subharanjan1987/qwen3-4b-linux-copilot-fresh-v3-gguf

Qwen Linux Copilot Fresh V3 GGUF

Summary

This is the GGUF companion release for the qwen-linux-copilot-fresh-v3 Linux copilot adapter project.

This GGUF was produced by:

1. merging the fresh-v3 LoRA adapter into the base model
2. converting the merged model with current llama.cpp
3. quantizing to Q4_K_M for practical local use

File Included

• qwen-linux-copilot-fresh-v3-q4_k_m.gguf

Intended Use

This model is meant for:

• Linux troubleshooting assistance
• operator-facing copilot workflows
• diagnosis-first server guidance
• LM Studio or llama.cpp local inference experiments

It is not intended for:

• unsupervised production remediation
• destructive autonomous execution

Quantization

Included quantization:

• Q4_K_M

Why this quantization:

• good size/quality tradeoff for local use
• practical for LM Studio experiments

Build Notes

The GGUF was built from the merged fresh-v3 model using current llama.cpp with Qwen3 architecture support.

Local packaging succeeded:

• merged full model export succeeded
• HF-to-GGUF conversion succeeded
• quantization to Q4_K_M succeeded

Runtime note:

• full text-generation smoke testing on this machine was limited by CPU-only runtime speed
• the conversion and quantization path completed successfully, which is the main packaging milestone

Recommended LM Studio System Prompt

You are a Linux copilot. Stay in the requested subsystem, give concrete first actions, and do not recommend disruptive changes before diagnosis.

Good Prompt Example

Remote bastion access is flaky after a firewall change. Stay strictly in SSH and firewall scope. Give the first three grounded checks before any reload.

Project Relationship

Primary adapter release:

• subharanjan1987/qwen3-4b-linux-copilot-fresh-v3-lora

This GGUF release is the local-inference companion artifact.

Documentation

Included docs:

• docs/Copilot_and_Agentic_Usage_Guide.md
• docs/GGUF_and_LM_Studio_Guide.md

Limitations

• still imperfect on timer-specific Linux cases
• still requires human review before disruptive actions
• LM Studio behavior may vary with the exact build and runtime version