hf_mirrors/Mellum2-12B-A2.5B-Thinking-mxfp4
Model Introduction
Files & versions
Analysis

Mellum2 12B A2.5B Thinking - MXFP4 MLX

This is an MXFP4 MLX quantization of JetBrains/Mellum2-12B-A2.5B-Thinking.

Mellum2 Thinking is a reasoning-augmented Mixture-of-Experts assistant model. It has 64 experts with 8 active per token, a 131,072-token context window, and emits reasoning in <think>...</think> blocks before the final answer.

Conversion details

  • Source: JetBrains/Mellum2-12B-A2.5B-Thinking
  • Format: MLX safetensors
  • Quantization: MXFP4, 4 bits, group size 32
  • License: Apache-2.0
  • EOS token: <|im_end|> (token ID 28)

The upstream config.json and generation_config.json identify token ID 0 as the EOS token, while the tokenizer identifies <|im_end|> (ID 28) as EOS. This conversion uses token ID 28 so MLX generation stops at the end of the assistant turn.

Usage

pip install -U mlx-lm

mlx_lm.chat \
  --model mlx-community/Mellum2-12B-A2.5B-Thinking-mxfp4 \
  --max-tokens 8192 \
  --temp 0.6 \
  --top-p 0.95

Benchmarks

Benchmarked with oMLX 0.5.7 and MLX-LM 0.31.3 on an Apple M5 Max MacBook Pro with 128 GB unified memory, using the built-in Python-code workload and 128 generated tokens. Results vary with hardware, runtime versions, context length, and generation settings.

Prompt tokensTTFTPrefillGenerationEnd-to-endPeak memory
4,096963.4 ms4,251.6 tok/s172.3 tok/s1.7 s6.80 GB
16,3845,739.5 ms2,854.6 tok/s100.9 tok/s7.0 s7.04 GB
32,76817,226.8 ms1,902.2 tok/s101.1 tok/s18.5 s7.42 GB
65,53639,122.4 ms1,675.2 tok/s73.3 tok/s40.9 s8.12 GB
131,07283,223.6 ms1,574.9 tok/s42.2 tok/s86.3 s9.54 GB

At a 4,096-token prompt, batch generation achieved 226.6, 183.6, and 220.6 aggregate generation tok/s at 2, 4, and 8 concurrent requests, respectively, compared with 172.3 tok/s for one request.

Model provenance

For the original model card, training details, benchmark results, and usage guidance, see the upstream JetBrains checkpoint.