veyra-ai/Veyra2-Apricot-50M-Base

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Upload banner.jpg

Delete banner.png

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Rename kmwPxmD5XrDdK6z_lZG0D.png to banner.png

Upload kmwPxmD5XrDdK6z_lZG0D.png

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Update README.md

Add model card README

Ignore missing tied lm_head weight on load

Fix RoPE to match original split-half reference implementation

Use exact SDPA reference path for non-cache attention

Fix cache_position slicing during generation

Fix DynamicCache update path

Replace modeling code with clean modern Cache implementation

Switch Veyra2-Apricot to modern HF Cache only

Use SDPA enable_gqa for cached generation

Fix cached GQA K/V head handling

Fix cached generation attention mask handling

Add DynamicCache support to Veyra2-Apricot modeling code