SmolLM2-135M-V2-QuadOrbit / training_metadata.json
Argo1-OOAS's picture
Upload 16 files
9a2fa50 verified
Raw History Blame Contribute Delete
708 Bytes
{
"artifact": "SmolLM2-135M-V2-QuadOrbit",
"release_date": "2026-08-26",
"base_model": "HuggingFaceTB/SmolLM2-135M-Instruct",
"base_core_frozen": true,
"checkpoint_step_zero_indexed": 1999,
"completed_updates": 2000,
"tokens_per_update": 131072,
"token_presentations": 262144000,
"sequence_length": 8192,
"parameter_count": 134939649,
"branch_parameter_count": 424641,
"prepared_train_tokens": 420314292,
"prepared_validation_tokens": 838170,
"terminal_sampled_validation_loss": 1.6958,
"terminal_sampled_validation_perplexity": 5.5,
"validation_note": "Ten random validation batches from the training loop; branch-training objective, not a standard full-stream benchmark."
}