py:840

↓ 1 callersMethod_experimental_pull_model_checkpoint

Pull a model checkpoint from W&B artifacts to a local path. For ServerlessBackend, this downloads the checkpoint from W&B artifact storage.

src/art/serverless/backend.py:572

↓ 1 callersFunction_extract_anthropic_token_counts

(response: Any)

src/art/api_costs.py:178

↓ 1 callersFunction_extract_direct_response_cost

(response: Any)

src/art/api_costs.py:139

↓ 1 callersFunction_extract_openai_token_counts

(response: Any)

src/art/api_costs.py:157

↓ 1 callersFunction_extract_scenario_id

(group: TrajectoryGroup)

src/art/metrics_taxonomy.py:110

↓ 1 callersFunction_extract_scenario_id_from_metadata

( metadata: dict[str, Any], )

src/art/metrics_taxonomy.py:121

↓ 1 callersFunction_extract_step_from_wandb_artifact

Extract step number from a W&B artifact's aliases.

src/art/serverless/backend.py:29

↓ 1 callersFunction_extract_tool_name

(tool_id: str)

src/art/tinker/cookbook_v/renderers/kimi_k2.py:55

↓ 1 callersFunction_fallback_scrape

Fallback scraping method using BeautifulSoup for basic text extraction

examples/just-the-facts/just_the_facts/utils.py:57

↓ 1 callersFunction_find_existing_job_id

Finds an existing model deployment job in Together.

src/art/utils/deployment/together.py:129

↓ 1 callersFunction_flatten_message

Convert a message or Choice to flat parquet format.

src/art/utils/trajectory_logging.py:21

↓ 1 callersMethod_format_condensed_line

(self)

src/art/pipeline_trainer/status.py:195

↓ 1 callersMethod_format_full_log

(self)

src/art/pipeline_trainer/status.py:260

↓ 1 callersFunction_format_parameters_block

Format function parameters as a TypeScript-style block.

src/art/tinker/cookbook_v/renderers/gpt_oss.py:102

↓ 1 callersMethod_format_response_model

(self, model_name: str, step: int)

src/art/tinker_native/backend.py:745

↓ 1 callersMethod_format_thinking_text

Format a ThinkingPart payload for rendering.

src/art/tinker/cookbook_v/renderers/qwen3.py:436

↓ 1 callersMethod_format_tool_call_xml

Format a single tool call in Qwen3.5's XML parameter format.

src/art/tinker/cookbook_v/renderers/qwen3_5.py:209

↓ 1 callersMethod_format_tool_calls_chunks

Format tool_calls as output chunks. Override in subclasses for different formats.

src/art/tinker/cookbook_v/renderers/qwen3.py:499

↓ 1 callersFunction_format_tool_definition

Format a single tool as a Harmony TypeScript-style definition.

src/art/tinker/cookbook_v/renderers/gpt_oss.py:131

↓ 1 callersMethod_get_generation_suffix

Return tokens to append to the prompt for generation. This is called by build_generation_prompt to add the role header that precedes

src/art/tinker/cookbook_v/renderers/base.py:917

↓ 1 callersFunction_get_hf_tokenizer

(model_name: str)

src/art/tinker/cookbook_v/tokenizer_utils.py:84

↓ 1 callersFunction_get_hidden_size

(model_name: str)

src/art/tinker/cookbook_v/hyperparam_utils.py:77

↓ 1 callersMethod_get_next_scenario

(self)

src/art/pipeline_trainer/trainer.py:297

↓ 1 callersMethod_get_optimizer_state_path

(self)

src/art/megatron/service.py:57

↓ 1 callersMethod_get_packed_tensors

( self, model: AnyTrainableModel, trajectory_groups: list[TrajectoryGroup], ad

src/art/local/backend.py:290

↓ 1 callersMethod_get_sampler_client

( self, state: ModelState, step: int | None )

src/art/tinker_native/backend.py:670

↓ 1 callersMethod_get_state

(self)

src/art/tinker/service.py:197

↓ 1 callersMethod_get_step

(self, model: AnyTrainableModel)

src/art/backend.py:26

↓ 1 callersMethod_get_step

(self, model: "Model")

src/art/serverless/backend.py:152

↓ 1 callersMethod_get_step

(self, model: AnyTrainableModel)

src/art/local/backend.py:356

↓ 1 callersFunction_group_is_trainable

(group: TrajectoryGroup)

src/art/metrics_taxonomy.py:105

↓ 1 callersMethod_group_zero_variance

(self, group: TrajectoryGroup)

src/art/pipeline_trainer/status.py:332

↓ 1 callersFunction_handle_request

( obj: object, request: Request, responses: mp.Queue, generators: dict[int, AsyncGenerator[Any

src/mp_actors/move.py:303

↓ 1 callersFunction_handle_requests

( obj: object, requests: mp.Queue, responses: mp.Queue )

src/mp_actors/move.py:290

↓ 1 callersFunction_has_fused_moe_lora

Check if the adapter contains fused MoE LoRA tensors.

src/art/utils/convert_moe_lora.py:23

↓ 1 callersFunction_infer_moe_params

Infer num_experts, rank, intermediate_size, hidden_size from tensor shapes.

src/art/utils/convert_moe_lora.py:31

↓ 1 callersMethod_is_scalar_metadata

(value: object)

src/art/pipeline_trainer/trainer.py:782

↓ 1 callersFunction_iter

()

src/art/pipeline_trainer/trainer.py:28

↓ 1 callersFunction_json_schema_to_typescript

Convert JSON schema to an inline TypeScript-ish type string.

src/art/tinker/cookbook_v/renderers/gpt_oss.py:67

↓ 1 callersFunction_judge_provider

(judge_model: str)

src/art/rewards/ruler.py:53

↓ 1 callersFunction_load_history_timestamps

Load recorded_at timestamps from history.jsonl files.

src/art/utils/benchmarking/aggregate_trajectories.py:92

↓ 1 callersMethod_load_samplable_model

( self, model_path_or_base_model: str )

src/art/tinker/server.py:345

↓ 1 callersMethod_log_zero_variance_groups

(self, step: int)

src/art/pipeline_trainer/trainer.py:723

↓ 1 callersFunction_max_fit_font_size

Binary search the largest font that fits the canvas with wrapping. This aggressively grows the font size to fill the available space (subject

dev/yes-no-maybe-vision/generate_images.py:80

↓ 1 callersFunction_merge_consecutive_text_parts

Merge consecutive TextParts into single parts. This ensures text is tokenized as a single string, matching HuggingFace's apply_chat_template

src/art/tinker/cookbook_v/renderers/qwen3.py:38

↓ 1 callersMethod_merge_lora_adapter

Merge sharded LoRA adapters from distributed training.

src/art/megatron/service.py:307

↓ 1 callersFunction_merge_token_pricing

( *, base_pricing: TokenPricing, prompt_price_per_million: float | None, completion_price_per_

src/art/api_costs.py:288

↓ 1 callersMethod_merge_wandb_config

( existing: dict[str, Any], updates: dict[str, Any], *, path: str = "", )

src/art/model.py:377

↓ 1 callersMethod_model_inference_name

( self, model: AnyModel, step: int | None = None )

src/art/backend.py:18

↓ 1 callersMethod_model_inference_name

(self, model: Model, step: int | None = None)

src/art/tinker_native/backend.py:363

↓ 1 callersMethod_model_inference_name

Return the inference name for a model checkpoint. For LocalBackend with vLLM, the base model is served under its HF name, and LoRA ad

src/art/local/backend.py:210

↓ 1 callersMethod_monitor_openai_server

( self, model: AnyTrainableModel, base_url: str, api_key: str )

src/art/local/backend.py:416

↓ 1 callersFunction_new_pending_metrics_state

()

src/art/metrics.py:62

↓ 1 callersFunction_new_shared_metrics_state

()

src/art/metrics.py:66

↓ 1 callersMethod_next_lora_id

(self)

src/art/megatron/service.py:53

↓ 1 callersMethod_node_count

(self)

src/art/tinker/prefix_cache.py:201

↓ 1 callersMethod_normalize_eval_items

( items: list[art.Trajectory | art.TrajectoryGroup], )

src/art/pipeline_trainer/trainer.py:634

↓ 1 callersMethod_normalize_messages

(self, messages: Iterable[Any])

src/art/tinker_native/backend.py:700

↓ 1 callersMethod_normalize_tools

( self, tools: Iterable[Any] | None )

src/art/tinker_native/backend.py:709

↓ 1 callersMethod_normalize_trajectory_groups

( self, trajectories: Iterable[Trajectory | BaseException] | Iterable[TrajectoryGroup], )

src/art/model.py:693

↓ 1 callersFunction_openai_function_to_typescript_style

Convert OpenAI function definition (dict) to TypeScript style string.

src/art/tinker/cookbook_v/renderers/kimi_k2_5_tool_declaration_ts.py:409

↓ 1 callersFunction_openai_server_coroutine

( config: OpenAIServerConfig, )

src/art/vllm/server.py:134

↓ 1 callersFunction_override_variables_in_notebook

Replace top-level assignments inside code cells. Replaces entire assignment statements (including multi-line values) using AST ranges to avoi

tests/integration.py:82

↓ 1 callersMethod_parse_deepseek_tool_calls

Parse tool calls from DeepSeek V3.1 format. Expected format (per HuggingFace model card and chat template): <｜tool▁calls▁begin｜><

src/art/tinker/cookbook_v/renderers/deepseek_v3.py:202

↓ 1 callersFunction_parse_guess_args

(arguments: str | None)

src/art/pipeline_trainer/binary_prefix_tool_pipeline.py:91

↓ 1 callersMethod_parse_harmony_messages

Parse Harmony format content into a list of message dicts. Uses manual string parsing (find/rfind) rather than regex. This approach i

src/art/tinker/cookbook_v/renderers/gpt_oss.py:557

↓ 1 callersMethod_parse_harmony_output

( self, content: str )

src/art/tinker/cookbook_v/renderers/gpt_oss.py:512

↓ 1 callersFunction_parse_jsonl_line

Parse a JSONL line into a Trajectory object. Args: line: A JSON string containing trajectory data with 'messages' and optional 'tools'.

src/art/utils/sft.py:24

↓ 1 callersMethod_parse_qwen3_5_tool_call_xml

Parse Qwen3.5 XML-style tool calls from a raw <tool_call> block.

src/art/tinker/cookbook_v/renderers/qwen3_5.py:89

↓ 1 callersFunction_parse_tool_call_json

Parse tool call JSON. Returns UnparsedToolCall on failure.

src/art/tinker/cookbook_v/renderers/base.py:473

↓ 1 callersMethod_preprocess_message_parts

Convert message content to list form for VL rendering. Converts ThinkingPart to <think>...</think> text (or strips if strip_thinking=True).

src/art/tinker/cookbook_v/renderers/qwen3.py:444

↓ 1 callersMethod_process_error

(self)

src/mp_actors/move.py:129

↓ 1 callersMethod_prune

(self, node: _RadixNode)

src/art/tinker/prefix_cache.py:155

↓ 1 callersMethod_put_output_group

(self, group: TrajectoryGroup)

src/art/pipeline_trainer/trainer.py:785

↓ 1 callersMethod_read_pipeline_state

(self)

src/art/pipeline_trainer/trainer.py:768

↓ 1 callersFunction_read_usage_nested_field

(usage: Any, *fields: str)

src/art/api_costs.py:95

↓ 1 callersFunction_record_ruler_cost

(judge_model: str, response: ModelResponse)

src/art/rewards/ruler.py:63

↓ 1 callersFunction_register_proxy_for_atexit

(proxy: "Proxy")

src/mp_actors/move.py:46

↓ 1 callersMethod_render_tool_calls

Render tool calls in Harmony commentary channel format. Each tool call becomes a separate commentary message: to=functions.name<|chan

src/art/tinker/cookbook_v/renderers/gpt_oss.py:284

↓ 1 callersMethod_render_tool_result_message

Render a tool result message. Format: <|start|>functions.name to=assistant<|channel|>commentary<|message|>{result}<|end|> IMPORTANT:

src/art/tinker/cookbook_v/renderers/gpt_oss.py:304

↓ 1 callersMethod_resolve_gpu_cost_per_hour_usd

(self)

src/art/local/backend.py:121

↓ 1 callersFunction_resolve_layer_spec

( base_layer_spec: ModuleSpec | Callable[[GPTModelProvider], ModuleSpec], config: GPTModelProvider,

src/art/megatron/provider.py:16

↓ 1 callersFunction_resolve_registered_or_default_pricing

( model_name: str, *, model_pricing: Mapping[str, TokenPricing], )

src/art/api_costs.py:277

↓ 1 callersFunction_resolve_token_pricing

( *, provider: str, model_name: str, prompt_price_per_million: float | None, completion_pr

src/art/api_costs.py:335

↓ 1 callersMethod_rollout_stage

(self)

src/art/pipeline_trainer/trainer.py:364

← previous next →401–500 of 1,502, ranked by callers