sdxs-2b-sdnq-t4-tebf16 / quantization_summary.json
WaveCut's picture
Add files using upload-large-folder tool
1b8bd6c verified
{
"source": "/root/sdxs-sdnq-lab/upstream",
"source_model": "AiArtLab/sdxs-2b",
"variant": "t4-tebf16",
"repo_suffix": "sdxs-2b-sdnq-t4-tebf16",
"description": "Cosmos transformer uint4, text encoder bf16, VAE bf16.",
"timings": {
"load_seconds": 12.35,
"quantize_transformer_seconds": 0.962,
"save_seconds": 4.913,
"total_seconds": 18.229
},
"cuda_mem_final": {
"allocated_gb": 3.0434,
"reserved_gb": 3.125,
"peak_allocated_gb": 5.6538
},
"components": {
"transformer": {
"layers": {
"uint4": 442
},
"weight_elements": {
"uint4": 965738496
},
"class": "CosmosTransformer3DModel",
"target_dtype": "uint4",
"cuda_mem_after": {
"allocated_gb": 3.0434,
"reserved_gb": 3.125,
"peak_allocated_gb": 5.6538
}
}
},
"output_size_bytes": 3237051220,
"output_size_gb": 3.0147
}