Update app.py

app.py

@@ -46,7 +46,7 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 
 # model_id = "mistralai/Mistral-7B-v0.3"
 
-model_id = "CohereForAI/c4ai-command-r-plus-4bit"
+model_id = "CohereForAI/c4ai-command-r-v01"
 
 
 tokenizer = AutoTokenizer.from_pretrained(
@@ -55,15 +55,15 @@ tokenizer = AutoTokenizer.from_pretrained(
     , token= token,)
 
 
-# accelerator = Accelerator()
+accelerator = Accelerator()
 
 model = AutoModelForCausalLM.from_pretrained(model_id, token= token, 
                                                  # torch_dtype= torch.uint8, 
-                                             # torch_dtype=torch.float16,
-                                             #  load_in_4bit=True,
+                                             torch_dtype=torch.float16,
+                                              load_in_4bit=True,
                                              # #  # torch_dtype=torch.fl,
-                                             # attn_implementation="flash_attention_2",
-                                             # low_cpu_mem_usage=True,
+                                             attn_implementation="flash_attention_2",
+                                             low_cpu_mem_usage=True,
                                              # device_map='cuda',
                                              # device_map=accelerator.device_map,
                                              
@@ -71,7 +71,7 @@ model = AutoModelForCausalLM.from_pretrained(model_id, token= token,
 
 
 # 
-# model = accelerator.prepare(model)
+model = accelerator.prepare(model)
 
 
 # device_map = infer_auto_device_map(model, max_memory={0: "79GB", "cpu":"65GB" })
@@ -102,7 +102,7 @@ def respond(
     
     messages= json_obj
 
-    input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to('cuda') #.to(accelerator.device)
+    input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to(accelerator.device)
     input_ids2 = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, return_tensors="pt") #.to('cuda')
     print(f"Converted input_ids dtype: {input_ids.dtype}")
     input_str= str(input_ids2)