GithubStar/ktransformers
1
0
Fork 0
mirror of https://github.com/kvcache-ai/ktransformers.git synced 2026-10-28 05:06:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #71 from Azure-Tang/main

Browse Source 
[fix] Fix qlen > chunk_size mask is none error
This commit is contained in:
UnicornChan
2024-09-02 14:51:05 +08:00
committed by GitHub
parent 022b893819 c55de02f7b
commit be81269e38
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ktransformers/operators/attention.py
+2 -2
View File
@@ -195,11 +195,11 @@ class KDeepseekV2Attention(BaseInjectedModule, DeepseekV2Attention):
[:,:min(self.chunck_size, min(past_key_value.max_cache_len-cur_idx, self.chunck_size))]
self.attn_mask[:, :, :, cur_idx+self.chunck_size:] = -1e+38
self.attn_mask[:, :, :, :cur_idx] = 0
chunck_mask = torch.narrow(self.attn_mask, 2, 0, min(self.chunck_size, q_len-cur_idx))
chunk_mask = torch.narrow(self.attn_mask, 2, 0, min(self.chunck_size, q_len-cur_idx))
cur_output, _, _ = self.forward_chunck(
hidden_states[:, cur_idx:min(cur_idx + self.chunck_size, q_len), ...],
chunck_mask,
chunk_mask,
position_ids[:, cur_idx:min(cur_idx + self.chunck_size, q_len)],
past_key_value,
output_attentions,