AesSedai/MiniMax-M3-GGUF overview
This repo contains specialized MoE quants for MiniMaxAI/MiniMax M3. The idea being that given the huge size of the FFN tensors compared to the rest of the tens…
Runs locally from ~7.9 MB disk (4 GB VRAM class GPUs with llama.cpp / guIDE).
Repository Files & Downloads
| File | Type | Quantization | Size | Link |
|---|---|---|---|---|
| IQ2_S/MiniMax-M3-IQ2_S-00001-of-00004.gguf | GGUF | IQ2_S | 7.9 MB | Download |
| IQ2_S/MiniMax-M3-IQ2_S-00002-of-00004.gguf | GGUF | IQ2_S | 46.51 GB | Download |
| IQ2_S/MiniMax-M3-IQ2_S-00003-of-00004.gguf | GGUF | IQ2_S | 46.09 GB | Download |
| IQ2_S/MiniMax-M3-IQ2_S-00004-of-00004.gguf | GGUF | IQ2_S | 41.40 GB | Download |
| IQ3_S/MiniMax-M3-IQ3_S-00001-of-00005.gguf | GGUF | IQ3_S | 7.9 MB | Download |
| IQ3_S/MiniMax-M3-IQ3_S-00002-of-00005.gguf | GGUF | IQ3_S | 46.09 GB | Download |
| IQ3_S/MiniMax-M3-IQ3_S-00003-of-00005.gguf | GGUF | IQ3_S | 45.91 GB | Download |
| IQ3_S/MiniMax-M3-IQ3_S-00004-of-00005.gguf | GGUF | IQ3_S | 45.91 GB | Download |
| IQ3_S/MiniMax-M3-IQ3_S-00005-of-00005.gguf | GGUF | IQ3_S | 10.12 GB | Download |
| IQ4_XS/MiniMax-M3-IQ4_XS-00001-of-00006.gguf | GGUF | IQ4_XS | 7.9 MB | Download |
| IQ4_XS/MiniMax-M3-IQ4_XS-00002-of-00006.gguf | GGUF | IQ4_XS | 46.46 GB | Download |
| IQ4_XS/MiniMax-M3-IQ4_XS-00003-of-00006.gguf | GGUF | IQ4_XS | 45.80 GB | Download |
| IQ4_XS/MiniMax-M3-IQ4_XS-00004-of-00006.gguf | GGUF | IQ4_XS | 45.80 GB | Download |
| IQ4_XS/MiniMax-M3-IQ4_XS-00005-of-00006.gguf | GGUF | IQ4_XS | 45.80 GB | Download |
| IQ4_XS/MiniMax-M3-IQ4_XS-00006-of-00006.gguf | GGUF | IQ4_XS | 5.25 GB | Download |
| Q4_K_M/MiniMax-M3-Q4_K_M-00001-of-00007.gguf | GGUF | Q4_K_M | 7.9 MB | Download |
| Q4_K_M/MiniMax-M3-Q4_K_M-00002-of-00007.gguf | GGUF | Q4_K_M | 46.10 GB | Download |
| Q4_K_M/MiniMax-M3-Q4_K_M-00003-of-00007.gguf | GGUF | Q4_K_M | 45.43 GB | Download |
| Q4_K_M/MiniMax-M3-Q4_K_M-00004-of-00007.gguf | GGUF | Q4_K_M | 45.55 GB | Download |
| Q4_K_M/MiniMax-M3-Q4_K_M-00005-of-00007.gguf | GGUF | Q4_K_M | 45.27 GB | Download |
| Q4_K_M/MiniMax-M3-Q4_K_M-00006-of-00007.gguf | GGUF | Q4_K_M | 45.43 GB | Download |
| Q4_K_M/MiniMax-M3-Q4_K_M-00007-of-00007.gguf | GGUF | Q4_K_M | 18.33 GB | Download |
| Q5_K_M/MiniMax-M3-Q5_K_M-00001-of-00008.gguf | GGUF | Q5_K_M | 7.9 MB | Download |
| Q5_K_M/MiniMax-M3-Q5_K_M-00002-of-00008.gguf | GGUF | Q5_K_M | 46.34 GB | Download |
| Q5_K_M/MiniMax-M3-Q5_K_M-00003-of-00008.gguf | GGUF | Q5_K_M | 46.07 GB | Download |
| Q5_K_M/MiniMax-M3-Q5_K_M-00004-of-00008.gguf | GGUF | Q5_K_M | 46.07 GB | Download |
| Q5_K_M/MiniMax-M3-Q5_K_M-00005-of-00008.gguf | GGUF | Q5_K_M | 46.07 GB | Download |
| Q5_K_M/MiniMax-M3-Q5_K_M-00006-of-00008.gguf | GGUF | Q5_K_M | 46.07 GB | Download |
| Q5_K_M/MiniMax-M3-Q5_K_M-00007-of-00008.gguf | GGUF | Q5_K_M | 46.07 GB | Download |
| Q5_K_M/MiniMax-M3-Q5_K_M-00008-of-00008.gguf | GGUF | Q5_K_M | 18.51 GB | Download |
| mmproj-MiniMax-M3-BF16.gguf | GGUF | BF16 | 1.62 GB | Download |
| mmproj-MiniMax-M3-F16.gguf | GGUF | F16 | 1.61 GB | Download |
| mmproj-MiniMax-M3-F32.gguf | GGUF | F32 | 3.22 GB | Download |
| mmproj-MiniMax-M3-Q8_0.gguf | GGUF | Q8_0 | 882.9 MB | Download |
Model Details
Model README
---
base_model:
- MiniMaxAI/MiniMax-M3
---
This repo contains specialized MoE-quants for MiniMaxAI/MiniMax-M3. The idea being that given the huge size of the FFN tensors compared to the rest of the tensors in the model, it should be possible to achieve a better quality while keeping the overall size of the entire model smaller compared to a similar naive quantization. To that end, the quantization type default is kept in high quality and the FFN UP + FFN GATE tensors are quanted down along with the FFN DOWN tensors.
Additionally, I've provided a J-Space lens in the lens/ folder, trained on ~128 text prompts.
| Quant | Size | Mixture | PPL | 1-(Mean PPL(Q)/PPL(base)) | KLD |
| :----- | :-------------------- | :------------------------------- | :------------------ | :------------------------ | :------------------ |
| Q8_0 | 421.84 GiB (8.50 BPW) | Q8_0 | 5.202737 ± 0.034827 | +0.3357% | 0.030136 ± 0.000884 |
| Q5_K_M | 295.20 GiB (5.95 BPW) | Q8_0 / Q5_K / Q5_K / Q6_K | 5.218907 ± 0.034969 | +0.6475% | 0.041986 ± 0.001016 |
| Q4_K_M | 246.11 GiB (4.96 BPW) | Q8_0 / Q4_K / Q4_K / Q5_K | 5.282848 ± 0.035408 | +1.8807% | 0.068863 ± 0.000982 |
| IQ4_XS | 189.12 GiB (3.81 BPW) | Q6_K / IQ3_S / IQ3_S / IQ4_XS | 5.657391 ± 0.038606 | +9.1038% | 0.151112 ± 0.001437 |
| IQ3_S | 148.04 GiB (2.98 BPW) | Q6_K / IQ2_S / IQ2_S / IQ3_S | 6.417468 ± 0.044920 | +23.7620% | 0.321256 ± 0.002052 |
| IQ2_S | 134.01 GiB (2.70 BPW) | Q6_K / IQ2_XS / IQ2_XS / IQ3_XXS | 6.636852 ± 0.046125 | +27.9929% | 0.397892 ± 0.002336 |
Run AesSedai/MiniMax-M3-GGUF with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models