Text Generation
Transformers
Safetensors
English
llama
micro
nano
small
supra
SupraLabs
gtx
rtx
nvidia
lh-tech
axionlab
text-generation-inference
Instructions to use SupraLabs/MicroSupra-1k with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use SupraLabs/MicroSupra-1k with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="SupraLabs/MicroSupra-1k")# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("SupraLabs/MicroSupra-1k") model = AutoModelForCausalLM.from_pretrained("SupraLabs/MicroSupra-1k") - Notebooks
- Google Colab
- Kaggle
- Local Apps
- vLLM
How to use SupraLabs/MicroSupra-1k with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "SupraLabs/MicroSupra-1k" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "SupraLabs/MicroSupra-1k", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/SupraLabs/MicroSupra-1k
- SGLang
How to use SupraLabs/MicroSupra-1k 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/MicroSupra-1k" \ --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/MicroSupra-1k", "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/MicroSupra-1k" \ --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/MicroSupra-1k", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use SupraLabs/MicroSupra-1k with Docker Model Runner:
docker model run hf.co/SupraLabs/MicroSupra-1k
Update README.md
Browse files
README.md
CHANGED
|
@@ -12,12 +12,58 @@ tags:
|
|
| 12 |
- supra
|
| 13 |
- SupraLabs
|
| 14 |
- gtx
|
| 15 |
-
- gpu
|
| 16 |
- rtx
|
| 17 |
- nvidia
|
| 18 |
- llama
|
|
|
|
|
|
|
|
|
|
| 19 |
---
|
| 20 |
|
| 21 |
## **MicroSupra-1k**
|
| 22 |
|
| 23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
- supra
|
| 13 |
- SupraLabs
|
| 14 |
- gtx
|
|
|
|
| 15 |
- rtx
|
| 16 |
- nvidia
|
| 17 |
- llama
|
| 18 |
+
- lhtech
|
| 19 |
+
- axionlab
|
| 20 |
+
library_name: transformers
|
| 21 |
---
|
| 22 |
|
| 23 |
## **MicroSupra-1k**
|
| 24 |
|
| 25 |
+
So... have you ever seen a model that runs on a 3 dollars hardware? No? If no, Now you're seeing!
|
| 26 |
+
|
| 27 |
+
MicroSupra-1k is a 1k parameters model trained on 300M Tokens of FineWeb-Edu for 1 minute(Yes! 59 seconds!) on a GTX 750Ti 4GB(AxionLab Hardware)
|
| 28 |
+
|
| 29 |
+
**Some model outputs:**
|
| 30 |
+
|
| 31 |
+
[*] Prompt: The main concept of physics is
|
| 32 |
+
|
| 33 |
+
[*] Output: The main concept of physics is a,s and the. thet to, theing.... the,a then,c,i to, thee in b. toed.,,e theyalp the in,er thees- s,el,,,,
|
| 34 |
+
and, the of ine,,s the of cs of thesss the. f. to. thesining andor dar,,al the,. of p.
|
| 35 |
+
the.s the.,,s. anded,e. of, ofed, l toinging and themsr the of of. to
|
| 36 |
+
to thes thes aen,., ofes of a.
|
| 37 |
+
|
| 38 |
+
[*] Prompt: My name is
|
| 39 |
+
|
| 40 |
+
[*] Output: My name is ed and. as the, to. the, iningt thee the ofingi in
|
| 41 |
+
the., anda.-eo
|
| 42 |
+
ofles, b the,er,s fing.ssp the the
|
| 43 |
+
, of of, the,al, d to the m, the, to toed,
|
| 44 |
+
seng,,.y. in the,., in and them the thened.sing to
|
| 45 |
+
the of of andan the the,, the
|
| 46 |
+
to..,,sing,,.aring the the. of.al.,s ofcal ar s..e and.sssor of, and and.
|
| 47 |
+
|
| 48 |
+
[*] Prompt: Question: What is the capital of France?\nAnswer:
|
| 49 |
+
|
| 50 |
+
[*] Output: Question: What is the capital of France?
|
| 51 |
+
|
| 52 |
+
Answer:,. and to the. toc. ofs the m,a thee.. the, f ofling. as.,,y bt, the p, in, the,,ees toed ing to.o,
|
| 53 |
+
thes. the..,s the.ed and andang,,ed the of,,ms. of, thei the, the,ey,,s l.ing toe the the,se the to, the, the,aror, the of-. in the. the. the,e the of ds to,ic the the aal at the..
|
| 54 |
+
ingssy s and and
|
| 55 |
+
|
| 56 |
+
**🚫What the model CAN'T do:**
|
| 57 |
+
Think
|
| 58 |
+
Chat
|
| 59 |
+
even predict the next token correctly lol
|
| 60 |
+
|
| 61 |
+
**Why SupraLabs created this???**
|
| 62 |
+
|
| 63 |
+
Because we are experimenting sizes, experiments(like 1Bit quant, distillation(NEW THINGS ARE COMING WITH DISTILLATION! GET TUNED!), pruning, all to better your experience! We are working to big things!)
|
| 64 |
+
|
| 65 |
+
**Final thought**
|
| 66 |
+
|
| 67 |
+
Even without any intelligence, it shows that scaling laws are real. This ant model doesn't know how to talk, but we all know it emotions 🤖🫶
|
| 68 |
+
|
| 69 |
+
SupraLabs are excited to work at more things to you!
|