Update app.py

app.py

@@ -40,15 +40,15 @@ tokenizer = AutoTokenizer.from_pretrained(
     , token= token,)
 
 
-with init_empty_weights():
-    model = AutoModelForCausalLM.from_pretrained(model_id, token= token, 
+# with init_empty_weights():
+model = AutoModelForCausalLM.from_pretrained(model_id, token= token, 
                                                  # torch_dtype= torch.uint8, 
                                              torch_dtype=torch.float16,
                                               # torch_dtype=torch.fl,
                                              attn_implementation="flash_attention_2",
-                                             # low_cpu_mem_usage=True,
+                                             low_cpu_mem_usage=True,
                                              # llm_int8_enable_fp32_cpu_offload=True,
-                                             # device_map=device_map,
+                                             device_map='cuda',
                                              
                                             )
 
@@ -56,11 +56,11 @@ with init_empty_weights():
 # 
 
 
-device_map = infer_auto_device_map(model, max_memory={0: "80GB", 1: "80GB", "cpu": "65GB"})
+# device_map = infer_auto_device_map(model, max_memory={0: "80GB", 1: "80GB", "cpu": "65GB"})
 
 # Load the model with the inferred device map
 # model = load_checkpoint_and_dispatch(model, model_id, device_map=device_map, no_split_module_classes=["GPTJBlock"])
-# model.half()
+model.half()
 
 
 
@@ -77,8 +77,8 @@ def respond(
     messages = [{"role": "user", "content": "Hello, how are you?"}]
     input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to('cuda')
 ## <BOS_TOKEN><|START_OF_TURN_TOKEN|><|USER_TOKEN|>Hello, how are you?<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>
-    with autocast():
-        gen_tokens = model.generate(
+    # with autocast():
+    gen_tokens = model.generate(
     input_ids, 
     max_new_tokens=100, 
     # do_sample=True,