[MIEB] Fix get_fused_emebddings

[izhx]

In PR #1583 , I only changed the arguments of get_fused_emebddings, but forgot to update the internal implementation logic accordingly, which caused runtime errors. (Sorry)

Now, they are fixed.

I will run i, t, and it encodings for each model to check.

Checklist

• Run tests locally to make sure nothing is broken using make test.
• Run the formatter to format the code using make lint.

---

Checking progress:

code	i2t - Flickr30kI2TRetrieval	it2t - LLaVAIT2TRetrieval PR #1611	note
align_models.py	✓	✓
blip2_models.py			salesforce-lavis is conflict with sentence_transformers>=3.0
blip_models.py	✓	✓
clip_models.py	✓	✓
cohere_v.py			I don't have a key
dino_models.py			DINO models only support image encoding.
e5_v.py	✓	✓
evaclip_models.py			Code error when init model
jina_clip.py	✓	✓
nomic_models_vision.py	✓	✓
openclip_models.py	✓	✓
siglip_models.py	✓	✓
vista_models.py			Code error when init model
vlm2vec_models.py	✓	✓
voyage_v			I don't have a key

[gowitheflow-1998]

no worries thanks a lot!

[izhx]

@gowitheflow-1998 Hi, I fixed vista model loading and a runtime error related to max_length (the code did not truncate to max input length).

May I submit a PR with these fixes?

Commit: izhx@e849843

---

Error massage:

[WARNING|tokenization_utils_base.py:3928] 2024-12-26 19:53:40,650 >> Token indices sequence length is longer than the specified maximum sequence length for this model (537 > 512). Running this sequence through the model will result in indexing errors

ERROR:mteb.evaluation.MTEB:Error while evaluating OKVQAIT2TRetrieval: The size of tensor a (537) must match the size of tensor b (512) at non-singleton dimension 1

[gowitheflow-1998]

@gowitheflow-1998 Hi, I fixed vista model loading and a runtime error related to max_length (the code did not truncate to max input length).

May I submit a PR with these fixes?

Yeah, will be great to have your fix merged to MIEB! Also I think ours was from the earliest version when the import wasn't moved to visual_bge from FlagEmbedding. Will be nice to have the latest!