--- license: apache-2.0 library_name: mlx pipeline_tag: text-generation base_model: Jackrong/Qwopus3.8-27B-Coder-Flash base_model_relation: quantized language: - en - zh tags: - mlx - mixed-precision - qwen3_5 - coding - tool-use - experimental inference: false --- # Qwopus3.8-27B Coder Flash · MLX Balanced **23.74 GB weights · 6/8 executable coding smoke · 64K/128K retrieval, tools and cached coding passed.** Select Balanced for the higher local coding score, Fast for lower memory and faster decode. 4-bit MLP group 32; 14 low-sensitivity projections use group 64. Attention, embeddings, head and eight tail MLP layers use 8-bit group 64. Both retain 96 BF16 recurrent gating projections, publisher tokenizer/template, BF16 vision and native MTP. [Exact precision recipe](quantization_recipe.json). | Local replay | 64K | 128K | |---|---:|---:| | Prompt tokens | 61,441 | 126,976 | | Cold turn | 192.87 s | 579.09 s | | Cached coding decode | 78.4 tok/s | 61.9 tok/s | M4 Max 40-core GPU / 64 GiB; Kapitsa 0.2.27 + DFlash2; one slot; disk cache off; greedy; thinking off. One run per context cell. The eight-task coding smoke missed weighted interval scheduling and TTL-cache semantics. [Evaluation contract](evaluation_summary.json) · [Weight hashes](weights_manifest.json). ```bash hf download Shiftedx/Qwopus3.8-27B-Coder-Flash-MLX-Balanced --local-dir ./qwopus-balanced hf download incoai/Qwen3.8-27B-DFlash2 --local-dir ./dflash2 kapitsa serve --private --local --model ./qwopus-balanced \ --model-id qwopus-coder --context 131072 --slots 1 \ --dflash ./dflash2 --disk-cache 0 ``` Use --context 65536 for 64K. Send temperature=0, top_p=1, top_k=0 and chat_template_kwargs={"enable_thinking":false} to /v1/chat/completions. Prompt and output share the limit. DFlash2 is separate. **Structured JSON / vision.** With the request schemas, both models passed **8/8 original cases twice, 12/12 held-out cases and 8/8 with native MTP3**, graded as exact bare JSON. Apply the matching response_format from [structured_output_profiles.json](structured_output_profiles.json) for OCR, chart, spatial-array or two-image replies. The schemas constrain structure without supplying answers. This request profile fixes the previously observed fences and nested arrays. [Validation](structured_output_validation.json). | Kapitsa 0.2.31 + DFlash2 | Balanced | Fast | |---|---:|---:| | Structured JSON decode | 128.1 tok/s | 165.6 tok/s | | Code/prose decode | 90.2 tok/s | 108.0 tok/s | Median of three repetitions on M4 Max / 64 GiB. Native MTP depths 1–3 and DFlash2 each matched plain greedy text on 4/4 short probes. Use --dflash none --spec 3 for native MTP or --spec 0 for plain decoding. [Earlier capability checks](capability_validation.json). Experimental refusal-direction edit, strength 0.5 on attention outputs in layers 10–17. Refusal reduction and BF16 fidelity unverified. Text, still-image and short MTP checks only; long-context MTP untested. Converted from pinned Apache-2.0 BF16 using streamed weights. [Jackrong source](https://huggingface.co/Jackrong/Qwopus3.8-27B-Coder-Flash/tree/c7d46e5dadf43ab838f2d594839805b2c2f6ccd3) · [Kapitsa v0.2.27](https://github.com/nace-ai/kapitsa/tree/v0.2.27) · [Other variant](https://huggingface.co/Shiftedx/Qwopus3.8-27B-Coder-Flash-MLX-Fast).