Instructions to use SupraLabs/Supra-Mini-v4-2M with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SupraLabs/Supra-Mini-v4-2M with Transformers:

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

pipe = pipeline("text-generation", model="SupraLabs/Supra-Mini-v4-2M")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("SupraLabs/Supra-Mini-v4-2M")
model = AutoModelForCausalLM.from_pretrained("SupraLabs/Supra-Mini-v4-2M")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use SupraLabs/Supra-Mini-v4-2M with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SupraLabs/Supra-Mini-v4-2M"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SupraLabs/Supra-Mini-v4-2M",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/SupraLabs/Supra-Mini-v4-2M

SGLang

How to use SupraLabs/Supra-Mini-v4-2M 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 "SupraLabs/Supra-Mini-v4-2M" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SupraLabs/Supra-Mini-v4-2M",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "SupraLabs/Supra-Mini-v4-2M" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SupraLabs/Supra-Mini-v4-2M",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use SupraLabs/Supra-Mini-v4-2M with Docker Model Runner:
```
docker model run hf.co/SupraLabs/Supra-Mini-v4-2M
```

Awesome

by Datdanboi25 - opened 8 days ago

Discussion

Datdanboi25

8 days ago

Cool implementation, could prob get away with a higher learning rate and more depth over width (would also help with tokenizer size).
Whats ur activation?

LH-Tech-AI

SupraLabs org 8 days ago

Hey thanks!
I use standard HF Transformers with Llama architecture.
See the full code in this repos' files list as train.py 😃

Datdanboi25

8 days ago

•

edited 8 days ago

Ahh sweet, would probably be worth dropping ur intermediate dim multipler to 2.67-3x and investing further into depth, swiglu at 4x is pretty overweighted

Also big fan of the little competition going on, cool to see some action in the small model community, almost tempted to train a competitor haha. What parameter limit are you guys using?