-
Delete tokenizer/training_corpus.txt
-
SAGE model repository : Updating some model checkpoints
-
SAGE model repository : Updating some model checkpoints
-
SAGE model repository : Updating some model checkpoints
-
feat: add authenticated remote control UI and ngrok launcher
-
feat: add authenticated remote control UI and ngrok launcher
-
feat: add authenticated remote control UI and ngrok launcher
-
feat: add authenticated remote control UI and ngrok launcher
-
feat: add authenticated remote control UI and ngrok launcher
-
feat: add authenticated remote control UI and ngrok launcher
-
feat: add authenticated remote control UI and ngrok launcher
-
feat: add authenticated remote control UI and ngrok launcher
-
feat: add authenticated remote control UI and ngrok launcher
-
feat: add authenticated remote control UI and ngrok launcher
-
feat: add authenticated remote control UI and ngrok launcher
-
feat: add authenticated remote control UI and ngrok launcher
-
feat: rewrite SAGE 1B architecture and replace legacy repo contents
-
feat: implement multi-GPU support via DataParallel
-
feat: implement multi-GPU support via DataParallel
-
feat: implement multi-GPU support via DataParallel
-
feat: implement multi-GPU support via DataParallel
-
feat: implement multi-GPU support via DataParallel
-
feat: implement multi-GPU support via DataParallel
-
feat: implement multi-GPU support via DataParallel
-
feat: implement multi-GPU support via DataParallel
-
perf: SAGE Turbo Mode optimizations (TF32, torch.compile, Fused AdamW)
-
docs: add Kaggle/Colab guide and SAGE v3 roadmap
-
fix: restore missing signature in model.py
-
feat: SAGE v2 upgrade (GQA, CoT, W&B, FineWeb-Edu)
-
fix: ensure freqs_cis is complex in apply_rotary_emb
-
feat: add manual attention fallback for older CUDA architectures
-
feat: implement multi-GPU support via DataParallel
-
feat: add non-interactive one-liner training commands
-
fix: LoRALinear device mismatch during finetuning
-
Initial codebase upload: SAGE Transformer Models & Scripts
-
Upload 4 files
Adding the main model files -
initial commit