[RL][Frontend] Return weight-version spans on OpenAI completions and chat (#61347)

Signed-off-by: Shuolei Wang <shuoleiwang123@gmail.com>
Co-authored-by: aoshen02 <aoshen@inferact.ai>
This commit is contained in:
Shuolei Wang
2026-10-17 02:08:07 +00:00
committed by GitHub
co-authored by aoshen02
parent 6b6a6c5c98
commit b3a7e2ff6e
7 changed files with 139 additions and 4 deletions
@@ -376,6 +376,7 @@ class OpenAIServingCompletion(GenerateBaseServing):
not delta_text
and not delta_token_ids
and not previous_num_tokens[i]
and finish_reason is None
):
# Chunked prefill case, don't return empty chunks
continue
@@ -421,6 +422,11 @@ class OpenAIServingCompletion(GenerateBaseServing):
if request.return_token_ids
else None
),
weight_versions=(
output.weight_versions
if request.return_weight_versions
else None
),
)
],
)
@@ -594,6 +600,11 @@ class OpenAIServingCompletion(GenerateBaseServing):
as_list(output.token_ids) if request.return_token_ids else None
),
routed_experts=routed_experts_b64,
weight_versions=(
output.weight_versions
if request.return_weight_versions
else None
),
)
choices.append(choice_data)