← back to Exo
Revert dumb merge mistake
612f58c78de71d39476e886f2beca177580c68f3 · 2025-11-06 18:39:08 -0800 · rltakashige
Files touched
M src/exo/worker/runner/generate.py
Diff
commit 612f58c78de71d39476e886f2beca177580c68f3
Author: rltakashige <rl.takashige@gmail.com>
Date: Thu Nov 6 18:39:08 2025 -0800
Revert dumb merge mistake
---
src/exo/worker/runner/generate.py | 13 ++++---------
1 file changed, 4 insertions(+), 9 deletions(-)
diff --git a/src/exo/worker/runner/generate.py b/src/exo/worker/runner/generate.py
index 9fe58d40..e6821fd0 100644
--- a/src/exo/worker/runner/generate.py
+++ b/src/exo/worker/runner/generate.py
@@ -215,11 +215,9 @@ def generate_step(
y, logprobs = _step(input_tokens=prompt, input_embeddings=input_embeddings)
- prompt_progress_callback(total_prompt_tokens, total_prompt_tokens)
-
mx.async_eval(y, logprobs)
- next_y, next_logprobs = _step(y)
- mx.async_eval(next_y, next_logprobs)
+ next_y: array | None = None
+ next_logprobs: array | None = None
n = 0
while True:
@@ -236,11 +234,8 @@ def generate_step(
yield int(y.item()), logprobs
if n % 256 == 0:
mx.clear_cache()
- if n == max_tokens:
- break
- y, logprobs = next_y, logprobs
- next_y, next_logprobs = _step(y)
- mx.async_eval(next_y, next_logprobs)
+ y, logprobs = next_y, next_logprobs
+ n += 1
def stream_generate(
← 6bcac37d stop benching on all pushes
·
back to Exo
·
pipeline parallel fix 9058b117 →