tinker-rl-bench-scale_gsm8k_qwen3-8b – AI Model by arvindcr4 | AlphaNeural AI

You can deploy this model and start earning money today!

arvindcr4/
tinker-rl-bench-scale_gsm8k_qwen3-8b

like

0

reinforcement-learning

tinker-rl-bench

Views

No views yet

scale_gsm8k_qwen3-8b

GRPO experiment from TinkerRL-Bench world-class experiment suite.

Training Details

Base model: Qwen/Qwen3-8B
Method: GRPO (Group Relative Policy Optimization)
Platform: Tinker API v0.18.1
Task: gsm8k
Seed: 42
LoRA rank: 32
Learning rate: 3e-05
Group size: 8
Steps: 30

Results

First-5 avg reward: 31.2%
Last-10 avg reward: 34.4%
Peak reward: 62.5%
Zero-loss steps: 7%
Tinker Run ID: 3154dee5-4fd7-59b6-ba3b-721eef675bfc:train:0

Reward Trace

[
  0.375,
  0.0625,
  0.5625,
  0.1875,
  0.375,
  0.3125,
  0.125,
  0.125,
  0.1875,
  0.125,
  0.25,
  0.4375,
  0.625,
  0.0,
  0.4375,
  0.0625,
  0.1875,
  0.0,
  0.375,
  0.3125,
  0.3125,
  0.1875,
  0.375,
  0.375,
  0.5,
  0.25,
  0.4375,
  0.1875,
  0.625,
  0.1875
]

Citation

bibtex

@misc{tinker-rl-bench-2026,
  title={TinkerRL-Bench: A Unified Benchmark for RL Post-Training},
  author={Arvind C R and Sandhya Jeyaraj and Madhu Kumara L and Mohammad Rafi and Dhruva N Murthy and Arumugam K},
  year={2026},
  url={https://github.com/arvindcr4/tinker-rl-lab}
}