Skip to content

Fix device error when using cuda device other than cuda:0 #216

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

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Fix device error when using cuda device other than cuda:0 #216

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
4a4e681
Fix device error when using cuda device other than cuda:0
cornzz Aug 28, 2024
e00ce7c
Merge branch 'mistralai:main' into main
cornzz Aug 28, 2024
83fb40a
Fix formatting in BufferCache.get_input_metadata()
cornzz Aug 29, 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
7 changes: 6 additions & 1 deletion src/mistral_inference/cache.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -177,19 +177,24 @@ def get_input_metadata(self, seqlens: List[int]) -> CacheInputMetadata:
subsequent_prefill = any(seqlen > 1 for seqlen in seqlens)
if first_prefill:
assert all([pos == 0 for pos in seqpos]), seqpos
mask = BlockDiagonalCausalMask.from_seqlens(seqlens).make_local_attention(self.max_seq_len)
mask = BlockDiagonalCausalMask.from_seqlens(
seqlens,
device=self.device,
).make_local_attention(self.max_seq_len)
elif subsequent_prefill:
mask = BlockDiagonalMask.from_seqlens(
q_seqlen=seqlens,
kv_seqlen=[
s + cached_s.clamp(max=self.max_seq_len).item() for (s, cached_s) in zip(seqlens, self.kv_seqlens)
],
device=self.device,
).make_local_attention_from_bottomright(self.max_seq_len)
else:
mask = BlockDiagonalCausalWithOffsetPaddedKeysMask.from_seqlens(
q_seqlen=seqlens,
kv_padding=self.max_seq_len,
kv_seqlen=(self.kv_seqlens + cached_elements).clamp(max=self.max_seq_len).tolist(),
device=self.device,
)

return CacheInputMetadata(
Expand Down