Add supported models documentation

Co-authored-by: Ona <no-reply@ona.com>

Author: gHashTag

docs/SUPPORTED_MODELS.md

@@ -0,0 +1,96 @@
+# TRINITY Supported Models
+
+## Quick Start
+
+```bash
+# General chat (fast, small)
+./bin/vibee chat --model models/smollm-135m-instruct-q8_0.gguf --prompt "Hello"
+
+# Coding (requires more RAM)
+./bin/vibee chat --model models/qwen2.5-coder-0.5b-instruct-q8_0.gguf --prompt "Write fibonacci in Python"
+```
+
+## Supported Models
+
+### General Chat
+
+| Model | Size | RAM | Speed | Quality |
+|-------|------|-----|-------|---------|
+| SmolLM-135M-Instruct | 139 MB | 1 GB | 12 tok/s | ★★☆☆☆ |
+| TinyLlama-1.1B | 1.1 GB | 4 GB | 4 tok/s | ★★★☆☆ |
+
+### Coding
+
+| Model | Size | RAM | Speed | Quality |
+|-------|------|-----|-------|---------|
+| Qwen2.5-Coder-0.5B | 600 MB | 2 GB | 6 tok/s | ★★★☆☆ |
+| Qwen2.5-Coder-1.5B | 1.8 GB | 6 GB | 3 tok/s | ★★★★☆ |
+
+## Download Models
+
+```bash
+cd models/
+
+# SmolLM-135M (recommended for demos)
+curl -L -o smollm-135m-instruct-q8_0.gguf \
+  "https://huggingface.co/HuggingFaceTB/smollm-135M-instruct-v0.2-Q8_0-GGUF/resolve/main/smollm-135m-instruct-add-basics-q8_0.gguf"
+
+# Qwen2.5-Coder-0.5B (coding, small)
+curl -L -o qwen2.5-coder-0.5b-instruct-q8_0.gguf \
+  "https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B-Instruct-GGUF/resolve/main/qwen2.5-coder-0.5b-instruct-q8_0.gguf"
+
+# Qwen2.5-Coder-1.5B (coding, better quality)
+curl -L -o qwen2.5-coder-1.5b-instruct-q8_0.gguf \
+  "https://huggingface.co/Qwen/Qwen2.5-Coder-1.5B-Instruct-GGUF/resolve/main/qwen2.5-coder-1.5b-instruct-q8_0.gguf"
+```
+
+## Quantization Support
+
+| Format | Supported | Notes |
+|--------|-----------|-------|
+| Q8_0 | ✅ Yes | Best quality, larger size |
+| Q4_0 | ✅ Yes | Good balance |
+| Q4_K_M | ⚠️ Partial | K-quants, experimental |
+| F32 | ✅ Yes | Full precision |
+| F16 | ✅ Yes | Half precision |
+
+## Chat Templates
+
+TRINITY auto-detects the model and uses appropriate chat template:
+
+| Model | Template |
+|-------|----------|
+| Qwen* | ChatML (`<\|im_start\|>`) |
+| SmolLM* | ChatML |
+| TinyLlama* | TinyLlama format |
+| Llama-2* | Llama2 format |
+
+## Fly.io Deployment
+
+Current deployment uses SmolLM-135M (fits in 2GB RAM):
+
+```bash
+# API endpoint
+curl https://trinity-llm.fly.dev/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{"messages":[{"role":"user","content":"Hello"}]}'
+```
+
+For coding models, deploy with more RAM:
+
+```toml
+# fly.toml
+[[vm]]
+  size = "shared-cpu-4x"
+  memory = "4gb"
+```
+
+## Ternary Mode
+
+Enable 16x memory savings (experimental):
+
+```bash
+./bin/vibee chat --model models/smollm-135m-instruct-q8_0.gguf --ternary
+```
+
+Note: Quality may degrade for non-BitNet models.