load weights manually

generate_summaries_uv.py

@@ -25,7 +25,7 @@ os.environ["VLLM_ATTENTION_BACKEND"] = "FLASHINFER"
 
 import polars as pl
 from datasets import Dataset, load_dataset
-from huggingface_hub import login, dataset_info
+from huggingface_hub import login, dataset_info, snapshot_download
 from tqdm.auto import tqdm
 from transformers import AutoTokenizer
 from vllm import LLM, SamplingParams
@@ -106,10 +106,18 @@ def generate_summaries(
         input_dataset_id, card_type, min_likes, min_downloads
     )
 
-    # Initialize model and tokenizer
-    logger.info(f"Initializing vLLM model: {model_id}")
-    llm = LLM(model=model_id)
-    tokenizer = AutoTokenizer.from_pretrained(model_id)
+    # Download model to local directory first
+    logger.info(f"Downloading model {model_id} to local directory...")
+    local_model_path = snapshot_download(
+        repo_id=model_id,
+        resume_download=True
+    )
+    logger.info(f"Model downloaded to: {local_model_path}")
+    
+    # Initialize model and tokenizer from local path
+    logger.info(f"Initializing vLLM model from local path: {local_model_path}")
+    llm = LLM(model=local_model_path)
+    tokenizer = AutoTokenizer.from_pretrained(local_model_path)
     sampling_params = SamplingParams(
         temperature=temperature,
         max_tokens=max_tokens,