Update handler.py

handler.py

@@ -1,15 +1,22 @@
-from typing import Dict, List, Any
-import torch
-from modelscope import AutoTokenizer
-from modelscope import AutoModelForCausalLM
+# Code to inference Hermes with HF Transformers
+# Requires pytorch, transformers, bitsandbytes, sentencepiece, protobuf, and flash-attn packages
 
-device = "cuda" # the device to load the model onto
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from transformers import LlamaTokenizer, MixtralForCausalLM
+import bitsandbytes, flash_attn
 
 class EndpointHandler:
     def __init__(self, path=""):
-        self.model = AutoModelForCausalLM.from_pretrained(path, torch_dtype=torch.bfloat16, device_map=device)
-        self.tokenizer = AutoTokenizer.from_pretrained(path)
-
+        self.tokenizer = LlamaTokenizer.from_pretrained(path, trust_remote_code=True)
+        self.model = MixtralForCausalLM.from_pretrained(
+            path,
+            torch_dtype=torch.float16,
+            device_map="auto",
+            load_in_8bit=False,
+            load_in_4bit=True,
+            use_flash_attention_2=True
+            )
     def __call__(self, data: Any) -> List[List[Dict[str, float]]]:
         sys_prompt=data["prompt"]
         list=data["inputs"]
@@ -23,12 +30,19 @@ class EndpointHandler:
                 prompt+=f"<|im_start|>user\n{content}<|im_end|>\n"
         prompt+="<|im_start|>assistant\n"
 
+        #for chat in prompts:
+            #print(chat)
+        input_ids = self.tokenizer(prompt, return_tensors="pt").input_ids.to("cuda")
+        generated_ids = self.model.generate(input_ids, max_new_tokens=750, temperature=0.8, repetition_penalty=1.1, do_sample=True, eos_token_id=self.tokenizer.eos_token_id)
+        response = self.tokenizer.decode(generated_ids[0][input_ids.shape[-1]:], skip_special_tokens=True, clean_up_tokenization_space=True)
+        return response
+
+        """
         encodeds = self.tokenizer.encode(prompt, return_tensors="pt")
         model_inputs = encodeds.to(device)
         self.model.to(device)
         generated_ids = self.model.generate(model_inputs, max_new_tokens=1000, do_sample=True)
         decoded = self.tokenizer.decode(generated_ids[0])
         return decoded
-
-
+        """