[load] using MP=2 (auto-detected) [load] merged 65982 amax entries [load] fused w1/w3 input amax on 10997 experts [load] fused w1/w3 weight amax on 10997 experts [load] input-fallback projections: 3 populated [config] 48 input shards device=cuda [shard 1/48] model-00001-of-00048.safetensors [shard 2/48] model-00002-of-00048.safetensors [shard 3/48] model-00003-of-00048.safetensors [shard 4/48] model-00004-of-00048.safetensors [shard 5/48] model-00005-of-00048.safetensors [shard 6/48] model-00006-of-00048.safetensors [shard 7/48] model-00007-of-00048.safetensors [shard 8/48] model-00008-of-00048.safetensors [shard 9/48] model-00009-of-00048.safetensors [shard 10/48] model-00010-of-00048.safetensors [shard 11/48] model-00011-of-00048.safetensors [shard 12/48] model-00012-of-00048.safetensors [shard 13/48] model-00013-of-00048.safetensors [shard 14/48] model-00014-of-00048.safetensors [shard 15/48] model-00015-of-00048.safetensors [shard 16/48] model-00016-of-00048.safetensors [shard 17/48] model-00017-of-00048.safetensors [shard 18/48] model-00018-of-00048.safetensors [shard 19/48] model-00019-of-00048.safetensors [shard 20/48] model-00020-of-00048.safetensors [shard 21/48] model-00021-of-00048.safetensors [shard 22/48] model-00022-of-00048.safetensors [shard 23/48] model-00023-of-00048.safetensors [shard 24/48] model-00024-of-00048.safetensors [shard 25/48] model-00025-of-00048.safetensors [shard 26/48] model-00026-of-00048.safetensors [shard 27/48] model-00027-of-00048.safetensors [shard 28/48] model-00028-of-00048.safetensors [shard 29/48] model-00029-of-00048.safetensors [shard 30/48] model-00030-of-00048.safetensors [shard 31/48] model-00031-of-00048.safetensors [shard 32/48] model-00032-of-00048.safetensors [shard 33/48] model-00033-of-00048.safetensors [shard 34/48] model-00034-of-00048.safetensors [shard 35/48] model-00035-of-00048.safetensors [shard 36/48] model-00036-of-00048.safetensors [shard 37/48] model-00037-of-00048.safetensors [shard 38/48] model-00038-of-00048.safetensors [shard 39/48] model-00039-of-00048.safetensors [shard 40/48] model-00040-of-00048.safetensors [shard 41/48] model-00041-of-00048.safetensors [shard 42/48] model-00042-of-00048.safetensors [shard 43/48] model-00043-of-00048.safetensors [shard 44/48] model-00044-of-00048.safetensors [shard 45/48] model-00045-of-00048.safetensors [shard 46/48] model-00046-of-00048.safetensors [shard 47/48] model-00047-of-00048.safetensors [shard 48/48] model-00048-of-00048.safetensors [stats] cast_blocks_lossless 8657043456 cast_blocks_total 8657043456 experts_hash 2304 experts_score 30720 experts_total 33024 input_fallback_score 33 passthrough 6585 [cast] lossless MXFP4->NVFP4 blocks: 8657043456/8657043456 (100.0000%) [index] wrote model.safetensors.index.json (138681 keys) [config] rewriting config.json (marking moe_quant_algo=NVFP4) [aux] linking ancillary files from /source [done] /candidate (43 quantized routed-expert modules)