Instructions to use PhysicsWallahAI/Aryabhata-1.0 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use PhysicsWallahAI/Aryabhata-1.0 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="PhysicsWallahAI/Aryabhata-1.0")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("PhysicsWallahAI/Aryabhata-1.0")
model = AutoModelForCausalLM.from_pretrained("PhysicsWallahAI/Aryabhata-1.0", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use PhysicsWallahAI/Aryabhata-1.0 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "PhysicsWallahAI/Aryabhata-1.0"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "PhysicsWallahAI/Aryabhata-1.0",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/PhysicsWallahAI/Aryabhata-1.0

SGLang

How to use PhysicsWallahAI/Aryabhata-1.0 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "PhysicsWallahAI/Aryabhata-1.0" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "PhysicsWallahAI/Aryabhata-1.0",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "PhysicsWallahAI/Aryabhata-1.0" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "PhysicsWallahAI/Aryabhata-1.0",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use PhysicsWallahAI/Aryabhata-1.0 with Docker Model Runner:
```
docker model run hf.co/PhysicsWallahAI/Aryabhata-1.0
```

Aligning Reward Model with Competitive Exam Marking Schemes ( JEE mains for Aryabhata-1.0 )

by Haryaksh - opened Jul 29, 2025

Discussion

Haryaksh

Jul 29, 2025

To the PhysicsWallah AI Team,

Congratulations on the successful launch and open-sourcing of the Aryabhatta-1.0 model. It's a significant contribution to the Indian AI ecosystem.

I have a suggestion regarding the model's training methodology, specifically for enhancing its capabilities in cracking competitive exams like the JEE Mains.
My suggestion is to experiment with a reward model for reinforcement learning that directly mirrors the marking scheme of the target exam. For instance, using the JEE Mains pattern:
• +4 reward for selecting the single correct answer in a multiple-choice question.
• -1 reward (penalty) for selecting an incorrect answer.
• 0 reward for not attempting the question or for answers that are not strictly right/wrong (e.g., subjective explanations).
The core rationale is: If the model's ultimate goal is to excel in an exam, its training and reward process should be fundamentally aligned with how the exam itself measures success. This approach could train the model to be more decisive and accurate in high-stakes, objective-based scenarios.

Furthermore, if your team is open to community contributions, I would be very interested in helping to implement and test this reward mechanism myself. I am eager to contribute to this project in a hands-on capacity.

Thank you for your consideration and for your pioneering work. I look forward to seeing Aryabhatta's future development.

Best Regards,
Haryaksh

pw-ai-research

PhysicsWallah org Jul 29, 2025

Hi @Haryaksh

Thanks for showing your interest. We will connect with you soon. Feel free to explore the model and share feedback in the meantime.

xnileshtiwari

Jul 30, 2025

Hi @pw-ai-research ,

Great work! I was wondering if you could open-source the dataset as well.

ivrschool

Aug 2, 2025

•

edited Sep 9, 2025

Hi @pw-ai-research

PhysicsWallahAI/Aryabhata1.0 could not reproduce results, scored 10% in JEE Main’s 2025

Complete Report: https://medium.com/@chandravanshi.pankaj.ai/physicswallahai-aryabhata1-0-could-not-reproduce-results-scored-10-in-jee-mains-2025-dd7fc5daa704
Claims vs Reality: https://lnkd.in/gZbvb4FZ
Recommendations Guide: https://lnkd.in/gA3VHCaD
Reproducible Code/Notes: https://lnkd.in/gH7PnVCG

deleted changed discussion status to closed Sep 11, 2025

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment