tahamajs/RR / Docs /module_metadata /hashed_net_compressor.json
tahamajs's picture
download
raw
1.02 kB
{
"module_name": "hashed_net_compressor",
"class_name": "HashedNetCompressor",
"category": "compression",
"family": "compression",
"params": [
"model",
"rank",
"n_buckets",
"use_sign",
"cfg",
"target_layers"
],
"ops": [
"linear",
"matmul",
"norm",
"softmax"
],
"docstring": "Compress a model by replacing every nn.Linear layer with HashedLinear,\nwhich shares weights globally through compact hash tables.\n\nArgs:\n model: PyTorch model.\n rank: Rank parameter (optionally maps to bucket budget).\n n_buckets: Number of hash buckets (size of shared weight table).\n use_sign: Whether to use auxiliary sign hashing (unbiased trick).\n cfg: Optional configuration object.\n target_layers: Optional list of layer names or substrings to compress.\n **kwargs: Passed to BaseCompressor.",
"rank": 32,
"d_state": 1,
"num_experts": 1,
"top_k": 1,
"sparsity_pct": 0.0,
"bits": 4,
"hash_val": 92335457557404950782183145875741016578
}

Xet Storage Details

Size:
1.02 kB
·
Xet hash:
d4799a69593c129c203f14b9ce9764718d41e71779567dc6a9eabe9d3131594d

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