hyjhyj233/fwe-1b-adamw-step100000

FWE 1B AdamW — step 100000

Private research checkpoint converted to Hugging Face format.

• —Optimizer: AdamW
• —Training step: 100000
• —Learning rate: 0.003
• —Sequence length: 32768
• —Model weights, configuration, and tokenizer are included.
• —Training optimizer state and intermediate checkpoints are excluded.