---
library_name: transformers
pipeline_tag: text-generation
tags:
- verl
- math
- grpo
- transfer
- qwen2
- 3b
- 7bi-to-3bi
- pool7to1
---

# VERL Math Transfer 7B to 3B fix07 pool7to1

Math transfer experiment trained with verl. This repo groups all exported Hugging Face checkpoints for the 7B-to-3B fix_0_7 pool7to1 configuration.

## Layout

- `main`: latest exported checkpoint, currently `step-070`
- step revisions: `step-010, step-020, step-030, step-040, step-050, step-060, step-070`

## Usage

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

repo_id = "hyunseoki/verl-math-transfer-7bi-to-3bi-fix07-pool7to1"
tokenizer = AutoTokenizer.from_pretrained(repo_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(repo_id, trust_remote_code=True)
```

Load a specific checkpoint revision:

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

repo_id = "hyunseoki/verl-math-transfer-7bi-to-3bi-fix07-pool7to1"
revision = "step-070"
tokenizer = AutoTokenizer.from_pretrained(repo_id, revision=revision, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(repo_id, revision=revision, trust_remote_code=True)
```

## Notes

- Architecture detected from the exported config: `Qwen2ForCausalLM`
- The original base model Hub ID is not encoded in these local checkpoints, so `base_model` metadata is not set automatically.
- Checkpoints were exported from verl FSDP shards into Hugging Face `safetensors` format.