mirror of
https://github.com/samvallad33/vestige.git
synced 2026-06-14 20:55:14 +02:00
14b061f124
Hardens Sanhedrin Receipt Lock for model-agnostic use, adds fail-open telemetry and receipt docs, fixes smart_ingest batch safety, wires opt-in CUDA Qwen3 device selection, and refreshes dashboard/release assets.\n\nFixes #54\nFixes #58\nFixes #60\nRefs #59
103 lines
4.4 KiB
JSON
103 lines
4.4 KiB
JSON
{
|
|
"schema": "vestige.sanhedrin.presets.v2",
|
|
"defaultPreset": null,
|
|
"description": "Model-agnostic Sanhedrin backend recipes. Presets are suggestions only; users may set any OpenAI-compatible endpoint and model name.",
|
|
"presets": {
|
|
"custom-openai-compatible": {
|
|
"label": "Custom OpenAI-compatible endpoint",
|
|
"tier": "custom",
|
|
"bestFor": "Any model/server the user already trusts",
|
|
"requiresUserModel": true,
|
|
"endpointPlaceholder": "http://127.0.0.1:8000/v1/chat/completions",
|
|
"modelPlaceholder": "your-model-name",
|
|
"env": {
|
|
"VESTIGE_SANHEDRIN_ENDPOINT": "",
|
|
"VESTIGE_SANHEDRIN_MODEL": "",
|
|
"VESTIGE_SANHEDRIN_BACKEND": "openai-compatible",
|
|
"VESTIGE_SANHEDRIN_TIMEOUT": "45"
|
|
}
|
|
},
|
|
"small-laptop-ollama": {
|
|
"label": "Small laptop Ollama",
|
|
"tier": "small-local",
|
|
"bestFor": "8-16 GB RAM laptops that need a lightweight offline verifier",
|
|
"setup": "Install Ollama, then pull any small instruct model you trust, for example: ollama pull llama3.2:3b or ollama pull qwen2.5:7b",
|
|
"tradeoffs": ["fast and accessible", "weaker contradiction judgment than larger models"],
|
|
"env": {
|
|
"VESTIGE_SANHEDRIN_ENDPOINT": "http://127.0.0.1:11434/v1/chat/completions",
|
|
"VESTIGE_SANHEDRIN_MODEL": "your-ollama-model",
|
|
"VESTIGE_SANHEDRIN_BACKEND": "ollama",
|
|
"VESTIGE_SANHEDRIN_TIMEOUT": "60"
|
|
}
|
|
},
|
|
"balanced-local-ollama": {
|
|
"label": "Balanced local Ollama",
|
|
"tier": "balanced-local",
|
|
"bestFor": "16-32 GB RAM machines using 7B-14B local models",
|
|
"setup": "Install Ollama and pull a balanced verifier model such as qwen3:14b, llama3.1:8b, or another OpenAI-compatible local model.",
|
|
"tradeoffs": ["good first local choice", "model quality depends on the exact model selected"],
|
|
"env": {
|
|
"VESTIGE_SANHEDRIN_ENDPOINT": "http://127.0.0.1:11434/v1/chat/completions",
|
|
"VESTIGE_SANHEDRIN_MODEL": "your-ollama-model",
|
|
"VESTIGE_SANHEDRIN_BACKEND": "ollama",
|
|
"VESTIGE_SANHEDRIN_TIMEOUT": "60"
|
|
}
|
|
},
|
|
"mlx-qwen3.6-apple-silicon": {
|
|
"label": "MLX Qwen3.6 35B A3B, Apple Silicon local",
|
|
"tier": "strong-local",
|
|
"bestFor": "High-memory Apple Silicon users who explicitly choose the strong MLX path",
|
|
"env": {
|
|
"VESTIGE_SANHEDRIN_ENDPOINT": "http://127.0.0.1:8080/v1/chat/completions",
|
|
"VESTIGE_SANHEDRIN_MODEL": "mlx-community/Qwen3.6-35B-A3B-4bit",
|
|
"VESTIGE_SANHEDRIN_BACKEND": "mlx",
|
|
"VESTIGE_SANHEDRIN_TIMEOUT": "45"
|
|
}
|
|
},
|
|
"vllm-openai-compatible": {
|
|
"label": "vLLM OpenAI-compatible server",
|
|
"tier": "workstation",
|
|
"bestFor": "GPU workstations and team servers",
|
|
"env": {
|
|
"VESTIGE_SANHEDRIN_ENDPOINT": "http://127.0.0.1:8000/v1/chat/completions",
|
|
"VESTIGE_SANHEDRIN_MODEL": "your-vllm-model",
|
|
"VESTIGE_SANHEDRIN_BACKEND": "vllm",
|
|
"VESTIGE_SANHEDRIN_TIMEOUT": "45"
|
|
}
|
|
},
|
|
"llama-cpp-openai-compatible": {
|
|
"label": "llama.cpp server",
|
|
"tier": "small-local",
|
|
"bestFor": "CPU or small-GPU local deployments",
|
|
"env": {
|
|
"VESTIGE_SANHEDRIN_ENDPOINT": "http://127.0.0.1:8081/v1/chat/completions",
|
|
"VESTIGE_SANHEDRIN_MODEL": "your-gguf-model",
|
|
"VESTIGE_SANHEDRIN_BACKEND": "llama.cpp",
|
|
"VESTIGE_SANHEDRIN_TIMEOUT": "90"
|
|
}
|
|
},
|
|
"hosted-openai-compatible": {
|
|
"label": "Hosted OpenAI-compatible API",
|
|
"tier": "hosted",
|
|
"bestFor": "Users who want zero local model setup",
|
|
"requires": "VESTIGE_SANHEDRIN_API_KEY exported in the hook environment and a model chosen by the user/provider",
|
|
"env": {
|
|
"VESTIGE_SANHEDRIN_ENDPOINT": "https://api.openai.com/v1/chat/completions",
|
|
"VESTIGE_SANHEDRIN_MODEL": "your-hosted-model",
|
|
"VESTIGE_SANHEDRIN_BACKEND": "openai",
|
|
"VESTIGE_SANHEDRIN_TIMEOUT": "45"
|
|
}
|
|
},
|
|
"anthropic-via-litellm": {
|
|
"label": "Anthropic through LiteLLM OpenAI-compatible proxy",
|
|
"bestFor": "Claude users who already run LiteLLM",
|
|
"setup": "Run LiteLLM locally with an Anthropic model, then point Sanhedrin at the proxy.",
|
|
"env": {
|
|
"VESTIGE_SANHEDRIN_ENDPOINT": "http://127.0.0.1:4000/v1/chat/completions",
|
|
"VESTIGE_SANHEDRIN_MODEL": "anthropic/claude-3-5-haiku-latest",
|
|
"VESTIGE_SANHEDRIN_BACKEND": "litellm",
|
|
"VESTIGE_SANHEDRIN_TIMEOUT": "45"
|
|
}
|
|
}
|
|
}
|
|
}
|