Qwen3.8-27B-opencode-adapted adapter for mlx-lm

The LoRA from harness-hub/Qwen3.8-27B-opencode-adapted (revision b4171c0) in mlx-lm's adapter format: rank 32, scale 2.0 (alpha 64), on q, k, v, o of the 16 full-attention layers and gate, up, down of all 64 layers. The weights are the original float32 values, transposed to mlx-lm's layout. Size: 638 MB.

harness-hub trained this LoRA with reinforcement learning (AsyncGRPO in TRL) to be the model behind the OpenCode coding agent on software-engineering and terminal tasks. Their published results, under OpenCode with thinking off: SWE-bench Pro (hard 51) 35.3% โ†’ 51.0% and Terminal-Bench 2.1 46.4% โ†’ 52.4% against the base model, and with thinking off it matches the base model thinking at its highest effort while generating about a sixth of the tokens. See the original adapter card for the method, data and full results.

Use it on top of any MLX build of Qwen/Qwen3.8-27B to keep one base model and switch the adapter on and off. For a single ready-to-run model, the merged builds are simpler and slightly faster:

Build Size Same next token as bf16 KL vs bf16
netprojects/Qwen3.8-27B-opencode-adapted-MLX-6bit (recommended) 23.1 GB 98.7% 0.0014
netprojects/Qwen3.8-27B-opencode-adapted-MLX-8bit 30.0 GB 99.2% 0.0004
netprojects/Qwen3.8-27B-opencode-adapted-MLX-4bit 16.3 GB 96.7% 0.0203
netprojects/Qwen3.8-27B-opencode-adapted-MLX-bf16 55.6 GB reference reference
pip install -U "mlx-lm>=0.32.2"
mlx_lm.generate --model mlx-community/Qwen3.8-27B-4bit --adapter-path netprojects/Qwen3.8-27B-opencode-adapted-MLX-adapter \
  --chat-template-config '{"enable_thinking": false}' --prompt "Write a bash one-liner that counts lines in all .py files." --max-tokens 300
from huggingface_hub import snapshot_download
from mlx_lm import load, generate

adapter = snapshot_download("netprojects/Qwen3.8-27B-opencode-adapted-MLX-adapter")
model, tokenizer = load("mlx-community/Qwen3.8-27B-4bit", adapter_path=adapter)
prompt = tokenizer.apply_chat_template([{"role": "user", "content": "Hello"}], add_generation_prompt=True,
                                       enable_thinking=False, tokenize=False)
print(generate(model, tokenizer, prompt=prompt, max_tokens=200))

mlx_lm.server --model mlx-community/Qwen3.8-27B-4bit --adapter-path <local adapter dir> serves it for OpenCode (see the 6-bit card for the OpenCode configuration; keep thinking off).

What was checked

On a MacBook Pro M5 Max (128 GB), 18 November 2026, mlx 0.32.4 / mlx-lm 0.32.2.

  • Loaded in mlx-lm on mlx-community/Qwen3.8-27B-4bit, it creates exactly 256 LoRA modules with scale 2.0, and each module's update equals 2.0 ยท B @ A from harness-hub's original file exactly.
  • With the included chat template and thinking off, base 4-bit + adapter produced a well-formed tool call and passed 3/3 short coding tasks.

License

Apache-2.0, as Qwen3.8-27B and the harness-hub adapter. LICENSE is the Apache-2.0 text from the Qwen3.8-27B repository and NOTICE describes the conversion and credits harness-hub and the Qwen Team (Alibaba Cloud).

Published by NetProjects (Internet Projects Ltd). Not affiliated with Hugging Face, harness-hub or Alibaba.

More MLX builds from NetProjects: huggingface.co/netprojects ยท netprojects.com

Downloads last month

-

Downloads are not tracked for this model. How to track
MLX
Hardware compatibility
Log In to add your hardware

Quantized

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for netprojects/Qwen3.8-27B-opencode-adapted-MLX-adapter

Base model

Qwen/Qwen3.8-27B
Quantized
(6)
this model

Collection including netprojects/Qwen3.8-27B-opencode-adapted-MLX-adapter