File size: 1,629 Bytes
310d8db cf43b8c 310d8db cf43b8c 310d8db cf43b8c 310d8db cf43b8c 310d8db | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 | {
"format": "minimax_h3_te_clean_external_vram_v1",
"method": "Text-encoder peak VRAM from a fresh child process per variant; nvidia-smi sampled every 100 ms; identical seven-token prompt. Unrelated pre-existing GPU use was excluded and is not recorded.",
"variants": [
{
"id": "official_nvfp4_awq",
"te_peak_vram_mib": 16430,
"exit_code": 0
},
{
"id": "pruned24_int8_convrot",
"te_peak_vram_mib": 15525,
"exit_code": 0
},
{
"id": "pruned24_nvfp4_awq",
"te_peak_vram_mib": 10150,
"exit_code": 0
}
],
"limitations": [
"External process-wide GPU memory, not tensor residency attribution.",
"Short text-conditioning smoke; longer multimodal conditioning can require more memory.",
"Current local ComfyUI/PyTorch/comfy-kitchen stack on NVIDIA GeForce RTX 5090 Laptop GPU."
],
"artifact": {
"file": "text_encoders/pruned24_balanced/qwen3vl_32b_minimax_h3_pruned24_balanced_nvfp4_awq.safetensors",
"bytes": 8553755869,
"sha256": "9e3c36932e10849fd2a946bb45d3e22834c95485d412a9293d1fcd082eb698e9",
"source_sha256": "35a88d51044231fe332301d7a62aa81e3f2cba62febeb446e2c1e3e0ef76f2c6",
"source_revision": "0543966fbdce5ba05709a8f2031c94bdba629b4a",
"source_keys": 2054,
"output_keys": 1170,
"removed_keys": 884,
"retained_tensor_mismatches": 0,
"nvfp4_linear_count": 168
},
"conditioning": {
"shape": [
1,
7,
5120
],
"pruned_int8_vs_nvfp4_cosine": 0.9999949932098389,
"mean_abs_difference": 0.3806621730327606,
"max_abs_difference": 56.3076171875
}
}
|