---
license: apache-2.0
base_model:
  - Qwen/Qwen2.5-Coder-1.5B
  - Qwen/Qwen2.5-Coder-1.5B-Instruct
tags:
  - gguf
  - ollama
  - qwen2
  - code
  - rust
  - rust-lightning
  - code2lora
---

# rust-lightning-code2lora GGUF Models

Ollama-ready GGUF exports of Code2LoRA-generated rust-lightning models.

## Files

- `rust-lightning-code2lora-instruct-q4_K_M.gguf`: chat/instruct variant. Use
  this for OpenWebUI and normal chat.
- `rust-lightning-code2lora-q4_K_M.gguf`: raw completion variant. This is more
  likely to continue code/text rather than answer chat prompts.

Both were produced by merging the Code2LoRA-generated PEFT adapter into the
corresponding Qwen2.5-Coder 1.5B base, converting to GGUF, and quantizing to
`Q4_K_M`.

## Ollama

Create the chat model:

```bash
cat > Modelfile <<'EOF'
FROM ./rust-lightning-code2lora-instruct-q4_K_M.gguf
SYSTEM """
You are Qwen2.5-Coder-1.5B-Instruct with a Code2LoRA-generated rust-lightning
adapter merged into the weights. Answer as a practical Rust and Lightning
Development Kit assistant. Prefer rust-lightning terminology and be explicit
when uncertain.
"""
EOF

ollama create rust-lightning-code2lora-chat -f Modelfile
ollama run rust-lightning-code2lora-chat
```

## Important Caveats

This is not a conventional supervised fine-tune on rust-lightning examples. It
is a repository-conditioned adapter generated by the Code2LoRA hypernetwork and
then merged into Qwen2.5-Coder. The released Code2LoRA checkpoint was
trained/evaluated on Python repositories, so Rust/LDK quality should be treated
as experimental.

## Provenance

- Target repository: `lightningdevkit/rust-lightning`
- Local source commit used during generation: `5049f7c02`
- Code2LoRA checkpoint: `code2lora/code2lora-direct`
- Quantization: `Q4_K_M`