diff --git a/unsloth/models/_utils.py b/unsloth/models/_utils.py index ce644f8540..7cbecfa76a 100644 --- a/unsloth/models/_utils.py +++ b/unsloth/models/_utils.py @@ -1130,6 +1130,8 @@ def unsloth_compile_transformers( fuse_lm_head = True, gradient_checkpointing = True, manual_replacements = True, + fast_lora_forwards = True, + fast_residual_stream = True, epilogue_fusion = True, max_autotune = False, shape_padding = True, @@ -1174,6 +1176,8 @@ def unsloth_compile_transformers( fuse_lm_head = fuse_lm_head, gradient_checkpointing = gradient_checkpointing, manual_replacements = manual_replacements, + fast_lora_forwards = fast_lora_forwards, + fast_residual_stream = fast_residual_stream, epilogue_fusion = epilogue_fusion, max_autotune = max_autotune, shape_padding = shape_padding, diff --git a/unsloth/models/loader.py b/unsloth/models/loader.py index 9c5ea5baca..d1aae95a52 100644 --- a/unsloth/models/loader.py +++ b/unsloth/models/loader.py @@ -470,6 +470,8 @@ class FastVisionModel(FastBaseVisionModel): fuse_lm_head = True, gradient_checkpointing = True, manual_replacements = True, + fast_lora_forwards = True, + fast_residual_stream = True, epilogue_fusion = True, max_autotune = False, shape_padding = True,