1

app.py

@@ -126,7 +126,7 @@ def download_and_merge_model(base_model_name, lora_model_name, output_dir, devic
     log.info("Loading base model...")
     model = AutoModelForCausalLM.from_pretrained(base_model_name, low_cpu_mem_usage=True, device_map="auto", force_download=True, trust_remote_code=True, torch_dtype=torch.float16)
     log.info("Loading adapter tokenizer...")
-    adapter_tokenizer = AutoTokenizer.from_pretrained(lora_model_name, trust_remote_code=True, device_map="auto", force_download=True, trust_remote_code=True, torch_dtype=torch.float16)
+    adapter_tokenizer = AutoTokenizer.from_pretrained(lora_model_name, trust_remote_code=True, device_map="auto", force_download=True, trust_remote_code=True)
     log.info("Resizing token embeddings...")
     added_tokens_decoder = adapter_tokenizer.added_tokens_decoder
     model.resize_token_embeddings(adapter_tokenizer.vocab_size + len(added_tokens_decoder))