jayzou3773/less-is-moe-qwen3.5-122b-a10b-gpqa-main-64-intdim-e-50

IntDim-E 50% pruned Qwen/Qwen3.5-122B-A10B

This checkpoint was structurally pruned with the released Less-is-MoE mean-absolute-gradient method. It removes exactly 50% of routed-expert FFN neurons using 64 calibration samples from the gpqa_main configuration of Idavidrein/gpqa revision 633f5ee89ab8ad4522a9f850766b73f62147ffdd. The released loader settings are preserved: train, Question plus shuffled choices, Explanation, selection_seed=1234, BF16, and no optimizer step. The samples are full length: no tokenizer max_length, truncation, or padding. The longest input for this tokenizer is 1,632 tokens. The source checkpoint was loaded and pruned in BF16.

The source-row selection hash is 790c4c22309def44542965fdde7c5f38f1d8e354602640cfb31518134b8d92e6 and the model-specific token-file hash is 4cecf02da096c0d1c1f8f01bbdf8867186ab3064eccbfbb34cd9c16a89564d62. Full export and zero-mask equivalence metadata are in experiment-export.json. The exact calibration and held-out test rows are in the private dataset jayzou3773/less-is-moe-gpqa-main-calibration-64 revision b9596e85179b3017f77ba1436a5d2e61b6a61a5b, following GPQA's access terms.

Inference requires stock vLLM from the unified Less-is-MoE GPU image. IntDim-E has one uniform expert width. IntDim-L/G retain the routed MoE topology and store compact per-expert widths in config.json.