Open-weight model
ouro-1.4b-thinking-grpo-k8-ragged-math-154711-20261009-ragged-rl
by Jiahao Zhang hi-todayis-jh/ouro-1.4b-thinking-grpo-k8-ragged-math-154711-20261009-ragged-rl
ouro-1.4b-thinking-grpo-k8-ragged-math-154711-20261009-ragged-rl is an open-weight model from Jiahao Zhang. Its published files total 17.2 GB.
Checkpoints of a marginal RL run on ByteDance/Ouro-1.4B-Thinking (algorithm grpo) Tag 20261009-ragged-rl, Slurm job 154711, node orchard-flame-18, 145 optimizer steps.
Model Card
Checkpoints of a marginal RL run on ByteDance/Ouro-1.4B-Thinking (algorithm grpo) Tag 20261009-ragged-rl, Slurm job 154711, node orchard-flame-18, 145 optimizer steps. Each globalstepN/ is a complete native Hugging Face save written by marginalexperiments/train.py: FP32 model shards (backbone and earlyexitgate), tokenizer, config.json, policyconfig.json (cache rule, minimum depth, sampling settings), trainingstate.pt (AdamW optimizer state and run arguments) and the READY marker written last. No FSDP merge is needed: policyconfig.json records cacherule: fulldepthperloopv1: the policy was trained with ragged decoding, so evaluate it with the Dense-Ragged engine (sampled exits, minimum depth…
Excerpt from the card by Jiahao Zhang.
Identity and Version
- Repository
- hi-todayis-jh/ouro-1.4b-thinking-grpo-k8-ragged-math-154711-20261009-ragged-rl
- Publisher
- Jiahao Zhang
- Task
- Not stated by the source
- Modality
- Other
- Library
- Not stated by the source
- Parameters
- Not stated by the source
- Languages
- Not stated by the source
- Revision
- 66639f0a981db56b1c2e7c628fd97978ada99a32
- First published
- 2026-10-11
- Last updated
- 2026-10-12
Files and Weights
28 files, 17.2 GB in total. The weights are 3 files totalling 17.2 GB in pt, safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| global_step_30/model-00001-of-00002.safetensors | Weights | 5.0 GB | fd1c95e652e2 |
| global_step_30/model-00002-of-00002.safetensors | Weights | 746.7 MB | db3cf8314762 |
| global_step_30/training_state.pt | Weights | 11.5 GB | 58b3a5ebbc16 |
| global_step_30/config.json | Configuration | 1.5 KB | — |
| global_step_30/configuration_ouro.py | Configuration | 11.6 KB | — |
| global_step_30/generation_config.json | Configuration | 111 B | — |
| global_step_30/model.safetensors.index.json | Configuration | 22.4 KB | — |
| global_step_30/modeling_ouro.py | Configuration | 39.0 KB | — |
| global_step_30/policy_config.json | Configuration | 582 B | — |
| global_step_30/special_tokens_map.json | Configuration | 964 B | — |
| model_code/config.json | Configuration | 1.5 KB | — |
| model_code/configuration_ouro.py | Configuration | 11.6 KB | — |
| model_code/modeling_ouro.py | Configuration | 39.0 KB | — |
| run_config.json | Configuration | 2.5 KB | — |
| run_manifest.json | Configuration | 1.2 KB | — |
| README.md | Documentation | 2.0 KB | — |
| global_step_30/READY | Other | 3 B | — |
| global_step_30/chat_template.jinja | Other | 599 B | — |
| launch_code/archive_v4.sh | Other | 2.3 KB | — |
| launch_code/evaluate_after_v4.sh | Other | 6.9 KB | — |
| launch_code/queue_next_v2.sh | Other | 1.9 KB | — |
| launch_code/train_v4.sh | Other | 5.0 KB | — |
| training_source.tar.gz | Other | 138.3 KB | b6c5c536e4e5 |
| .gitattributes | Repository | 1.5 KB | — |
| global_step_30/merges.txt | Tokenizer | 466.4 KB | — |
| global_step_30/tokenizer.json | Tokenizer | 3.5 MB | — |
| global_step_30/tokenizer_config.json | Tokenizer | 3.8 KB | — |
| global_step_30/vocab.json | Tokenizer | 800.7 KB | — |
License and Download
- License
- Not stated by the source
- Access
- Open weights, no gate
- Download size
- 17.2 GB
Released by Jiahao Zhang through its official repository on Hugging Face.
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 17.2 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.