fixes info page display for gemini flash/ultra

src/proxy/middleware/request/preprocessors/validate-context-size.ts

@@ -81,7 +81,7 @@ export const validateContextSize: RequestPreprocessor = async (req) => {
     modelMax = 200000;
   } else if (model.match(/^claude-3/)) {
     modelMax = 200000;
-  } else if (model.match(/^gemini-\d{3}$/)) {
+  } else if (model.match(/^gemini-$/)) {
     modelMax = 1024000;
   } else if (model.match(/^anthropic\.claude-3/)) {
     modelMax = 200000;

src/proxy/middleware/response/error-generator.ts

@@ -143,8 +143,6 @@ export function sendErrorToClient({
     res.setHeader("x-oai-proxy-error-status", redactedOpts.statusCode || 500);
   }
 
-  req.log.info({ statusCode: res.statusCode, isStreaming, format, redactedOpts, event }, "Sending error response");
-  
   if (isStreaming) {
     if (!res.headersSent) {
       initializeSseStream(res);
@@ -222,19 +220,16 @@ export function buildSpoofedCompletion({
         stop_sequence: null,
       };
     case "google-ai":
-      // TODO: Native Google AI non-streaming responses are not supported, this
-      // is an untested guess at what the response should look like.
       return {
         candidates: [
           {
-            content: { parts: [{ text: content }], role: "assistant" },
+            content: { parts: [{ text: content }], role: "model" },
             finishReason: title,
             index: 0,
             tokenCount: null,
             safetyRatings: [],
           },
         ],
-        usage: { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 },
       };
     case "openai-image":
       return obj;

src/service-info.ts

@@ -363,12 +363,14 @@ function addKeyToAggregates(k: KeyPoolKey) {
     }
     case "google-ai": {
       if (!keyIsGoogleAIKey(k)) throw new Error("Invalid key type");
-      const family = "gemini-pro";
-      sumTokens += k["gemini-proTokens"];
-      sumCost += getTokenCostUsd(family, k["gemini-proTokens"]);
-      increment(modelStats, `${family}__active`, k.isDisabled ? 0 : 1);
-      increment(modelStats, `${family}__revoked`, k.isRevoked ? 1 : 0);
-      increment(modelStats, `${family}__tokens`, k["gemini-proTokens"]);
+      k.modelFamilies.forEach((family) => {
+        const tokens = k[`${family}Tokens`];
+        sumTokens += tokens;
+        sumCost += getTokenCostUsd(family, tokens);
+        increment(modelStats, `${family}__tokens`, tokens);
+        increment(modelStats, `${family}__active`, k.isDisabled ? 0 : 1);
+        increment(modelStats, `${family}__revoked`, k.isRevoked ? 1 : 0);
+      });
       break;
     }
     case "mistral-ai": {