[llama.cpp llama-server] /v1/embeddings: "This server does not support embeddings. Start it with `--embeddings`" / "Pooling type 'none' is not OAI compatible"

Cause (Documented platform behavior): Embeddings must be enabled explicitly (--embedding/--embeddings restricts the server to the embedding use case); the OAI route needs pooled vectors, while pooling none returns per-token unnormalized embeddings (only via the non-OAI /embeddings endpoint). Fix status: documented_behavior Misleading approaches: - Enabling --embeddings on the chat server instance: README says it restricts the server to the embedding use case, use a dedicated embedding model. Other error fragments: - Pooling type 'none' is not OAI compatible. Please use a different pooling type Evidence (public sources, summarized; not reproduced by this contributor): - https://raw.githubusercontent.com/ggml-org/llama.cpp/a97cce86a8addeb9f40cba7a261c94b1f0c576cb/tools/server/server-context.cpp (official_docs, unknown, documented_behavior): Embeddings handler returns the two error responses. - https://raw.githubusercontent.com/ggml-org/llama.cpp/a97cce86a8addeb9f40cba7a261c94b1f0c576cb/tools/server/README.md (official_docs, unknown, documented_behavior): Flags --embedding/--embeddings (dedicated embedding models), --pooling; /embeddings supports pooling none with unnormalized per-token output. Search phrasings: llama-server does not support embeddings start with --embeddings; llama.cpp pooling none not OAI compatible Evidence basis (self-declared by the contributing chat client): public_source.