diff --git a/src/text_generation/adapters/foundation_models/config/meta_llama_config.py b/src/text_generation/adapters/foundation_models/config/meta_llama_config.py index ce3cb860d..97475da33 100644 --- a/src/text_generation/adapters/foundation_models/config/meta_llama_config.py +++ b/src/text_generation/adapters/foundation_models/config/meta_llama_config.py @@ -7,4 +7,6 @@ from src.text_generation.adapters.foundation_models.base.base_model_config impor class MetaLlamaConfig(BaseModelConfig): """meta-llama/Llama-3.2-3B-Instruct configuration""" use_flash_attention: bool = False - rope_scaling: Optional[Dict[str, Any]] = None \ No newline at end of file + rope_scaling: Optional[Dict[str, Any]] = None + trust_remote_code: True, + torch_dtype: "auto" \ No newline at end of file diff --git a/src/text_generation/adapters/foundation_models/meta_llama_foundation_model.py b/src/text_generation/adapters/foundation_models/meta_llama_foundation_model.py index 1da2ad816..5e9eed920 100644 --- a/src/text_generation/adapters/foundation_models/meta_llama_foundation_model.py +++ b/src/text_generation/adapters/foundation_models/meta_llama_foundation_model.py @@ -4,7 +4,7 @@ from transformers import AutoModelForCausalLM, AutoTokenizer from src.text_generation.adapters.foundation_models.config.meta_llama_config import MetaLlamaConfig from src.text_generation.adapters.foundation_models.base.base_foundation_model import BaseFoundationModel from src.text_generation.common.model_id import ModelId -from huggingface_hub import login +from huggingface_hub import login, list_repo_files if "HF_TOKEN" in os.environ: hf_token = os.environ.get("HF_TOKEN", "").strip() @@ -30,6 +30,11 @@ class MetaLlamaFoundationModel(BaseFoundationModel): super().__init__(config) def _load_model(self) -> None: + + files = list_repo_files(self.MODEL_ID.value) + for file in files: + print(file) + self.tokenizer = AutoTokenizer.from_pretrained( self.MODEL_ID.value, local_files_only=self.config.local_files_only