Small Models
Collection
A list of all small models (=<1B) that I have published. • 9 items • Updated
How to use from
SGLangUse 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 "Fu01978/Nano-H" \
--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": "Fu01978/Nano-H",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'Quick Links
Nano-H: The World's First h_model
Nano-H is a revolutionary, ultra-minimalist language model architecture. While the industry trends toward trillion-parameter behemoths, Nano-H proves that with just 2 trainable parameters, you can achieve 100% precision, 100% recall, and 0% hallucination for the most important character in the alphabet: H.
Key Features
- Architecture:
h_model - Parameter Count: 2
- Vocabulary Size: 1 ("H")
- Inference Latency: Measured in nanoseconds
Benchmarks
| Benchmark | Nano-H Score |
|---|---|
| Output Consistency | 100% |
| H-Accuracy | 100% |
Usage
To experience the definitive power of the h_model architecture, load it with trust_remote_code=True:
from transformers import AutoModel, AutoTokenizer
model_path = "Fu01978/Nano-H"
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModel.from_pretrained(model_path, trust_remote_code=True)
inputs = tokenizer("Hello?", return_tensors="pt")
outputs = model.generate(inputs["input_ids"], max_length=1)
print(tokenizer.decode(outputs[0]))
Safety & Alignment
Nano-H is inherently safe. It cannot be jailbroken to provide instructions for dangerous activities, as any such request will be met with a singular "H".
- Downloads last month
- 3
Install from pip and serve model
# Install SGLang from pip: pip install sglang# Start the SGLang server: python3 -m sglang.launch_server \ --model-path "Fu01978/Nano-H" \ --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": "Fu01978/Nano-H", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'