[Fix] fix input_audio wrong prefix (#1152) (#1174)

labAxiaoming · gemini-code-assist[bot] · web-flow · commit b419590dc087 · 2026-01-30T17:00:04.000+08:00
Update evalscope/models/utils/openai.py

Co-authored-by: gemini-code-assist[bot] &lt;176961590+gemini-code-assist[bot]@users.noreply.github.com&gt;
diff --git a/evalscope/models/utils/openai.py b/evalscope/models/utils/openai.py
@@ -55,7 +55,7 @@
     as_stop_reason,
 )
 from evalscope.api.tool import ToolCall, ToolChoice, ToolFunction, ToolInfo, parse_tool_call
-from evalscope.utils.url_utils import file_as_data_uri, is_http_url
+from evalscope.utils.url_utils import data_uri_to_base64, file_as_data_uri, is_http_url
 
 BASE_64_DATA_REMOVED = '<base64-data-removed>'
 
@@ -103,10 +103,11 @@ def openai_chat_completion_part(content: Content) -> ChatCompletionContentPartPa
             image_url=dict(url=image_url, detail=detail),
         )
     elif content.type == 'audio':
-        audio_data_uri = file_as_data_uri(content.audio)
 
+        # remove prefix
+        audio_uri = data_uri_to_base64(file_as_data_uri(content.audio))
         return ChatCompletionContentPartInputAudioParam(
-            type='input_audio', input_audio=dict(data=audio_data_uri, format=content.format)
+            type='input_audio', input_audio=dict(data=audio_uri, format=content.format)
         )
 
     else:

Original file line number	Diff line number	Diff line change
`@@ -55,7 +55,7 @@`
`55`	`55`	`as_stop_reason,`
`56`	`56`	`)`
`57`	`57`	`from evalscope.api.tool import ToolCall, ToolChoice, ToolFunction, ToolInfo, parse_tool_call`
`58`		`-from evalscope.utils.url_utils import file_as_data_uri, is_http_url`
	`58`	`+from evalscope.utils.url_utils import data_uri_to_base64, file_as_data_uri, is_http_url`
`59`	`59`
`60`	`60`	`BASE_64_DATA_REMOVED = '<base64-data-removed>'`
`61`	`61`
`@@ -103,10 +103,11 @@ def openai_chat_completion_part(content: Content) -> ChatCompletionContentPartPa`
`103`	`103`	`image_url=dict(url=image_url, detail=detail),`
`104`	`104`	`)`
`105`	`105`	`elif content.type == 'audio':`
`106`		`- audio_data_uri = file_as_data_uri(content.audio)`
`107`	`106`
	`107`	`+ # remove prefix`
	`108`	`+ audio_uri = data_uri_to_base64(file_as_data_uri(content.audio))`
`108`	`109`	`return ChatCompletionContentPartInputAudioParam(`
`109`		`- type='input_audio', input_audio=dict(data=audio_data_uri, format=content.format)`
	`110`	`+ type='input_audio', input_audio=dict(data=audio_uri, format=content.format)`
`110`	`111`	`)`
`111`	`112`
`112`	`113`	`else:`