README.md

---
base_model: MiniMaxAI/MiniMax-M2.7
library_name: mlx
pipeline_tag: text-generation
license: other
license_name: minimax-model-license
license_link: https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/main/LICENSE
tags:
  - minimax
  - m2.7
  - moe
  - quantized
  - rotorquant
  - kv-cache-quantization
  - mlx
---

> [!TIP]
> **KV-cache quantization without any fork (recommended, 2026):** upstream
> llama.cpp/Ollama now cover this natively — use `-ctk q8_0 -ctv q8_0`
> (~half KV memory, negligible quality loss: perplexity +0.002–0.05) or
> `-ctk q4_0 -ctv q4_0` (~quarter memory, ≈7.6% perplexity increase). In
> Ollama: `OLLAMA_KV_CACHE_TYPE=q8_0` with `OLLAMA_FLASH_ATTENTION=1`. Keep
> K and V types symmetric to stay on the fast fused Flash-Attention path.
> Since April 2026, mainline llama.cpp also applies Hadamard rotation to
> KV activations ([PR #21038](https://github.com/ggml-org/llama.cpp/pull/21038)),
> which greatly improves low-bit KV quality (opt-out:
> `LLAMA_ATTN_ROT_DISABLE=1`).
>
> The RotorQuant/TurboQuant fork flow below is **experimental/legacy**: the
> TurboQuant llama.cpp PR was closed without merging (June 2026) and the fork
> is unmaintained relative to mainline. It is NOT required to use this model.
<!-- kv-upstream-note -->

# MiniMax-M2.7-RotorQuant-MLX-3bit

**MLX 3-bit quantized variant of [MiniMaxAI/MiniMax-M2.7](https://huggingface.co/MiniMaxAI/MiniMax-M2.7) with the legacy RotorQuant KV-cache fork (superseded by upstream llama.cpp KV options), optimized for Apple Silicon.**

## Overview

MiniMax-M2.7 is a massive 256-expert Mixture-of-Experts (MoE) model with 8 experts active per token, totaling approximately 456 billion parameters. This variant combines **3-bit MLX weight quantization** with **RotorQuant** KV-cache quantization for deployment on Apple Silicon hardware.

RotorQuant applies a learned Hadamard rotation matrix to keys and values before quantization, smoothing the activation distribution for better quality retention. At 3-bit, RotorQuant's rotation-based approach is particularly valuable for preserving output quality where naive quantization would noticeably degrade.

| Property | Value |
|---|---|
| Architecture | MoE (256 experts, 8 active/token) |
| Total Parameters | ~456B |
| Layers | 62 |
| Hidden Size | 3072 |
| Attention Heads | 48 |
| Weight Quantization | 3-bit (MLX) |
| KV-Cache Quantization | RotorQuant |
| Estimated Size | ~170 GB |
| Base Model | MiniMaxAI/MiniMax-M2.7 |

## Quickstart

```bash
pip install mlx-lm
```

```python
from mlx_lm import load, generate

model, tokenizer = load("majentik/MiniMax-M2.7-RotorQuant-MLX-3bit")

prompt = "What is a Comprehensive Geriatric Assessment?"
messages = [{"role": "user", "content": prompt}]
text = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)

response = generate(
    model,
    tokenizer,
    prompt=text,
    max_tokens=512,
)
print(response)
```

## RotorQuant vs TurboQuant

| Feature | RotorQuant | TurboQuant |
|---|---|---|
| Technique | Rotation-based KV quantization (Hadamard transform) | Asymmetric per-channel KV quantization |
| Throughput | Slightly lower throughput (rotation overhead) | Higher throughput, lower latency |
| Quality | Better quality preservation at low bit-widths | Good quality preservation |
| Best For | Quality-sensitive tasks, research | High-throughput serving, long contexts |

> At 3-bit quantization, RotorQuant provides meaningfully better quality than TurboQuant due to its rotation-based outlier smoothing.

## Memory Estimates (Apple Silicon)

| Variant | Estimated Size | Minimum Unified Memory |
|---|---|---|
| MLX 8-bit | ~456 GB | 512 GB (Mac Studio M2/M3/M4 Ultra) |
| MLX 5-bit | ~280 GB | 384 GB |
| MLX 4-bit | ~225 GB | 256 GB |
| MLX 3-bit | ~170 GB | 192 GB |
| MLX 2-bit | ~110 GB | 128 GB |

> **Note**: 3-bit quantization requires Apple Silicon with 192 GB+ unified memory, such as a Mac Studio with M2/M3/M4 Ultra.

## See Also

- [MiniMaxAI/MiniMax-M2.7](https://huggingface.co/MiniMaxAI/MiniMax-M2.7) -- Base model
- [majentik/MiniMax-M2.7-TurboQuant-MLX-3bit](https://huggingface.co/majentik/MiniMax-M2.7-TurboQuant-MLX-3bit) -- TurboQuant MLX 3-bit
- [majentik/MiniMax-M2.7-RotorQuant-MLX-4bit](https://huggingface.co/majentik/MiniMax-M2.7-RotorQuant-MLX-4bit) -- MLX 4-bit

## Quant trade-off (MLX lane)

| Bits | Approx size | Use case | Recommendation |
|---|---|---|---|
| 2-bit | ~119 GB | Aggressive quantization | Very low-RAM Macs |
| **3-bit** | ~164 GB | Lossy but small | **Low-RAM Macs** |
| 4-bit | ~192 GB | Balanced default | Recommended for most Macs |
| 5-bit | ~228 GB | Higher fidelity | Quality-sensitive |
| 6-bit | ~274 GB | Approaching FP16 quality | High-fidelity |
| 8-bit | ~347 GB | Near-lossless reference | Fidelity-critical work |

(Current variant — **3bit** — is bolded.)

## Variants in this family

(Showing 12 sibling variants under `majentik/minimax-m2.7-*`. The current variant — `RotorQuant-MLX-3bit` — is **bolded**.)

| Variant | Runtime | Approx size | Use case |
|---|---|---|---|
| **RotorQuant-MLX-3bit** | mlx-lm | ~1.2 GB | Apple Silicon, small |
| [RotorQuant-MLX-4bit](https://huggingface.co/majentik/minimax-m2.7-rotorquant-mlx-4bit) | mlx-lm | ~1.7 GB | Apple Silicon balanced |
| [RotorQuant-MLX-5bit](https://huggingface.co/majentik/minimax-m2.7-rotorquant-mlx-5bit) | mlx-lm | ~2.1 GB | Apple Silicon, higher fidelity |
| [TurboQuant-MLX-3bit](https://huggingface.co/majentik/minimax-m2.7-turboquant-mlx-3bit) | mlx-lm | ~1.2 GB | Apple Silicon, small |
| [TurboQuant-MLX-4bit](https://huggingface.co/majentik/minimax-m2.7-turboquant-mlx-4bit) | mlx-lm | ~1.7 GB | Apple Silicon balanced |
| [TurboQuant-MLX-5bit](https://huggingface.co/majentik/minimax-m2.7-turboquant-mlx-5bit) | mlx-lm | ~2.1 GB | Apple Silicon, higher fidelity |

## About the RotorQuant / TurboQuant labels

RotorQuant and TurboQuant are this project's **release labels**, not distinct
quantization algorithms — for any given tier, both brand repos carry
byte-identical weights produced with the standard MLX / llama.cpp quantizers.
No brand-specific speedup is claimed or measured.