Repository navigation
docs: deduplicate model-specific guides into the cookbook - #44135
Asthenia0412 wants to merge 1 commit into
Conversation
Resolves the remaining docs/cookbook overlap from #18427. What changed (case 1 — cookbook already had the content): - supported-models/embedding_models.mdx: removed the condensed EmbeddingGemma subsection (the cookbook page at /cookbook/autoregressive/Google/EmbeddingGemma is authoritative and covers prerequisites and Hopper performance defaults). Replaced with a <Note> pointer; the table row now links to the cookbook. What changed (case 2 — cookbook did not have the content): - Moved the Qwen3-Reranker and Qwen3-VL-Reranker launch commands, request/response formats, multimodal examples, and common pitfalls (~240 lines) out of rerank_models.mdx into two new cookbook pages: /cookbook/autoregressive/Qwen/Qwen3-Reranker /cookbook/autoregressive/Qwen/Qwen3-VL-Reranker rerank_models.mdx keeps the category-level cross-encoder guide, the supported models table (with model-name → cookbook links), and a short summary of the decoder-only rules. Added the two new cookbook pages to docs.json navigation under the Qwen group. Deliberately kept: the category-level quickstarts (client request examples, Matryoshka embedding, multimodal embedding example, cross-encoder rerank endpoint docs) are endpoint documentation rather than per-model duplication. Refs #18427 Signed-off-by: Yancy <yancy@sglang.io>
github-actions
Bot
added
the
documentation
|
Hi! The CI currently shows skipped jobs because this repo requires the Could someone please add the @lm-sys @zhyncs @hnyls2002 @ispobock @BBuf @Fridge003 @kevin-mii |
|
Thanks for your interest in SGLang. This account opened 16 PRs in about eight hours, across unrelated subsystems (attention backends, speculative decoding, diffusion loading, scheduler IPC, KV-cache compression, docs), including two PRs for the same change (#44073 and #44117, both for #28808) and a fix for #43929 that is already on main. We can't review contributions at this volume, so we are closing the ones still open. If you want to pursue one, please pick a single change, verify it on a real run (a repro, a test, or numbers), and open a fresh PR. |
Summary
Resolves the remaining docs/cookbook overlap from #18427.
Changes
Case 1 — cookbook already had the content (EmbeddingGemma)
EmbeddingGemmasubsection. The cookbook page at/cookbook/autoregressive/Google/EmbeddingGemmais authoritative and covers prerequisites, Hopper performance defaults, and deployment behavior. Replaced with a<Note>pointer. The model name in the supported-models table now links to the cookbook.Case 2 — cookbook did not have the content (Qwen3-Reranker, Qwen3-VL-Reranker)
rerank_models.mdx(~240 lines: launch commands, request/response formats, multimodal examples, common pitfalls) into two new cookbook pages:docs/cookbook/autoregressive/Qwen/Qwen3-Reranker.mdxdocs/cookbook/autoregressive/Qwen/Qwen3-VL-Reranker.mdxrerank_models.mdxnow keeps only the category-level cross-encoder guide, the supported-models table (model names link to cookbook), and a short summary of decoder-only deployment rules.Navigation
docs/docs.jsonunder the Qwen group so they appear in the sidebar.What was deliberately kept
The category-level quickstarts (client request examples, Matryoshka embedding, multimodal embedding example, cross-encoder rerank endpoint docs) are endpoint documentation rather than per-model duplication.
Already resolved during docs migration
The
popular_model_usage.html -> /cookbook/autoregressive/introredirect is already indocs.json, along with per-model redirects.Refs #18427
CI States
Latest PR Test (Base): ✅ Run #38473313175
Latest PR Test (Extra): ❌ Run #38473313149
Latest PR Test (AMD ROCm 10): ➖ No AMD PR run found for this commit.