tahamajs/RR / Docs /module_metadata /frandi_compressor.json
tahamajs's picture
download
raw
1.09 kB
{
"module_name": "frandi_compressor",
"class_name": "FRanDICompressor",
"category": "compression",
"family": "compression",
"params": [
"in_features",
"out_features",
"rank",
"bias",
"dtype",
"device",
"model",
"energy_threshold",
"calibration_loader",
"cfg",
"target_layers"
],
"ops": [
"reshape",
"linear",
"view"
],
"docstring": "FRanDI (Feature-Regression aligned SVD) Compressor.\nCompresses linear layers using singular value decomposition with optional\nactivation alignment and adaptive energy-preservation thresholding.\n\nArgs:\n model: Model to compress.\n rank: Target rank when energy_threshold is None.\n energy_threshold: Energy ratio threshold (0 < energy_threshold <= 1.0).\n calibration_loader: Optional DataLoader for calibration statistics.\n cfg: Optional configuration object.\n target_layers: Optional target layer filter.",
"rank": 32,
"d_state": 1,
"num_experts": 1,
"top_k": 1,
"sparsity_pct": 0.0,
"bits": 4,
"hash_val": 45561839979659658853908864236249565620
}

Xet Storage Details

Size:
1.09 kB
·
Xet hash:
a7edc55985456b8d20e0a0b5030c0fc2f7c0a4781e6a5b954f4d76f930ac48da

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.