Qwen2.5-1.5B-Indonesian-Assistant-GRPO/README.md

base_model, tags, license, language

base_model	tags	license	language
kendrickfff/Qwen2.5-1.5B-Indonesian-Assistant	text-generation-inference transformers unsloth qwen2	apache-2.0	en

Uploaded finetuned model

• Developed by: kendrickfff
• License: apache-2.0
• Finetuned from model : kendrickfff/Qwen2.5-1.5B-Indonesian-Assistant

Qwen2.5-1.5B Indonesian Assistant (GRPO)

Training Method

• Type: Group Relative Policy Optimization (GRPO)
• Base Model: kendrickfff/Qwen2.5-1.5B-Indonesian-Assistant (SFT model)
• Steps: 100
• Reward Functions: 4 (format, reasoning length, correctness, language)
• Key Difference: Model learns to use ... reasoning tags

This qwen2 model was trained 2x faster with Unsloth and Huggingface's TRL library.