chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[Bugfix] Fix KeyError: 'prompt_tokens' when streaming requests are aborted (#19514)

Browse Source
This commit is contained in:
Jun Liu
2026-03-01 13:21:35 +09:00
committed by GitHub
parent 4ec450e97b
commit 38dc372dae
3 changed files with 12 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
python/sglang/srt/entrypoints/openai/serving_chat.py
View File

@@ -632,8 +632,10 @@ class OpenAIServingChat(OpenAIServingBase):
):
index = content.get("index", 0)
prompt_tokens[index] = content["meta_info"]["prompt_tokens"]
completion_tokens[index] = content["meta_info"]["completion_tokens"]
prompt_tokens[index] = content["meta_info"].get("prompt_tokens", 0)
completion_tokens[index] = content["meta_info"].get(
"completion_tokens", 0
)
cached_tokens[index] = content["meta_info"].get("cached_tokens", 0)
hidden_states[index] = content["meta_info"].get("hidden_states", None)
routed_experts[index] = content["meta_info"].get("routed_experts", None)

6
python/sglang/srt/entrypoints/openai/serving_completions.py
View File

@@ -209,8 +209,10 @@ class OpenAIServingCompletion(OpenAIServingBase):
index = content.get("index", 0)
text = content["text"]
prompt_tokens[index] = content["meta_info"]["prompt_tokens"]
completion_tokens[index] = content["meta_info"]["completion_tokens"]
prompt_tokens[index] = content["meta_info"].get("prompt_tokens", 0)
completion_tokens[index] = content["meta_info"].get(
"completion_tokens", 0
)
cached_tokens[index] = content["meta_info"].get("cached_tokens", 0)
hidden_states[index] = content["meta_info"].get("hidden_states", None)
routed_experts[index] = content["meta_info"].get("routed_experts", None)

6
python/sglang/srt/entrypoints/openai/usage_processor.py
View File

@@ -20,10 +20,12 @@ class UsageProcessor:
n_choices: int = 1,
enable_cache_report: bool = False,
) -> UsageInfo:
completion_tokens = sum(r["meta_info"]["completion_tokens"] for r in responses)
completion_tokens = sum(
r["meta_info"].get("completion_tokens", 0) for r in responses
)
prompt_tokens = sum(
responses[i]["meta_info"]["prompt_tokens"]
responses[i]["meta_info"].get("prompt_tokens", 0)
for i in range(0, len(responses), n_choices)
)