---
license: apache-2.0
base_model: Qwen/Qwen2.5-Coder-14B-Instruct
tags:
  - bitnet
  - quantization
  - ternary
  - 1.58-bit
  - qwen
  - qwen2.5
  - code
  - experimental
  - 14b-architecture
library_name: safetensors
pipeline_tag: text-generation
language:
  - en
  - zh
model_name: Qwen2.5-Coder-14B-BitNet-1.58b
datasets: []
metrics: []
---

# Qwen2.5-Coder-14B-Instruct-BitNet-1.58b

**Architecture: 14.7 Billion Parameters** | BitNet 1.58-bit Ternary Quantization

---

> **IMPORTANT: Parameter Count Display**
>
> HuggingFace displays a reduced parameter count because it counts packed bytes, not actual parameters.
> This model has the **full 14.7B parameter Qwen2.5-Coder architecture**.
> The weights are stored as ternary values ({-1, 0, +1}) packed 4 per byte, which reduces
> storage to 4.6 GB but preserves all 14.7 billion parameters.

---

## Overview

This is an **experimental** BitNet 1.58-bit quantization of the Qwen2.5-Coder-14B-Instruct model using absmean scaling with group-wise quantization. The model stores weights as ternary values ({-1, 0, +1}) packed 4 values per byte.

**This is research/experimental work. Quality and performance have not been formally benchmarked.**

## Specifications

| Property | Value |
|----------|-------|
| Base Model | [Qwen/Qwen2.5-Coder-14B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-14B-Instruct) |
| Architecture | Qwen2 (Qwen2ForCausalLM) |
| Parameters | 14.7B |
| Quantization | BitNet 1.58-bit ternary |
| Bits per Weight | ~1.58 |
| Group Size | 64 |
| Original Size | 29.54 GB (BF16) |
| Quantized Size | 4.62 GB (SafeTensors) |
| GGUF Size | 6.52 GB (TQ2_0) |
| Compression | ~6.4x |

## Formats

| Format | File | Description |
|--------|------|-------------|
| SafeTensors | `model-*.safetensors` | Sharded quantized weights + scales |
| GGUF | `qwen-coder-14b-tq2.gguf` | llama.cpp compatible |

## Quantization Method

### Algorithm
1. Reshape weights into groups of 64
2. Compute per-group scale: `scale = mean(|weights|)`
3. Normalize and round to nearest ternary: `q = round(w / scale)` clamped to {-1, 0, +1}
4. Map to unsigned: {-1, 0, +1} → {0, 1, 2}
5. Pack 4 values per byte: `v0 + v1*3 + v2*9 + v3*27`

### Tooling
- **Quantization**: Custom Rust tool using [Candle](https://github.com/huggingface/candle)
- **GGUF Conversion**: [llama.cpp](https://github.com/ggerganov/llama.cpp) convert_hf_to_gguf.py

### Hardware Used
- GPU: NVIDIA RTX 5080 (16GB VRAM)
- Quantization time: ~99 seconds
- Memory: Streaming mode with CPU fallback for large tensors

## Usage

### With Ollama/llama.cpp
```bash
# llama.cpp
./llama-cli -m qwen-coder-14b-tq2.gguf -p "Write a Python function:"
```

### Unpacking Weights (Python)
```python
def unpack_ternary(packed_byte):
    """Unpack 4 ternary values from byte."""
    values = []
    val = packed_byte
    for _ in range(4):
        values.append((val % 3) - 1)  # {0,1,2} → {-1,0,+1}
        val //= 3
    return values
```

## Limitations

- **Quality not benchmarked** - May have significant degradation vs original
- **Requires custom runtime** - Standard transformers doesn't support ternary weights
- **Experimental** - Not intended for production use without evaluation
- GGUF keeps embeddings/lm_head at F16, hence larger than SafeTensors

## License

Apache 2.0 (inherited from [Qwen2.5-Coder-14B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-14B-Instruct))

## Citation

```bibtex
@misc{qwen-coder-bitnet-2025,
  title={Qwen2.5-Coder-14B-BitNet-1.58b: Experimental BitNet Quantization},
  author={Tzervas},
  year={2025},
  url={https://huggingface.co/tzervas/qwen2.5-coder-14b-bitnet-1.58b}
}
```