vodorod/Dorod-Sky
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

fix cua engine (#4036)

Browse Source
This commit is contained in:
LawyZheng
2025-11-20 02:24:00 +08:00
committed by GitHub
parent 31a79a1313
commit 0b47dff89d
1 changed files with 25 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

49
skyvern/forge/sdk/api/llm/api_handler_factory.py
View File

@@ -1121,6 +1121,31 @@ class LLMCaller:
thought_cost=call_stats.llm_cost, thought_cost=call_stats.llm_cost,
) )
organization_id = organization_id or (
step.organization_id if step else (thought.organization_id if thought else None)
)
# Track LLM API handler duration, token counts, and cost
duration_seconds = time.perf_counter() - start_time
LOG.info(
"LLM API handler duration metrics",
llm_key=self.llm_key,
prompt_name=prompt_name,
model=self.llm_config.model_name,
duration_seconds=duration_seconds,
step_id=step.step_id if step else None,
thought_id=thought.observer_thought_id if thought else None,
organization_id=organization_id,
input_tokens=call_stats.input_tokens if call_stats and call_stats.input_tokens else None,
output_tokens=call_stats.output_tokens if call_stats and call_stats.output_tokens else None,
reasoning_tokens=call_stats.reasoning_tokens if call_stats and call_stats.reasoning_tokens else None,
cached_tokens=call_stats.cached_tokens if call_stats and call_stats.cached_tokens else None,
llm_cost=call_stats.llm_cost if call_stats and call_stats.llm_cost else None,
)
# Raw response is used for CUA engine LLM calls.
if raw_response:
return response.model_dump(exclude_none=True)
parsed_response = parse_api_response(response, self.llm_config.add_assistant_prefix) parsed_response = parse_api_response(response, self.llm_config.add_assistant_prefix)
parsed_response_json = json.dumps(parsed_response, indent=2) parsed_response_json = json.dumps(parsed_response, indent=2)
if step and not is_speculative_step: if step and not is_speculative_step:
@@ -1149,27 +1174,6 @@ class LLMCaller:
ai_suggestion=ai_suggestion, ai_suggestion=ai_suggestion,
) )
organization_id = organization_id or (
step.organization_id if step else (thought.organization_id if thought else None)
)
# Track LLM API handler duration, token counts, and cost
duration_seconds = time.perf_counter() - start_time
LOG.info(
"LLM API handler duration metrics",
llm_key=self.llm_key,
prompt_name=prompt_name,
model=self.llm_config.model_name,
duration_seconds=duration_seconds,
step_id=step.step_id if step else None,
thought_id=thought.observer_thought_id if thought else None,
organization_id=organization_id,
input_tokens=call_stats.input_tokens if call_stats and call_stats.input_tokens else None,
output_tokens=call_stats.output_tokens if call_stats and call_stats.output_tokens else None,
reasoning_tokens=call_stats.reasoning_tokens if call_stats and call_stats.reasoning_tokens else None,
cached_tokens=call_stats.cached_tokens if call_stats and call_stats.cached_tokens else None,
llm_cost=call_stats.llm_cost if call_stats and call_stats.llm_cost else None,
)
if step and is_speculative_step: if step and is_speculative_step:
step.speculative_llm_metadata = SpeculativeLLMMetadata( step.speculative_llm_metadata = SpeculativeLLMMetadata(
prompt=llm_prompt_value, prompt=llm_prompt_value,
@@ -1187,9 +1191,6 @@ class LLMCaller:
llm_cost=call_stats.llm_cost, llm_cost=call_stats.llm_cost,
) )
if raw_response:
return response.model_dump(exclude_none=True)
return parsed_response return parsed_response
def get_screenshot_resize_target_dimension(self, window_dimension: Resolution | None) -> Resolution: def get_screenshot_resize_target_dimension(self, window_dimension: Resolution | None) -> Resolution: