ory / lumen / 23907082126

69%

push

github

feat: add LUMEN_EMBED_DIMS/LUMEN_EMBED_CTX env vars for custom models (#80)

Adds two env vars that bypass the KnownModels registry, allowing any
model served by an existing backend to be used without modifying source:

  LUMEN_EMBED_DIMS=4096   — embedding vector dimensionality (required)
  LUMEN_EMBED_CTX=40960   — context length in tokens (optional, default 8192)

When LUMEN_EMBED_DIMS is set, LUMEN_EMBED_MODEL can be any string and
the registry lookup is skipped entirely. This is useful for models not
yet in KnownModels (e.g. quantized MLX variants, local fine-tunes) served
via the existing lmstudio or ollama backends.

Example with omlx/vLLM/LocalAI on a non-standard port:

  LUMEN_BACKEND=lmstudio
  LM_STUDIO_HOST=http://localhost:8801
  LUMEN_EMBED_MODEL=Qwen3-Embedding-8B-4bit-DWQ
  LUMEN_EMBED_DIMS=4096
  LUMEN_EMBED_CTX=40960

0 of 16 new or added lines in 1 file covered. (0.0%)

2579 of 3774 relevant lines covered (68.34%)

47.09 hits per line