---
title: Tmax-9B-MLX-6bit
canonical_url: "https://www.modelscope.cn/models/mlx-community/Tmax-9B-MLX-6bit"
md_url: "https://www.modelscope.cn/models/mlx-community/Tmax-9B-MLX-6bit.md"
repository: mlx-community/Tmax-9B-MLX-6bit
last_updated: 2026-06-25
license: apache-2.0
pipeline_tag: text-generation
tasks:
  - text-generation
model_type:
  - qwen3_5
architectures:
  - Qwen3_5ForCausalLM
base_model:
  - allenai/tmax-9b
base_model_relation: quantized
parameters: 2.0B
tensor_type:
  - BF16
  - U32
library_name:
  - mlx
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 26
stars: 0
tags:
  - mlx
  - qwen3_5
  - text-generation
---

# Tmax-9B-MLX-6bit

> Tmax-9B-MLX-6bit - mlx-community 在 ModelScope 开源的模型。MLX-converted text-only weights of allenai/tmax-9b.

mlx-community/Tmax-9B-MLX-6bit 是 ModelScope 魔搭社区上的 2.0B 参数text-generation模型，采用 apache-2.0 许可，基于 allenai/tmax-9b 构建。

- **Repository**: mlx-community/Tmax-9B-MLX-6bit
- **License**: apache-2.0
- **Tasks**: text-generation
- **Parameters**: 2.0B
- **Base model**: allenai/tmax-9b
- **Tags**: mlx, qwen3_5, text-generation
- **Downloads**: 26
- **Stars**: 0
- **Last updated**: 2026-06-25

Source: https://www.modelscope.cn/models/mlx-community/Tmax-9B-MLX-6bit

---

# Tmax-9B MLX (6bit)

MLX-converted text-only weights of [`allenai/tmax-9b`](https://huggingface.co/allenai/tmax-9b).

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-9b`
- **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-9B-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-9B (6-bit MLX) | 79.8 | 140 | 1,032 | 1,097 | 1,064 | 814 ms (OK) |

Full results (all 7 Tmax MLX variants + 2 Qwen3.5 controls): [rapid-mlx docs](https://github.com/raullenchai/Rapid-MLX/blob/main/docs/benchmarks/tmax-m3-ultra.md).

Reproduce:

```bash
pip install rapid-mlx==0.8.18
rapid-mlx serve tmax-9b-6bit --port 8765
```