File size: 1,409 Bytes
1164913 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 | ---
license: other
library_name: transformers
tags: [meta-reasoning, rl, grpo, qwen3_5, layer-audit, research-checkpoint]
---
# mr9b meta-reasoning RL — layer-audit — 8-node/215 checkpoints
Mid-training research checkpoints from a fully-async GRPO run with the
**pipelined frozen-E layer audit** enabled (`meta_reasoning.layer_audit.enable=true`).
- **Base:** Qwen3.5 (~9B hybrid, `model_type: qwen3_5`, 32 layers)
- **Method:** fully-async GRPO meta-reasoning RL (VERL fully_async_policy)
- **Layer audit:** after each completed reasoning layer, a frozen executor audits it
(classify explorations as new/refinement/verification/duplicate, diagnose tunnel
vision + gaps) and injects one-step control feedback into the next MR step.
- **Topology 215:** 2 trainer / 1 rollout / 5 frozen-E nodes (8 nodes, TP8/PP1/DP2)
- **Experiment:** `mr9b_layeraudit_8node_215_20260718_231300`
- **W&B project:** `meta_reason_rl_h100_5node`
## Contents
Each subfolder is a standalone HF model (weights + tokenizer + chat template):
- `global_step_68/`
- `global_step_69/`
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
m = AutoModelForCausalLM.from_pretrained("aviralku/mr9b-layeraudit-8node-215", subfolder="global_step_69")
t = AutoTokenizer.from_pretrained("aviralku/mr9b-layeraudit-8node-215", subfolder="global_step_69")
```
_Private research checkpoints; not an official release._
|