---
library_name: transformers
license: apache-2.0
pipeline_tag: image-text-to-text
base_model:
- google/gemma-4-26B-A4B-it
tags:
- sft
- Moe
- Pruned
datasets:
- huihui-ai/GLM-5.1-Multilingual-STEM
---

# 🤖 Huihui4-8B-A4B-v2 Model Card

## 📌 Overview
`Huihui4-8B-A4B-v2` is a lightweight MoE (Mixture of Experts) conversational model optimized from Google's `gemma-4-26B-A4B-it` architecture. Through expert pruning and supervised fine-tuning on high-quality dialogue data, the dataset adopts the thinking mode in GLM-5.1 format. This way, in thinking mode, it better reflects the thinking mode of GLM-5.1.
 this model significantly reduces computational overhead while preserving core reasoning and interaction capabilities. It is specifically designed for deployment on consumer-grade hardware and code-related conversational tasks.

**This model is not an ablation variant.**


## 🧱 Architecture & Configuration
| Parameter | Description |
|:---|:---|
| **Base Model** | `google/gemma-4-26B-A4B-it` |
| **Total MoE Experts** | 32 (pruned from the original 128) |
| **Active Experts per Token** | 8 (maintaining the A4B activation scale) |
| **Model Positioning** | Lightweight MoE conversational base / Consumer-hardware friendly |

## 📊 Training Data & Methodology
- **Data Source:** **[huihui-ai/GLM-5.1-Multilingual-STEM](https://huggingface.co/datasets/huihui-ai/GLM-5.1-Multilingual-STEM)** carefully extracted from code preference data.
- **Training Method:** Supervised Fine-Tuning (SFT).
- **Optimization Goal:** Maintain semantic coherence, instruction-following capability, and code context understanding post-pruning.

## 📈 Evaluation & Performance
- **Evaluation Tool:** Quantitative perplexity assessment using the `calculate_perplexity` script.
- **Test Results:** Preliminary dialogue tests indicate smooth interactions and stable logic. The model performs reliably in daily conversations and code-assistance tasks, with no significant performance degradation observed after pruning.

## 💻 Inference & Deployment Recommendations
- **Recommended Frameworks:** `vLLM` / `llama.cpp` / `HuggingFace Transformers`
- **VRAM Requirements:**
  - `FP16`: < 18GB
  - `INT4/INT8 Quantized`: < 6~9GB (compatible with mainstream single consumer GPUs)
- **Use Cases:** Code conversation assistants, lightweight task planning, local deployment prototyping, and baseline validation for MoE pruning/merging techniques.

## 🗺️ Roadmap
1. **Multi-Domain Fine-Tuning:** Further SFT on **four distinct datasets** to enhance the generalization capabilities of this 32-expert model.
2. **Expert Merging Validation:** Experiment with merging the four independently fine-tuned models back into a **128-expert architecture**, validating the feasibility of a `"prune → fine-tune → merge"` pipeline.
3. **Core Objective:** Ultimately verify the engineering viability of **training and iterating on large-scale MoE models using only consumer-grade hardware**.
4. If you're interested, feel free to fine-tune this model on your own datasets. We plan to merge all resulting models into a unified version at the end.

## 📝 Notes
- This model represents the initial pruned and fine-tuned iteration of the `Huihui` series. Future updates will involve multi-dataset integration and expert merging.

## Citation

```
@misc{huihui4-8b-a4b-v2,
      title  = {{Huihui4-8B-A4B-v2}: A lightweight MoE (Mixture of Experts) conversational model},
      author = {Huihui-ai},
      year   = {2026},
      url    = {https://hf.co/huihui-ai/Huihui4-8B-A4B-v2}
}
```

## Contact

If you have any questions, please raise an issue or contact us at [support@huihui.ai](support@huihui.ai).