Difference between revisions of "Gallery Container"
Jump to navigation
Jump to search
Ominiverdi (talk | contribs) |
|||
| Line 14: | Line 14: | ||
* '''BGE-M3 Embedding Server''' - Semantic search embeddings via llama.cpp | * '''BGE-M3 Embedding Server''' - Semantic search embeddings via llama.cpp | ||
** Port: 8094 (localhost only), Model: bge-m3-Q8_0.gguf | ** Port: 8094 (localhost only), Model: bge-m3-Q8_0.gguf | ||
| − | * | + | * [[Osgeo-library]] |
| − | * | + | * [[Osgeo-knowledge-base]] |
| − | |||
| − | |||
| − | |||
| − | |||
The embedding model running on osgeo7-gallery is BGE-M3 (bge-m3-Q8_0.gguf), a multilingual model from BAAI supporting 100+ languages. It produces 1024-dimensional vectors and runs via llama-server on port 8094 (localhost only). The Q8_0 quantization keeps it light: 606 MB on disk, ~457 MB RAM, ~12ms per query. | The embedding model running on osgeo7-gallery is BGE-M3 (bge-m3-Q8_0.gguf), a multilingual model from BAAI supporting 100+ languages. It produces 1024-dimensional vectors and runs via llama-server on port 8094 (localhost only). The Q8_0 quantization keeps it light: 606 MB on disk, ~457 MB RAM, ~12ms per query. | ||
Revision as of 09:11, 5 June 2026
Debian 12 (bookworm) container for AI/ML services, media, and experimental applications. Gallery is a container on osgeo7 host.
Resources: 8 vCPUs, 12GB RAM System services: nginx, redis, PostgreSQL 17, turnserver, php-fpm
Brian's services (darkblueb):
- GalleryVM media library
- llamafile experiments
- photoprism configuration
Lorenzo's services (ominiverdi):
- LLM-based Chat Assistant
- BGE-M3 Embedding Server - Semantic search embeddings via llama.cpp
- Port: 8094 (localhost only), Model: bge-m3-Q8_0.gguf
- Osgeo-library
- Osgeo-knowledge-base
The embedding model running on osgeo7-gallery is BGE-M3 (bge-m3-Q8_0.gguf), a multilingual model from BAAI supporting 100+ languages. It produces 1024-dimensional vectors and runs via llama-server on port 8094 (localhost only). The Q8_0 quantization keeps it light: 606 MB on disk, ~457 MB RAM, ~12ms per query.
Contact: darkblueb (Brian Hamlin), ominiverdi (Lorenzo Becchi) or SAC channel