---
library_name: transformers
pipeline_tag: text-generation
license: apache-2.0
base_model:
  - tsinghua-sigs-robot-lab/VeriLoop-E2
base_model_relation: quantized
language:
  - en
  - zh
tags:
  - veriloop
  - veriloop-coder
  - code
  - coding-agent
  - software-engineering
  - mathematical-reasoning
  - gguf
  - llama.cpp
  - imatrix
  - code-optimized
  - quantization
  - open-source
  - apache-2.0
  - qwen3_5
  - self-harness
  - harness-engineering
  - surface-host-adapter
  - evidence-binding
  - rollback
  - uncertainty-calibration
  - long-context
  - vertical-code-model
  - recursive-improvement
---

<div align="center">
  <h1>VeriLoop E2 · GGUF</h1>
  <p><strong>Coding-Optimized Quantized Models</strong></p>
  <p>
    <a href="https://huggingface.co/tsinghua-sigs-robot-lab/VeriLoop-E2">Original Model ↗</a>
    ·
    <a href="https://github.com/rodrigoramosrs">GitHub</a>
    ·
    Apache-2.0
  </p>
</div>

---

## Overview

This repository contains **GGUF quantizations** of [VeriLoop E2](https://huggingface.co/tsinghua-sigs-robot-lab/VeriLoop-E2), an open 27B post-trained model built on **Qwen3.8-27B** for code, mathematics, and physics. Its core reasoning discipline is **VeriLoop-Governed Recurrence (VGR)**: candidate states are recursively proposed, externally checked, and retained only when the protected evidence state improves without regression.

Quantized by [Rodrigo Ramos](https://github.com/rodrigoramosrs).

## Quantization Approach

All quants were produced with [llama.cpp](https://github.com/ggml-org/llama.cpp) using a **code-specialized importance matrix (imatrix)**. Unlike generic imatrix datasets, this one was curated from software engineering corpora: repository-level code, patches, test suites, and agentic coding traces, ensuring that quantization preserves fidelity on the distributions that matter most for coding tasks.

The result is a set of GGUF files that retain the original model's strong software-engineering capabilities while being deployable via `llama.cpp`, `llama-cpp-python`, `Ollama`, `LM Studio`, and other GGUF-compatible runtimes.

## Available Quants

| File | Quant Type | Size | Notes |
|---|---|---|---|
| `LoopCoder-VeriLoop-E2-BF16.gguf` | BF16 | 50.9 GB | Full-precision reference |
| `LoopCoder-VeriLoop-E2-Q8_0.gguf` | Q8_0 | 27.0 GB | High quality, larger file |
| `LoopCoder-VeriLoop-E2-Q6_K.gguf` | Q6_K | 20.9 GB | Excellent quality / size trade-off |
| `LoopCoder-VeriLoop-E2-Q5_K_M.gguf` | Q5_K_M | 18.2 GB | Strong quality, reduced size |
| `LoopCoder-VeriLoop-E2-Q4_K_M.gguf` | Q4_K_M | 15.7 GB | Balanced quality / size |
| `LoopCoder-VeriLoop-E2-Q3_K_M.gguf` | Q3_K_M | 12.6 GB | Smaller, good for limited RAM |
| `LoopCoder-VeriLoop-E2-IQ4_XS.gguf` | IQ4_XS | 14.3 GB | Extra-small 4-bit |
| `LoopCoder-VeriLoop-E2-IQ3_XS.gguf` | IQ3_XS | 11.6 GB | Extra-small 3-bit |

## Usage

### llama.cpp

```bash
./llama-cli \
  -m LoopCoder-VeriLoop-E2-Q4_K_M.gguf \
  -p "Your coding prompt here" \
  -n 2048 \
  -t 8
```

### llama-cpp-python

```python
from llama_cpp import Llama

llm = Llama(
    model_path="LoopCoder-VeriLoop-E2-Q4_K_M.gguf",
    n_ctx=32768,
    n_threads=8,
)

output = llm(
    "Write a Python function to merge two sorted lists.",
    max_tokens=1024,
    temperature=0.2,
)
print(output["choices"][0]["text"])
```

### Ollama

```bash
ollama modelfile from ./LoopCoder-VeriLoop-E2-Q4_K_M.gguf
ollama create veriloop-e2:q4_k_m -f Modelfile
ollama run veriloop-e2:q4_k_m
```

## Acknowledgements

- **Libo Wang** and the **Intelligent Robotics Laboratory, Tsinghua SIGS** for developing the original VeriLoop E2 model.
- The **llama.cpp** community for the quantization and inference tooling.
- The original model repository: [tsinghua-sigs-robot-lab/VeriLoop-E2](https://huggingface.co/tsinghua-sigs-robot-lab/VeriLoop-E2)

## License

Apache-2.0. The weights are quantized from the original Apache-2.0 licensed model. See the [original repository](https://huggingface.co/tsinghua-sigs-robot-lab/VeriLoop-E2) for full licensing details and third-party notices.