demon-zombie
Initial release: W4A16 quant of MiniMax-M2.7
82d1e56
|
Raw
History Blame Contribute Delete
1.14 kB
---
license: other
license_name: modified-mit
license_link: https://github.com/MiniMax-AI/MiniMax-M2.7/blob/main/LICENSE
base_model: MiniMaxAI/MiniMax-M2.7
base_model_relation: quantized
library_name: vllm
pipeline_tag: text-generation
tags:
- vllm
- compressed-tensors
- awq
- w4a16
- 4bit
- marlin
- minimax
- quantization
- moe
---
# MiniMax-M2.7 AWQ 4bit (W4A16)
W4A16 quantization of [`MiniMaxAI/MiniMax-M2.7`](https://huggingface.co/MiniMaxAI/MiniMax-M2.7),
produced with [`llm-compressor`](https://github.com/vllm-project/llm-compressor).
- **Format:** `compressed-tensors` pack-quantized, int4 weights / fp16 activations
- **Group size:** 128, symmetric
- **Calibration:** data-free, MSE observer
- **Kept in BF16:** MoE routing gates and `lm_head` only — every other `Linear` is quantized,
matching the ignore list from [`cyankiwi/MiniMax-M2.5-AWQ-4bit`](https://huggingface.co/cyankiwi/MiniMax-M2.5-AWQ-4bit).
## vLLM
```
vllm serve demon-zombie/MiniMax-M2.7-AWQ-4bit \
--tensor-parallel-size 4 \
--tool-call-parser minimax_m2 \
--reasoning-parser minimax_m2 \
--enable-auto-tool-choice
```