Qwen-3.5 Collection
Collection
Quantized Qwen3.5 models for efficient image-text understanding (AutoRound W4A16). • 7 items • Updated • 1
Vishva007/Qwen3.5-0.8B-W4A16-AutoRound-GPTQ
This is a W4A16 (4-bit weight, 16-bit activation) GPTQ-format quantized version of Qwen/Qwen3.5-0.8B, produced using AutoRound — Intel's sign gradient descent based quantization method designed for production-grade accuracy retention.
Quantization Details
| Parameter | Value |
|---|---|
| Method | AutoRound (W4A16, GPTQ format) |
| Group Size | 128 |
| Symmetric | Yes |
| Iterations | 1000 |
| Calibration Samples | 512 |
| Sequence Length | 2048 |
| Torch Compile | Enabled |
Key Notes
- GPTQ format — Exported in the standard GPTQ format for broad ecosystem compatibility.
- Ultra-high accuracy configuration — 1000 iterations with 512 calibration samples ensures near-lossless quantization, especially critical at this model scale where parameter budget is tight.
- W4A16 — Weights are quantized to 4-bit integers; activations remain in FP16 for inference stability.
- Extremely lightweight — The quantized 0.8B model is suitable for edge deployment, low-latency inference, and resource-constrained environments.
- ~50% memory reduction compared to the FP16 base model.
Usage
This model is compatible with transformers, AutoGPTQ, vLLM, and SGLang — any backend supporting GPTQ-format weights works out of the box. For full model details, architecture, and capabilities, refer to the base model page.
- Downloads last month
- 115
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support