Skip to content

perf: vectorize KV cache prefix matching with numpy #2179

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
nausicaalii wants to merge 1 commit into
base: main
Choose a base branch
from
+18 −13
Open

perf: vectorize KV cache prefix matching with numpy #2179

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
31 changes: 18 additions & 13 deletions llama_cpp/llama.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -887,12 +887,18 @@ def generate(

# Check for kv cache prefix match
if reset and self.n_tokens > 0:
longest_prefix = 0
for a, b in zip(self._input_ids, tokens[:-1]):
if a == b:
longest_prefix += 1
else:
break
cached = self._input_ids
n = min(len(cached), len(tokens) - 1)
if n > 0:
eq = np.asarray(cached[:n]) == np.asarray(
tokens[:n]
)
mismatch = np.argmin(eq)
longest_prefix = (
int(n) if eq[mismatch] else int(mismatch)
)
else:
longest_prefix = 0
if longest_prefix > 0:
if self._ctx.kv_cache_seq_rm(-1, longest_prefix, -1):
reset = False
Expand Down Expand Up @@ -2252,13 +2258,12 @@ def logits_to_logprobs(

@staticmethod
def longest_token_prefix(a: Sequence[int], b: Sequence[int]):
longest_prefix = 0
for _a, _b in zip(a, b):
if _a == _b:
longest_prefix += 1
else:
break
return longest_prefix
n = min(len(a), len(b))
if n == 0:
return 0
eq = np.asarray(a[:n]) == np.asarray(b[:n])
mismatch = np.argmin(eq)
return int(n) if eq[mismatch] else int(mismatch)

@classmethod
def from_pretrained(
Expand Down