Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Transformers-compatible library for applying various compression algorithms to LLMs for optimized deployment with vLLM
| Date | Stars |
|---|---|
| 2026-07-24 | 3578 |
| 2026-07-25 | 3581 |
| 2026-07-28 | 3581 |
| 2026-07-30 | 3581 |
| 2026-07-31 | 3604 |
| 2026-08-10 | 3648 |
| 2026-08-18 | 3694 |
| 2026-08-19 | 3701 |
| 2026-08-20 | 3706 |
| 2026-08-21 | 3707 |
| 2026-08-22 | 3710 |
| 2026-08-23 | 3711 |
| 2026-08-24 | 3717 |
| 2026-08-25 | 3725 |
| 2026-08-26 | 3727 |
| 2026-08-27 | 3731 |
| 2026-08-28 | 3735 |
| 2026-08-29 | 3736 |
| 2026-08-30 | 3739 |
| 2026-08-31 | 3744 |
| 2026-09-01 | 3748 |
| 2026-09-02 | 3751 |
| 2026-09-03 | 3756 |
| 2026-09-04 | 3759 |
| 2026-09-05 | 3760 |
| 2026-09-06 | 3762 |
| 2026-09-07 | 3764 |
| 2026-09-08 | 3766 |
| 2026-09-09 | 3769 |
| 2026-09-10 | 3771 |
| 2026-09-11 | 3775 |
| 2026-09-12 | 3776 |
| 2026-09-13 | 3778 |
| 2026-09-14 | 3780 |
| 2026-09-15 | 3785 |
| 2026-09-16 | 3788 |
| 2026-09-17 | 3789 |
| 2026-09-18 | 3794 |
| 2026-09-19 | 3797 |
| 2026-09-20 | 3801 |
Today
+4 stars today
This week
+23 stars this week
This month
+94 stars this month
Momentum
54.0
growth rate 0.61%/day
<div align="center">
<picture>
<source
media="(prefers-color-scheme: dark)"
srcset="docs/assets/llmcompressor-icon-name-dark.png"
>
<source
media="(prefers-color-scheme: light)"
srcset="docs/assets/llmcompressor-icon-name-light.png"
>
<img
src="docs/assets/llmcompressor-icon-name-light.png"
alt="LLM Compressor"
width="420"
>
</picture>
[](https://docs.vllm.ai/projects/llm-compressor/en/latest/) [](https://pypi.org/project/llmcompressor/)
</div>
`llmcompressor` is the fast, efficient, and 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 with hardware efficiency
**✨ Read the announcement blog [here](https://neuralmagic.com/blog/llm-compressor-is-here-faster-inference-with-vllm/)! ✨**
<p align="center">
<img alt="LLM Compressor Flow" src="https://github.com/user-attachments/assets/adf07594-6487-48ae-af62-d9555046d51b" width="80%" />
</p>
---
📊 Help us improve by taking our [1-minute user survey](https://red.ht/llm-compressor-user-survey)
💬 Join us on the [vLLM Community Slack](https://inviter.co/vllm-slack) and share your questions, thoughts, or ideas in:
- `#sig-quantization`
- `#llm-compressor`
---
## 🚀 What's New!
Big updates have landed in LLM Compressor! To get a more in-depth look, check out the [LLM Compressor overview](https://docs.google.com/presentation/d/1WNkYBKv_CsrYs69lb7bJKjh2dWt8U1HXUw7Gr4Wn3gE/edit?usp=sharing).
Since the v0.13.0 release, a number of meaningful improvements have landed:
* **Batched GPTQ quantization with a new Triton GPTQ kernel**: GPTQ now ships a Triton-based quantization kernel (~15x faster than the previous eager path) together with the ability to batch layers that share the same shape (up to ~1.67x per batch, roughly ~30x end-to-end on MoE workloads). Activation-order (act-order) calibration is supported, hessian offloading has been removed, and the remaining eager path was also sped up by 1.5-2x on its own.
* **Expanded MSE and iMatrix observers for FP4, with a new `fouroversix` default**: The MSE observer and the iMatrix observer gained a grid-search expansion factor that makes the search a strict superset of *fouroversix* (which chooses between the full `absmax` and `absmax * 1.5` scales for FP4 blocks). A new `fouroversix` observer built on this expanded search is now the **default observer for NVFP4 quantization**, and it outperforms GPTQ for NVFP4 on average across our internal perplexity benchmarks.
* **Triton grid-search kernel for the MSE observer**: A Triton kernel now performs the MSE observer's scale grid search using buffered per-qparam patience and adaptive 512-value tiling. It reaches bitwise parity with the eager path when configured for full evaluation, supports INT, FP4, FP8, and FP16/BF16 (with E8M0 scales), and defaults `triton_error_buffer` to 100% for FP4 and 30% otherwise.
### Model highlights
The Red Hat AI team has been using LLM Compressor to produce a fresh batch of production-ready quantized checkpoints:
* **GLM-5.3 MXFP4**: An MXFP4 quantized checkpoint for [GLM-5.3](https://huggingface.co/zai-org/GLM-5.3). The linear operators within the transformer blocks are quantized to MXFP4, while the MoE router, embeddings, DSA indexer, and output head are kept in their original precision to maintain accuracy recovery.
- [RedHatAI/GLM-5.3-MXFP4](https://huggingface.co/RedHatAI/GLM-5.3-MXFP4)
- [GLM-5.3 MXFP4 Example](examples/model_free_ptq/glm_5_3_mxfp4.py)
* **GLM-5.3-Flash NVFP4**: An NVFP4 quantized checkpoint for [GLM-5.3-Flash](httpsExcerpt of 14,984 characters
Read on GitHub
Kyle Sayers · @RedHatOfficial · United States
326
Mark Kurtz · Red Hat · United States
297
293
276
209
117
111
Tuan Nguyen · United States
91
Brian Dellabetta · Red Hat · Israel
89
Kevin Escobar Rodriguez · Block
81
Eldar Kurtić
80
64
Michael Goin · @vllm-project @redhatofficial · United States
63
52
Neural Magic · United States
48
43
35
Robert Shaw · @RedHatOfficial · United States
26
25
24
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:633302a7c13d503d, topic:quantization, readme:quantization, readme:quantized