cmndcntrlcyber/qwen14b-code-rtpi-tools-gguf

qwen14b-code-trainer-v6-gguf

GGUF quantizations of the Code-Trainer V6 fine-tuned model. The Phase 4A LoRA adapter qwen14b-code-trainer-v6-aggressive has been merged into Qwen/Qwen2.5-Coder-14B-Instruct and quantized via llama.cpp.

This is Phase 5 of the Code-Trainer V6 / RTPI pipeline. The conversion runs as an HF Job on a100-large — the GPU sits idle, we use that flavor only for its 144 GB system RAM during the float16 merge step.

Files

File	Quantization	Size (≈)	Notes
Qwen2.5-Coder-14B-Instruct-Q4_K_M.gguf	Q4_K_M	~9 GB	Recommended default — balanced quality / footprint

Additional quantizations (Q5_K_M, Q8_0, F16) can be produced by passing --quants to launch_convert.py; this repo currently ships only Q4_K_M because that is the architecture-doc target for the Phase 6 hot-swap inference stack.

Intended use

• Local inference via llama-cli, llama-server, Ollama, LM Studio, or text-generation-webui.
• Phase 6 hot-swap target for the project's vLLM + Qwen-Agent stack — swapped in for compiled-language tasks alongside a smaller primary model.
• Out of scope: anything the upstream qwen14b-code-trainer-v6-aggressive card flags as out of scope (no safety tuning, no non-code tasks).

Source

Stage	Repo / artifact
Base model	Qwen/Qwen2.5-Coder-14B-Instruct
LoRA adapter	cmndcntrlcyber/qwen14b-code-trainer-v6-aggressive
Converter	llama.cpp (convert_hf_to_gguf.py + llama-quantize)
Conversion runtime	HF Job, a100-large, ~1 h on the merge + quantize path

Evaluation

Quality is inherited from the source LoRA adapter (eval_loss = 0.4724 on the 3,265-row validation split — see the upstream model card). Quantization to Q4_K_M typically introduces a small additional perplexity penalty (~1 – 3 %) for 14 B coder models; we have not separately re-measured this here because the adapter eval is the canonical signal.

Quick start

llama-server

llama-server \
  -m Qwen2.5-Coder-14B-Instruct-Q4_K_M.gguf \
  --host 0.0.0.0 --port 8080 \
  --ctx-size 4096 --n-gpu-layers 999

Ollama Modelfile

FROM ./Qwen2.5-Coder-14B-Instruct-Q4_K_M.gguf
TEMPLATE """{{ if .System }}<|im_start|>system
{{ .System }}<|im_end|>
{{ end }}{{ if .Prompt }}<|im_start|>user
{{ .Prompt }}<|im_end|>
{{ end }}<|im_start|>assistant
"""
PARAMETER stop "<|im_start|>"
PARAMETER stop "<|im_end|>"
PARAMETER num_ctx 4096

llama-cpp-python

from llama_cpp import Llama

llm = Llama(
    model_path="Qwen2.5-Coder-14B-Instruct-Q4_K_M.gguf",
    n_ctx=4096,
    n_gpu_layers=999,
)
print(llm.create_chat_completion(messages=[
    {"role": "user", "content": "Write a Go function that reverses a UTF-8 string."},
])["choices"][0]["message"]["content"])

Limitations

• Lossy quantization. Q4_K_M is a 4-bit-mixed format; expect minor degradation vs. the unquantized adapter on long-form code.
• No safety tuning. Inherits all caveats from the source adapter.
• Single quant shipped. If you need Q5_K_M / Q8_0 / F16, regenerate with python -m src.phase5_deployment.scripts.launch_convert --quants Q5_K_M Q8_0.

Reproducibility

set -a && source .env && set +a
python -m src.phase5_deployment.scripts.launch_convert \
    --config src/config/v6_config.yaml --wait

• Code: github.com/cmndcntrlcyber/code-trainer-offsec-pipeline (src/phase5_deployment/)
• Cost: ~$2 on a100-large once the job runs.