Instructions to use GulkoA/stilt.1-124m-it with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use GulkoA/stilt.1-124m-it with Transformers:

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

pipe = pipeline("text-generation", model="GulkoA/stilt.1-124m-it", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("GulkoA/stilt.1-124m-it", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use GulkoA/stilt.1-124m-it with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "GulkoA/stilt.1-124m-it"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "GulkoA/stilt.1-124m-it",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/GulkoA/stilt.1-124m-it

SGLang

How to use GulkoA/stilt.1-124m-it 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 "GulkoA/stilt.1-124m-it" \
    --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": "GulkoA/stilt.1-124m-it",
		"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 "GulkoA/stilt.1-124m-it" \
        --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": "GulkoA/stilt.1-124m-it",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use GulkoA/stilt.1-124m-it with Docker Model Runner:
```
docker model run hf.co/GulkoA/stilt.1-124m-it
```

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

stilt.1-124m-it

Instruction-tuned variant of stilt.1-124m — a 124M research language model with a custom attention mechanism. Fine-tuned on smol-smoltalk with a plain-text chat template.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "GulkoA/stilt.1-124m-it"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, trust_remote_code=True)

prompt = "<|user|>\nWhat is the capital of France?\n<|assistant|>\n"
ids = tok(prompt, return_tensors="pt")
out = model.generate(**ids, max_new_tokens=60, do_sample=False)
print(tok.decode(out[0, ids.input_ids.shape[1]:], skip_special_tokens=True))

The chat template is plain text: <|user|>\n{message}\n<|assistant|>\n, turns separated by the GPT-2 end-of-text token.

Family

model	params	notes
stilt.1-124m	124M	base
stilt.1-124m-it	124M	this repo
stilt.1-355m	355M	GulkoA/stilt.1-355m
stilt.1-355m-it	355M	GulkoA/stilt.1-355m-it

Endpoint (OpenAI-compatible bodies)

Deployed as an Inference Endpoint (Default container, handler.py), the model accepts OpenAI chat-completion request bodies at the endpoint URL and returns OpenAI-shaped responses:

curl <ENDPOINT_URL> -H "Authorization: Bearer <HF_TOKEN>" \
  -H "Content-Type: application/json" -d '{
    "messages": [{"role": "user", "content": "What is the capital of France?"}],
    "max_tokens": 60, "temperature": 0.7}'

(The classic HF {"inputs": ..., "parameters": ...} format also still works. Note: the OpenAI SDK appends /v1/chat/completions to its base URL, which the handler container does not route — use plain HTTP as above, or any client that posts to the URL you give it.)

Token limits: none enforced — the tokenizer does not truncate and generation length is caller-controlled. The model was trained at context 1024; beyond that, positions saturate and it keeps generating with gradually degrading quality (attention itself has no length limit).

Research artifact — small model, minimal alignment; expect 124M-class limitations.

Downloads last month: 54

Safetensors

Model size

0.1B params

Tensor type

F32

Space using GulkoA/stilt.1-124m-it 1

Collection including GulkoA/stilt.1-124m-it

Stilt

Collection

Family of models using sharp attention mechanism • 5 items • Updated 5 days ago