README.md

metadata

title: JuliaSLM
emoji: 🏛️
colorFrom: purple
colorTo: blue
sdk: docker
app_port: 7860
pinned: false
license: mit
tags:
  - julia
  - lux
  - slm
  - philosophy
  - openai-compatible
  - bpe
  - rope
  - rmsnorm
  - swiglu

JuliaSLM

A decoder-only transformer (RoPE, RMSNorm, SwiGLU) trained on classical philosophy texts, implemented in Julia with Lux.jl. Serves an OpenAI-compatible API with streaming support.

Endpoints

• GET / — Health check and model info
• GET /v1/models — List available models
• POST /v1/chat/completions — Generate text (supports streaming, top-k, top-p)

Usage

# Non-streaming
curl -X POST https://your-space.hf.space/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"messages": [{"role": "user", "content": "the nature of"}], "max_tokens": 200}'

# Streaming
curl -X POST https://your-space.hf.space/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"messages": [{"role": "user", "content": "the nature of"}], "stream": true, "temperature": 0.7, "top_k": 40}'

Architecture

• Model: ~5M params, 256d embed, 6 layers, 4 heads
• Tokenizer: BPE (2000 tokens)
• Framework: Lux.jl (explicit parameter/state management)
• Positional encoding: Rotary Position Embeddings (RoPE)
• Normalization: RMSNorm (pre-norm)
• Feed-forward: SwiGLU activation
• Weight tying: Shared embedding/output projection
• Inference: CPU-only, no Lux dependency at runtime (pure NNlib)

Required HF Model Repo Files

Upload these to LisaMegaWatts/JuliaSLM (or set HF_REPO env var):

• final.jld2 — Trained model checkpoint (parameters)
• config.toml — Model architecture configuration (from 5m.toml)
• vocab.json — BPE vocabulary (dict format: {"token": id, ...})
• merges.txt — BPE merge rules

Environment Variables

• HF_REPO — HuggingFace model repo (default: LisaMegaWatts/JuliaSLM)
• PORT — Server port (default: 7860)