majentik/gpt-oss-120b-RotorQuant-MLX-8bit

GPT-OSS-120B - RotorQuant MLX 8-bit

8-bit weight-quantized MLX version of openai/gpt-oss-120b with RotorQuant KV-cache quantization. Optimized for Apple Silicon inference via the MLX framework. RotorQuant delivers 5.3x faster prefill and 28% faster decode compared to TurboQuant. GPT-OSS-120B is OpenAI's flagship open-weights Mixture-of-Experts model (Apache 2.0), approaching o4-mini quality for reasoning tasks.

Approximate model size: ~120 GB

Model Specifications

Property	Value
Base Model	openai/gpt-oss-120b
Parameters	120 billion (MoE)
Architecture	Mixture-of-Experts (MoE) Transformer
License	Apache 2.0 (commercial use OK)
Weight Quantization	8-bit (~120 GB)
KV-Cache Quantization	RotorQuant
Framework	MLX (Apple Silicon)

Quickstart

from mlx_lm import load, generate
from rotorquant import IsoQuantCache

model, tokenizer = load("majentik/gpt-oss-120b-RotorQuant-MLX-8bit")

prompt = "Explain the theory of relativity."
response = generate(model, tokenizer, prompt=prompt, max_tokens=512)
print(response)

What is RotorQuant?

RotorQuant applies block-diagonal rotations (Clifford algebra) for KV cache compression. Combined with 8-bit weight quantization in MLX, this provides a dual compression strategy with superior KV-cache performance.

Key advantages over TurboQuant:

• 5.3x faster prefill
• 28% faster decode
• Equivalent memory savings

KV-Cache Quantization Comparison

Method	Prefill Speed	Decode Speed	Memory Savings	Reference
TurboQuant	1x (baseline)	1x (baseline)	High	arXiv: 2504.19874
RotorQuant	5.3x faster	28% faster	High	GitHub

Memory Estimates (GPT-OSS-120B)

Precision	Approximate Size	MLX Variant
BF16 (original)	~240 GB	--
8-bit quantized	~120 GB	This model
4-bit quantized	~65 GB	RotorQuant-MLX-4bit
2-bit quantized	~30 GB	RotorQuant-MLX-2bit

Hardware Requirements

This model requires approximately 120 GB of unified memory. Recommended hardware:

• Apple M2 Ultra (192 GB)
• Apple M3 Ultra (192 GB or 512 GB)
• Mac Studio M4 Ultra (192 GB+)
• Multi-device MLX inference for smaller Macs

See Also

• openai/gpt-oss-120b -- Base model
• majentik/gpt-oss-120b-RotorQuant -- RotorQuant KV-cache only (transformers)
• majentik/gpt-oss-120b-RotorQuant-MLX-4bit -- MLX 4-bit variant
• majentik/gpt-oss-120b-RotorQuant-MLX-2bit -- MLX 2-bit variant
• majentik/gpt-oss-120b-TurboQuant-MLX-8bit -- TurboQuant MLX 8-bit variant
• RotorQuant GitHub
• MLX Framework

Quant trade-off (MLX lane)

Bits	Approx size	Use case	Recommendation
2-bit	~31 GB	Aggressive quantization	Very low-RAM Macs
3-bit	~43 GB	Lossy but small	Low-RAM Macs
4-bit	~50 GB	Balanced default	Recommended for most Macs
5-bit	~60 GB	Higher fidelity	Quality-sensitive
6-bit	~72 GB	Approaching FP16 quality	High-fidelity
8-bit	~91 GB	Near-lossless reference	Fidelity-critical work

(Current variant — 8bit — is bolded.)

Variants in this family

(Showing 14 sibling variants under majentik/gpt-oss-120b-*. The current variant — RotorQuant-MLX-8bit — is bolded.)

Variant	Runtime	Approx size	Use case
RotorQuant	runtime modifier	n/a	KV-cache root (weight-agnostic)
RotorQuant-GGUF-IQ4_XS	llama.cpp	~103 GB	Lossy 4-bit, low-RAM CPU/edge
RotorQuant-GGUF-Q2_K	llama.cpp	~72 GB	Lossy, low-RAM CPU/edge
RotorQuant-GGUF-Q3_K_M	llama.cpp	~94 GB	Smaller 3-bit, CPU-friendly
RotorQuant-GGUF-Q4_K_M	llama.cpp	~132 GB	Balanced default
RotorQuant-GGUF-Q5_K_M	llama.cpp	~158 GB	Higher fidelity, more RAM
RotorQuant-GGUF-Q8_0	llama.cpp	~252 GB	Near-lossless reference
RotorQuant-MLX-2bit	mlx-lm	~38 GB	Apple Silicon, smallest
RotorQuant-MLX-4bit	mlx-lm	~74 GB	Apple Silicon balanced
RotorQuant-MLX-8bit	mlx-lm	~142 GB	Apple Silicon reference
TurboQuant	runtime modifier	n/a	KV-cache root (weight-agnostic)
TurboQuant-MLX-2bit	mlx-lm	~38 GB	Apple Silicon, smallest
TurboQuant-MLX-4bit	mlx-lm	~74 GB	Apple Silicon balanced
TurboQuant-MLX-8bit	mlx-lm	~142 GB	Apple Silicon reference