You can deploy this model and start earning money today!

mlx-community/
Tmax-27B-MLX-6bit

Name: Tmax-27B-MLX-6bit
Availability: InStock

Views

No views yet

Tmax-27B MLX (6bit)

MLX-converted text-only weights of allenai/tmax-27b.

The upstream base ships as a multimodal Qwen3_5ForConditionalGeneration config but contains zero vision tensors in its safetensors — i.e. it is already a text-only checkpoint with stub vision metadata. This release strips the residual vision_config / image-token entries so it loads cleanly via mlx_lm without a vision tower.

Source: allenai/tmax-27b
License: Apache-2.0
Variant: 6bit
Quantized by: raullenchai
Tooling: mlx-lm 0.31.3 (the upstream mlx_vlm 0.3.12 qwen3_5 loader hard-requires vision-tower weights that this base does not ship, so the text-only mlx_lm.convert path is used instead)
Chat template: ships with the source repo (chat_template.jinja)
Tool format: qwen3_xml-compatible (<tool_call>{json}</tool_call>)

Usage

python

from mlx_lm import load, generate

model, tokenizer = load("mlx-community/Tmax-27B-MLX-6bit")
print(generate(model, tokenizer, prompt="Hello", max_tokens=32))

Notes

This is a pure text-generation MLX release. No vision/image inputs.
For best chat behavior, use the chat template that ships with this repo.

Benchmarks

Measured on M3 Ultra Studio (28 (20 Performance and 8 Efficiency) CPU, 60-core GPU, 256 GB unified memory) via rapid-mlx 0.8.18. Medians of 3 runs.

Variant	Decode tok/s	TTFT (ms)	Prefill 1k (tok/s)	Prefill 4k (tok/s)	Prefill 16k (tok/s)	Tool-call e2e
Tmax-27B (6-bit MLX)	26.8	288	305	314	303	2489 ms (OK)

Architecture note: Tmax-27B uses a hybrid Gated-DeltaNet design (3:1 linear-attention to full-attention layer mix). 16k-context prefill is bandwidth-bound at ~310 tok/s regardless of quantization bit width — ~54 s wall to first token at 16k. This is an architectural property of hybrid linear-attention models on Apple Silicon, not a regression, and not a rapid-mlx bug. Decode and short-context (≤4k) tool-call performance are competitive with the dense Qwen3.5-27B-4bit control on the same hardware.

Full results (all 7 Tmax MLX variants + 2 Qwen3.5 controls): rapid-mlx docs.

Reproduce:

bash

pip install rapid-mlx==0.8.18
rapid-mlx serve tmax-27b-6bit --port 8765