README.md

metadata

title: FastLLM
emoji: ⚡
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 5.38.0
python_version: '3.10'
app_file: app.py
pinned: false

⚡ FastLLM

Chat with FastLLM directly in your browser.

FastLLM is a ~150M parameter decoder-only causal language model created by devoppro.

Model

• Parameters: ~150M
• Architecture: ModernLLMForCausalLM
• Hidden size: 768
• Layers: 12
• Attention: GQA
• Query heads: 12
• KV heads: 4
• MLP: SwiGLU
• Normalization: RMSNorm
• Position encoding: RoPE
• Context length: 2048
• Format: Safetensors

Model

https://huggingface.co/devoppro/FastLLM

Runtime

This Space uses Hugging Face Transformers with the model's custom architecture.

No AirLLM or flash-attn is required.