Merge pull request #10362 from Seniorsimo/usage-openai-compatible

**fix** Added OpenAI usage standard keys to API signature

backend/open_webui/utils/response.py

@@ -23,18 +23,8 @@ def convert_ollama_tool_call_to_openai(tool_calls: dict) -> dict:
         openai_tool_calls.append(openai_tool_call)
     return openai_tool_calls
 
-
-def convert_response_ollama_to_openai(ollama_response: dict) -> dict:
-    model = ollama_response.get("model", "ollama")
-    message_content = ollama_response.get("message", {}).get("content", "")
-    tool_calls = ollama_response.get("message", {}).get("tool_calls", None)
-    openai_tool_calls = None
-
-    if tool_calls:
-        openai_tool_calls = convert_ollama_tool_call_to_openai(tool_calls)
-
-    data = ollama_response
-    usage = {
+def convert_ollama_usage_to_openai(data: dict) -> dict:
+    return {
         "response_token/s": (
             round(
                 (
@@ -66,14 +56,37 @@ def convert_response_ollama_to_openai(ollama_response: dict) -> dict:
         "total_duration": data.get("total_duration", 0),
         "load_duration": data.get("load_duration", 0),
         "prompt_eval_count": data.get("prompt_eval_count", 0),
+        "prompt_tokens": int(data.get("prompt_eval_count", 0)), # This is the OpenAI compatible key
         "prompt_eval_duration": data.get("prompt_eval_duration", 0),
         "eval_count": data.get("eval_count", 0),
+        "completion_tokens": int(data.get("eval_count", 0)), # This is the OpenAI compatible key
         "eval_duration": data.get("eval_duration", 0),
         "approximate_total": (lambda s: f"{s // 3600}h{(s % 3600) // 60}m{s % 60}s")(
             (data.get("total_duration", 0) or 0) // 1_000_000_000
         ),
+        "total_tokens": int( # This is the OpenAI compatible key
+            data.get("prompt_eval_count", 0) + data.get("eval_count", 0)
+        ),
+        "completion_tokens_details": { # This is the OpenAI compatible key
+            "reasoning_tokens": 0,
+            "accepted_prediction_tokens": 0,
+            "rejected_prediction_tokens": 0
+        }
     }
 
+def convert_response_ollama_to_openai(ollama_response: dict) -> dict:
+    model = ollama_response.get("model", "ollama")
+    message_content = ollama_response.get("message", {}).get("content", "")
+    tool_calls = ollama_response.get("message", {}).get("tool_calls", None)
+    openai_tool_calls = None
+
+    if tool_calls:
+        openai_tool_calls = convert_ollama_tool_call_to_openai(tool_calls)
+
+    data = ollama_response
+
+    usage = convert_ollama_usage_to_openai(data)
+
     response = openai_chat_completion_message_template(
         model, message_content, openai_tool_calls, usage
     )
@@ -96,45 +109,7 @@ async def convert_streaming_response_ollama_to_openai(ollama_streaming_response)
 
         usage = None
         if done:
-            usage = {
-                "response_token/s": (
-                    round(
-                        (
-                            (
-                                data.get("eval_count", 0)
-                                / ((data.get("eval_duration", 0) / 10_000_000))
-                            )
-                            * 100
-                        ),
-                        2,
-                    )
-                    if data.get("eval_duration", 0) > 0
-                    else "N/A"
-                ),
-                "prompt_token/s": (
-                    round(
-                        (
-                            (
-                                data.get("prompt_eval_count", 0)
-                                / ((data.get("prompt_eval_duration", 0) / 10_000_000))
-                            )
-                            * 100
-                        ),
-                        2,
-                    )
-                    if data.get("prompt_eval_duration", 0) > 0
-                    else "N/A"
-                ),
-                "total_duration": data.get("total_duration", 0),
-                "load_duration": data.get("load_duration", 0),
-                "prompt_eval_count": data.get("prompt_eval_count", 0),
-                "prompt_eval_duration": data.get("prompt_eval_duration", 0),
-                "eval_count": data.get("eval_count", 0),
-                "eval_duration": data.get("eval_duration", 0),
-                "approximate_total": (
-                    lambda s: f"{s // 3600}h{(s % 3600) // 60}m{s % 60}s"
-                )((data.get("total_duration", 0) or 0) // 1_000_000_000),
-            }
+            usage = convert_ollama_usage_to_openai(data)
 
         data = openai_chat_chunk_message_template(
             model, message_content if not done else None, openai_tool_calls, usage