[Docs] Add blog: training embedding models and publishing to Hugging Face - #2670
[Docs] Add blog: training embedding models and publishing to Hugging Face#2670TwilightTechie wants to merge 2 commits into
Conversation
✅ Deploy Preview for vllm-semantic-router ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
…Face Walkthrough of the cache-LoRA and domain-adapted embedding pipelines, evaluation, and publishing to the llm-semantic-router Hugging Face org. Fixes vllm-project#2667 Signed-off-by: Anup Sharma <anupnewsmail@gmail.com>
92c4a58 to
129b6bf
Compare
github-actions
Bot
assigned
AayushSaini101,
drivebyer,
FAUST-BENCHOU,
ramkrishs,
shraderdm and
WUKUNTAI-0211
👥 vLLM Semantic Team NotificationThe following members have been identified for the changed files in this PR and have been automatically assigned when their GitHub accounts are assignable in this repository: 📁
|
drivebyer
left a comment
drivebyer
left a comment
There was a problem hiding this comment.
Thanks for putting this together — the walkthrough of both pipelines is genuinely useful, and I checked the CLI flags, hyperparameters, margin formula, triplet field semantics, and the +71% MRR@5 figure against the code and READMEs: they all match.
Two factual issues worth fixing before merge, both flagged inline: three of the Hugging Face links don't resolve, and the cited arXiv paper title doesn't match the actual paper.
| - [multi-domain-cache-lora-L12](https://huggingface.co/llm-semantic-router/multi-domain-cache-lora-L12) | ||
| - [medical-cache-lora](https://huggingface.co/llm-semantic-router/medical-cache-lora) | ||
| - Dataset: [cache-embedding-test-sets](https://huggingface.co/datasets/llm-semantic-router/cache-embedding-test-sets) |
There was a problem hiding this comment.
All three links return 401, and none of them appear in the org's public listing (58 models / 13 datasets) — same for law-cache-lora and programming-cache-lora. Either re-publish the repos or drop the links. The mmbert-embed-* links in Pipeline B resolve fine.
| not only cache LoRA. | ||
|
|
||
| Inspired by | ||
| [Distilling an LLM’s Wisdom…](https://arxiv.org/abs/2512.08088). |
There was a problem hiding this comment.
arXiv:2512.08088 is actually titled "Adaptation of Embedding Models to Financial Filings via LLM Distillation" (v1 only, never renamed). No paper by the cited name exists. Same typo in domain_adapted_embeddings/README.md:7 — worth fixing both.
7 participants
What type of PR is this?
/kind documentation
What this PR does / why we need it
Adds a blog post walking through how vLLM Semantic Router trains its embedding models and publishes them to Hugging Face:
llm-semantic-routerorg, and consuming the adapter with PEFTIncludes a hero image (1200x630) used for the blog card and social preview.
Which issue(s) this PR fixes
Fixes #2667
Test Plan
markdownlint -c tools/linter/markdown/markdownlint.yamlpasses on the new postauthors.yml), tags, tables, code blocks, and hero image all display correctly<!-- truncate -->marker added so the blog index shows a short preview