From 63a7799550e1b49cb42eecb79e30106bd3d3f1d1 Mon Sep 17 00:00:00 2001 From: yuanlehome Date: Fri, 28 Nov 2025 10:25:26 +0800 Subject: [PATCH 1/3] fix tsp bias add --- fastdeploy/model_executor/load_weight_utils.py | 1 + 1 file changed, 1 insertion(+) diff --git a/fastdeploy/model_executor/load_weight_utils.py b/fastdeploy/model_executor/load_weight_utils.py index 6bf85c0681d..27c409237e1 100644 --- a/fastdeploy/model_executor/load_weight_utils.py +++ b/fastdeploy/model_executor/load_weight_utils.py @@ -289,6 +289,7 @@ def get_expert_ranges(fd_config): if fd_config.parallel_config.use_sequence_parallel_moe: for i in range(fd_config.model_config.moe_layer_start_index, fd_config.model_config.num_hidden_layers): k = f"ernie.{prefix_layer_name}.{i}.self_attn.o_proj.weight" + k = f"ernie.{prefix_layer_name}.{i}.self_attn.o_proj.weight.bias" if k in weight_list: no_tp_action_keys.append(k) tp_actions = cls._get_tensor_parallel_mappings(fd_config.model_config.pretrained_config) From fff8739f7e007f1ee42064ce5c4dfb13db81082b Mon Sep 17 00:00:00 2001 From: yuanlehome Date: Fri, 28 Nov 2025 10:27:41 +0800 Subject: [PATCH 2/3] fix --- fastdeploy/model_executor/load_weight_utils.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/fastdeploy/model_executor/load_weight_utils.py b/fastdeploy/model_executor/load_weight_utils.py index 27c409237e1..8d6bcd8e13e 100644 --- a/fastdeploy/model_executor/load_weight_utils.py +++ b/fastdeploy/model_executor/load_weight_utils.py @@ -289,7 +289,7 @@ def get_expert_ranges(fd_config): if fd_config.parallel_config.use_sequence_parallel_moe: for i in range(fd_config.model_config.moe_layer_start_index, fd_config.model_config.num_hidden_layers): k = f"ernie.{prefix_layer_name}.{i}.self_attn.o_proj.weight" - k = f"ernie.{prefix_layer_name}.{i}.self_attn.o_proj.weight.bias" + k = f"ernie.{prefix_layer_name}.{i}.self_attn.o_proj.bias" if k in weight_list: no_tp_action_keys.append(k) tp_actions = cls._get_tensor_parallel_mappings(fd_config.model_config.pretrained_config) From 9a57fe0f4c6f47eb88628a0413cb95282501251d Mon Sep 17 00:00:00 2001 From: yuanlehome Date: Fri, 28 Nov 2025 10:31:30 +0800 Subject: [PATCH 3/3] fix --- fastdeploy/model_executor/load_weight_utils.py | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-) diff --git a/fastdeploy/model_executor/load_weight_utils.py b/fastdeploy/model_executor/load_weight_utils.py index 8d6bcd8e13e..69c91384fa7 100644 --- a/fastdeploy/model_executor/load_weight_utils.py +++ b/fastdeploy/model_executor/load_weight_utils.py @@ -288,10 +288,13 @@ def get_expert_ranges(fd_config): no_tp_action_keys = copy.deepcopy(num_local_ffn_keys) if fd_config.parallel_config.use_sequence_parallel_moe: for i in range(fd_config.model_config.moe_layer_start_index, fd_config.model_config.num_hidden_layers): - k = f"ernie.{prefix_layer_name}.{i}.self_attn.o_proj.weight" - k = f"ernie.{prefix_layer_name}.{i}.self_attn.o_proj.bias" - if k in weight_list: - no_tp_action_keys.append(k) + no_tp_keys = [ + f"ernie.{prefix_layer_name}.{i}.self_attn.o_proj.weight", + f"ernie.{prefix_layer_name}.{i}.self_attn.o_proj.bias", + ] + for k in no_tp_keys: + if k in weight_list: + no_tp_action_keys.append(k) tp_actions = cls._get_tensor_parallel_mappings(fd_config.model_config.pretrained_config) new_actions = {k: v for k, v in tp_actions.items() if k not in no_tp_action_keys}