GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
adimunot
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
adimunot
Model Tree
Qwen/Qwen2.5-1.5B
Input Modalities
Output Modalities
Supported Functionality
| GSM8K | MATH | |
|---|---|---|
| SFT (before) | 39.0% | 40.0% |
| **GRPO (this) ** | 5.5% | 5.5% |
On a single GPU we couldn't run vLLM rollouts, so GRPO used HF generation that produced fixed-length completions which never stopped at the answer. The verifiable reward (which reads the last boxed value) then frequently graded correct solutions as wrong, so GRPO optimized a corrupted signal and collapsed the policy. The lesson — GRPO is only as good as its reward's fidelity — is the most useful outcome of the run.
Full post-mortem: https://github.com/adimunot21/mathnano/blob/main/RESULTS.md