This commit is contained in:
wangbluo
2024-09-25 18:56:18 +08:00
parent 65c8297710
commit 6fb1322db1
3 changed files with 13 additions and 9 deletions

View File

@@ -859,6 +859,7 @@ def get_gpt2_flash_attention_forward(shard_config: Optional[ShardConfig] = None)
sp_mode = shard_config.sequence_parallelism_mode
sp_group = shard_config.sequence_parallel_process_group
tp_group = shard_config.tensor_parallel_process_group
if sp_mode == "ring_attn":
attn_output = RingAttention.attention(
query,