Spaces:
Sleeping
Sleeping
File size: 1,995 Bytes
f84e083 caa64e7 f84e083 215f4a9 f84e083 d0c61b6 215f4a9 f84e083 d0c61b6 215f4a9 d0c61b6 215f4a9 d0c61b6 f84e083 215f4a9 f84e083 d0c61b6 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 |
from fastapi import FastAPI
from fastapi.responses import StreamingResponse
from pydantic import BaseModel
from huggingface_hub import InferenceClient
import uvicorn
import json # Make sure to import json
app = FastAPI()
client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
class Item(BaseModel):
prompt: str
history: list
system_prompt: str
temperature: float = 0.0
max_new_tokens: int = 1048
top_p: float = 0.15
repetition_penalty: float = 1.0
def format_prompt(message, history):
prompt = "<s>"
for user_prompt, bot_response in history:
prompt += f"[INST] {user_prompt} [/INST]"
prompt += f" {bot_response}</s> "
prompt += f"[INST] {message} [/INST]"
return prompt
import json # Import the JSON module
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=42,
)
formatted_prompt = format_prompt(f"{item.system_prompt}, {item.prompt}", item.history)
stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
# Convert stream to a list to check if it's the last element
responses = list(stream)
for i, response in enumerate(responses):
# Prepare the chunk as a JSON object
chunk = {
"text": response.token.text,
"complete": i == len(responses) - 1 # True if this is the last chunk
}
# Yield the JSON-encoded string with a newline to separate chunks
yield json.dumps(chunk).encode("utf-8") + b"\n"
@app.post("/generate/")
async def generate_text(item: Item):
return StreamingResponse(generate(item), media_type="application/x-ndjson")
|