athrael-soju/colqwen3.5-4.5B-v3

Add Sentence Transformers usage section (#3)

Integrate with Sentence Transformers via MultiVectorEncoder (#2)

Declare bidirectional retrieval attention

Card: mention embed_dim synonym alongside dim

Add embed_dim: 320 synonym for cross-ecosystem crawler compatibility

Simplify dim note; config.json already has dim: 320

Fix loading note: default from_pretrained reinits proj to random 128-dim; workaround requires config.json dim field or manual Linear override

Clarify embedding dim: weights [320, 2560] but effective retrieval dim is 128 via colpali-engine default

Revert: retrieval dim is 128 per MTEB; 320 weight shape is internal

Fix embed dim: 128 -> 320 (matches custom_text_proj weight shape [320, 2560])

Update ViDoRe V3 leaderboard table (2026-04-20 snapshot); soften rank claims

Add datasets tag for dataset-page backlinks

Add Archived Versions section (v1 and v2 consolidated into v3 repo)

Consolidate v2/tokenizer_config.json into v3 repo

Consolidate v2/tokenizer.json into v3 repo

Consolidate v2/processor_config.json into v3 repo

Consolidate v2/model.safetensors into v3 repo

Consolidate v2/config.json into v3 repo

Consolidate v2/chat_template.jinja into v3 repo

Consolidate v2/README.md into v3 repo

Consolidate v2/.gitattributes into v3 repo

Consolidate v1/tokenizer_config.json into v3 repo

Consolidate v1/tokenizer.json into v3 repo

Consolidate v1/processor_config.json into v3 repo

Consolidate v1/model.safetensors into v3 repo

Consolidate v1/config.json into v3 repo

Consolidate v1/chat_template.jinja into v3 repo

Consolidate v1/README.md into v3 repo

Consolidate v1/.gitattributes into v3 repo

Update README.md

Update README.md

Update README.md

Update V3 table with full leaderboard data from CSV

Restore per-task V3 table with official #3 ranking

Update V3 leaderboard with official #3 ranking

Update README.md

Update README.md

Update README.md

Update V3 table with top 5 leaderboard ranking

Update V3 benchmark table with ViDoRe(v3) results

Update model card with ViDoRe(v3) benchmark results

Add MTEB evaluation results (V1+V2+V3)

Update README.md

Update config.json

Update config.json

Update benchmark tables: add English and Multilingual splits

Update tokenizer_config.json

Update README.md

Update README.md

Upload README.md with huggingface_hub