embeddings: default check_embedding_ctx_length=False for OpenAI-compa…

[kkampli-singlestore]

Don’t use OpenAI’s tokenizer for Qwen — LangChain’s default (check_embedding_ctx_length=True) turns text into OpenAI token IDs. Qwen can’t read those, so embeddings are garbage. Send raw text instead (False) so the server tokenizes correctly.

Keep OpenAI/Azure on the default — Real OpenAI/Azure models still use True (tiktoken + LangChain’s built-in long-text handling). Only non-OpenAI/Nova models get the raw-text default.

Chunk long Qwen inputs ourselves — With raw text, LangChain won’t split long docs, so the server 400s or truncates. For those models, split into character chunks, embed each, then average into one vector so long text still works.

Test cases:
1.
Before:
label cosine_vs_anchor
anchor 1.0
unrelated2 0.3536
same_intent 0.3461
paraphrase 0.304
unrelated 0.2523

After

Chunking Long Text

---

Note

Medium Risk
Changes default embedding behavior and vector composition for non-Azure OpenAI-compatible models, which can alter embedding outputs and API call patterns for long documents; Azure and Amazon paths are explicitly preserved.

Overview
SingleStoreEmbeddingsFactory now picks the embeddings client by hosting platform instead of always returning stock OpenAIEmbeddings.

For Azure, behavior stays aligned with real OpenAI models: check_embedding_ctx_length=True and LangChain’s tiktoken-based client-side handling.

For other OpenAI-compatible hosts (e.g. Qwen on Nova), the factory defaults check_embedding_ctx_length=False so requests send raw text for server-side tokenization, and returns a new _ChunkedOpenAIEmbeddings wrapper. That class splits inputs longer than max_chunk_chars (default 24k), embeds each chunk via sync/async embed_documents, then length-weighted averages chunk vectors and L2-normalizes them back to one vector per input so long text avoids server reject/truncate limits.

The Bedrock/Amazon path is unchanged.

^{Reviewed by Cursor Bugbot for commit 6def72b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[mgiannakopoulos]

🚢 Ship IT!

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 6def72b. Configure here.}

[cursor]

Character chunks exceed token limit

Medium Severity

_ChunkedOpenAIEmbeddings caps chunks at max_chunk_chars (24000), but Nova-style servers enforce a token context window (often 8192). For scripts where characters map roughly one-to-one to tokens, a single chunk can still exceed that limit, so the server may reject the request or truncate input while the client treats the embedding as complete.

 

^{Reviewed by Cursor Bugbot for commit 6def72b. Configure here.}