Instructions to use SupraLabs/Supra2-100M-Base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use SupraLabs/Supra2-100M-Base with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="SupraLabs/Supra2-100M-Base") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("SupraLabs/Supra2-100M-Base") model = AutoModelForCausalLM.from_pretrained("SupraLabs/Supra2-100M-Base", 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]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use SupraLabs/Supra2-100M-Base with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "SupraLabs/Supra2-100M-Base" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "SupraLabs/Supra2-100M-Base", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/SupraLabs/Supra2-100M-Base
- SGLang
How to use SupraLabs/Supra2-100M-Base 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/Supra2-100M-Base" \ --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": "SupraLabs/Supra2-100M-Base", "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 "SupraLabs/Supra2-100M-Base" \ --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": "SupraLabs/Supra2-100M-Base", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use SupraLabs/Supra2-100M-Base with Docker Model Runner:
docker model run hf.co/SupraLabs/Supra2-100M-Base
Question
How many total steps?
114 440
lr?
I don't remember right now.
Please don't ask me questions right now. This isn't finished and i will release it once it's done. Stay tuned 🚀
100K steps ONLY? for a 100M, dang...
Either way, good luck! very excited to see a new model from SupraLabs since like... a lot of time haha
100K steps ONLY? for a 100M, dang...
Either way, good luck! very excited to see a new model from SupraLabs since like... a lot of time haha
Thank you 🤗
100K steps ONLY? for a 100M, dang...
Either way, good luck! very excited to see a new model from SupraLabs since like... a lot of time hahaIt's not the total steps i think, it is the current step.
Nope, it's total steps
lr?
0.001, but there's warming (warmup_ratio=0.0175)
That's not even correct 😭😭😭
lr?
0.001, but there's warming (warmup_ratio=0.0175)
That's not even correct 😭😭😭
ik
Benchmarking access?
bro... that's barely any progress. The 100M model should be able to beat our 50M on all benchmarks.
Why did my comment get marked as low quality if someone just asking for benchmarks was retained?
Why did my comment get marked as low quality if someone just asking for benchmarks was retained?
Maybe the HF system flagged it as the R word. Lol sometimes AI just go wild, but it could be someone on SupraLabs too lol
Why did my comment get marked as low quality if someone just asking for benchmarks was retained?
Maybe the HF system flagged it as the R word. Lol sometimes AI just go wild, but it could be someone on SupraLabs too lol
We didn't hided nothing.
Dang.... 1b model soon?
Dang.... 1b model soon?
dont think so not alot compute
I mean... maybe like a B200 for like 10 hours, that's like 70 bucks, but if yall want, its alays an option
I mean... maybe like a B200 for like 10 hours, that's like 70 bucks, but if yall want, its alays an option
yea but for chinchilla its 10 hours but for overtrain to actually good more than 500$
I mean... maybe like a B200 for like 10 hours, that's like 70 bucks, but if yall want, its alays an option
but a 200M that i would like
also where did Qyrou 1 go? its gone from thhe org
I mean... maybe like a B200 for like 10 hours, that's like 70 bucks, but if yall want, its alays an option
yea but for chinchilla its 10 hours but for overtrain to actually good more than 500$
nvm not that expensive,
For 1B with 100B tokens
Compute: 6 × 1e9 × 100e9 = 6e20 FLOPs
Time on a single B300:
BF16, 40% MFU (6 PFLOPS effective): 6e20 / 6e15 ≈ 28 hours (1.15 days)
FP8, 35% MFU (10 PFLOPS effective): 6e20 / 10e15 ≈ ~17 hours
also where did Qyrou 1 go? its gone from thhe org
need to do a bit more testing
also @Banaxi-Tech , u should prop join the SFZ discord server. thats the place we all hangout lol, its only u who's not there
