From 9217bcbae7dbf86283b746e1074ef39710f3508d Mon Sep 17 00:00:00 2001 From: 3outeille Date: Mon, 11 Aug 2025 16:30:46 +0000 Subject: [PATCH] remove sequence parallel in llama4 --- src/transformers/models/llama4/configuration_llama4.py | 3 --- 1 file changed, 3 deletions(-) diff --git a/src/transformers/models/llama4/configuration_llama4.py b/src/transformers/models/llama4/configuration_llama4.py index 76162ee25964..d74107c2610a 100644 --- a/src/transformers/models/llama4/configuration_llama4.py +++ b/src/transformers/models/llama4/configuration_llama4.py @@ -251,9 +251,6 @@ class Llama4TextConfig(PretrainedConfig): "layers.*.self_attn.k_proj": "colwise", "layers.*.self_attn.v_proj": "colwise", "layers.*.self_attn.o_proj": "rowwise", - "layers.*.input_layernorm.weight": "sequence_parallel", - "layers.*.post_attention_layernorm.weight": "sequence_parallel", - "norm.weight": "sequence_parallel", "layers.*.feed_forward.shared_expert.gate_proj": "local_colwise", "layers.*.feed_forward.shared_expert.up_proj": "local_colwise", "layers.*.feed_forward.shared_expert.down_proj": "local_rowwise",