github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-07-28 05:49:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix pp

Browse Source
This commit is contained in:
flybird11111 2025-06-25 09:48:50 +08:00
parent 7891cc7ddb
commit bd8b57156b
1 changed files with 9 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
colossalai/shardformer/policies/qwen2.py
View File

@ -302,6 +302,15 @@ class Qwen2Policy(Policy):
) )
if self.shard_config.enable_flash_attention or self.shard_config.enable_sequence_parallelism: if self.shard_config.enable_flash_attention or self.shard_config.enable_sequence_parallelism:
if not self.shard_config.enable_tensor_parallelism:
decoder_attribute_replacement = {
"self_attn.hidden_size": self.model.config.hidden_size,
"self_attn.num_heads": self.model.config.num_attention_heads,
}
if getattr(self.model.config, "num_key_value_heads", False):
decoder_attribute_replacement["self_attn.num_key_value_heads"] = self.model.config.num_key_value_heads
policy[Qwen2DecoderLayer] = ModulePolicyDescription(attribute_replacement=decoder_attribute_replacement)
self.append_or_create_method_replacement( self.append_or_create_method_replacement(
description={ description={
"forward": get_qwen2_flash_attention_npu_forward(self.shard_config, sp_mode, sp_size, sp_group), "forward": get_qwen2_flash_attention_npu_forward(self.shard_config, sp_mode, sp_size, sp_group),