File size: 3,191 Bytes
6e98acb f453fc2 6e98acb |
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 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 |
# transformers_backend.py
import time, logging
from typing import Any, Dict, AsyncIterable
from transformers import AutoTokenizer, AutoModelForCausalLM
from backends_base import ChatBackend, ImagesBackend
from config import settings
logger = logging.getLogger(__name__)
try:
import spaces
except ImportError:
spaces = None
class TransformersChatBackend(ChatBackend):
"""
Lightweight backend for Hugging Face Spaces (ZeroGPU).
Reloads model on every request using Transformers, not vLLM.
"""
async def stream(self, request: Dict[str, Any]) -> AsyncIterable[Dict[str, Any]]:
messages = request.get("messages", [])
prompt = messages[-1]["content"] if messages else "(empty)"
# Config-driven defaults
model_id = request.get("model") or settings.LlmHFModelID
temperature = float(request.get("temperature", settings.LlmTemp or 0.7))
max_tokens = int(request.get("max_tokens", settings.LlmOpenAICtxSize or 512))
rid = f"chatcmpl-transformers-{int(time.time())}"
now = int(time.time())
# Run inside ZeroGPU lease
if spaces:
@spaces.GPU(duration=300)
def run_once(prompt: str) -> str:
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(
**inputs,
max_new_tokens=max_tokens,
temperature=temperature,
do_sample=True,
)
return tokenizer.decode(outputs[0], skip_special_tokens=True)
else:
def run_once(prompt: str) -> str:
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(
**inputs,
max_new_tokens=max_tokens,
temperature=temperature,
do_sample=True,
)
return tokenizer.decode(outputs[0], skip_special_tokens=True)
try:
text = run_once(prompt)
yield {
"id": rid,
"object": "chat.completion.chunk",
"created": now,
"model": model_id,
"choices": [
{"index": 0, "delta": {"content": text}, "finish_reason": "stop"}
],
}
except Exception:
logger.exception("Transformers inference failed")
raise
class StubImagesBackend(ImagesBackend):
"""
Image generation stub — returns a transparent PNG placeholder.
"""
async def generate_b64(self, request: Dict[str, Any]) -> str:
logger.warning("Image generation not supported in Transformers backend.")
return (
"iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR4nGP4BwQACfsD/etCJH0AAAAASUVORK5CYII="
)
|