Upload README.md with huggingface_hub

README.md

@@ -1,8 +1,8 @@
 ---
 library_name: pytorch
 ---
-# Slayer 11B pilot training checkpoint
+# Slayer 11B pilot weights
 
 Private backup of an experimental approximately 11.18B-parameter model trained from scratch on a subset of SlayerLab/slayer-pl-8x3b. This is an early pilot, not a finished or evaluated model.
 
-`checkpoint.pt` is a PyTorch training checkpoint containing model and optimizer state, intended for resuming the custom trainer. It is not a Transformers-format model. Load PyTorch checkpoints only from trusted sources. Training code and data provenance are included for reproducibility. The checkpoint file will appear after transfer and upload finish.
+`weights-only.pt` is a PyTorch dictionary containing `model`, `config`, and `step`. It preserves the original weight precision, but does not contain the optimizer state and cannot resume the original training state exactly. This uses the custom `train_gpt.py` architecture, not Transformers format; the reference tokenizer is SlayerLab/Slayer1B-1.00. Load PyTorch files only from trusted sources. The file appears after export and upload finish.