mirror of
https://github.com/exo-explore/exo.git
synced 2026-09-12 21:38:59 -04:00
qwen3_5_moe_split: change async_eval to eval for matches/all_next/target_hidden
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
1 parent
adfcbd8da1
commit
6f623c5647
1 file changed
+1
-1
@@ -117,7 +117,7 @@ def make_split_speculative_next(group): # type: ignore[no-untyped-def]
|
||||
target_tokens = mx.argmax(verify_logits[:, :verify_len, :], axis=-1)
|
||||
matches = mx.equal(target_tokens, drafts_arr).squeeze(0)
|
||||
all_next = mx.argmax(verify_logits[0], axis=-1)
|
||||
mx.async_eval(matches, all_next, target_hidden)
|
||||
mx.eval(matches, all_next, target_hidden)
|
||||
n_accepted = 0
|
||||
for i in range(verify_len):
|
||||
if matches[i].item():
|
||||
|
||||
Reference in new issue
Block a user