CodeWarden-7B (GGUF)

100% Offline PR Security & Architecture Auditor fine-tuned from Qwen/Qwen2.5-Coder-7B-Instruct via 4-bit QLoRA (Unsloth).

Hold-Out Benchmark Results (RTX 5070 Ti 12GB)

Metric Base (Qwen2.5-Coder-7B) CodeWarden-7B
Audit Schema Adherence 0.0% 86.7%
Verdict Classification 0.0% 86.7%
Avg Latency / Audit 4.51s 2.69s (-40.4%)
Throughput 84.3 tok/s 68.7 tok/s

Quickstart with Ollama

huggingface-cli download nihalmotukuri/codewarden-7b-gguf --local-dir ./codewarden-7b
cd codewarden-7b
ollama create codewarden:7b -f Modelfile
Downloads last month
109
GGUF
Model size
8B params
Architecture
qwen2
Hardware compatibility
Log In to add your hardware

4-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for nihalmotukuri/codewarden-7b-gguf

Base model

Qwen/Qwen2.5-7B
Quantized
(289)
this model