Merge branch 'main' into featherless-provider

Author: VisargD

src/providers/anthropic/chatComplete.ts

@@ -43,7 +43,16 @@ interface AnthropicTool extends PromptCache {
 interface AnthropicToolResultContentItem {
   type: 'tool_result';
   tool_use_id: string;
-  content?: string;
+  content?:
+    | {
+        type: string;
+        text?: string;
+        cache_control?: {
+          type: string;
+          ttl?: number;
+        };
+      }[]
+    | string;
 }
 
 interface AnthropicBase64ImageContentItem {
@@ -138,6 +147,9 @@ const transformAssistantMessage = (msg: Message): AnthropicMessage => {
         input: toolCall.function.arguments?.length
           ? JSON.parse(toolCall.function.arguments)
           : {},
+        ...(toolCall.cache_control && {
+          cache_control: toolCall.cache_control,
+        }),
       });
     });
   }
@@ -155,7 +167,7 @@ const transformToolMessage = (msg: Message): AnthropicMessage => {
       {
         type: 'tool_result',
         tool_use_id,
-        content: msg.content as string,
+        content: msg.content,
       },
     ],
   };
@@ -250,6 +262,9 @@ export const AnthropicChatCompleteConfig: ProviderConfig = {
 
             if (msg.role === 'assistant') {
               messages.push(transformAssistantMessage(msg));
+            } else if (msg.role === 'tool') {
+              // even though anthropic supports images in tool results, openai doesn't support it yet
+              messages.push(transformToolMessage(msg));
             } else if (
               msg.content &&
               typeof msg.content === 'object' &&
@@ -275,9 +290,6 @@ export const AnthropicChatCompleteConfig: ProviderConfig = {
                 }
               });
               messages.push(transformedMessage as AnthropicMessage);
-            } else if (msg.role === 'tool') {
-              // even though anthropic supports images in tool results, openai doesn't support it yet
-              messages.push(transformToolMessage(msg));
             } else {
               messages.push({
                 role: msg.role,

src/providers/bedrock/utils.ts

@@ -407,11 +407,26 @@ export const populateHyperParameters = (value: FinetuneRequest) => {
 
 export const getInferenceProfile = async (
   inferenceProfileIdentifier: string,
-  awsRegion: string,
-  awsAccessKeyId: string,
-  awsSecretAccessKey: string,
-  awsSessionToken?: string
+  providerOptions: Options,
+  c: Context
 ) => {
+  if (providerOptions.awsAuthType === 'assumedRole') {
+    const { accessKeyId, secretAccessKey, sessionToken } =
+      (await getAssumedRoleCredentials(
+        c,
+        providerOptions.awsRoleArn || '',
+        providerOptions.awsExternalId || '',
+        providerOptions.awsRegion || ''
+      )) || {};
+    providerOptions.awsAccessKeyId = accessKeyId;
+    providerOptions.awsSecretAccessKey = secretAccessKey;
+    providerOptions.awsSessionToken = sessionToken;
+  }
+
+  const awsRegion = providerOptions.awsRegion || 'us-east-1';
+  const awsAccessKeyId = providerOptions.awsAccessKeyId || '';
+  const awsSecretAccessKey = providerOptions.awsSecretAccessKey || '';
+  const awsSessionToken = providerOptions.awsSessionToken || '';
   const url = `https://bedrock.${awsRegion}.amazonaws.com/inference-profiles/${encodeURIComponent(decodeURIComponent(inferenceProfileIdentifier))}`;
 
   const headers = await generateAWSHeaders(
@@ -463,10 +478,8 @@ export const getFoundationModelFromInferenceProfile = async (
 
     const inferenceProfile = await getInferenceProfile(
       inferenceProfileIdentifier || '',
-      providerOptions.awsRegion || '',
-      providerOptions.awsAccessKeyId || '',
-      providerOptions.awsSecretAccessKey || '',
-      providerOptions.awsSessionToken || ''
+      { ...providerOptions },
+      c
     );
 
     // modelArn is always like arn:aws:bedrock:us-east-1::foundation-model/anthropic.claude-v2:1

src/providers/groq/index.ts

@@ -10,7 +10,11 @@ import { GROQ } from '../../globals';
 
 const GroqConfig: ProviderConfigs = {
   api: GroqAPIConfig,
-  chatComplete: chatCompleteParams(['logprobs', 'logits_bias', 'top_logprobs']),
+  chatComplete: chatCompleteParams(
+    ['logprobs', 'logits_bias', 'top_logprobs'],
+    undefined,
+    { service_tier: { param: 'service_tier', required: false } }
+  ),
   createTranscription: {},
   createTranslation: {},
   createSpeech: createSpeechParams([]),