Fix: enumerate typo and sliding_window key name in modeling_trillion.py
#4
by AeonProtocol - opened
Two bug fixes in modeling_trillion.py:
enumrateβenumerate(NameError during forward pass)"sliding_window"β"sliding_attention"(KeyError in causal_mask_mapping)
These bugs cause crashes when running the model with output_attentions=True or during LoRA fine-tuning. Discovered during cooperative basin research experiments.
WonsukYangTL changed pull request status to merged