MCPcopy Create free account
hub / github.com/NVIDIA/TensorRT-LLM / load_from_hf

Method load_from_hf

tensorrt_llm/lora_manager.py:877–1179  ·  view source on GitHub ↗

Returns the adapter UIDs that were loaded by this call. Note that when an adapter was already loaded before this call, it would not be included in the returned list of UIDs. Lora config of https://huggingface.co/hfl/chinese-alpaca-2-lora-7b. { "base_mod

(
        self,
        model_dirs: List[str],
        model_config: Union["ModelConfig", LoraModelConfig],
        uids: Optional[List[str]] = None,
        component: Optional[str] = None,
    )

Source from the content-addressed store, hash-verified

875 return new_uids
876
877 def load_from_hf(
878 self,
879 model_dirs: List[str],
880 model_config: Union["ModelConfig", LoraModelConfig],
881 uids: Optional[List[str]] = None,
882 component: Optional[str] = None,
883 ) -> List[str]:
884 """Returns the adapter UIDs that were loaded by this call.
885
886 Note that when an adapter was already loaded before this call, it would not be
887 included in the returned list of UIDs.
888
889 Lora config of https://huggingface.co/hfl/chinese-alpaca-2-lora-7b.
890
891 {
892 "base_model_name_or_path": "/Llama-2-7b-hf",
893 "bias": "none",
894 "enable_lora": null,
895 "fan_in_fan_out": false,
896 "inference_mode": true,
897 "lora_alpha": 128.0,
898 "lora_dropout": 0.05,
899 "merge_weights": false,
900 "modules_to_save": [
901 "embed_tokens",
902 "lm_head"
903 ],
904 "peft_type": "LORA",
905 "r": 64,
906 "target_modules": [
907 "q_proj",
908 "v_proj",
909 "k_proj",
910 "o_proj",
911 "gate_proj",
912 "down_proj",
913 "up_proj"
914 ],
915 "task_type": "CAUSAL_LM"
916
917 }
918
919 keys in adapter_model.bin:
920 base_model.model.model.layers.0.self_attn.q_proj.lora_A.weight torch.Size([64, 4096])
921 base_model.model.model.layers.0.self_attn.q_proj.lora_B.weight torch.Size([4096, 64])
922 base_model.model.model.layers.0.self_attn.k_proj.lora_A.weight torch.Size([64, 4096])
923 base_model.model.model.layers.0.self_attn.k_proj.lora_B.weight torch.Size([4096, 64])
924 base_model.model.model.layers.0.self_attn.v_proj.lora_A.weight torch.Size([64, 4096])
925 base_model.model.model.layers.0.self_attn.v_proj.lora_B.weight torch.Size([4096, 64])
926 base_model.model.model.layers.0.self_attn.o_proj.lora_A.weight torch.Size([64, 4096])
927 base_model.model.model.layers.0.self_attn.o_proj.lora_B.weight torch.Size([4096, 64])
928 base_model.model.model.layers.0.mlp.gate_proj.lora_A.weight torch.Size([64, 4096])
929 base_model.model.model.layers.0.mlp.gate_proj.lora_B.weight torch.Size([11008, 64])
930 base_model.model.model.layers.0.mlp.up_proj.lora_A.weight torch.Size([64, 4096])
931 base_model.model.model.layers.0.mlp.up_proj.lora_B.weight torch.Size([11008, 64])
932 base_model.model.model.layers.0.mlp.down_proj.lora_A.weight torch.Size([64, 11008])
933 base_model.model.model.layers.0.mlp.down_proj.lora_B.weight torch.Size([4096, 64])
934 ...

Callers 2

load_from_ckptMethod · 0.95
__init__Method · 0.80

Calls 6

_generate_uidMethod · 0.95
invert_module_mappingFunction · 0.85
release_gcFunction · 0.85
appendMethod · 0.45
loadMethod · 0.45
keysMethod · 0.45

Tested by

no test coverage detected