Commit History
9 days ago
-
Card: link to squad-0.8b (was qev-s-0.8b)
Verified signaturefcd607a -
Card: link to qev-s-0.8b (MoQLE)
Verified signature9f49c6b
10 days ago
-
Correct: benchmarks guided checkpoint choice and some training data shares benchmark formats; release is SG50
Verified signaturecb077b4
11 days ago
-
Card: rename the Images section to 'Image input' (the Hub drops id=images, so #images links did not resolve)
Verified signature72d1d1e -
Card: 'Which version should I use?', per_question recommendation (JevVision fast engine 0.783), completeness fixes
Verified signaturedead354 -
Images on the fast path: compiled + batched image requests in decide_batch, vision tower batched by grid size, per_question / cache_mb image prefixes (cache keyed by image content), image-embedding LRU; card speed (Linux) and numbers
Verified signaturef65934e -
Image input: vision tower + merger in MLX (vision.safetensors, bf16, also in q8/), qev_mlx.py images=, card
Verified signature565aa62 -
SHA256SUMS: the updated README.md
Verified signature5de5ac9 -
Card: image input is in the transformers checkpoint labqoat/qev-0.8b (qev_image.py); this build is text-only
Verified signatured4a995a
12 days ago
-
Model card: the full qev-0.8b card (demo, benchmarks, evaluation, training, data, limitations) with the MLX parts
Verified signature799c195 -
qev-0.8b-MLX: initial release
Verified signatureffc2fee -
tokenizer.json (root and q8) matches the training pre-tokenizer
Verified signaturef692bac -
qev-0.8b-MLX: initial release
Verified signaturea6d299e -
initial commit
Verified signature295b83e