| from transformers import AutoTokenizer, AutoModelForCausalLM | |
| import torch | |
| import os | |
| hf_token = os.environ.get("HF_TOKEN") | |
| MODEL_REPO = "Rahul-8799/project_manager_gemma3" | |
| tokenizer = AutoTokenizer.from_pretrained(MODEL_REPO, trust_remote_code=True, token=hf_token) | |
| model = AutoModelForCausalLM.from_pretrained(MODEL_REPO, device_map="auto", torch_dtype=torch.float16, token=hf_token) | |
| model.eval() | |
| def run(prompt): | |
| inputs = tokenizer(prompt, return_tensors="pt").to(model.device) | |
| with torch.no_grad(): | |
| outputs = model.generate(**inputs, max_new_tokens=512) | |
| return tokenizer.decode(outputs[0], skip_special_tokens=True) |