Skip to content
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

Update IPEX_LLM_PERFORMANCE_MODE with input length threshold #11908

Merged
merged 5 commits into from
Aug 23, 2024
Merged

Update IPEX_LLM_PERFORMANCE_MODE with input length threshold #11908

Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
5fd6412
Update IPEX_LLM_PERFORMANCE_MODE with input length threshold
Oscilloscope98 Aug 23, 2024
e631ecd
Update based on comments. And and judgement for inputs_embeds
Oscilloscope98 Aug 23, 2024
0f34656
Fix for benchmarking purposes
Oscilloscope98 Aug 23, 2024
0d09144
Update based on comments
Oscilloscope98 Aug 23, 2024
1379be2
Small fix
Oscilloscope98 Aug 23, 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 python/llm/src/ipex_llm/transformers/lookup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,9 @@
original_generate = GenerationMixin.generate
query_group_size = 16

# may tune it with more tested data
PERFORMANCE_MODE_LOOKUP_INPUT_THRESHOLD = 100


@torch.no_grad()
def generate(
Expand All @@ -54,10 +57,12 @@ def generate(
streamer: Optional["BaseStreamer"] = None,
**kwargs,
):
device_name = get_xpu_device_type(inputs)
lookahead = kwargs.pop("lookahead", None)
perf_mode = os.environ.get("IPEX_LLM_PERFORMANCE_MODE", None)
if perf_mode == "1" and lookahead is None:
lookahead = 2 # default to 2 now
if device_name != 'mtl' or inputs.shape[-1] >= PERFORMANCE_MODE_LOOKUP_INPUT_THRESHOLD:
Oscilloscope98 marked this conversation as resolved.
Show resolved Hide resolved
lookahead = 2 # default to 2 now
if lookahead:
from ipex_llm.transformers.convert import get_enable_ipex
_enable_ipex = get_enable_ipex()
Expand Down
Loading