Galvanize-60M / head.json
tljohnsilver's picture
Publish the FP32 source weights of the INT8 model and correct the model card
7bf8df0 verified
Raw History Blame Contribute Delete
345 Bytes
{
"model_name": "Galvanize-60M",
"model_type": "Galvanize-60M",
"base_model": "answerdotai/ModernBERT-base",
"layers": 4,
"hidden_dim": 768,
"heads": 12,
"num_queries": 4,
"scale": 0.7,
"shift": -17.60634142857143,
"raw_threshold": -13.4,
"probability": "sigmoid(scale * (logit - shift))",
"threshold_probability": 0.95
}