---
license: other
base_model:
- MiniMaxAI/MiniMax-M2.5
language:
- en
tags:
- gguf
- minimax
- moe
- reap
- text-generation
pipeline_tag: text-generation
---

# MiniMax-M2.5-REAP-139B-A10B-GGUF

This is the REAP model in practical pants: high quality GGUF quants for local inference without setting your workstation on fire.

Built from:
- Base: `MiniMaxAI/MiniMax-M2.5`
- REAP source: `tomngdev/MiniMax-M2.5-REAP-139B-A10B-GGUF` (BF16 split)
- Quantized locally with `llama.cpp` on Strix Halo + high RAM mode.

## Available Quants

| Quant | Status | Size (GiB) | Notes |
|---|---|---:|---|
| `Q8_0` | uploaded | 137.78 | Highest quality quant in this pack |
| `Q5_K_M` | processing/uploading | TBD | Better quality/size balance |
| `Q4_K_M` | uploaded | 78.83 | Strong practical default |

## File Layout

All quants are split GGUF sets (`00001-of-00007` etc.) for safer handling of very large models.

## Quality Notes

- These are generated from BF16 REAP GGUF, not requantized from lower precision.
- Token embedding and output tensors are kept at `Q8_0` during quantization for quality retention.

## Usage

Use any first shard with `llama.cpp`; it auto-discovers sibling shards:

```bash
llama-cli -m MiniMax-M2.5-REAP-Q4_K_M-00001-of-00007.gguf -ngl 0 -c 8192
```

## Credits

- `MiniMaxAI` for MiniMax-M2.5
- `tomngdev` for the BF16 REAP GGUF release
- `BennyDaBall` for this quant pack

## Disclaimer

You are responsible for your own use, outputs, and compliance with applicable laws and platform policies.