adds gpt4-turbo model family and support for gpt-4-1106-preview model

src/proxy/middleware/request/transform-outbound-payload.ts

@@ -65,6 +65,7 @@ const OpenAIV1ChatCompletionSchema = z.object({
   presence_penalty: z.number().optional().default(0),
   logit_bias: z.any().optional(),
   user: z.string().optional(),
+  seed: z.number().int().optional(),
 });
 
 const OpenAIV1TextCompletionSchema = z

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

@@ -42,6 +42,8 @@ export const validateContextSize: RequestPreprocessor = async (req) => {
   let modelMax: number;
   if (model.match(/gpt-3.5-turbo-16k/)) {
     modelMax = 16384;
+  } else if (model.match(/gpt-4-1106(-preview)?/)) {
+    modelMax = 131072;
   } else if (model.match(/gpt-3.5-turbo/)) {
     modelMax = 4096;
   } else if (model.match(/gpt-4-32k/)) {
@@ -60,8 +62,6 @@ export const validateContextSize: RequestPreprocessor = async (req) => {
     // Not sure if AWS Claude has the same context limit as Anthropic Claude.
     modelMax = 100000;
   } else {
-    // Don't really want to throw here because I don't want to have to update
-    // this ASAP every time a new model is released.
     req.log.warn({ model }, "Unknown model, using 100k token limit.");
     modelMax = 100000;
   }

src/proxy/openai.ts

@@ -41,6 +41,7 @@ function getModelsResponse() {
 
   // https://platform.openai.com/docs/models/overview
   const knownModels = [
+    "gpt-4-1106-preview",
     "gpt-4",
     "gpt-4-0613",
     "gpt-4-0314", // EOL 2024-06-13

src/proxy/queue.ts

@@ -211,6 +211,7 @@ function processQueue() {
 
   // TODO: `getLockoutPeriod` uses model names instead of model families
   // TODO: genericize this it's really ugly
+  const gpt4TurboLockout = keyPool.getLockoutPeriod("gpt-4-1106");
   const gpt432kLockout = keyPool.getLockoutPeriod("gpt-4-32k");
   const gpt4Lockout = keyPool.getLockoutPeriod("gpt-4");
   const turboLockout = keyPool.getLockoutPeriod("gpt-3.5-turbo");
@@ -219,6 +220,9 @@ function processQueue() {
   const awsClaudeLockout = keyPool.getLockoutPeriod("anthropic.claude-v2");
 
   const reqs: (Request | undefined)[] = [];
+  if (gpt4TurboLockout === 0) {
+    reqs.push(dequeue("gpt4-turbo"));
+  }
   if (gpt432kLockout === 0) {
     reqs.push(dequeue("gpt4-32k"));
   }