| ══════════════════════════════════════════════════════════════ |
| JARVIS → HUGGINGFACE UPLOAD GUIDE |
| 16 GB RAM · Local inference · No GPU required |
| ══════════════════════════════════════════════════════════════ |
|
|
| WHAT'S IN THIS FOLDER (hf_jarvis/) |
| ──────────────────────────────────────────────────────────── |
| app.py → Gradio UI (entry point for HF Spaces) |
| brain.py → HuggingFace Transformers inference engine |
| memory.py → SQLite memory (facts + conversation history) |
| thinking.py → Background anticipation engine |
| voice_tts.py → Edge-TTS voice output |
| config.py → All settings via environment variables |
| requirements.txt→ All pip packages needed |
| README.md → HuggingFace model card (auto-displayed) |
| .env.example → Local dev environment template |
| .gitignore → Keeps your DB and secrets out of git |
|
|
| ════════════════════════════════════════════════════════════ |
| STEP 1: CREATE A HUGGINGFACE ACCOUNT / LOG IN |
| ════════════════════════════════════════════════════════════ |
| Go to: https://huggingface.co |
| Sign up (free) or log in. |
|
|
| ════════════════════════════════════════════════════════════ |
| STEP 2: CREATE A NEW SPACE |
| ════════════════════════════════════════════════════════════ |
| 1. Click your profile icon → New Space |
| 2. Space name: jarvis (or any name you like) |
| 3. License: MIT |
| 4. SDK: Gradio ← IMPORTANT: select Gradio |
| 5. Hardware: CPU (upgraded) ← free tier is fine for 3.8B 4-bit |
| For Mistral-7B choose "CPU upgrade" or "A10G small" if available |
| 6. Visibility: Private (or Public — your choice) |
| 7. Click: Create Space |
|
|
| ════════════════════════════════════════════════════════════ |
| STEP 3: UPLOAD FILES |
| ════════════════════════════════════════════════════════════ |
| METHOD A — Drag-and-drop in browser (easiest) |
| ────────────────────────────────────────────── |
| 1. Open your Space page → click "Files" tab |
| 2. Click "Add file" → "Upload files" |
| 3. Select ALL files from this folder: |
| app.py |
| brain.py |
| memory.py |
| thinking.py |
| voice_tts.py |
| config.py |
| requirements.txt |
| README.md |
| .env.example |
| .gitignore |
| 4. Click "Commit changes to main" |
| 5. HuggingFace will install requirements.txt and launch app.py |
| automatically. Wait ~2-5 minutes for the build to finish. |
|
|
| METHOD B — Git push (recommended for updates) |
| ─────────────────────────────────────────────── |
| Run in PowerShell from the hf_jarvis\ folder: |
|
|
| git init |
| git remote add origin https://huggingface.co/spaces/<YOUR_USERNAME>/jarvis |
| git add . |
| git commit -m "Initial JARVIS upload" |
| git push origin main |
|
|
| (You may need: git lfs install if you ever add large model files) |
|
|
| ════════════════════════════════════════════════════════════ |
| STEP 4: SET ENVIRONMENT VARIABLES (optional but recommended) |
| ════════════════════════════════════════════════════════════ |
| In your Space page → Settings tab → Variables & Secrets |
|
|
| Add these if you want to customize: |
| JARVIS_MODEL = microsoft/Phi-3-mini-4k-instruct (default) |
| JARVIS_QUANT = 4bit (default) |
| JARVIS_THINKING = true |
|
|
| Leave blank to use the safe defaults already in config.py. |
|
|
| ════════════════════════════════════════════════════════════ |
| STEP 5: WAIT FOR BUILD & OPEN JARVIS |
| ════════════════════════════════════════════════════════════ |
| HuggingFace will show build logs. When it says "Running", |
| your JARVIS is live at: |
|
|
| https://huggingface.co/spaces/<YOUR_USERNAME>/jarvis |
|
|
| First response takes ~30-90 seconds (model loading). |
| Every subsequent message is fast. |
|
|
| ════════════════════════════════════════════════════════════ |
| WHAT JARVIS CAN DO ON HUGGINGFACE |
| ════════════════════════════════════════════════════════════ |
| ✅ Chat with 3.8B / 7B local model — no external API costs |
| ✅ Voice replies (Edge-TTS, Microsoft GuyNeural voice) |
| ✅ Remember facts: type "remember name is Tony" |
| ✅ Check status: type "status" |
| ✅ See predictions: type "anticipate" |
| ✅ Persistent memory across sessions (SQLite DB, local to the Space) |
| ✅ Runs on 16 GB RAM — Phi-3-mini 4-bit uses only ≈3-4 GB |
|
|
| ════════════════════════════════════════════════════════════ |
| RAM & MODEL OPTIONS |
| ════════════════════════════════════════════════════════════ |
| Your 16 GB gives you a LOT of headroom: |
|
|
| Model │ Quant │ RAM used |
| ───────────────────────────────────┼───────┼────────── |
| Phi-3-mini-4k-instruct (3.8B) │ 4-bit │ ~3-4 GB ← default (fast) |
| Phi-3-mini-4k-instruct (3.8B) │ fp16 │ ~7.6 GB |
| Mistral-7B-Instruct-v0.2 (7B) │ 4-bit │ ~4-5 GB ← deep reasoning |
| Mistral-7B-Instruct-v0.2 (7B) │ fp16 │ ~14 GB ← max quality |
|
|
| To switch: set JARVIS_MODEL in Space Settings. |
|
|
| ════════════════════════════════════════════════════════════ |
| TROUBLESHOOTING |
| ════════════════════════════════════════════════════════════ |
| "Model loading..." forever |
| → First load downloads from HF Hub. Wait 3-5 minutes. |
| The model (~2-7 GB) downloads once and is cached. |
|
|
| "bitsandbytes not available" |
| → 4-bit quant needs a recent version. Set JARVIS_QUANT=none |
| if bitsandbytes fails (fp32, uses more RAM but works). |
|
|
| Out of memory |
| → Set JARVIS_QUANT=4bit (default) to reduce RAM to ~3-4 GB. |
|
|
| Voice not working |
| → edge-tts needs internet. On fully offline HF Spaces, |
| TTS falls back silently to text-only — chat still works. |
|
|
| ══════════════════════════════════════════════════════════════ |
| SIR — YOUR JARVIS IS READY FOR THE WORLD |
| ══════════════════════════════════════════════════════════════ |
|
|
