This repository contains the oQ8 MLX conversion of Qwen3.8-27B. Its stored recipe is uniform 8-bit affine with group size 64 and contains no per-module precision overrides. The upstream model is a dense, native vision-language model with flexible thinking control and support for text, images, and video. Its tokenizer, processor configuration, chat template, and generation configuration are preserved.
This checkpoint was load-tested and generation-tested on the following machine:
Hardware
Configuration
Host
Mac Studio
Chip
Apple M3 Ultra
CPU
32 cores (24 performance + 8 efficiency)
Unified memory
256 GB
Runtime
MLX-VLM 0.6.3 / MLX 0.32.0
Measurement
Result
Decode (median)
24.08 tokens/s
Reported peak memory
35.61 GB
Timed runs
3 × 256 generated tokens
Warm-up
256 generated tokens
Prompt
81 tokens after chat templating
The decode figure is the median of three greedy 256-token runs after a 256-token Metal-kernel warm-up. Individual runs measured 24.09, 24.08, and 24.07 tokens/s. This is a practical local reference, not a controlled cross-platform benchmark; prompt length, context growth, sampler settings, memory pressure, thermal state, and runtime versions can materially change performance.
Quick start with MLX-VLM
bash
1python -m pip install -U mlx-vlm huggingface_hub
23python -m mlx_vlm.generate \4 --model Vontra/Qwen3.8-27B-oQ8 \5 --prompt "Explain the difference between linear and full attention."\6 --max-tokens 512
Place the model at ~/.omlx/models/Vontra/Qwen3.8-27B-oQ8.
Refresh the oMLX model registry.
Load Qwen3.8-27B-oQ8 and use the chat UI or OpenAI-compatible endpoint.
bash
1curl"$OMLX_BASE_URL/v1/chat/completions"\2 -H "Content-Type: application/json"\3 -H "Authorization: Bearer $OMLX_API_KEY"\4 -d '{
5 "model": "Qwen3.8-27B-oQ8",
6 "messages": [{"role": "user", "content": "Write a short Swift actor example."}],
7 "temperature": 1.0,
8 "top_p": 0.95,
9 "max_tokens": 256
10 }'
For long prompts, begin with a conservative context limit and increase it while watching memory pressure. The configured context is a model capability, not a guarantee that every host can prefill it within available unified memory.
Architecture
Qwen3.8-27B is a dense causal language model with a vision encoder. It uses the Qwen3.5 architectural foundation, interleaving Gated DeltaNet linear-attention blocks with periodic full-attention blocks.
Architecture detail
Upstream value
Parameters
27B
Language layers
64
Hidden size
5,120
Attention heads / KV heads
24 / 4
Linear-attention V / QK heads
48 / 16
FFN intermediate size
17,408
Vocabulary / padded embeddings
248,320
Configured context
262,144 tokens
For upstream evaluations, usage guidance, intended use, limitations, safety information, and the full architecture discussion, see the original model card.
Conversion and validation notes
Source weights: the official Qwen checkpoint.
Quantization: uniform 8-bit affine weights with group size 64 and no per-module overrides.
The upstream tokenizer, processor files, chat template, and generation configuration are preserved.
Both quantization and quantization_config preserve the oQ8 recipe.
All 2,180 converted tensors and all six indexed shards were checked locally.
Quantization can reduce output quality relative to the source weights; use a higher-precision variant when quality matters more than memory use.
The model was loaded and exercised through end-to-end generation on Apple silicon.
This is a community conversion, not an official Qwen release. Validate quality and numerical behaviour on your own representative workload before production use.
Licence and attribution
The upstream model is released under the Apache License 2.0. A copy is included in this repository; review it before use or redistribution.
All model design, training, benchmark, and upstream documentation credit belongs to Qwen and the original contributors. The MLX conversion, Apple-silicon validation, compatibility work, and packaging are provided by Vontra.
