Qwen3.5-0.8B-grok

Fine-tuned version of Qwen/Qwen3.5-0.8B for Grok parser pattern generation.

Description

Given a raw log line, generates the appropriate parse_grok!() pattern that extracts structured fields.

Training Details

• Base Model: Qwen/Qwen3.5-0.8B
• Method: LoRA (r=64, alpha=64) + 4-bit quantization
• Training Samples: 646973
• Epochs: 1
• Learning Rate: 0.0002
• Final Training Loss: 0.36663773090502155
• Training Time: 422.1 minutes
• Hardware: NVIDIA H100 80GB
• Framework: Unsloth + TRL

Usage

Model card will be updated with benchmark results after evaluation.