Falcon3-3B-Instruct-LiteRT / model.litertlm

Commit History

Rebuild via stock litert-torch (recipe.json, no custom code); same blockwise-128 int4, GSM8K 77%
1e553b4
verified

mlboydaisuke commited on

Switch to blockwise-128 int4 + cache2048 (iPhone 27 tok/s, GSM8K 77% parity)
07c5809
verified

mlboydaisuke commited on

Replace channelwise int4 with blockwise int4 + cache4096 (GSM8K 79% ~ bf16 75%)
5a6d84a
verified

mlboydaisuke commited on

Upload model.litertlm with huggingface_hub
247ba94
verified

mlboydaisuke commited on