MCPcopy
hub / github.com/NVIDIA/TensorRT-LLM / get_model_format

Function get_model_format

tensorrt_llm/llmapi/llm_args.py:3380–3407  ·  view source on GitHub ↗

Get the format of the model.

(model_dir: str,
                     trust_remote_code: bool = False)

Source from the content-addressed store, hash-verified

3378
3379
3380def get_model_format(model_dir: str,
3381 trust_remote_code: bool = False) -> _ModelFormatKind:
3382 ''' Get the format of the model. '''
3383 if not (Path(model_dir) / 'config.json').exists():
3384 raise ValueError(
3385 f"Failed to infer model format because no config.json exists in {model_dir}"
3386 )
3387
3388 with open(Path(model_dir) / 'config.json') as f:
3389 config = json.load(f)
3390
3391 try:
3392 if 'pretrained_config' in config and 'build_config' in config:
3393 model_format = _ModelFormatKind.TLLM_ENGINE
3394 EngineConfig.from_json_file(Path(model_dir) / 'config.json')
3395 elif 'architecture' in config and 'dtype' in config:
3396 model_format = _ModelFormatKind.TLLM_CKPT
3397 PretrainedConfig.from_checkpoint(model_dir)
3398 else:
3399 model_format = _ModelFormatKind.HF
3400 AutoConfig.from_hugging_face(model_dir,
3401 trust_remote_code=trust_remote_code)
3402 except Exception as e:
3403 raise ValueError(
3404 f"Inferred model format {model_format}, but failed to load config.json: {e}"
3405 )
3406 else:
3407 return model_format
3408
3409
3410LlmArgs = TorchLlmArgs

Callers 2

test_CachedModelLoaderFunction · 0.85

Calls 4

loadMethod · 0.45
from_json_fileMethod · 0.45
from_checkpointMethod · 0.45
from_hugging_faceMethod · 0.45

Tested by 1

test_CachedModelLoaderFunction · 0.68