Model Card for Llama-3.2-3B-Instruct-lora-nvidia-blackwell-vllm

This model is a fine-tuned version of meta-llama/Llama-3.2-3B-Instruct. It has been trained using TRL. It was trained with Low-Rank Adaptation (LoRA) and a supervised fine-tuning method within the Parameter-Efficient Fine-Tuning (PEFT) framework. This LoRA was optimized for use with vLLM.

Quick start

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
device = "cuda:0" if torch.cuda.is_available() else "cpu"
HF_TOKEN = "<YOUR_HF_TOKEN_GOES_HERE>"
base_model_id = "meta-llama/Llama-3.2-3B-Instruct"
peft_model_id = "garystafford/Llama-3.2-3B-Instruct-lora-nvidia-blackwell"
base_model = AutoModelForCausalLM.from_pretrained(base_model_id, token=HF_TOKEN)
model = PeftModel.from_pretrained(base_model, peft_model_id).to(device)
tokenizer = AutoTokenizer.from_pretrained(base_model_id, token=HF_TOKEN)
test_prompt = [
  {
      "role": "user",
      "content": "Describe the NVIDIA Blackwell architecture.",
  }
]
inputs = tokenizer.apply_chat_template(
    test_prompt,
    tokenize=True,
    add_generation_prompt=True,
    return_tensors="pt",
).to(device)
output = model.generate(
    input_ids=inputs,
    max_new_tokens=128,
    temperature=0.1,
    pad_token_id=tokenizer.pad_token_id,
).to(device)
output = tokenizer.decode(output[0], skip_special_tokens=True)
output = output.split('assistant\n\n')[1].strip()
print(output)

Training procedure

This model was trained with SFT.

Framework versions

• TRL: 0.17.0
• Transformers: 4.51.3
• Pytorch: 2.7.0+cu128
• Datasets: 3.6.0
• Tokenizers: 0.21.1

Citations

Cite TRL as:

@misc{vonwerra2022trl,
    title        = {{TRL: Transformer Reinforcement Learning}},
    author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
    year         = 2020,
    journal      = {GitHub repository},
    publisher    = {GitHub},
    howpublished = {\url{https://github.com/huggingface/trl}}
}