--- tags: - fp4 - vllm language: - en - de - fr - it - pt - hi - es - th pipeline_tag: text-generation license: llama3.3 base_model: meta-llama/Meta-Llama-3.3-70B-Instruct --- # Meta-Llama-3.3-70B-Instruct-NVFP4 ## Model Overview - **Model Architecture:** Meta-Llama-3.3 - **Input:** Text - **Output:** Text - **Model Optimizations:** - **Weight quantization:** FP4 - **Activation quantization:** FP4 - **Intended Use Cases:** Intended for commercial and research use in multiple languages. Similarly to [Meta-Llama-3.3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.3-8B-Instruct), this models is intended for assistant-like chat. - **Out-of-scope:** Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in languages other than English. - **Release Date:** 6/25/2025 - **Version:** 1.0 - **License(s):** [llama3.3](https://huggingface.co/meta-llama/Meta-Llama-3.3-8B/blob/main/LICENSE) - **Model Developers:** RedHatAI This model is a quantized version of [Meta-Llama-3.3-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.3-70B-Instruct). It was evaluated on a several tasks to assess the its quality in comparison to the unquatized model. ### Model Optimizations This model was obtained by quantizing the weights and activations of [Meta-Llama-3.3-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.3-70B-Instruct) to FP4 data type, ready for inference with vLLM>=0.9.1 This optimization reduces the number of bits per parameter from 16 to 4, reducing the disk size and GPU memory requirements by approximately 75%. Only the weights and activations of the linear operators within transformers blocks are quantized using [LLM Compressor](https://github.com/vllm-project/llm-compressor). ## Deployment ### Use with vLLM This model can be deployed efficiently using the [vLLM](https://docs.vllm.ai/en/latest/) backend, as shown in the example below. ```python from vllm import LLM, SamplingParams from transformers import AutoTokenizer model_id = "RedHatAI/Meta-Llama-3.3-70B-Instruct-NVFP4" number_gpus = 2 sampling_params = SamplingParams(temperature=0.6, top_p=0.9, max_tokens=256) tokenizer = AutoTokenizer.from_pretrained(model_id) messages = [ {"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"}, {"role": "user", "content": "Who are you?"}, ] prompts = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False) llm = LLM(model=model_id, tensor_parallel_size=number_gpus) outputs = llm.generate(prompts, sampling_params) generated_text = outputs[0].outputs[0].text print(generated_text) ``` vLLM aslo supports OpenAI-compatible serving. See the [documentation](https://docs.vllm.ai/en/latest/) for more details. ## Creation This model was created by applying [LLM Compressor with calibration samples from UltraChat](https://github.com/vllm-project/llm-compressor/blob/main/examples/quantization_w4a4_fp4/llama3_example.py), as presented in the code snipet below. ```python from datasets import load_dataset from transformers import AutoModelForCausalLM, AutoTokenizer from llmcompressor import oneshot from llmcompressor.modifiers.quantization import QuantizationModifier from llmcompressor.utils import dispatch_for_generation MODEL_ID = "meta-llama/Meta-Llama-3-8B-Instruct" # Load model. model = AutoModelForCausalLM.from_pretrained(MODEL_ID, torch_dtype="auto") tokenizer = AutoTokenizer.from_pretrained(MODEL_ID) DATASET_ID = "HuggingFaceH4/ultrachat_200k" DATASET_SPLIT = "train_sft" # Select number of samples. 512 samples is a good place to start. # Increasing the number of samples can improve accuracy. NUM_CALIBRATION_SAMPLES = 512 MAX_SEQUENCE_LENGTH = 2048 # Load dataset and preprocess. ds = load_dataset(DATASET_ID, split=f"{DATASET_SPLIT}[:{NUM_CALIBRATION_SAMPLES}]") ds = ds.shuffle(seed=42) def preprocess(example): return { "text": tokenizer.apply_chat_template( example["messages"], tokenize=False, ) } ds = ds.map(preprocess) # Tokenize inputs. def tokenize(sample): return tokenizer( sample["text"], padding=False, max_length=MAX_SEQUENCE_LENGTH, truncation=True, add_special_tokens=False, ) ds = ds.map(tokenize, remove_columns=ds.column_names) # Configure the quantization algorithm and scheme. # In this case, we: # * quantize the weights to fp4 with per group 16 via ptq # * calibrate a global_scale for activations, which will be used to # quantize activations to fp4 on the fly recipe = QuantizationModifier(targets="Linear", scheme="NVFP4", ignore=["lm_head"]) # Save to disk in compressed-tensors format. SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-NVFP4" # Apply quantization. oneshot( model=model, dataset=ds, recipe=recipe, max_seq_length=MAX_SEQUENCE_LENGTH, num_calibration_samples=NUM_CALIBRATION_SAMPLES, output_dir=SAVE_DIR, ) print("\n\n") print("========== SAMPLE GENERATION ==============") dispatch_for_generation(model) input_ids = tokenizer("Hello my name is", return_tensors="pt").input_ids.to("cuda") output = model.generate(input_ids, max_new_tokens=100) print(tokenizer.decode(output[0])) print("==========================================\n\n") model.save_pretrained(SAVE_DIR, save_compressed=True) tokenizer.save_pretrained(SAVE_DIR) ``` ## Evaluation This model was evaluated on the well-known OpenLLM v1, OpenLLM v2, HumanEval, and HumanEval_64 benchmarks. All evaluations were conducted using [lm-evaluation-harness](https://github.com/neuralmagic/lm-evaluation-harness). ### Accuracy
| Category | Metric | Meta-Llama-3.3-70B-Instruct | RedHatAI/Llama-3.3-70B-Instruct-NVFP4 (this model) | Recovery |
|---|---|---|---|---|
| OpenLLM V1 | mmlu_llama | 83.40 | 81.28 | 97.46 |
| mmlu_cot_llama (0-shot) | 86.42 | 84.77 | 98.09 | |
| arc_challenge_llama (0-shot) | 93.39 | 92.62 | 99.18 | |
| gsm8k_llama (8-shot, strict-match) | 85.22 | 77.10 | 90.47 | |
| hellaswag (10-shot) | 70.23 | 75.40 | 107.36 | |
| winogrande (5-shot) | 73.48 | 72.53 | 98.71 | |
| truthfulQA (0-shot, mc2) | 66.26 | 66.26 | 100.00 | |
| Average | 79.77 | 78.57 | 98.49 | |
| OpenLLM V2 | MMLU-Pro (5-shot) | 53.48 | 50.26 | 93.98 |
| IFEval (0-shot) | 92.45 | 91.49 | 98.96 | |
| BBH (3-shot) | 69.05 | 66.31 | 96.03 | |
| Math-|v|-5 (4-shot) | 47.96 | 43.81 | 91.35 | |
| GPQA (0-shot) | 31.63 | 32.21 | 101.83 | |
| MuSR (0-shot) | 44.18 | 44.18 | 100.00 | |
| Average | 57.05 | 55.60 | 97.45 | |
| Coding | HumanEval_64 pass@2 | 88.69 | 84.77 | 95.58 |