80 9 months ago

Qwen 2.5 Coder 7b Finetuned on MainframeBench for Cobol Coding

tools
ollama run sammcj/qwen2.5-cobol-coder-7b-instruct:q6_k

Applications

Claude Code
Claude Code ollama launch claude --model sammcj/qwen2.5-cobol-coder-7b-instruct:q6_k
Codex
Codex ollama launch codex --model sammcj/qwen2.5-cobol-coder-7b-instruct:q6_k
OpenCode
OpenCode ollama launch opencode --model sammcj/qwen2.5-cobol-coder-7b-instruct:q6_k
OpenClaw
OpenClaw ollama launch openclaw --model sammcj/qwen2.5-cobol-coder-7b-instruct:q6_k

Models

View all →

Readme

SFT+DPO trained on dataset: https://huggingface.co/datasets/Fsoft-AIC/MainframeBench/

Training Config

  • Training Set Size: 2,270
  • Test Set Size: 253
  • learning_rate_multiplier: 1
  • num_epochs: 4
  • batch_size: auto
  • preference_tuning_variant: APO Zero
  • preference_tuning_learning_rate_multiplier: 1
  • preference_tuning_num_epochs: 1
  • preference_tuning_training_beta: 0.1
  • preference_tuning_adapter_weight: 0.5