General Q&A · Coding
Qwen3.8-27B · SGLang 0.5.19
A 27B single-device inference setup for Q&A, coding, and agent applications. Uses Unsloth NVFP4 weights and SGLang 0.5.19, available through the PINVOU Model Store.
01 / Model overview
Model architecture & capabilities
Qwen3.8-27B is Qwen's dense multimodal model for everyday Q&A, document understanding, coding, and tool use. It combines linear and full attention with native multi-token prediction. Inference speed and memory use depend on the quantized weights and inference engine, even when the base model is the same.
- Publisher / base model
- Alibaba Qwen · Qwen3.8-27B
- Parameter count / architecture
- 27B dense model; 64 layers; Gated DeltaNet + Gated Attention hybrid architecture
- Model context
- Native 262,144 tokens, expandable to 1M; actual service length depends on runtime settings
- Input / output
- Text, image, and video input; text and tool-call output
- Reasoning capabilities
- Supports thinking mode, agent tool calling, and native multi-token prediction (MTP)
- License
- Apache-2.0
- Quantization used here
- Unsloth Dynamic V3 NVFP4 preview; FP8 lm_head; requires SGLang 0.5.19 or later
- Weight download size
- Approx. 23.4 GB; includes a separate MTP prediction component
02 / Use cases
Use cases
Everyday Q&A and writing
Answer workplace questions, draft copy, organize key points, and adjust wording to your requirements.
Code development and tool calling
Write, explain, and debug code, and use tools through compatible applications to complete multi-step tasks.
Document understanding
Understand information in text, images, and documents to help extract content, summarize, and answer questions.
Plan model deployment and application integration
Evaluate device configurations and inference services for your model size, workloads, and concurrency needs.