license: apache-2.0 library_name: splash pipeline_tag: image-text-to-text inference: false base_model:

• orcarouter/Qwen3.8-27B-Uncensored-MLX
• incoai/Qwen3.8-27B-DFlash2 tags:
• splash
• apple-silicon
• dflash2
• 4-bit
• uncensored

Qwen3.8-27B-Uncensored-Splash

Splash conversion of the OrcaRouter community model, using its 4-bit affine weights and Inco AI's 4-bit affine DFlash2 draft. Includes vision and the Splash chat template. The draft is reused from the standard Qwen3.8-27B package, not retrained for this variant.

Apple M3 or newer, macOS 26.4+, 36 GB memory minimum (48 GB+ recommended).

brew install incoai/tap/splash
splash serve --model jianchen0311/Qwen3.8-27B-Uncensored-Splash

The first run downloads about 17.4 GB. Open http://127.0.0.1:8000, or connect an agent you have already installed:

splash claude  # or: splash codex / splash opencode / splash hermes

Community model; refusal behavior is modified upstream. Serving compatibility has been tested; no broad model-quality benchmark is claimed. Component revisions and artifact hashes are recorded in manifest.json.

Source model · Splash · Blog