tinyllms/qwen2.5-7b-instruct-sft-aime-grpo

Qwen2.5-7B-Instruct SFT+GRPO — AIME

GRPO-trained from tinyllms/qwen2.5-7b-instruct-sft-aime (itself SFT'd from Qwen/Qwen2.5-7B-Instruct) using QLoRA (4-bit NF4 quantization + LoRA adapters).

Training Configuration

• Algorithm: GRPO (Group Relative Policy Optimization)
• Learning rate: 5e-6
• Batch size: 2 per device
• Group size: 2 (completions per prompt)
• Max steps: 20
• Max completion length: 2048
• Precision: bf16
• Gradient checkpointing: enabled

QLoRA

• Quantization: 4-bit NF4 with double quantization
• LoRA rank: 64
• LoRA alpha: 128
• LoRA dropout: 0.05
• Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj

Reward Functions

• correctness_reward: task-specific answer verification (binary 0/1)
• format_reward: checks for proper <answer> tag formatting, penalises excessive length

Dataset

Trained on tinyllms/aime-1983-2023-trajectories (204 prompts).

Infrastructure

• GPU: NVIDIA H100 80GB (2 workers, STRICT_PACK)
• Framework: TRL + Ray Train (DDP)
• Tracking: Weights & Biases (project: pocket-sheet-grpo)
• Ray Job ID: raysubmit_iHjwAgfrbFLnsCgN