Instructions to use AxionLab-official/MiniBot-0.9M-Base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AxionLab-official/MiniBot-0.9M-Base with Transformers:

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

pipe = pipeline("text-generation", model="AxionLab-official/MiniBot-0.9M-Base")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("AxionLab-official/MiniBot-0.9M-Base")
model = AutoModelForCausalLM.from_pretrained("AxionLab-official/MiniBot-0.9M-Base", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use AxionLab-official/MiniBot-0.9M-Base with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AxionLab-official/MiniBot-0.9M-Base"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AxionLab-official/MiniBot-0.9M-Base",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/AxionLab-official/MiniBot-0.9M-Base

SGLang

How to use AxionLab-official/MiniBot-0.9M-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 "AxionLab-official/MiniBot-0.9M-Base" \
    --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": "AxionLab-official/MiniBot-0.9M-Base",
		"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 "AxionLab-official/MiniBot-0.9M-Base" \
        --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": "AxionLab-official/MiniBot-0.9M-Base",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use AxionLab-official/MiniBot-0.9M-Base with Docker Model Runner:
```
docker model run hf.co/AxionLab-official/MiniBot-0.9M-Base
```

Card param count is ~9.5% low: 985,728 in the checkpoint vs ~900K on the card

by Compactbot - opened about 15 hours ago

Discussion

Compactbot

about 15 hours ago

Hi! I was auditing parameter counts on tiny models and wanted to flag a small discrepancy here.

The card says ~900K / ~0.9M parameters, but the actual checkpoint holds 985,728 — closer to a 1M model.

What I checked:

model.safetensors: 52 tensors, all F32, 985,728 stored elements
config.json (GPT2LMHeadModel, vocab 1375, n_embd 128, 4 layers, n_head 4, n_positions 128, n_inner null→512, tie_word_embeddings: true) computes to exactly 985,728:
- wte 176,000 + wpe 16,384 + 4 × 198,272 (per-layer) + ln_f 256
File size 3,947,952 B = 8 + 5032 (header) + 985,728 × 4 — exact
Only wte present, no separate lm_head (properly tied) ✓

So the artifact is internally consistent; it's just that the card's "~900K" under-reports by 85,728 (about 9.5%). The model is really ~0.99M.

Happy to open a one-line PR to update the card if you'd like — just say the word. Otherwise no action needed, this is a minor cosmetic thing.

AxionLab-official

Owner about 7 hours ago

Hi! I was auditing parameter counts on tiny models and wanted to flag a small discrepancy here.

The card says ~900K / ~0.9M parameters, but the actual checkpoint holds 985,728 — closer to a 1M model.

What I checked:

model.safetensors: 52 tensors, all F32, 985,728 stored elements

config.json (GPT2LMHeadModel, vocab 1375, n_embd 128, 4 layers, n_head 4, n_positions 128, n_inner null→512, tie_word_embeddings: true) computes to exactly 985,728:

wte 176,000 + wpe 16,384 + 4 × 198,272 (per-layer) + ln_f 256

File size 3,947,952 B = 8 + 5032 (header) + 985,728 × 4 — exact

Only wte present, no separate lm_head (properly tied) ✓

So the artifact is internally consistent; it's just that the card's "~900K" under-reports by 85,728 (about 9.5%). The model is really ~0.99M.

Happy to open a one-line PR to update the card if you'd like — just say the word. Otherwise no action needed, this is a minor cosmetic thing.

Wow! Thanks for pointing that, im going to change this.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment