Spaces:
Running
Running
Commit History
-
Create requirements.txt Running
Verified signature0517216 -
Create app.py
Verified signature99441f8 -
Create requirements.txt
Verified signature362b919 -
Create app.py
Verified signature6658332 -
Create app.py
Verified signature# app.py - Gradio correct binding and port import os import gradio as gr from transformers import AutoTokenizer, AutoModelForCausalLM import torch MODEL_ID = os.environ.get("MODEL_ID", "bigcode/starcoder") DEVICE = "cuda" if torch.cuda.is_available() else "cpu" # Lazy load to avoid failing at import time def load_model(): tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True) model = AutoModelForCausalLM.from_pretrained(MODEL_ID, trust_remote_code=True) model.to(DEVICE) model.eval() return tokenizer, model tokenizer, model = load_model() def generate(prompt, max_new_tokens=64, temperature=0.2): inputs = tokenizer(prompt, return_tensors="pt", truncation=True).to(DEVICE) with torch.no_grad(): out = model.generate(**inputs, max_new_tokens=int(max_new_tokens), temperature=float(temperature)) text = tokenizer.decode(out[0], skip_special_tokens=True) if text.startswith(prompt): text = text[len(prompt):].lstrip() …6d38079 -
Create app.py
Verified signature# app.py - FastAPI proxy to Hugging Face Inference API import os from fastapi import FastAPI from pydantic import BaseModel import httpx app = FastAPI() HF_TOKEN = os.environ.get("HF_INFERENCE_TOKEN", "") HF_MODEL = os.environ.get("MODEL_ID", "bigcode/starcoder") HF_URL = f"https://api-inference.huggingface.co/models/{HF_MODEL}" HEADERS = {"Authorization": f"Bearer {HF_TOKEN}"} if HF_TOKEN else {} class GenRequest(BaseModel): prompt: str max_new_tokens: int = 128 temperature: float = 0.2 @app.post("/generate") async def generate(req: GenRequest): payload = { "inputs": req.prompt, "parameters": { "max_new_tokens": req.max_new_tokens, "temperature": req.temperature }, "options": {"wait_for_model": True} } async with httpx.AsyncClient(timeout=120) as client: r = await client.post(HF_URL, headers={**HEADERS, "Content-Type":"application/json"}, json=payload) if r.status_code != 200: retur…2b3e869 -
Create requirements.txt
Verified signature# For Gradio local option gradio>=3.30 transformers>=4.30.0 torch>=2.0.0 sentencepiece # For FastAPI proxy option fastapi uvicorn[standard] httpxffa94da -
Create README.md
Verified signature3a9ac4c -
Create requirements.txt
Verified signature# For FastAPI proxy option fastapi uvicorn[standard] httpx # For Gradio local option gradio>=3.30 transformers>=4.30.0 torch>=2.0.0 sentencepiece0570c0c -
Create README.md
Verified signature3949e9e -
Create README.md
Verified signature24209a5
-
Update app.py
Verified signature8c71198 -
Update requirements.txt
Verified signature719b62a -
Update app.py
Verified signaturefbdf71e -
Update app.py
Verified signaturea990a02 -
Update app.py
Verified signature6f6d64e -
Create README.md
Verified signature54be2f7 -
Create README.md
Verified signature7a40d5e -
Create README.md
Verified signature8a87694 -
Create README.md
Verified signature42b261e -
Create app.py
Verified signature193b261 -
Create requirements.txt
Verified signature89f1f25 -
initial commit
Verified signature50c18d4