51 lines
1.8 KiB
Python
51 lines
1.8 KiB
Python
|
from typing import Dict, List, Any
|
||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
|
import torch
|
||
|
|
|
||
|
|
class EndpointHandler:
|
||
|
|
def __init__(self, path=""):
|
||
|
|
self.tokenizer = AutoTokenizer.from_pretrained(path)
|
||
|
|
self.model = AutoModelForCausalLM.from_pretrained(
|
||
|
|
path,
|
||
|
|
torch_dtype=torch.bfloat16,
|
||
|
|
device_map="auto"
|
||
|
|
)
|
||
|
|
|
||
|
|
def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
|
||
|
|
inputs = data.pop("inputs", data)
|
||
|
|
parameters = data.pop("parameters", {})
|
||
|
|
|
||
|
|
# Handle chat format
|
||
|
|
if isinstance(inputs, list) and len(inputs) > 0 and isinstance(inputs[0], dict):
|
||
|
|
text = self.tokenizer.apply_chat_template(
|
||
|
|
inputs,
|
||
|
|
tokenize=False,
|
||
|
|
add_generation_prompt=True
|
||
|
|
)
|
||
|
|
else:
|
||
|
|
text = inputs
|
||
|
|
|
||
|
|
encoded = self.tokenizer(text, return_tensors="pt").to(self.model.device)
|
||
|
|
|
||
|
|
# Default generation parameters
|
||
|
|
gen_kwargs = {
|
||
|
|
"max_new_tokens": parameters.get("max_new_tokens", 512),
|
||
|
|
"temperature": parameters.get("temperature", 0.7),
|
||
|
|
"top_p": parameters.get("top_p", 0.9),
|
||
|
|
"do_sample": parameters.get("do_sample", True),
|
||
|
|
"pad_token_id": self.tokenizer.eos_token_id,
|
||
|
|
}
|
||
|
|
|
||
|
|
with torch.no_grad():
|
||
|
|
outputs = self.model.generate(**encoded, **gen_kwargs)
|
||
|
|
|
||
|
|
decoded = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
|
||
|
|
|
||
|
|
# Return only the generated part (remove input)
|
||
|
|
if isinstance(inputs, str):
|
||
|
|
generated = decoded[len(inputs):].strip()
|
||
|
|
else:
|
||
|
|
generated = decoded
|
||
|
|
|
||
|
|
return [{"generated_text": generated}]
|