NisargUpadhyay/manim-coder-grpo

Merged vLLM Model

This folder contains a merged vLLM model export produced by the project's merging workflow.

Contents typically include:

• One or more .safetensors shard files (e.g. model-00001-of-00004.safetensors)
• model.safetensors.index.json (index for shards)
• Tokenizer files: tokenizer.json, tokenizer_config.json, special_tokens_map.json
• config.json and generation_config.json

How to push this folder to Hugging Face Hub

1. Install dependencies:

pip install huggingface-hub

2. From the repository root run the helper script:

python scripts/push_to_hf.py --repo-id YOUR_USERNAME/YOUR_MODEL_NAME
# or with env token:
HF_TOKEN=xxx python scripts/push_to_hf.py --repo-id YOUR_USERNAME/YOUR_MODEL_NAME --private

Loading the model

With vLLM (recommended for inference server):

from vllm import Model
model = Model.from_pretrained("YOUR_USERNAME/YOUR_MODEL_NAME")
# then use vLLM APIs to run inference

With Hugging Face Transformers (if applicable):

from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("YOUR_USERNAME/YOUR_MODEL_NAME")
model = AutoModelForCausalLM.from_pretrained("YOUR_USERNAME/YOUR_MODEL_NAME", trust_remote_code=True)

Notes

• For large .safetensors shards, ensure you have sufficient bandwidth and storage.
• If you plan to host the model publicly, review license and privacy requirements.
• If you encounter upload limits, generate an access token with appropriate scopes and pass it via HF_TOKEN or --token.

If you want me to also commit the repo on Hugging Face with git lfs (instead of API uploads), say so and I can add an alternative script.