Downloads · 30 days
16
1% of all-time downloads
Qwopus3.5-27B-v3-int4-autoround
Lorbus/Qwopus3.5-27B-v3-int4-autoround
Qwopus3.5-27B-v3-int4-autoround is a machine learning model from Lorbus. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
[](https://huggingface.co/Jackrong/Qwopus3.5-27B-v3)
All-time downloads
2.4K
Public
Parameters
6.3B
31.5 GB on disk
Likes
4
Public
Repo makeup
Click a slice to open those files.
.safetensors18.7 GB · 100%
Parameter types
How the weights are stored.
I323.1B · 49%
Base models
Model card
From the Hugging Face model README
Qwopus3.5-27B-v3 AutoRound Int4
This is a quantized version of Jackrong/Qwopus3.5-27B-v3
Quantization Details
| Parameter | Value |
|---|---|
| Method | AutoRound W4A16 (4-bit weights, 16-bit activations) |
| Mode | Full tuning (iters=200) - best accuracy |
| Group size | 128 |
| Symmetric | Yes |
| Quantization time | ~7 hours on NVIDIA GB10 |
Model Size
- Original BF16: ~54 GB
- Quantized Int4: ~17 GB
Accuracy
- Best accuracy mode uses sign gradient descent to optimize rounding values
- 200 iterations per layer for optimal quantization quality
- Some layers kept FP16 (linear_attn.in_proj_a/b, visual modules) due to shape constraints
Usage with vLLM
vllm serve Lorbus/Qwopus3.5-27B-v3-int4-autoround \
--host 0.0.0.0 --port 8000 \
--tensor-parallel-size 1 \
--gpu-memory-utilization 0.90
About Qwopus3.5-27B-v3
Qwopus3.5-27B-v3 is a reasoning-enhanced model based on Qwen3.5-27B, designed to improve reasoning stability and correctness while optimizing inference efficiency. It excels at programming, math, and analytical tasks.
Acknowledgments
- Quantization method: Intel AutoRound
- Base model: Jackrong/Qwopus3.5-27B-v3
- Type
- qwen3_5
- License
- apache-2.0
- Created
- Apr 3, 2026
- Updated
- Apr 3, 2026