Overview
This repository contains GGUF quantizations of VeriLoop Coder-E1, an open-source vertical coding model built on Qwen3.6-27B. The original model introduces the Self-Harness paradigm — an evidence-bound execution substrate that turns model generation into a recursive engineering loop of falsification, exploration, and repair.
Quantized by Rodrigo Ramos.
Quantization Approach
All quants were produced with 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 | Notes |
|---|---|---|
LoopCoder-Qwen3.6-27B-BF16.gguf | BF16 | Full-precision reference |
LoopCoder-Qwen3.6-27B-Q8_0.gguf | Q8_0 | High quality, larger file |
LoopCoder-Qwen3.6-27B-Q6_K.gguf | Q6_K | Excellent quality / size trade-off |
LoopCoder-Qwen3.6-27B-Q5_K_M.gguf | Q5_K_M | Strong quality, reduced size |
LoopCoder-Qwen3.6-27B-Q4_K_M.gguf | Q4_K_M | Balanced quality / size |
LoopCoder-Qwen3.6-27B-Q3_K_M.gguf | Q3_K_M | Smaller, good for limited RAM |
LoopCoder-Qwen3.6-27B-IQ4_XS.gguf | IQ4_XS | Extra-small 4-bit |
LoopCoder-Qwen3.6-27B-IQ3_XS.gguf | IQ3_XS | Extra-small 3-bit |
Usage
llama.cpp
./llama-cli \
-m LoopCoder-Qwen3.6-27B-Q4_K_M.gguf \
-p "Your coding prompt here" \
-n 2048 \
-t 8
llama-cpp-python
from llama_cpp import Llama
llm = Llama(
model_path="LoopCoder-Qwen3.6-27B-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
ollama modelfile from ./LoopCoder-Qwen3.6-27B-Q4_K_M.gguf
ollama create veriloop-coder-e1:q4_k_m -f Modelfile
ollama run veriloop-coder-e1:q4_k_m
Acknowledgements
- Libo Wang and the Intelligent Robotics Laboratory, Tsinghua SIGS for developing the original VeriLoop Coder-E1 model.
- The llama.cpp community for the quantization and inference tooling.
- The original model repository: tsinghua-sigs-robot-lab/veriloop-coder-e1
License
Apache-2.0. The weights are quantized from the original Apache-2.0 licensed model. See the original repository for full licensing details and third-party notices.