[07:18:57][DEBUG] hyper_parallel.fully_shard: hook=forward_pre enter module=HSDPGlmImageMoeForConditionalGeneration@fffe834590d0
[07:18:57][DEBUG] hyper_parallel.fully_shard: hook=forward_pre action=unshard module=HSDPGlmImageMoeForConditionalGeneration@fffe834590d0
[07:18:57][DEBUG] hyper_parallel.fully_shard: action=unshard module=HSDPGlmImageMoeForConditionalGeneration@fffe834590d0 async_op=False shard_params=Embedding.weight@fffe832dc850, Glm4MoeLiteRMSNorm.weight@fffe832dcc10, Linear.weight@fffe832dd010, Linear.weight@fffe832dd310, Linear.bias@fffe832dd650 replicate_params=- unshard_replicate=True
[07:18:58][DEBUG] hyper_parallel.fully_shard: action=wait_unshard module=HSDPGlmImageMoeForConditionalGeneration@fffe834590d0 shard_params=Embedding.weight@fffe832dc850, Glm4MoeLiteRMSNorm.weight@fffe832dcc10, Linear.weight@fffe832dd010, Linear.weight@fffe832dd310, Linear.bias@fffe832dd650 replicate_params=- wait_for_replicate=True
[07:18:59][DEBUG] hyper_parallel.fully_shard: hook=forward_pre enter module=HSDPGlm4MoeLiteDecoderLayer@fffe82d7c390
[07:18:59][DEBUG] hyper_parallel.fully_shard: hook=forward_pre action=unshard module=model.language_model.layers.0
[07:18:59][DEBUG] hyper_parallel.fully_shard: action=unshard module=model.language_model.layers.0 async_op=False shard_params=Linear.weight@fffe8329d3d0, Glm4MoeLiteRMSNorm.weight@fffeb817b3d0, Linear.weight@fffeda95e250, Linear.weight@fffe8329e5d0, Glm4MoeLiteRMSNorm.weight@fffe8329e950, Linear.weight@fffe8329ee10, Linear.weight@fffeb817b550, Linear.weight@fffe8321b1d0, Linear.weight@fffe8329fa50, Linear.weight@fffe8329edd0, Glm4MoeLiteRMSNorm.weight@fffe832ac410, Glm4MoeLiteRMSNorm.weight@fffe832ac950 replicate_params=- unshard_replicate=True
[07:18:59][DEBUG] hyper_parallel.fully_shard: action=wait_unshard module=model.language_model.layers.0 shard_params=Linear.weight@fffe8329d3d0, Glm4MoeLiteRMSNorm.weight@fffeb817b3d0, Linear.weight@fffeda95e250, Linear.weight@fffe8329e5d0, Glm4MoeLiteRMSNorm.weight@fffe8329e950, Linear.weight@fffe8329ee10, Linear.weight@fffeb817b550, Linear.weight@fffe8321b1d0, Linear.weight@fffe8329fa50, Linear.weight@fffe8329edd0, Glm4MoeLiteRMSNorm.weight@fffe832ac410, Glm4MoeLiteRMSNorm.weight@fffe832ac950 replicate_params=- wait_for_replicate=True
[07:18:59][DEBUG] hyper_parallel.fully_shard: hook=forward_pre action=prefetch module=model.language_model.layers.0 target=HSDPGlm4MoeLiteDecoderLayer@fffe83477290
[07:18:59][DEBUG] hyper_parallel.fully_shard: action=prefetch module=HSDPGlm4MoeLiteDecoderLayer@fffe83477290 shard_params=Linear.weight@fffe832adf10, Glm4MoeLiteRMSNorm.weight@fffe83213190, Linear.weight@fffe832ae810, Linear.weight@fffe832aec50, Glm4MoeLiteRMSNorm.weight@fffe832af090, Linear.weight@fffe832af590, Linear.weight@fffe832afa90, Glm4MoeLiteFusedExperts.gate_proj@fffe832afe90, Glm4MoeLiteFusedExperts.up_proj@fffeda9270d0, Glm4MoeLiteFusedExperts.down_proj@fffe832af550, Glm4MoeLiteTopkRouter.weight@fffe832b5110, Linear.weight@fffe832b5450, Linear.weight@fffe832b5810, Linear.weight@fffe832b5c50, Glm4MoeLiteRMSNorm.weight@fffe832b6090, Glm4MoeLiteRMSNorm.weight@fffe832b6590 replicate_params=- unshard_replicate=True
[07:18:59][DEBUG] hyper_parallel.fully_shard: action=unshard module=HSDPGlm4MoeLiteDecoderLayer@fffe83477290 async_op=True shard_params=Linear.weight@fffe832adf10, Glm4MoeLiteRMSNorm.weight@fffe83213190, Linear.weight@fffe832ae810, Linear.weight@fffe832aec50, Glm4MoeLiteRMSNorm.weight@fffe832af090, Linear.weight@fffe832af590, Linear.weight@fffe832afa90, Glm4MoeLiteFusedExperts.gate_proj@fffe832afe90, Glm4MoeLiteFusedExperts.up_proj@fffeda9270d0, Glm4MoeLiteFusedExperts.down_proj@fffe832af550, Glm4MoeLiteTopkRouter.weight@fffe832b5110, Linear.weight@fffe832b5450, Linear.weight@fffe832b5810, Linear.weight@fffe832b5c50, Glm4MoeLiteRMSNorm.weight@fffe832b6090, Glm4MoeLiteRMSNorm.weight@fffe832b6590 replicate_params=- unshard_replicate=True
[07:19:00][DEBUG] hyper_parallel.fully_shard: hook=forward enter module=model.language_model.layers.0
[07:19:00][DEBUG] hyper_parallel.fully_shard: hook=forward action=reshard module=model.language_model.layers.0
[07:19:00][DEBUG] hyper_parallel.fully_shard: action=reshard module=model.language_model.layers.0 shard_params=Linear.weight@fffe8329d3d0, Glm4MoeLiteRMSNorm.weight@fffeb817b3d0, Linear.weight@fffeda95e250, Linear.weight@fffe8329e5d0, Glm4MoeLiteRMSNorm.weight@fffe8329e950, Linear.weight@fffe8329ee10, Linear.weight@fffeb817b550, Linear.weight@fffe8321b1d0, Linear.weight@fffe8329fa50, Linear.weight@fffe8329edd0, Glm4MoeLiteRMSNorm.weight@fffe832ac410, Glm4MoeLiteRMSNorm.weight@fffe832ac950 replicate_params=- shard_replicate=False
[07:19:00][DEBUG] hyper_parallel.fully_shard: hook=forward_pre enter module=HSDPGlm4MoeLiteDecoderLayer@fffe83477290
[07:19:00][DEBUG] hyper_parallel.fully_shard: hook=forward_pre action=unshard module=model.language_model.layers.1
[07:19:00][DEBUG] hyper_parallel.fully_shard: action=unshard module=model.language_model.layers.1 async_op=False shard_params=Linear.weight@fffe832adf10, Glm4MoeLiteRMSNorm.weight@fffe83213190, Linear.weight@fffe832ae810, Linear.weight@fffe832aec50, Glm4MoeLiteRMSNorm.weight@fffe832af090, Linear.weight@fffe832af590, Linear.weight@fffe832afa90, Glm4MoeLiteFusedExperts.gate_proj@fffe832afe90, Glm4MoeLiteFusedExperts.up_proj@fffeda9270d0, Glm4MoeLiteFusedExperts.down_proj@fffe832af550, Glm4MoeLiteTopkRouter.weight@fffe832b5110, Linear.weight@fffe832b5450, Linear.weight@fffe832b5810, Linear.weight@fffe832b5c50, Glm4MoeLiteRMSNorm.weight@fffe832b6090, Glm4MoeLiteRMSNorm.weight@fffe832b6590 replicate_params=- unshard_replicate=True
[07:19:00][DEBUG] hyper_parallel.fully_shard: action=wait_unshard module=model.language_model.layers.1 shard_params=Linear.weight@fffe832adf10, Glm4MoeLiteRMSNorm.weight@fffe83213190, Linear.weight@fffe832ae810, Linear.weight@fffe832aec50, Glm4MoeLiteRMSNorm.weight@fffe832af090, Linear.weight@fffe832af590, Linear.weight@fffe832afa90, Glm4MoeLiteFusedExperts.gate_proj@fffe832afe90, Glm4MoeLiteFusedExperts.up_proj@fffeda9270d0, Glm4MoeLiteFusedExperts.down_proj@fffe832af550, Glm4MoeLiteTopkRouter.weight@fffe832b5110, Linear.weight@fffe832b5450, Linear.weight@fffe832b5810, Linear.weight@fffe832b5c50, Glm4MoeLiteRMSNorm.weight@fffe832b6090, Glm4MoeLiteRMSNorm.weight@fffe832b6590 replicate_params=- wait_for_replicate=True
[07:19:00][DEBUG] hyper_parallel.fully_shard: hook=forward_pre action=prefetch module=model.language_model.layers.1 target=model.language_model.layers.0
[07:19:00][DEBUG] hyper_parallel.fully_shard: action=prefetch module=model.language_model.layers.0 shard_params=Linear.weight@fffe8329d3d0, Glm4MoeLiteRMSNorm.weight@fffeb817b3d0, Linear.weight@fffeda95e250, Linear.weight@fffe8329e5d0, Glm4MoeLiteRMSNorm.weight@fffe8329e950, Linear.weight@fffe8329ee10, Linear.weight@fffeb817b550, Linear.weight@fffe8321b1d0, Linear.weight@fffe8329fa50, Linear.weight@fffe8329edd0, Glm4MoeLiteRMSNorm.weight@fffe832ac410, Glm4MoeLiteRMSNorm.weight@fffe832ac950 replicate_params=- unshard_replicate=True
[07:19:00][DEBUG] hyper_parallel.fully_shard: action=unshard module=model.language_model.layers.0 async_op=True shard_params=Linear.weight@fffe8329d3d0, Glm4MoeLiteRMSNorm.weight@fffeb817b3d0, Linear.weight@fffeda95e250, Linear.weight@fffe8329e5d0, Glm4MoeLiteRMSNorm.weight@fffe8329e950, Linear.weight@fffe8329ee10, Linear.weight@fffeb817b550, Linear.weight@fffe8321b1d0, Linear.weight@fffe8329fa50, Linear.weight@fffe8329edd0, Glm4MoeLiteRMSNorm.weight@fffe832ac410, Glm4MoeLiteRMSNorm.weight@fffe832ac950 replicate_params=- unshard_replicate=True
[WARNING][veomni.models.cogview.glm4_moe_lite_glm_image_ar.moe:327] 06/11/2026 07:19:00 >> You tried to access the ExpertsInterface with a config._experts_implementation set to None. This is expected if you use an Expert Module as a standalone Module. If this is not the case, something went wrong with the dispatch of config._experts_implementation
[07:19:00][DEBUG] hyper_parallel.fully_shard: hook=forward enter module=model.language_model.layers.1
[07:19:00][DEBUG] hyper_parallel.fully_shard: hook=forward action=reshard module=model.language_model.layers.1
[07:19:00][DEBUG] hyper_parallel.fully_shard: action=reshard module=model.language_model.layers.1 shard_params=Linear.weight@fffe832adf10, Glm4MoeLiteRMSNorm.weight@fffe83213190, Linear.weight@fffe832ae810, Linear.weight@fffe832aec50, Glm4MoeLiteRMSNorm.weight@fffe832af090, Linear.weight@fffe832af590, Linear.weight@fffe832afa90, Glm4MoeLiteFusedExperts.gate_proj@fffe832afe90, Glm4MoeLiteFusedExperts.up_proj@fffeda9270d0, Glm4MoeLiteFusedExperts.down_proj@fffe832af550, Glm4MoeLiteTopkRouter.weight@fffe832b5110, Linear.weight@fffe832b5450, Linear.weight@fffe832b5810, Linear.weight@fffe832b5c50, Glm4MoeLiteRMSNorm.weight@fffe832b6090, Glm4MoeLiteRMSNorm.weight@fffe832b6590 replicate_params=- shard_replicate=False
[07:19:00][DEBUG] hyper_parallel.fully_shard: hook=forward_pre enter module=HSDPGlm4MoeLiteDecoderLayer@fffe83485890
[07:19:00][DEBUG] hyper_parallel.fully_shard: hook=forward_pre action=unshard module=model.language_model.layers.2
[07:19:00][DEBUG] hyper_parallel.fully_shard: action=unshard module=model.language_model.layers.2 async_op=False shard_params=Linear.weight@fffe832b7a90, Glm4MoeLiteRMSNorm.weight@fffe832b7e10, Linear.weight@fffe832c0410, Linear.weight@fffe832c09d0, Glm4MoeLiteRMSNorm.weight@fffe832c0e10, Linear.weight@fffe832c1390, Linear.weight@fffe832c02d0, Glm4MoeLiteFusedExperts.gate_proj@fffe832c1c10, Glm4MoeLiteFusedExperts.up_proj@fffe832c2090, Glm4MoeLiteFusedExperts.down_proj@fffe832c1350, Glm4MoeLiteTopkRouter.weight@fffe832c2850, Linear.weight@fffe832c2dd0, Linear.weight@fffe832c3390, Linear.weight@fffe832c3990, Glm4MoeLiteRMSNorm.weight@fffe832c3ed0, Glm4MoeLiteRMSNorm.weight@fffe832c8450 replicate_params=- unshard_replicate=True