Skip to content

[BUGFIX] Restored handling of ROCM FA output as before adaptation of llama3.2 #241

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
maleksan85 merged 3 commits into from
Oct 23, 2024
Merged

[BUGFIX] Restored handling of ROCM FA output as before adaptation of llama3.2 #241

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
ef1f554
improved handling of output to be the same as before
Oct 22, 2024
c801632
Merge branch 'main' into fix_for_output_in_rocm_fa
Oct 22, 2024
ff36712
after merge correction
Oct 22, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 2 additions & 4 deletions vllm/attention/backends/rocm_flash_attn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -610,18 +610,17 @@ def forward(
assert attn_metadata.num_encoder_tokens is not None
num_prefill_tokens = attn_metadata.num_encoder_tokens

output = torch.empty_like(query)
# Query for decode. KV is not needed because it is already cached.
decode_query = query[num_prefill_tokens:]

# QKV for prefill.
query = query[:num_prefill_tokens]

if key is not None and value is not None:
key = key[:num_prefill_tokens]
value = value[:num_prefill_tokens]

if prefill_meta := attn_metadata.prefill_metadata:
output = torch.empty_like(query)

# Prompt run.
# normal attention and DECODER
if attn_type == AttentionType.DECODER and (
Expand Down Expand Up @@ -738,7 +737,6 @@ def forward(
if decode_meta := attn_metadata.decode_metadata:
# Decoding run.
# Whether to use rocm custom paged attention or not
output = torch.empty_like(decode_query)
num_seqs, num_heads, head_size = decode_query.shape
block_size = value_cache.shape[3]
gqa_ratio = num_heads // self.num_kv_heads
Expand Down