Instructions to use theschoolofai/LightningLM-0.1V-2B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use theschoolofai/LightningLM-0.1V-2B with Transformers:

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

pipe = pipeline("text-generation", model="theschoolofai/LightningLM-0.1V-2B")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("theschoolofai/LightningLM-0.1V-2B", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use theschoolofai/LightningLM-0.1V-2B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "theschoolofai/LightningLM-0.1V-2B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "theschoolofai/LightningLM-0.1V-2B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/theschoolofai/LightningLM-0.1V-2B

SGLang

How to use theschoolofai/LightningLM-0.1V-2B 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 "theschoolofai/LightningLM-0.1V-2B" \
    --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": "theschoolofai/LightningLM-0.1V-2B",
		"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 "theschoolofai/LightningLM-0.1V-2B" \
        --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": "theschoolofai/LightningLM-0.1V-2B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use theschoolofai/LightningLM-0.1V-2B with Docker Model Runner:
```
docker model run hf.co/theschoolofai/LightningLM-0.1V-2B
```

LightningLM 0.1V — weights-only staged release

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

Access to LightningLM 0.1V weights requires acceptance of the Apache 2.0 license.

This is a weights-only staged release: runnable inference code, model configuration, and custom kernels will be released as a coordinated update. AutoModelForCausalLM.from_pretrained() will not load this model until that update.

This work spans multiple LightningLM papers (tokenizer, Kronecker embeddings, recurrence/MoE architecture). The full citation list will be provided alongside the runnable-code release [URL TBD]. Please cite all relevant LightningLM papers in any derived work.

LightningLM-0.1V-2B

LightningLM 0.1V — pretrained 1.78B (dense) language model.

Staged release notice

This repository is a weights-only staged release. It contains model weights and tokenizer only. Runnable inference code, model configuration, and custom kernels will be released as a coordinated update; AutoModelForCausalLM.from_pretrained() will not load this model until that update. Weights are provided now for transparency and archival.

Architecture (one-liner)

LightningLM 0.1V: byte-level Kronecker product input embeddings + hybrid attention (gated-delta-net linear attention interleaved with gated sparse attention); vocabulary 131,072 (TSAI 131K BPE tokenizer); Multi-Token Prediction (MTP) head; multi-stream recurrence.

Total parameters: 1,781,570,624 (1.78B); all parameters active per token (dense)
Precision: bf16 (no upcasting in this release)

Files

model-*.safetensors + model.safetensors.index.json — bf16 weights, sharded
tokenizer.json, tokenizer_config.json, special_tokens_map.json — TSAI 131K tokenizer
LICENSE — Apache 2.0

Training data

(To be expanded.) Multi-lingual corpus covering English plus several Indic scripts (Bengali, Devanagari, Gujarati, Gurmukhi, Kannada, Malayalam, Odia, Tamil, Telugu). Full data description will be released with the accompanying papers.

Citation

This work spans multiple LightningLM papers (tokenizer, Kronecker embeddings, architecture, training methodology). The full list will be linked at the runnable-code release. Until then please cite the Kronecker embeddings paper:

@article{lightninglm_kronecker_2605.29459,
  title  = {Byte-Level Kronecker Product Embeddings},
  author = {TheSchoolOfAI},
  year   = {2026},
  eprint = {2605.29459},
  archivePrefix = {arXiv}
}

License

Apache 2.0. Access requires acceptance of the license and a request via the HuggingFace gated-model flow. Please cite the LightningLM papers in any derived work.

Downloads last month: -; Downloads are not tracked for this model. How to track

Safetensors

Model size

2B params

Tensor type

BF16

Paper for theschoolofai/LightningLM-0.1V-2B

Kronecker Embeddings: Byte-Level Structured Token Representations for Parameter-Efficient Language Models

Paper • 2605.29459 • Published 4 days ago