Qwen3.8-27B-oQ6e-mtp

[!IMPORTANT] This quantization was uploaded on 2026-08-15 and replaces a previous version. If you downloaded this model before this date, please re-download for the updated weights.

Qwen3.8-27B-oQ6e-mtp

This model was quantized using oQ (oMLX v0.6.0.dev1) mixed-precision quantization.

Quantization details

• Model type: qwen3_5
• Bits: 6
• Group size: 64
• Format: MLX safetensors

---

系統提示詞 (System prompt)

第一行必須是： (The first line must be:)

你是一個專業的助手，請以繁體中文思考，繁體中文回覆。
(You are a professional assistant. Please think and reply in Traditional Chinese.)
你是一個專業的助手，請以繁體中文思考，繁體中文回覆。
(You are a professional assistant. Please think and reply in Traditional Chinese.)

沒有它，這個模型表現會不佳。您可以在此行之後添加任何內容。 (The model underperforms without it. Append anything after that line.)

---

建議設定(按需開啟)

預填充優先順序(Prefill Priority): 速度(Speed)

• Lightning MTP ✅

或

預填充優先順序(Prefill Priority): 最大上下文(Max Context)

• VLM MTP ✅
• Drafter 模型: Qwen3.8-27B-MTP-mxfp8
• Draft 區塊大小: 2

或

預填充優先順序(Prefill Priority): 最大上下文(Max Context)

• Lightning MTP ✅
  • Adaptive max depth: 8 tokens
• Qwen ANE Prompt Processing ✅
  • Prompt block: 2048
  • Pad tails from: 2
  • MLP on ANE: 0.9
  • MLP layer limit: 1
• Use both ANEs ✅
• Accelerate GDN ✅
  • GDN on ANE: 0.05
  • GDN layer limit: 16

---