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

Finish parsing and yield any remaining content plus final Message. Call this after all tokens have been fed (either naturally when en

src/art/tinker/cookbook_v/renderers/kimi_k2.py:280

↓ 5 callersMethodget_active

()

src/art/metrics.py:235

↓ 5 callersFunctionget_last_checkpoint_dir

(output_dir: str)

src/art/local/checkpoints.py:23

↓ 5 callersFunctionget_model_pricing

Return pricing for a model or None if missing.

src/art/costs.py:71

↓ 5 callersFunctionget_renderer_name

(base_model: str)

src/art/tinker/renderers.py:1

↓ 5 callersFunctionget_step_from_dir

(output_dir: str)

src/art/utils/get_model_step.py:10

↓ 5 callersMethodlog

(self, message: str)

src/art/trajectories.py:50

↓ 5 callersMethodlookup

(self, rendered_tokens: Sequence[int])

src/art/tinker/prefix_cache.py:70

↓ 5 callersFunctionmigrate_model_dir

Migrate all trajectory files for a model. Args: model_dir: Path to the model directory (containing trajectories/ subdirectory).

src/art/utils/trajectory_migration.py:377

↓ 5 callersFunctionpacked_tensors_from_dir

(**kwargs: Unpack[DiskPackedTensors])

src/art/preprocessing/pack.py:177

↓ 5 callersFunctionparse_args

(argv: list[str] | None = None)

src/art/vllm/dedicated_server.py:16

↓ 5 callersFunctionparse_response_for_stop_token

Parse response for a single stop token. We expect a properly rendered response to have exactly one stop token; but it may have zero if e.g. the m

src/art/tinker/cookbook_v/renderers/base.py:1184

↓ 5 callersMethodping

(self)

tests/unit/test_mp_actors.py:25

↓ 5 callersFunctionpull_model_from_s3

Pull a model from S3 to the local directory. Args: model_name: The name of the model to pull. project: The project name.

src/art/utils/s3.py:161

↓ 5 callersMethodread_state

Read persistent state from the model directory. Returns: The state dictionary if it exists, or None if no state has been saved.

src/art/model.py:404

↓ 5 callersFunctionrollout

( x_model: art.Model, o_model: art.Model, scenario: TicTacToeScenario )

examples/tic_tac_toe_self_play/rollout.py:112

↓ 5 callersFunctions3_sync

Synchronise *source* and *destination* using the AWS CLI. Either *source* or *destination* (or both) can point to an S3 URI, making it possib

src/art/utils/s3.py:66

↓ 5 callersMethodstart

(self)

src/art/tinker/server.py:75

↓ 5 callersMethodtrain

( self, model: AnyTrainableModel, trajectory_groups: Iterable[TrajectoryGroup],

src/art/backend.py:38

↓ 5 callersFunctiontrain_one_step

Run one training step and return the result.

tests/integration/test_push_and_fork.py:49

↓ 5 callersFunctionwarn

(msg)

src/art/utils/logging.py:35

↓ 4 callersMethod__get_step

(self, model: Model)

src/art/local/backend.py:359

↓ 4 callersMethod__init__

(self, message: str, status_code: int)

src/art/errors.py:11

↓ 4 callersMethod__init__

( self, adapter_model_prefix: str, in_features: int, out_features: int,

src/art/megatron/lora.py:20

↓ 4 callersMethod_fmt_float

(value: object, decimals: int)

src/art/pipeline_trainer/status.py:394

↓ 4 callersMethod_fmt_float_compact

(value: object, decimals: int)

src/art/pipeline_trainer/status.py:402

↓ 4 callersMethod_fmt_int_compact

(value: object)

src/art/pipeline_trainer/status.py:388

↓ 4 callersMethod_get_service

(self, model: TrainableModel)

src/art/local/backend.py:244

↓ 4 callersFunction_get_trainer_optimizer

(trainer: GRPOTrainer)

src/art/unsloth/service.py:194

↓ 4 callersMethod_log_metrics

Log metrics to history.jsonl and optionally wandb.

src/art/model.py:510

↓ 4 callersMethod_persist_state

(self, training_step: int)

src/art/pipeline_trainer/trainer.py:772

↓ 4 callersMethod_save_sampler_weights

( self, training_client: tinker.TrainingClient, checkpoint_name: str, )

src/art/tinker_native/backend.py:806

↓ 4 callersMethodadd_user_timing

( self, step_wall_s: float | None = None, step_actor_s: float | None = None, s

src/art/metrics.py:159

↓ 4 callersMethodbackend

(self)

src/art/model.py:214

↓ 4 callersMethodbuild_generation_prompt

Generates tokens for sampling from the model. Args: messages: a list of messages to render. role: the role o

src/art/tinker/cookbook_v/renderers/base.py:941

↓ 4 callersFunctioncalculate_beat_comp

( groups: list[art.TrajectoryGroup], control_groups: list[art.TrajectoryGroup], control_first: boo

examples/mcp-rl/mcp_rl/benchmarks/generate_benchmarks.py:46

↓ 4 callersMethodclose

(self)

src/art/tinker_native/backend.py:178

↓ 4 callersMethodclose

Terminate vLLM subprocess if running.

src/art/unsloth/service.py:451

↓ 4 callersFunctioncondense_sequence

(sequence: list[int | None])

examples/2048/utils.py:89

↓ 4 callersFunctioncreate_temp_jsonl

Create a temporary JSONL file with dummy trajectories.

tests/unit/test_sft.py:17

↓ 4 callersMethodfrom_dict

Create a Tool from a dictionary.

src/art/mcp/types.py:19

↓ 4 callersFunctiongather_trajectory_groups_by_index

( grouped_trajectory_awaitables: Iterable[Awaitable[tuple[art.Trajectory, ...]]], *, pbar_desc: st

examples/tic_tac_toe_self_play/gather_trajectory_groups_by_index.py:10

↓ 4 callersFunctiongenerate_val_groups

( model: art.Model, val_scenarios: List[McpScenario] )

examples/mcp-rl/mcp_rl/benchmarks/generate_benchmarks.py:31

↓ 4 callersFunctionget_messages

(messages_and_choices: MessagesAndChoices)

src/art/trajectories.py:97

↓ 4 callersFunctioniterate_dataset

Generates batches from a dataset over multiple epochs with deterministic shuffling. Args: dataset: The list of data items. g

src/art/utils/iterate_dataset.py:21

↓ 4 callersMethodlitellm_completion_params

Return the parameters that should be sent to litellm.completion. Args: step: If provided, returns params for specific checkpoint

src/art/model.py:260

↓ 4 callersMethodload_state_dict

(self, state: dict[str, Any])

src/art/metrics.py:307

↓ 4 callersFunctionlog_timing

Context manager that logs a message with timestamp and duration.

src/art/tinker/service.py:27

↓ 4 callersFunctionnormalize_provider

(provider: str | None)

src/art/api_costs.py:74

↓ 4 callersFunctionrecord_metrics

(context: "GatherContext", trajectory: Trajectory)

src/art/gather.py:183

↓ 4 callersMethodregister_model_pricing

( self, model_name: str, *, prompt_per_million: float, completion_per_

src/art/metrics.py:268

↓ 4 callersFunctionrender_board

(game: TwentyFortyEightGame)

examples/2048/utils.py:59

↓ 4 callersFunctionrollout

(model: art.Model, scenario: FactsScenario)

examples/just-the-facts/just_the_facts/rollout.py:26

↓ 4 callersFunctionset_gather_context

(context: GatherContext)

src/art/gather.py:248

↓ 4 callersMethodstate_dict

(self)

src/art/metrics.py:301

↓ 3 callersMethod__init__

Initialize MCP server.

examples/mcp-rl/mcp_rl/mcp_server.py:16

↓ 3 callersFunction_check_job_status

Checks the status of a model deployment job in Together.

src/art/utils/deployment/together.py:147

↓ 3 callersMethod_ensure_lora_adapter_config

( self, lora_path: str, *, source_path: str | None = None )

src/art/megatron/service.py:124

↓ 3 callersFunction_ensure_message

Narrow a trajectory entry to a concrete message (not a Choice).

tests/unit/test_trajectory_parquet.py:58

↓ 3 callersFunction_extract_cost_value

(value: Any)

src/art/api_costs.py:126

↓ 3 callersFunction_format_description

(description: str, indent: str = "")

src/art/tinker/cookbook_v/renderers/kimi_k2_5_tool_declaration_ts.py:47

↓ 3 callersFunction_format_json

Format JSON data for display

examples/mcp-rl/servers/python/mcp_balldontlie/server.py:318

↓ 3 callersMethod_get_tenant

(self, api_key: str)

src/art/tinker/server.py:277

↓ 3 callersFunction_init_session

Initializes a session for interacting with Together.

src/art/utils/deployment/together.py:63

↓ 3 callersMethod_list_checkpoints

( self, rest_client: Any, tinker_run_ids: list[str] )

src/art/tinker_native/backend.py:642

↓ 3 callersFunction_load_font

(font_path: str | Path | None, preferred_size: int)

dev/yes-no-maybe-vision/generate_images.py:14

↓ 3 callersFunction_make_group

( rewards: list[float], *, initial_policy_version: int | None )

tests/unit/test_pipeline_trainer_metrics.py:13

↓ 3 callersMethod_metrics_builder_for_split

(self, split: str)

src/art/model.py:672

↓ 3 callersMethod_next_lora_id

Return a new unique LoRA ID to avoid collisions in vLLM.

src/art/unsloth/service.py:324

↓ 3 callersMethod_pending_state

(self)

src/art/metrics.py:340

↓ 3 callersMethod_persist_model_state

(self, model: TrainableModel, state: ModelState)

src/art/tinker_native/backend.py:832

↓ 3 callersMethod_reset_optimizer_if_mode_changed

Reset optimizer state if training mode changed. Uses a single shared optimizer (trainer.optimizer) for both SFT and RL. Resets optimi

src/art/unsloth/service.py:539

↓ 3 callersFunction_response_usage

(response: Any)

src/art/api_costs.py:133

↓ 3 callersMethod_update_ewa

(self, previous: float | None, new_value: float)

src/art/pipeline_trainer/status.py:359

↓ 3 callersMethodadd_idle_times

( self, step_trainer_idle_s: float | None = None, step_actor_idle_s: float | None = No

src/art/metrics.py:172

↓ 3 callersMethodadd_response_cost

( self, source: str, response: Any, *, provider: str, model_na

src/art/metrics.py:104

↓ 3 callersFunctionassert_chat_logprobs

( client: openai.AsyncOpenAI, model_name: str, )

tests/integration/test_vllm_contract.py:89

↓ 3 callersFunctionbenchmark_model

(model: art.Model)

examples/benchmarking_comparison_models.py:64

↓ 3 callersMethodbuild_supervised_example

Override to properly handle thinking preservation for the last assistant message. Also ensures default system message is prepended if

src/art/tinker/cookbook_v/renderers/kimi_k2.py:615

↓ 3 callersFunctioncheck_winner

(board: list[list[str]])

examples/tic_tac_toe_self_play/game_utils.py:82

↓ 3 callersFunctioncheck_winner

(board: list[list[str]])

examples/tic_tac_toe/game_utils.py:80

↓ 3 callersFunctioncompute_sample_costs

Compute Tinker prompt+completion costs for a single inference call.

src/art/costs.py:92

↓ 3 callersFunctionconvert_openai_messages_to_renderer_format

( messages: list[dict[str, Any]], tools: list[dict[str, Any]] | None, renderer: Any, )

src/art/tinker_native/data.py:38

↓ 3 callersFunctioncreate_sample_tools

Create sample tools for testing.

examples/mcp-rl/test_scenario_generation.py:15

↓ 3 callersMethoddelete_checkpoints

Delete all but the latest and best checkpoints. Args: best_checkpoint_metric: The metric to use to determine the best ch

src/art/model.py:1017

↓ 3 callersFunctiondeploy_model

Deploy a model checkpoint to a hosted inference endpoint. This function assumes the checkpoint is already available locally. Use Backend.pull

src/art/utils/deployment/common.py:28

↓ 3 callersFunctionensure_reward_variance

(groups)

tests/integration/test_tinker_native_backend.py:20

↓ 3 callersMethodexecute

()

src/art/langgraph/llm_wrapper.py:145

↓ 3 callersMethodfetch_data

Fetch data from Balldontlie API

examples/mcp-rl/servers/python/mcp_balldontlie/server.py:23

↓ 3 callersMethodfinish

(self)

src/art/trajectories.py:53

↓ 3 callersFunctionget_base_model

()

tests/integration/test_tinker_native_backend.py:16

↓ 3 callersFunctionget_gather_context

()

src/art/gather.py:256

↓ 3 callersFunctionget_latest_checkpoint_step_from_s3

Get the latest checkpoint step number from S3 without downloading files. Returns: The latest step number, or None if no checkpoints

src/art/utils/s3_checkpoint_utils.py:7

↓ 3 callersFunctionget_model_step

(model: "TrainableModel", art_path: str)

src/art/utils/get_model_step.py:26

↓ 3 callersFunctionget_model_trajectories

( trajectories: list[art.Trajectory], )

examples/tic_tac_toe_self_play/train_o4_mini.py:58

↓ 3 callersFunctionget_openai_server_config

( model_name: str, base_model: str, log_file: str, lora_path: str | None = None, config: "

src/art/dev/openai_server.py:8

↓ 3 callersFunctioninit_chat_completion

(chunk: ChatCompletionChunk)

src/art/openai.py:64

↓ 3 callersFunctionload_benchmarked_models

( project: str, benchmark_keys: list[BenchmarkedModelKey], metrics: list[str] = ["reward"], ap

src/art/utils/old_benchmarking/load_benchmarked_models.py:18

↓ 3 callersMethodmerge_state

Deep-merge state into the existing state and persist it. Args: state: A dictionary of JSON-serializable values to merge.

src/art/model.py:346

↓ 3 callersFunctionmigrate_trajectories_dir

Migrate all JSONL files in a trajectories directory (including subdirectories). Args: trajectories_dir: Path to the trajectories dir

src/art/utils/trajectory_migration.py:335

← previous next →101–200 of 1,502, ranked by callers