IntDim-L 50% pruned google/gemma-4-26B-A4B

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 128 calibration samples from yentinglin/s1K-1.1-trl-format revision 58a01564d278477da20ead1bcf1cde8e31f36251. The released loader settings are preserved: train, messages, shuffle_seed=1234, seq_length=8192, prefix truncation, no padding, and BF16, with no optimizer step. The exact model-specific token tensors are published at jayzou3773/less-is-moe-s1-calibration-128-seq8192 revision 2a49c84894fe97b29e84c6beb075e9249866ef01. The complete pretrained Gemma 4 causal language-model tower was loaded and pruned in BF16; the vision tower is intentionally excluded from this text-generation checkpoint.

The source-row selection hash is f261e952d4e6d5dec6d37db4ab22636761b215281d33896060fe4467bb352784 and the model-specific token-file hash is 6d1d9ebef111505060d508584dcfaf3f25498b7fc37834a22d3e7ecfa7f791df. Full export and zero-mask equivalence metadata are in experiment-export.json.

Inference requires the Less-is-MoE ragged vLLM plugin 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.