[Bugfix] Size and iterate w13 by shard count for non-gated MoE (#51125)

Signed-off-by: aoshen02 <aoshen@inferact.ai>
Co-authored-by: Matej Sirovatka <matej.sirovatka@gmail.com>
Co-authored-by: Rishi Puri <riship@nvidia.com>
Co-authored-by: Hsiao-Yuan Chen <littlecircle0730@users.noreply.github.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
aoshen02
2026-08-05 17:24:47 -07:00
committed by GitHub
co-authored by Matej Sirovatka Rishi Puri Hsiao-Yuan Chen Claude Opus 5
parent 811622c410
commit 7c77868cdf
16 changed files with 104 additions and 80 deletions
+1
View File
@@ -65,6 +65,7 @@ def test_auto_gptq_moe_creates_zero_initialized_expert_biases():
method.quant_config = AutoGPTQConfig(4, 128, False, True, False, {}, {})
method.input_dtype = None
method.experts_cls = None
method.moe = SimpleNamespace(w13_num_shards=2)
layer = torch.nn.Module()
method.create_weights(