32 lines
735 B
Markdown
32 lines
735 B
Markdown
|
---
|
||
|
|
base_model: meta-llama/Llama-3.2-3B-Instruct
|
||
|
|
library_name: transformers
|
||
|
|
tags:
|
||
|
|
- llama
|
||
|
|
- verl
|
||
|
|
- fsdp
|
||
|
|
---
|
||
|
|
|
||
|
|
# hill_8k_300
|
||
|
|
|
||
|
|
This repository contains the `global_step_300` actor checkpoint from the HiLL
|
||
|
|
8k run, converted from verl FSDP shards to Hugging Face Transformers format.
|
||
|
|
|
||
|
|
## Source
|
||
|
|
|
||
|
|
- Base model: `meta-llama/Llama-3.2-3B-Instruct`
|
||
|
|
- Training run: `HiLL-Llama-3.2-3B-Instruct-8k`
|
||
|
|
- Checkpoint: `global_step_300/actor`
|
||
|
|
- Conversion backend: verl FSDP model merger
|
||
|
|
|
||
|
|
## Loading
|
||
|
|
|
||
|
|
```python
|
||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
|
import torch
|
||
|
|
|
||
|
|
model_id = "sagnikM/hill_8k_300"
|
||
|
|
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
||
|
|
model = AutoModelForCausalLM.from_pretrained(model_id, dtype=torch.bfloat16)
|
||
|
|
```
|