-
feat: filter Ollama embedding dropdown to only show embedding models (#2290)
frostbyte_neo 发布于
2026-02-17 19:59:05 +00:00 - feat: filter Ollama embedding dropdown to only show embedding models
Ollama's model dropdown in the embedding settings page previously showed
all models (LLMs + embedding). Users had no way to tell which models
support embeddings, leading to confusing errors when selecting an LLM.Now queries each model's capabilities via /api/show and only returns
models with the "embedding" capability. Falls back to name heuristics
for older Ollama versions without the capabilities field.Also adds is_embedding_model() to the base embedding provider class so
any provider can implement model type detection.- feat: show all Ollama models with embedding compatibility labels
Instead of filtering out non-embedding models, show all models with
visual indicators: green "(Embedding)" for dedicated embedding models,
yellow "(LLM — may not support embeddings)" for others. Embedding
models are sorted first in the dropdown.Some LLMs (e.g. nemotron, deepseek-r1) can produce embeddings via
Ollama's /api/embed even without the "embedding" capability flag,
while others (e.g. qwen3) cannot — so we warn rather than block.Also adds 39 comprehensive tests for the capability detection,
filtering, and name heuristic fallback logic.- feat: add embedding model info hints and contextual guidance
Add contextual hints below the model dropdown that change based on
the selected model type (green for embedding, yellow for LLM).
Expand the Ollama provider info panel with detailed guidance about
dedicated embedding models vs LLMs that can also produce embeddings.下载附件