[Doc]: fix typos in various files (#24821)

Signed-off-by: Didier Durand <durand.didier@gmail.com>
This commit is contained in:
Didier Durand
2025-09-15 01:08:52 -07:00
committed by GitHub
parent a8c0f59973
commit 4979eb79da
7 changed files with 7 additions and 7 deletions
@@ -190,7 +190,7 @@ class MoeWNA16Method(FusedMoEMethodBase):
group_size = self.quant_config.group_size
group_size_div_factor = 1
# make intermediate_size and hidden_size diviable by group_size
# make intermediate_size and hidden_size divisible by group_size
# we reduce the group size to ensure that
# and we would repeat the loaded_weight later
while intermediate_size_per_partition % group_size or \
@@ -19,7 +19,7 @@ class MarlinWorkspace:
def __init__(self, out_features, min_thread_n, max_parallel):
assert (out_features % min_thread_n == 0), (
"out_features = {} is undivisible by min_thread_n = {}".format(
"out_features = {} is indivisible by min_thread_n = {}".format(
out_features, min_thread_n))
max_workspace_size = ((out_features // min_thread_n) * max_parallel)
+1 -1
View File
@@ -649,7 +649,7 @@ def _sample_with_torch(
else:
sampled_token_ids_tensor = None
# Counterintiutively, having two loops here is actually faster.
# Counterintuitively, having two loops here is actually faster.
# The first loop can run without waiting on GPU<->CPU sync.
for sampling_type in SamplingType:
sample_indices = categorized_sample_indices[sampling_type]