CQ3-Qwen3-4B-K16-CodeQuant / codequant_baseline.json
sy128's picture
CodeQuant baseline qwen3_4b K16, single-GPU build
9fe9671 verified
Raw History Blame Contribute Delete
346 Bytes
{
"baseline": "codequant",
"model": "Qwen/Qwen3-4B",
"bits": 4,
"cluster_num": 16,
"calibration_dataset": "mixed",
"max_samples": 512,
"rotation_max_samples": 512,
"sequence_length": 512,
"seconds": 4473.759551926516,
"rotation_seconds": 545.8282691957429,
"cluster_seconds": 3927.931282730773,
"folded_projections": 180
}