zhangyiss/mlx-examples
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx-examples.git synced 2025-08-31 11:54:37 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Make llm async eval less brittle

Browse Source
This commit is contained in:
Awni Hannun 2024-10-12 14:24:16 -07:00
parent d8611dd69f
commit 55bcfbc6a5
1 changed files with 5 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
llms/mlx_lm/utils.py
View File

@ -246,10 +246,10 @@ def generate_step(
y, logprobs = _step(y)
mx.async_eval(y)
mx.async_eval(y, logprobs)
while True:
next_y, next_logprobs = _step(y)
mx.async_eval(next_y)
mx.async_eval(next_y, next_logprobs)
yield y.item(), logprobs
y, logprobs = next_y, next_logprobs
@ -348,7 +348,9 @@ def generate(
if formatter:
# We have to finalize so that the prob corresponds to the last segment
detokenizer.finalize()
formatter(detokenizer.last_segment, mx.exp(logprobs[token]).item())
with mx.stream(mx.default_stream(mx.cpu)):
prob = mx.exp(logprobs[token]).item()
formatter(detokenizer.last_segment, prob)
else:
print(detokenizer.last_segment, end="", flush=True)