---
license: apache-2.0
library_name: transformers
base_model: Qwen/Qwen2.5-Math-1.5B
tags:
- math
- reinforcement-learning
- grpo
- checkpoint
---

# od2961/qwen2.5-Math-1.5b-drx-grpo-readmeflash-node302-seed43

This repo stores organized Dr.X-GRPO checkpoints for **seed 43** from our
`Qwen2.5-Math-1.5B` readme-flash math-reasoning training run.

## Layout

- `checkpoints/step_XXXXX/`: exported model checkpoints ready for inference
- `eval_results/`: per-benchmark JSON outputs saved at eval boundaries
- `metadata/checkpoints_index.json`: machine-readable checkpoint manifest
- `metadata/run_manifest.json`: local run provenance

## Source Run

- Local save root: `/n/fs/similarity/maxent-grpo/var/data/oat_zero_drx_grpo_1p5b_20260416_100347_seed43`
- Local run dir: `/n/fs/similarity/maxent-grpo/var/data/oat_zero_drx_grpo_1p5b_20260416_100347_seed43/qwen2.5-Math-1.5b-r1-zero-drx-grpo-node302-seed43_0416T10:04:41`
- Run name: `qwen2.5-Math-1.5b-r1-zero-drx-grpo-node302-seed43_0416T10:04:41`
- Objective: `Dr.X-GRPO`

## Available Checkpoints

| Step | Size (GiB) | Eval files |
| --- | ---: | --- |
| `step_00000` | `2.89` | aime, amc, math, minerva, olympiad_bench |
| `step_00016` | `2.89` | aime, amc, math, minerva, olympiad_bench |
| `step_00032` | `2.89` | aime, amc, math, minerva, olympiad_bench |
| `step_00048` | `2.89` | aime, amc, math, minerva, olympiad_bench |
| `step_00064` | `2.89` | aime, amc, math, minerva, olympiad_bench |
| `step_00080` | `2.89` | aime, amc, math, minerva, olympiad_bench |
| `step_00096` | `2.89` | aime, amc, math, minerva, olympiad_bench |
| `step_00112` | `2.89` | aime, amc, math, minerva, olympiad_bench |
| `step_00128` | `2.89` | aime, amc, math, minerva, olympiad_bench |
| `step_00144` | `2.89` | aime, amc, math, minerva, olympiad_bench |
| `step_00160` | `2.89` | aime, amc, math, minerva, olympiad_bench |
| `step_00176` | `2.89` | aime, amc, math, minerva, olympiad_bench |