hakatu/fpga-whale-training
0
Final validated training script - RoPE and mask fixes applied
Fix RoPE cos/sin broadcasting bug
Fix streaming dataset: use max_steps instead of num_train_epochs
Add self-contained training script
Fix TrainingArguments: remove project_name, fix report_to
Fix training script: correct TrainingArguments params, device handling
Add README
Add training script with knowledge distillation
Add FPGA-friendly model architecture
initial commit
