Instructions to use bharatgenai/Param2-17B-A2.4B-Thinking with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use bharatgenai/Param2-17B-A2.4B-Thinking with Transformers:

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

pipe = pipeline("text-generation", model="bharatgenai/Param2-17B-A2.4B-Thinking", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("bharatgenai/Param2-17B-A2.4B-Thinking", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use bharatgenai/Param2-17B-A2.4B-Thinking with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "bharatgenai/Param2-17B-A2.4B-Thinking"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "bharatgenai/Param2-17B-A2.4B-Thinking",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/bharatgenai/Param2-17B-A2.4B-Thinking

SGLang

How to use bharatgenai/Param2-17B-A2.4B-Thinking 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 "bharatgenai/Param2-17B-A2.4B-Thinking" \
    --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": "bharatgenai/Param2-17B-A2.4B-Thinking",
		"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 "bharatgenai/Param2-17B-A2.4B-Thinking" \
        --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": "bharatgenai/Param2-17B-A2.4B-Thinking",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use bharatgenai/Param2-17B-A2.4B-Thinking with Docker Model Runner:
```
docker model run hf.co/bharatgenai/Param2-17B-A2.4B-Thinking
```

Benchmark comparison with stronger models

by RhiteshKS - opened Mar 12

Discussion

RhiteshKS

Mar 12

•

edited Mar 12

Hi, thank you for releasing the model and the benchmark results.
I was wondering how Param-2-17B-A2.4B performs when compared with stronger models in the 15B–30B parameter range (for example Qwen2.5-14B/32B, Mixtral, etc.). The models currently shown in the benchmark table appear to be relatively lightweight or distilled variants, which makes it a bit difficult to understand where it stands.
Since Param is a MoE model with ~2.4B active parameters but larger total capacity, it would be interesting to see comparisons either with compute-matched models (like Qwen 2.5-3B) or with capacity-matched dense models in the mid-size range.
Could you share any results or evaluations against such models?
Thank you.

indiaiquest

Apr 3

Check this out - https://medium.com/@indiai/india-first-llms-on-indian-languages-sarvam-30b-and-param2-17b-6676b637f3ab

kundeshwar20

BharatGen AI org May 8

Check it here as well : https://github.com/nikhilvelpanur/constitutional-morality/blob/main/CMI-LEADERBOARD.md

Neural-Hacker

16 days ago

•

edited 16 days ago

It would be great to see Param-2 compared against better open models in the same size class, such as Sarvam-M, Qwen-3 14B, Gemma -3 27B, Mistral-3 14B, and Apriel-v1.6-15B-Thinker. The current benchmarks are not that helpful as the only reliable comparison is gpt-oss-20b, but these comparisons would better reflect the current state of mid-sized open reasoning models. @BharatGen-admin @kundeshwar20

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