Update main.py

main.py

@@ -1,21 +1,19 @@
 from fastapi import FastAPI
 from pydantic import BaseModel
-from llama_cpp import Llama
 import uvicorn
 import prompt_style
 import os
-from huggingface_hub import hf_hub_download
+from huggingface_hub import InferenceClient
 
 
-model_id = "failspy/Meta-Llama-3-8B-Instruct-abliterated-v3-GGUF"
-model_path = hf_hub_download(repo_id=model_id, filename="Meta-Llama-3-8B-Instruct-abliterated-v3_q6.gguf", token=os.environ['HF_TOKEN'])
-model = Llama(model_path=model_path, n_gpu_layers=-1, n_ctx=4096, verbose=False)
+model_id = "failspy/Meta-Llama-3-8B-Instruct-abliterated-v3"
+client = InferenceClient(model_id, token=os.environ['HF_TOKEN'])
 
 class Item(BaseModel):
     prompt: str
     history: list
     system_prompt: str
-    temperature: float = 0.6
+    temperature: float = 0.8
     max_new_tokens: int = 1024
     top_p: float = 0.95
     repetition_penalty: float = 1.0
@@ -34,14 +32,35 @@ def format_prompt(item: Item):
     return messages
 
 def generate(item: Item):
+    temperature = float(item.temperature)
+    if temperature < 1e-2:
+        temperature = 1e-2
+    top_p = float(item.top_p)
+
+    generate_kwargs = dict(
+        temperature=temperature,
+        max_new_tokens=item.max_new_tokens,
+        top_p=top_p,
+        repetition_penalty=item.repetition_penalty,
+        do_sample=True,
+        seed=item.seed,
+    )
+    
     formatted_prompt = format_prompt(item)
-    output = model.create_chat_completion(messages=formatted_prompt, seed=item.seed, 
-                                          temperature=item.temperature,
-                                          max_tokens=item.max_new_tokens)
+    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
+    output = ""
+
+    for response in stream:
+        output += response.token.text
+    return output
+    
+    # output = model.create_chat_completion(messages=formatted_prompt, seed=item.seed, 
+    #                                       temperature=item.temperature,
+    #                                       max_tokens=item.max_new_tokens)
 
 
-    out = output['choices'][0]['message']['content']
-    return out
+    # out = output['choices'][0]['message']['content']
+    # return out
 
 @app.post("/generate/")
 async def generate_text(item: Item):