GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
ahammad115566
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
ahammad115566
Model Tree
Qwen/Qwen3-8B
Input Modalities
Output Modalities
Supported Functionality
bash
pip install torch transformers bitsandbytes accelerate
python
import torchfrom transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfigfrom peft import PeftModelMODEL_NAME = "ahammad115566/smeft-qwen-7b"RESPONSE_PREFIX = "\n### Response:\n"bnb_config = BitsAndBytesConfig(load_in_4bit=True,bnb_4bit_quant_type="nf4",bnb_4bit_compute_dtype=torch.bfloat16,bnb_4bit_use_double_quant=True,)model = AutoModelForCausalLM.from_pretrained(MODEL_NAME,quantization_config=bnb_config,device_map={"": 0},trust_remote_code=True,)model.eval()tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, trust_remote_code=True)if tokenizer.pad_token is None:tokenizer.pad_token = tokenizer.eos_token# ================= Inference ==============def build_prompt(instruction: str, context: str = "") -> str:parts = [ f"\n### Instruction:\n{instruction}"]parts.append(RESPONSE_PREFIX)return "\n".join(parts)@torch.inference_mode()def ask(instruction: str, context: str = "") -> str:inputs = tokenizer(build_prompt(instruction, context),return_tensors="pt",add_special_tokens=True,).to("cuda:0")output_ids = model.generate(**inputs,max_new_tokens=512,do_sample=False,repetition_penalty=1.1,eos_token_id=tokenizer.eos_token_id,pad_token_id=tokenizer.pad_token_id,)new_tokens = output_ids[0][inputs["input_ids"].shape[-1]:]return tokenizer.decode(new_tokens, skip_special_tokens=True).strip()"""
#Training Details
| Parameter | Value |
|---|---|
| Base Model | Qwen3-8B |
| Fine-tuning Method | LoRA (merged) |
| Inference Quantization | 4-bit NF4 (bitsandbytes) |
| Domain | Standard Model Effective Field Theory |
| Training Corpus | Curated SMEFT and HEP preprints |
| Task Format | Instruction-following scientific QA |
This model may occasionally:
Hallucinate operator identities
Domain-locked by design. The model is not suitable for general-purpose tasks.
2,500 training examples. Coverage of the SMEFT operator space may be uneven; rare operators or non-Warsaw bases may be answered less reliably.
Omitting the system prompt will cause the model to behave like the base Qwen3-8B.
Ahmed Hammad
Assistant professor
Center of AI and natural science, KIAS, Seoul.
Veronica Sanz
Professor of Theoretical Physics
University of Valencia
A technical paper describing the dataset construction and fine-tuning procedure is forthcoming.
Please cite the model as:
bibtex
@misc{hammad2026smeftqwen,author = {Ahmed Hammad and Veronica Sanz},title = {SMEFT-Qwen-7B: A Domain-Adapted Language Model for Standard Model Effective Field Theory},year = {2026},howpublished = {\url{https://huggingface.co/ahammad115566/smeft-qwen-7b}}}