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

↓ 3 callersFunctionnormalize_model_name

(model_name: str | None)

src/art/api_costs.py:326

↓ 3 callersMethodoffload_to_cpu

Offload training model and optimizer to CPU using pinned memory for faster transfers.

src/art/unsloth/service.py:223

↓ 3 callersMethodparse_response

Parse sampled tokens back into a Message. Args: response: Token IDs returned from sampling. Returns:

src/art/tinker/cookbook_v/renderers/base.py:811

↓ 3 callersFunctionparse_think_blocks

Parse a string with only <think>...</think> tags into ThinkingPart/TextPart list. This is a simpler version of parse_content_blocks for rend

src/art/tinker/cookbook_v/renderers/base.py:583

↓ 3 callersFunctionpatch_tool_parser_manager

Patch ToolParserManager to support streaming tool call logprobs.

src/art/vllm/patches.py:38

↓ 3 callersFunctionpopulate_random_cell

(game: TwentyFortyEightGame)

examples/2048/utils.py:28

↓ 3 callersMethodpreview

Get a preview of the scenario task.

src/art/mcp/types.py:90

↓ 3 callersFunctionrecord_provenance

Record provenance on the latest artifact version's metadata.

src/art/utils/record_provenance.py:9

↓ 3 callersMethodrender_message

Render a single message into its header/output/stop_overlap components. This method breaks down a message into parts for loss maskin

src/art/tinker/cookbook_v/renderers/base.py:791

↓ 3 callersMethodreset

Clear any buffered tokens.

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

↓ 3 callersMethodreset_lora_parameters

Initialize LoRA weights (A=Kaiming, B=zeros) like PEFT defaults.

src/art/megatron/lora.py:54

↓ 3 callersFunctionrollout

( model: art.Model, step: int, is_validation: bool, verbose: bool = False )

examples/2048/rollout.py:26

↓ 3 callersFunctionrollout

(model: art.Model, scenario: TicTacToeScenario)

examples/tic_tac_toe/rollout.py:28

↓ 3 callersFunctionrollout

Run an MCP agent rollout with server parameters. Args: model: The ART model to use for the agent scenario: The MCP scenario to ru

examples/mcp-rl/mcp_rl/rollout.py:43

↓ 3 callersFunctionruler_score_group

Score a trajectory group using RULER for use in training loops. This is the recommended API for using RULER with ART. It integrates seamlessly

src/art/rewards/ruler.py:264

↓ 3 callersFunctionrun_training_loop

Run a simple training loop and return the TrainResults from each train call.

tests/integration/test_multi_checkpoint_training.py:63

↓ 3 callersFunctionsave_checkpoint

Save a checkpoint and return the checkpoint directory path.

src/art/unsloth/service.py:149

↓ 3 callersFunctionscore_title

Get the reward model score for a story asynchronously. Args: story_dict: Dictionary containing story data with keys: title, by, time, scr

examples/hn_title_generator/utils.py:95

↓ 3 callersFunctionsimple_rollout

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

tests/integration/test_tinker_native_backend.py:30

↓ 3 callersMethodstart

(self, *, initial_step: int | None = None)

src/art/pipeline_trainer/status.py:62

↓ 3 callersMethodstart

Start the MCP server connection.

examples/mcp-rl/mcp_rl/mcp_server.py:21

↓ 3 callersFunctionsubclass_chat_completion_request

Subclass ChatCompletionRequest so that logprobs are always returned.

src/art/vllm/patches.py:6

↓ 3 callersMethodto_dicts

Convert all scenarios to dictionaries.

src/art/mcp/types.py:119

↓ 3 callersMethodto_openai_message

Convert a Message to OpenAI chat completions API format. The returned object can be passed into the transformers library's a

src/art/tinker/cookbook_v/renderers/base.py:824

↓ 3 callersMethodtoo_many_exceptions

(self)

src/art/gather.py:228

↓ 3 callersMethodtrain

Train the model on the given trajectory groups. This is the recommended way to train models. Unlike model.train(), this method does N

src/art/serverless/backend.py:190

↓ 3 callersMethodtrain

( # type: ignore[override] self, model: TrainableModel, trajectory_groups: Iterable[T

src/art/tinker_native/backend.py:231

↓ 3 callersMethodtrain

Run the training pipeline over the configured scenario iterator.

src/art/pipeline_trainer/trainer.py:158

↓ 3 callersFunctionupdate_chat_completion

( chat_completion: ChatCompletion, chunk: ChatCompletionChunk )

src/art/openai.py:82

↓ 3 callersMethodvllm_engine_is_sleeping

(self)

src/art/local/service.py:23

↓ 3 callersFunctionwrap_group_awaitable

( awaitable: Awaitable[TrajectoryGroup], )

src/art/gather.py:140

↓ 2 callersMethod__aenter__

Async context manager entry.

examples/mcp-rl/mcp_rl/mcp_server.py:52

↓ 2 callersMethod__aexit__

Async context manager exit.

examples/mcp-rl/mcp_rl/mcp_server.py:57

↓ 2 callersMethod__init__

( self, *, name: str, project: str, entity: str | None = None,

src/art/model.py:126

↓ 2 callersMethod_adapter_has_weights

(self, lora_path: str)

src/art/megatron/service.py:81

↓ 2 callersMethod_add_lora_aliases

( self, llm: AsyncLLM, step: int, checkpoint_dir: str )

src/art/megatron/service.py:139

↓ 2 callersFunction_add_tensor_list

( packed_tensors: dict[str, Any], disk_packed_tensors: DiskPackedTensors, key: str, dtype: tor

src/art/preprocessing/pack.py:202

↓ 2 callersMethod_build_snapshot

(self)

src/art/pipeline_trainer/status.py:171

↓ 2 callersFunction_cacheable_prefix

(word_count: int = 1500)

tests/integration/test_live_api_cost.py:36

↓ 2 callersFunction_canonicalize_upstream_metric_key

(metric: str)

src/art/tinker_native/backend.py:75

↓ 2 callersFunction_canonicalize_upstream_metric_key

(metric: str)

src/art/unsloth/train.py:42

↓ 2 callersFunction_canonicalize_upstream_metrics

(metrics: dict[str, float])

src/art/serverless/backend.py:70

↓ 2 callersFunction_chat_with_skill

Run an interactive chat session using a skill as the system prompt.

src/art/cli.py:39

↓ 2 callersMethod_check_all_failed

Raise if all rollouts in a group failed with exceptions.

src/art/pipeline_trainer/trainer.py:539

↓ 2 callersMethod_close

(self)

src/art/local/backend.py:176

↓ 2 callersFunction_coerce_float

(value: Any)

src/art/api_costs.py:117

↓ 2 callersMethod_collect_automatic_backend_metrics

( self, *, split: str, step: int, provided_metric_keys: set[str],

src/art/model.py:597

↓ 2 callersFunction_common_prefix_len

( tokens: Sequence[int], start: int, label: tuple[int, ...] )

src/art/tinker/prefix_cache.py:49

↓ 2 callersMethod_create_training_client_from_checkpoint

( self, service_client: tinker.ServiceClient, checkpoint_state_path: str, base

src/art/tinker_native/backend.py:749

↓ 2 callersFunction_decode_model_input

(tokenizer: FakeTokenizer, model_input: object)

tests/unit/test_tinker_renderers.py:53

↓ 2 callersMethod_default_lora_adapter_config

(self)

src/art/megatron/service.py:64

↓ 2 callersMethod_delete_checkpoint_files

( self, model: AnyTrainableModel, steps_to_keep: list[int] )

src/art/backend.py:28

↓ 2 callersFunction_ensure_assistant_message

( item: MessageOrChoice, )

tests/unit/test_trajectory_parquet.py:64

↓ 2 callersMethod_ensure_system_message

Ensure a default system message is present if none exists. This matches the HuggingFace chat template behavior where a default system

src/art/tinker/cookbook_v/renderers/kimi_k2.py:376

↓ 2 callersFunction_ensure_user_message

(item: MessageOrChoice)

tests/unit/test_trajectory_parquet.py:78

↓ 2 callersMethod_evict

(self)

src/art/tinker/prefix_cache.py:148

↓ 2 callersMethod_experimental_fork_checkpoint

Fork a checkpoint from another model to initialize this model. Pulls the source checkpoint from W&B artifacts (or S3 if from_s3_bucket

src/art/serverless/backend.py:736

↓ 2 callersMethod_experimental_pull_model_checkpoint

Pull a model checkpoint to a local path. For LocalBackend, this: 1. When step is "latest" or None, checks both local storage and S3 (

src/art/local/backend.py:913

↓ 2 callersMethod_experimental_push_to_s3

Push model checkpoints from W&B artifacts to S3. Downloads checkpoint(s) from W&B and uploads them to S3. Args: model: T

src/art/serverless/backend.py:674

↓ 2 callersMethod_fail_pending

(self, error: Exception)

src/mp_actors/move.py:140

↓ 2 callersMethod_format_scaled

(value: int, scale: int, suffix: str)

src/art/pipeline_trainer/status.py:374

↓ 2 callersMethod_get_last_checkpoint_dir

(self)

src/art/tinker/service.py:251

↓ 2 callersMethod_get_qwen_role_for_message

Get the role to use for rendering a message in Qwen format. Per HuggingFace Qwen3 chat template, tool messages are rendered with role "user".

src/art/tinker/cookbook_v/renderers/qwen3.py:116

↓ 2 callersMethod_get_renderer

(self, base_model: str)

src/art/tinker/server.py:486

↓ 2 callersFunction_get_skill_path

Find a skill file, checking installed package first, then repo root.

src/art/cli.py:20

↓ 2 callersMethod_get_system_message

Return system message if configured, else None. Uses internal role to render as actual 'system' (not mapped to 'developer').

src/art/tinker/cookbook_v/renderers/gpt_oss.py:352

↓ 2 callersFunction_get_tensor_list_info

( tensors: list[torch.Tensor | None], )

src/art/preprocessing/pack.py:246

↓ 2 callersMethod_group_initial_version

(self, group: TrajectoryGroup)

src/art/pipeline_trainer/trainer.py:738

↓ 2 callersMethod_group_zero_variance

(group: TrajectoryGroup)

src/art/pipeline_trainer/trainer.py:731

↓ 2 callersFunction_history_rows

(history_path: Path)

tests/integration/test_live_api_cost.py:40

↓ 2 callersMethod_is_group_stale

(self, group: TrajectoryGroup, min_version: int)

src/art/pipeline_trainer/trainer.py:683

↓ 2 callersMethod_is_valid_decode

Check if decoded text represents a complete UTF-8 sequence. Returns False if the text ends with a replacement character, which indica

src/art/tinker/cookbook_v/renderers/base.py:216

↓ 2 callersFunction_json_type_to_typescript

Convert a single JSON schema type to TypeScript.

src/art/tinker/cookbook_v/renderers/gpt_oss.py:34

↓ 2 callersFunction_list_param_shapes_from_safetensors_remote

Returns {param_name: shape_tuple} by reading ONLY the safetensors header(s) over HTTP (ranged requests). No full file download.

src/art/tinker/cookbook_v/hyperparam_utils.py:17

↓ 2 callersFunction_load_json_payload

(payload: object | None)

src/art/utils/trajectory_logging.py:174

↓ 2 callersMethod_load_metrics_builder_state

(self)

src/art/model.py:679

↓ 2 callersMethod_log

(self, completion_id, input, output)

src/art/langgraph/llm_wrapper.py:137

↓ 2 callersFunction_longest_matching_suffix_prefix

Find longest suffix of text that matches a prefix of tag. This is used during streaming to determine how many characters at the end of accumu

src/art/tinker/cookbook_v/renderers/kimi_k2.py:101

↓ 2 callersFunction_model_checkpoint_id

Generates a unique ID for a model checkpoint.

src/art/utils/deployment/together.py:77

↓ 2 callersFunction_openai_completion

(*, api_key: str, prompt_cache_key: str, prefix: str)

tests/integration/test_live_api_cost.py:44

↓ 2 callersFunction_parse_tool_calls_section

( tool_section: str, )

src/art/tinker/cookbook_v/renderers/kimi_k2.py:64

↓ 2 callersMethod_persist_metrics_builder_state

(self)

src/art/model.py:688

↓ 2 callersMethod_pop_float_metadata

(group: TrajectoryGroup, key: str)

src/art/pipeline_trainer/trainer.py:808

↓ 2 callersFunction_post_json

(url: str, *, headers: dict[str, str], payload: dict)

tests/integration/test_live_api_cost.py:25

↓ 2 callersMethod_prepare_backend_for_training

( self, model: AnyTrainableModel, config: dev.OpenAIServerConfig | None, )

src/art/backend.py:32

↓ 2 callersFunction_record_api_cost

( *, result: Any, source: str, provider: str, response_getter: ResponseGetter | None,

src/art/api_costs.py:449

↓ 2 callersMethod_record_zero_variance

(self, group: TrajectoryGroup)

src/art/pipeline_trainer/trainer.py:689

↓ 2 callersMethod_reload_adapter

Reload LoRA adapter in vLLM subprocess via HTTP.

src/art/unsloth/service.py:426

↓ 2 callersFunction_require_live_test_env

(*required_vars: str)

tests/integration/test_live_api_cost.py:17

↓ 2 callersMethod_resolve_model_config

(self, model: TrainableModel)

src/art/tinker_native/backend.py:613

↓ 2 callersMethod_rollout_worker

(self, worker_id: int)

src/art/pipeline_trainer/trainer.py:318

↓ 2 callersMethod_run_eval

(self, step: int)

src/art/pipeline_trainer/trainer.py:579

↓ 2 callersMethod_save_checkpoint

Save checkpoint and return the sampler weights path.

src/art/tinker/service.py:260

↓ 2 callersFunction_set_cache_dir

(env_var: str, default_path: str)

src/art/megatron/train.py:5

↓ 2 callersFunction_sha256_file

(path: Path)

scripts/ci/compute_uv_fingerprint.py:13

↓ 2 callersFunction_split_tool_calls_section

(content: str)

src/art/tinker/cookbook_v/renderers/kimi_k2.py:47

↓ 2 callersFunction_stamp_pickle_metadata

Stamp renderer with metadata needed for pickle support.

src/art/tinker/cookbook_v/renderers/__init__.py:139

↓ 2 callersMethod_sync_wandb_config

( self, run: "Run", )

src/art/model.py:445

↓ 2 callersMethod_timestamp

(self)

src/art/tinker_native/backend.py:138

↓ 2 callersMethod_tinker_call

( self, label: str, awaitable: Awaitable[T], *, env_name: str,

src/art/tinker_native/backend.py:141

← previous next →201–300 of 1,502, ranked by callers