transfer-reasoning
Collection
27 items • Updated • 1
TTRL AIME2026 — random
LoRA adapter for Qwen/Qwen3-1.7B, trained with TTRL on AIME2026.
- Arm:
random - Seed: 42
- Training: 2 epochs, final checkpoint
global_step_2 - LoRA rank: 16; alpha: 32
This repository contains adapter weights only. Load with the base model using PEFT. Training and internal validation use the same adaptation problems.
- Downloads last month
- 12