Skip to content

feat: add jina-v3 into model list #1318

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
bwanglzu wants to merge 2 commits into from
Closed

feat: add jina-v3 into model list #1318

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b60c714
feat: add jina-v3 into model list
Oct 24, 2024
9302603
feat: add jina model to overview
Oct 24, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
154 changes: 154 additions & 0 deletions mteb/models/jina_models.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,154 @@
from __future__ import annotations

from functools import partial

from mteb.model_meta import ModelMeta

MODEL_NAME = "jinaai/jina-embeddings-v3"
REVISION = "fa78e35d523dcda8d3b5212c7487cf70a4b277da"
XLMR_LANGUAGES = [
"afr_Latn",
"amh_Latn",
"ara_Latn",
"asm_Latn",
"aze_Latn",
"bel_Latn",
"bul_Latn",
"ben_Latn",
"ben_Beng",
"bre_Latn",
"bos_Latn",
"cat_Latn",
"ces_Latn",
"cym_Latn",
"dan_Latn",
"deu_Latn",
"ell_Latn",
"eng_Latn",
"epo_Latn",
"spa_Latn",
"est_Latn",
"eus_Latn",
"fas_Latn",
"fin_Latn",
"fra_Latn",
"fry_Latn",
"gle_Latn",
"gla_Latn",
"glg_Latn",
"guj_Latn",
"hau_Latn",
"heb_Latn",
"hin_Latn",
"hin_Deva",
"hrv_Latn",
"hun_Latn",
"hye_Latn",
"ind_Latn",
"isl_Latn",
"ita_Latn",
"jpn_Latn",
"jav_Latn",
"kat_Latn",
"kaz_Latn",
"khm_Latn",
"kan_Latn",
"kor_Latn",
"kur_Latn",
"kir_Latn",
"lat_Latn",
"lao_Latn",
"lit_Latn",
"lav_Latn",
"mlg_Latn",
"mkd_Latn",
"mal_Latn",
"mon_Latn",
"mar_Latn",
"msa_Latn",
"mya_Latn",
"nep_Latn",
"nld_Latn",
"nob_Latn",
"orm_Latn",
"ori_Latn",
"pan_Latn",
"pol_Latn",
"pus_Latn",
"por_Latn",
"ron_Latn",
"rus_Latn",
"san_Latn",
"snd_Latn",
"sin_Latn",
"slk_Latn",
"slv_Latn",
"som_Latn",
"sqi_Latn",
"srp_Latn",
"sun_Latn",
"swe_Latn",
"swa_Latn",
"tam_Latn",
"tam_Taml",
"tel_Latn",
"tel_Telu",
"tha_Latn",
"tgl_Latn",
"tur_Latn",
"uig_Latn",
"ukr_Latn",
"urd_Latn",
"urd_Arab",
"uzb_Latn",
"vie_Latn",
"xho_Latn",
"yid_Latn",
"zho_Hant",
"zho_Hans",
]


model_prompts = {
"retrieval.query": "Represent the query for retrieving evidence documents: ",
"retrieval.passage": "Represent the document for retrieval: ",
"separation": "",
"classification": "",
"text-matching": "",
}

# Lora adaptation for specific downstream tasks.
# Empty represents for no-lora weights (or checkpoints after pair tuning).
supported_tasks = list(model_prompts.keys()) + [""]


def jina_embeddings_v3_loader(**kwargs):
class JinaV3Wrapper(Wrapper):
def encode(
self,
sentences: Sequence[str],
task: str,
*args,
**kwargs: Any,
) -> np.ndarray:
return super().encode(
Copy link
Member

@Samoed Samoed Oct 24, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm also working on this in #1319 too, and this implementation isn't quite right because MTEB will generate tasks that don't align with Jina tasks.

Copy link
Author

@bwanglzu bwanglzu Oct 24, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

okay okay @Samoed didn't aware you're working on it, i can imaging because now i'm still testing the implementation (marked as draft). What should we do now? shall i take over jina models?

Copy link
Author

@bwanglzu bwanglzu Oct 24, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

btw the reason i add here is because i want to give a full MMTEB evaluation of jina models

Copy link
Member

@Samoed Samoed Oct 24, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, I wanted too. I will provide my tests results in an hour

sentences, task=task, prompt_name=task, *args, **kwargs
)

return JinaV3Wrapper(**kwargs)


jina_embeddings_v3 = ModelMeta(
loader=partial(
jina_embeddings_v3_loader,
model_name=MODEL_NAME,
revision=REVISION,
model_prompts=model_prompts,
trust_remote_code=True,
),
name=MODEL_NAME,
languages=XLMR_LANGUAGES,
open_source=True, # CC-BY-NC-4.0
Copy link
Contributor

@KennethEnevoldsen KennethEnevoldsen Oct 24, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

in #1316 we update the metadata for all models:

It might be nice to add:

    max_tokens=...,
    embed_dim=...,
    n_parameters=...,
    memory_usage=...,
    license="cc-by-nc-4.0",
    reference=..., # you will need to pull from the PR to add this
    similarity_fn_name="cosine",
    framework=[...],
    use_instuctions=...,

Copy link
Author

@bwanglzu bwanglzu Oct 24, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

clear

revision=REVISION,
release_date="2024-09-18",
)
2 changes: 2 additions & 0 deletions mteb/models/overview.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@
google_models,
gritlm_models,
gte_models,
jina_models,
llm2vec_models,
mxbai_models,
nomic_models,
Expand All @@ -38,6 +39,7 @@
google_models,
gritlm_models,
gte_models,
jina_models,
llm2vec_models,
mxbai_models,
nomic_models,
Expand Down