feat(model-ports): add nomic_embed_v1, jina_v2_base_en, jina_v3 - #212
Open
DarthCeltic wants to merge 3 commits into
Open
feat(model-ports): add nomic_embed_v1, jina_v2_base_en, jina_v3#212DarthCeltic wants to merge 3 commits into
DarthCeltic wants to merge 3 commits into
Conversation
Three BERT-family text-embedding models, each a genuinely distinct execution family: arch=nomic-bert, arch=jina-bert-v2 (ALiBi, 8192-token context), arch=jina-bert-v3 (RoPE + bundled task-specific LoRA adapters). None collide with bert (distilbert_sst2) or modern-bert (modernbert_sentiment) already claimed in this campaign. Verified live via llama-server's /embedding endpoint (not GGUF-metadata- only) -- real embedding vectors returned for all 3. jina_v3 needed a longer load wait than the other two (bundles multiple LoRA adapters). Registered in artifacts.json only, not benchmark_config.json -- the shared runner doesn't speak /embedding, same limitation already documented for distilbert_sst2/modernbert_sentiment. jina_v3's license is CC-BY-NC-4.0 (non-commercial) -- flagged.
github-actions
Bot
added
track: model-ports
No pre-made GGUF exists anywhere for EuroBERT-210m, and self-conversion fails: the vendored convert_hf_to_gguf.py registers EuroBertModel and JinaEmbeddingsV5Model to the eurobert arch, but the only public EuroBERT-210m checkpoint declares architectures: [EuroBertForMaskedLM], which isn't recognized -- confirmed live, converter rejects it immediately. Also checked whether a real JinaEmbeddingsV5Model-class checkpoint could substitute: jinaai/jina-embeddings-v5-text-small-retrieval-GGUF exists and loads, but its GGUF reports arch=qwen3 (a Qwen3-backbone embedding fine-tune, not the EuroBert class path) -- already a seed identity, so it wouldn't create new credit even though it works.
Member
|
Manual ET review passed Nomic Embed and Jina v2/v3 with near-identical ET/CPU embeddings. What remains for model-port credit is to commit the deterministic embedding inputs/oracle vectors and comparison thresholds, pin any floating artifact revisions, and let maintainers adopt the three identities/contracts and approve the runner. Please also acknowledge the non-commercial Jina v3 license in the merge decision. EuroBERT is recipe-only here and needs its own benchmark/claim if intended to count. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds nomic_embed_v1, jina_v2_base_en, and jina_v3 text-embedding ports to llama.cpp-et, with model-ports claims for each. Verified live via /embedding endpoint responses, not just GGUF metadata.