trinity.common.models package

Subpackages

• trinity.common.models.vllm_patch package
  • Submodules
    • trinity.common.models.vllm_patch.api_patch module
    • trinity.common.models.vllm_patch.api_patch_v12 module
    • trinity.common.models.vllm_patch.worker_patch module
  • Module contents

Submodules

• trinity.common.models.mm_utils module
  • is_qwen_like_processor()
  • build_multi_modal_data()
  • build_mm_input_for_training()
  • build_mm_message()
  • has_multi_modal_content()
• trinity.common.models.model module
  • InferenceModel
    • InferenceModel.__init__()
    • InferenceModel.generate()
    • InferenceModel.chat()
    • InferenceModel.logprobs()
    • InferenceModel.convert_messages_to_experience()
    • InferenceModel.prepare()
    • InferenceModel.sync_model()
    • InferenceModel.get_model_version()
    • InferenceModel.get_available_address()
    • InferenceModel.get_api_server_url()
    • InferenceModel.get_api_key()
    • InferenceModel.get_model_config()
    • InferenceModel.get_model_path()
    • InferenceModel.shutdown()
  • BaseInferenceModel
    • BaseInferenceModel.__init__()
    • BaseInferenceModel.apply_chat_template()
    • BaseInferenceModel.convert_messages_to_experience()
  • ModelWrapper
    • ModelWrapper.__init__()
    • ModelWrapper.prepare()
    • ModelWrapper.generate()
    • ModelWrapper.generate_async()
    • ModelWrapper.chat()
    • ModelWrapper.chat_async()
    • ModelWrapper.logprobs()
    • ModelWrapper.logprobs_async()
    • ModelWrapper.convert_messages_to_experience()
    • ModelWrapper.convert_messages_to_experience_async()
    • ModelWrapper.api_key
    • ModelWrapper.model_version
    • ModelWrapper.model_version_async
    • ModelWrapper.model_path
    • ModelWrapper.model_path_async
    • ModelWrapper.model_name
    • ModelWrapper.model_config
    • ModelWrapper.generate_kwargs
    • ModelWrapper.get_lora_request()
    • ModelWrapper.get_lora_request_async()
    • ModelWrapper.get_message_token_len()
    • ModelWrapper.get_openai_client()
    • ModelWrapper.get_openai_async_client()
    • ModelWrapper.get_current_load()
    • ModelWrapper.sync_model_weights()
    • ModelWrapper.extract_experience_from_history()
    • ModelWrapper.set_workflow_state()
    • ModelWrapper.clean_workflow_state()
    • ModelWrapper.get_workflow_state()
    • ModelWrapper.clone_with_isolated_history()
  • convert_api_output_to_experience()
  • HistoryRecordingStream
    • HistoryRecordingStream.__init__()
    • HistoryRecordingStream.close()
    • HistoryRecordingStream.aclose()
  • extract_logprobs()
• trinity.common.models.tinker_model module
• trinity.common.models.utils module
  • tokenize_and_mask_messages_hf()
  • tokenize_and_mask_messages_default()
  • get_action_mask_method()
  • get_checkpoint_dir_with_step_num()
  • get_latest_state_dict()
  • load_state_dict()
  • get_verl_checkpoint_info()
  • load_fsdp_state_dict_from_verl_checkpoint()
  • load_huggingface_state_dict()
  • get_megatron_converter()
• trinity.common.models.vllm_model module
• trinity.common.models.vllm_worker module

Module contents

trinity.common.models.create_explorer_models(config: Config) → Tuple[List, List[List]]

Create rollout_models and auxiliary_models.

参数:

config -- The trinity configuration.

返回:

The rollout_models and auxiliary_models.

返回类型:

Tuple[List, List[List]]

trinity.common.models.create_vllm_inference_models(config: InferenceModelConfig, allocator: _BundleAllocator, actor_name: str) → List

async trinity.common.models.create_debug_explorer_model(config: Config) → None

Create explorer inference models for debugging.

trinity.common.models.get_debug_explorer_model(config: Config) → Tuple[InferenceModel, List[InferenceModel]]

Get the inference models for debugging. The models must be created by create_debug_explorer_model in another process first.

async trinity.common.models.get_auxiliary_model_wrappers(config: Config) → Dict[str | int, List[ModelWrapper]]

Get auxiliary models.

返回:

A dictionary mapping auxiliary model

index to a list of auxiliary model actor handlers.

返回类型:

Dict[str| int, List[ModelWrapper]]