Initial SurvivalAI Pro deploy

README.md

@@ -1,16 +1,41 @@
 ---
-title: Survival Model II
-emoji: 💬
-colorFrom: yellow
-colorTo: purple
+title: SurvivalAI Pro
+emoji: 🏕️
+colorFrom: green
+colorTo: gray
 sdk: gradio
-sdk_version: 6.5.1
+sdk_version: 5.0.0
 app_file: app.py
 pinned: false
-hf_oauth: true
-hf_oauth_scopes:
-- inference-api
 license: mit
+short_description: Off-grid survival AI — runs fully on CPU, no internet needed
 ---
 
-An example chatbot using [Gradio](https://gradio.app), [`huggingface_hub`](https://huggingface.co/docs/huggingface_hub/v0.22.2/en/index), and the [Hugging Face Inference API](https://huggingface.co/docs/api-inference/index).
+# SurvivalAI Pro
+
+A fine-tuned Phi-3-mini survival assistant running entirely on CPU.
+Trained on ~150,000 survival Q/A pairs covering medical, water, food,
+shelter, fire, navigation, signaling, foraging, hunting, and tools.
+
+## Setup secrets
+
+This Space loads its model from a private HF model repo. Set the following
+secret in **Settings → Variables and secrets**:
+
+- `HF_TOKEN` — a read-scope HF token with access to `Znilsson/survivalai-phi3-gguf`
+
+## Hardware
+
+Designed for **CPU Upgrade** tier (16 GB RAM, 8 vCPU). Performance:
+
+- Cold start: ~30–60 s (model download + load)
+- Inference: ~6–10 tokens/sec
+- Typical response: 30–60 sec for 200–400 tokens
+
+## Architecture
+
+- Base model: `microsoft/phi-3-mini-4k-instruct`
+- Fine-tune: QLoRA r=16, α=32, 3 epochs on ~150 k pairs
+- Quantization: Q4_K_M (2.4 GB)
+- Runtime: `llama-cpp-python`
+- UI: Gradio `ChatInterface` with streaming