llmcompressor is an easy-to-use library for optimizing models for deployment with vLLM, including:

• Comprehensive set of quantization algorithms and transforms for weight, activation, KV Cache, and attention quantization
• Seamless integration with Hugging Face models and repositories
• Models saved in the compressed-tensors format, compatible with vLLM
• DDP and disk offloading support for compressing very large models

✨ Read the announcement blog here! ✨

📊 Help us improve by taking our 1-minute user survey

💬 Join us on the vLLM Community Slack and share your questions, thoughts, or ideas in:

• #sig-quantization
• #llm-compressor

Big updates have landed in LLM Compressor! To get a more in-depth look, check out the LLM Compressor overview.

Some of the exciting new features include:

• GLM-5.2 NVFP4+FP8 Quantized Checkpoint: Quantized checkpoints for GLM-5.2 have been created by the Red Hat AI team using DDP + disk offloading in under 2 hours. The full precision model requires 1.6T of VRAM, but NVFP4 quantization of MoE layers and FP8 quantization of attention layers reduces the model size by >70% while maintaining state-of-the-art accuracy recovery on GPQA.
  • RedHatAI/GLM-5.2-NVFP4-FP8
  • GLM-5.2 Quantization Example
• REAP Expert Pruning Modifier: REAP reduces the VRAM requirements to run Mixture-of-Experts models by structurally removing less-relevant experts in each layer. With relevancy proxied by a saliency metric calculated from calibration forward pass data, REAP achieves a desired expert sparsity (set by the user) while aiming to minimize the impact of the pruned experts. The modifier implementation is in modifiers/pruning/reap and can be used as a template for implementing other expert pruning algorithms. Examples and additional documentation can be found below:
  • REAP Pruning README
  • REAP Prune Qwen/Qwen3-30B-A3B-Instruct-2507 to 25% Sparsity
  • REAP Prune moonshotai/Moonlight-16B-A3B-Instruct to 25% Sparsity
• Transformers v5 Support: LLM Compressor now supports Transformers v5, including updated MoE calibration workflows. Improved MoE calibration is powered by the modeling/moe classes, which provide linearization, expert-aware context management, and architecture-specific mappings for models like Llama 4 and GraniteMoE.
• Day-0 DiffusionGemma Support: LLM Compressor now supports quantization of DiffusionGemma models on day zero. Quantized checkpoints generated by the Red Hat team are available on the HF Hub:
  • diffusiongemma-26B-A4B-it-NVFP4
  • diffusiongemma-26B-A4B-it-FP8-dynamic
• Nemotron 3 Ultra Quantized Checkpoints: Quantized FP8 and Int4 checkpoints for Nemotron 3 Ultra have been created by the Red Hat team and posted to the HF Hub using a model_free_ptq example. Consider using:
  • Nemotron-3-Ultra-550B-A55B-FP8-Dynamic
  • Nemotron-3-Ultra-550B-A55B-BF16-FP8-BLOCK
  • Nemotron-3-Ultra-550B-A55B-BF16-W4A16-G128
• DeepSeek-V4-Flash and Kimi-K2.6 Quantized Checkpoints: Quantized checkpoints for DeepSeek-V4-Flash and Kimi-K2.6 have been generated by the Red Hat team and posted to the HF hub. Consider using:
  • DeepSeek-V4-Flash-NVFP4-FP8 — 163B DeepSeek-V4-Flash quantized to NVFP4 weights with FP8 KV cache
  • Kimi-K2.6-NVFP4 — Kimi-K2.6 quantized to NVFP4 (weights and activations), targeting NVIDIA Blackwell GPUs
  • Kimi-K2.6-FP8-BLOCK — 1T parameter Kimi-K2.6 quantized to FP8 block format (weights and activations), compatible with DeepGEMM FP8 kernels
• Qwen3.6 NVFP4 Generated Checkpoint: An NVFP4 quantized checkpoint has been generated by the RedHat team and posted to the HF hub. Qwen3.6 follows the same architecture as Qwen3.5, so existing LLM Compressor examples can be used for this model by swapping out the target model string.

• Activation Quantization: W8A8 (int8 and fp8), W4AFP8, Microscale (NVFP4, MXFP4, MXFP8)
• Mixed Precision: W4A16, W8A16, MXFP8A16, MXFP4A16, NVFP4A16
• Attention and KV Cache Quantization: FP8, NVFP4

• Simple PTQ
• GPTQ
• AWQ
• SmoothQuant
• AutoRound
• Rotation-based (SpinQuant, QuIP)

Please refer to our step-by-step compression guide for detailed information about selecting quantization schemes, algorithms, and their use cases.

Additional information about LLM Compressor functionality is also available in our User Guides

pip install llmcompressor

Applying quantization with llmcompressor:

• Activation quantization to int8
• Activation quantization to fp8
• Activation quantization to MXFP8
• Activation quantization to fp4 (NVFP4)
• Activation quantization to fp4 (MXFP4)
• Activation quantization to fp4 using AutoRound
• Activation quantization to fp8 and weight quantization to int4

• Weight only quantization to fp4 (NVFP4 format)
• Weight only quantization to fp4 (MXFP4 format)
• Weight only quantization to int4 using GPTQ
• Weight only quantization to int4 using AWQ
• Weight only quantization to int4 using AutoRound

• KV Cache quantization to fp8
• KV Cache quantization to fp8 using per-head
• Attention quantization to fp8
• Attention quantization to NVFP4 with SpinQuant (experimental)

• Quantizing MoE LLMs
• Quantizing Vision-Language Models
• Quantizing Audio-Language Models

• Quantizing Models Non-uniformly

• Quantizing large models with sequential onloading
• Quantizing large models with disk offloading

• Quantizing models without a Hugging Face model definition

• Distributed data parallel quantization with GPTQ

Let's quantize Qwen3-30B-A3B with FP8 weights and activations using the Round-to-Nearest algorithm.

Note that the model can be swapped for a local or remote HF-compatible checkpoint and the recipe may be changed to target different quantization algorithms or formats.

Quantization is applied by selecting an algorithm and calling the oneshot API.

from compressed_tensors.offload import dispatch_model
from transformers import AutoModelForCausalLM, AutoTokenizer

from llmcompressor import oneshot
from llmcompressor.modifiers.quantization import QuantizationModifier

MODEL_ID = "Qwen/Qwen3-30B-A3B"

# Load model.
model = AutoModelForCausalLM.from_pretrained(MODEL_ID)
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)

# Configure the quantization algorithm and scheme.
# In this case, we:
#   * quantize the weights to FP8 using RTN with block_size 128
#   * quantize the activations dynamically to FP8 during inference
recipe = QuantizationModifier(
    targets="Linear",
    scheme="FP8_BLOCK",
    ignore=["lm_head", "re:.*mlp.gate$"],
)

# Apply quantization.
oneshot(model=model, recipe=recipe)

# Confirm generations of the quantized model look sane.
print("========== SAMPLE GENERATION ==============")
dispatch_model(model)
input_ids = tokenizer("Hello my name is", return_tensors="pt").input_ids.to(
    model.device
)
output = model.generate(input_ids, max_new_tokens=20)
print(tokenizer.decode(output[0]))
print("==========================================")

# Save to disk in compressed-tensors format.
SAVE_DIR = MODEL_ID.split("/")[1] + "-FP8-BLOCK"
model.save_pretrained(SAVE_DIR)
tokenizer.save_pretrained(SAVE_DIR)

The checkpoints created by llmcompressor can be loaded and run in vllm:

Install:

pip install vllm

Run:

from vllm import LLM
model = LLM("Qwen/Qwen3-30B-A3B-FP8-BLOCK")
output = model.generate("My name is")

• If you have any questions or requests open an issue and we will add an example or documentation.
• We appreciate contributions to the code, examples, integrations, and documentation as well as bug reports and feature requests! Learn how here.

If you find LLM Compressor useful in your research or projects, please consider citing it:

@software{llmcompressor2024,
    title={{LLM Compressor}},
    author={Red Hat AI and vLLM Project},
    year={2024},
    month={8},
    url={https://github.com/vllm-project/llm-compressor},
}

!!! warning Sparse compression (24 sparsity) is no longer supported by LLM Compressor due to lack of hardware support and usage