13274243a0
Keep K3 suite selection and report-schema scoring in bash, merge K3/vision dataset_args into dpv4 yamls, and pin EvalScope at 735d920ee911 with local patches. Co-authored-by: Cursor <cursoragent@cursor.com>
118 lines
6.7 KiB
Diff
118 lines
6.7 KiB
Diff
--- /tmp/evalscope-sync/src-v191/evalscope/models/utils/openai.py 2026-07-21 03:18:18.000000000 +0000
|
|
+++ /data1/syy/evalscope/evalstone/evalscope/evalscope/models/utils/openai.py 2026-08-21 03:03:19.641323740 +0000
|
|
@@ -256,6 +256,8 @@
|
|
params['timeout'] = config.timeout
|
|
if config.max_tokens is not None:
|
|
params['max_tokens'] = config.max_tokens
|
|
+ if config.max_completion_tokens is not None:
|
|
+ params['max_completion_tokens'] = config.max_completion_tokens
|
|
if config.frequency_penalty is not None:
|
|
params['frequency_penalty'] = config.frequency_penalty
|
|
if config.stop_seqs is not None:
|
|
@@ -822,6 +824,8 @@
|
|
continue
|
|
collected_chunks.append(chunk)
|
|
for choice in chunk.choices:
|
|
+ # Some proxies (e.g. Vectron) omit choice.index; default to 0.
|
|
+ choice_index = choice.index if choice.index is not None else 0
|
|
reasoning_content = getattr(choice.delta, 'reasoning_content', None)
|
|
if reasoning_content is None or reasoning_content == '':
|
|
reasoning_content = getattr(choice.delta, 'reasoning', None)
|
|
@@ -835,11 +839,11 @@
|
|
|
|
# Handle reasoning content
|
|
if reasoning_content is not None:
|
|
- collected_reasoning[choice.index].append(str(reasoning_content))
|
|
+ collected_reasoning[choice_index].append(str(reasoning_content))
|
|
|
|
# Handle regular content
|
|
if choice.delta.content is not None:
|
|
- collected_messages[choice.index].append(choice.delta.content)
|
|
+ collected_messages[choice_index].append(choice.delta.content)
|
|
|
|
# Handle tool calls
|
|
if hasattr(choice.delta, 'tool_calls') and choice.delta.tool_calls:
|
|
@@ -847,8 +851,8 @@
|
|
tool_id = tool_call.index
|
|
|
|
# Initialize tool call if not present
|
|
- if tool_id not in collected_tool_calls[choice.index]:
|
|
- collected_tool_calls[choice.index][tool_id] = {
|
|
+ if tool_id not in collected_tool_calls[choice_index]:
|
|
+ collected_tool_calls[choice_index][tool_id] = {
|
|
'id': tool_call.id if hasattr(tool_call, 'id') and tool_call.id else None,
|
|
'type': tool_call.type if hasattr(tool_call, 'type') and tool_call.type else None,
|
|
'function': {
|
|
@@ -860,15 +864,15 @@
|
|
# Update tool call with new chunks
|
|
if hasattr(tool_call, 'function'):
|
|
if hasattr(tool_call.function, 'name') and tool_call.function.name:
|
|
- collected_tool_calls[choice.index][tool_id]['function']['name'] = tool_call.function.name
|
|
+ collected_tool_calls[choice_index][tool_id]['function']['name'] = tool_call.function.name
|
|
|
|
if hasattr(tool_call.function, 'arguments') and tool_call.function.arguments:
|
|
- collected_tool_calls[choice.index
|
|
+ collected_tool_calls[choice_index
|
|
][tool_id]['function']['arguments'] += tool_call.function.arguments
|
|
|
|
# Update ID if it was received later
|
|
if hasattr(tool_call, 'id') and tool_call.id:
|
|
- collected_tool_calls[choice.index][tool_id]['id'] = tool_call.id
|
|
+ collected_tool_calls[choice_index][tool_id]['id'] = tool_call.id
|
|
|
|
# Get all unique choice indices from all collections
|
|
all_indices = set(collected_messages.keys()) | set(collected_reasoning.keys()) | set(collected_tool_calls.keys())
|
|
@@ -953,6 +957,8 @@
|
|
continue
|
|
collected_chunks.append(chunk)
|
|
for choice in chunk.choices:
|
|
+ # Some proxies (e.g. Vectron) omit choice.index; default to 0.
|
|
+ choice_index = choice.index if choice.index is not None else 0
|
|
reasoning_content = getattr(choice.delta, 'reasoning_content', None)
|
|
if reasoning_content is None or reasoning_content == '':
|
|
reasoning_content = getattr(choice.delta, 'reasoning', None)
|
|
@@ -966,11 +972,11 @@
|
|
|
|
# Handle reasoning content
|
|
if reasoning_content is not None:
|
|
- collected_reasoning[choice.index].append(str(reasoning_content))
|
|
+ collected_reasoning[choice_index].append(str(reasoning_content))
|
|
|
|
# Handle regular content
|
|
if choice.delta.content is not None:
|
|
- collected_messages[choice.index].append(choice.delta.content)
|
|
+ collected_messages[choice_index].append(choice.delta.content)
|
|
|
|
# Handle tool calls
|
|
if hasattr(choice.delta, 'tool_calls') and choice.delta.tool_calls:
|
|
@@ -978,8 +984,8 @@
|
|
tool_id = tool_call.index
|
|
|
|
# Initialize tool call if not present
|
|
- if tool_id not in collected_tool_calls[choice.index]:
|
|
- collected_tool_calls[choice.index][tool_id] = {
|
|
+ if tool_id not in collected_tool_calls[choice_index]:
|
|
+ collected_tool_calls[choice_index][tool_id] = {
|
|
'id': tool_call.id if hasattr(tool_call, 'id') and tool_call.id else None,
|
|
'type': tool_call.type if hasattr(tool_call, 'type') and tool_call.type else None,
|
|
'function': {
|
|
@@ -991,15 +997,15 @@
|
|
# Update tool call with new chunks
|
|
if hasattr(tool_call, 'function'):
|
|
if hasattr(tool_call.function, 'name') and tool_call.function.name:
|
|
- collected_tool_calls[choice.index][tool_id]['function']['name'] = tool_call.function.name
|
|
+ collected_tool_calls[choice_index][tool_id]['function']['name'] = tool_call.function.name
|
|
|
|
if hasattr(tool_call.function, 'arguments') and tool_call.function.arguments:
|
|
- collected_tool_calls[choice.index
|
|
+ collected_tool_calls[choice_index
|
|
][tool_id]['function']['arguments'] += tool_call.function.arguments
|
|
|
|
# Update ID if it was received later
|
|
if hasattr(tool_call, 'id') and tool_call.id:
|
|
- collected_tool_calls[choice.index][tool_id]['id'] = tool_call.id
|
|
+ collected_tool_calls[choice_index][tool_id]['id'] = tool_call.id
|
|
|
|
# Get all unique choice indices from all collections
|
|
all_indices = set(collected_messages.keys()) | set(collected_reasoning.keys()) | set(collected_tool_calls.keys())
|