tighten chroma version to unbreak create-llama

Probably better performance since we're using the OpenAI compatible
endpoint. Also the model more people will be using.

.changeset/tender-horses-rhyme.md

@@ -0,0 +1,5 @@
+---
+"llamaindex": patch
+---
+
+fix bugs in Together.AI integration (thanks @Nutlope for reporting)

.prettierignore

@@ -1,4 +1,5 @@
 apps/docs/i18n
+apps/docs/docs/api
 pnpm-lock.yaml
 lib/
 dist/

examples/together-ai.ts

@@ -23,6 +23,7 @@ const together = new TogetherLLM({
   for await (const message of generator) {
     process.stdout.write(message.delta);
   }
+  console.log();
   const embedding = new TogetherEmbedding();
   const vector = await embedding.getTextEmbedding("Hello world!");
   console.log("vector:", vector);

packages/core/package.json

@@ -11,7 +11,7 @@
     "@qdrant/js-client-rest": "^1.7.0",
     "@xenova/transformers": "^2.10.0",
     "assemblyai": "^4.0.0",
-    "chromadb": "^1.7.3",
+    "chromadb": "~1.7.3",
     "file-type": "^18.7.0",
     "js-tiktoken": "^1.0.8",
     "lodash": "^4.17.21",

packages/core/src/embeddings/OpenAIEmbedding.ts

@@ -30,7 +30,7 @@ export class OpenAIEmbedding extends BaseEmbedding {
   constructor(init?: Partial<OpenAIEmbedding> & { azure?: AzureOpenAIConfig }) {
     super();
 
-    this.model = OpenAIEmbeddingModelType.TEXT_EMBED_ADA_002;
+    this.model = init?.model ?? OpenAIEmbeddingModelType.TEXT_EMBED_ADA_002;
 
     this.maxRetries = init?.maxRetries ?? 10;
     this.timeout = init?.timeout ?? 60 * 1000; // Default is 60 seconds

packages/core/src/embeddings/together.ts

@@ -1,16 +1,26 @@
 import { OpenAIEmbedding } from "./OpenAIEmbedding";
 
 export class TogetherEmbedding extends OpenAIEmbedding {
-  override model: string;
   constructor(init?: Partial<OpenAIEmbedding>) {
+    const {
+      apiKey = process.env.TOGETHER_API_KEY,
+      additionalSessionOptions = {},
+      model = "togethercomputer/m2-bert-80M-32k-retrieval",
+      ...rest
+    } = init ?? {};
+
+    if (!apiKey) {
+      throw new Error("Set Together Key in TOGETHER_API_KEY env variable"); // Tell user to set correct env variable, and not OPENAI_API_KEY
+    }
+
+    additionalSessionOptions.baseURL =
+      additionalSessionOptions.baseURL ?? "https://api.together.xyz/v1";
+
     super({
-      apiKey: process.env.TOGETHER_API_KEY,
-      ...init,
-      additionalSessionOptions: {
-        ...init?.additionalSessionOptions,
-        baseURL: "https://api.together.xyz/v1",
-      },
+      apiKey,
+      additionalSessionOptions,
+      model,
+      ...rest,
     });
-    this.model = init?.model ?? "togethercomputer/m2-bert-80M-32k-retrieval";
   }
 }

packages/core/src/llm/together.ts

@@ -2,13 +2,25 @@ import { OpenAI } from "./LLM";
 
 export class TogetherLLM extends OpenAI {
   constructor(init?: Partial<OpenAI>) {
+    const {
+      apiKey = process.env.TOGETHER_API_KEY,
+      additionalSessionOptions = {},
+      model = "togethercomputer/llama-2-7b-chat",
+      ...rest
+    } = init ?? {};
+
+    if (!apiKey) {
+      throw new Error("Set Together Key in TOGETHER_API_KEY env variable"); // Tell user to set correct env variable, and not OPENAI_API_KEY
+    }
+
+    additionalSessionOptions.baseURL =
+      additionalSessionOptions.baseURL ?? "https://api.together.xyz/v1";
+
     super({
-      ...init,
-      apiKey: process.env.TOGETHER_API_KEY,
-      additionalSessionOptions: {
-        ...init?.additionalSessionOptions,
-        baseURL: "https://api.together.xyz/v1",
-      },
+      apiKey,
+      additionalSessionOptions,
+      model,
+      ...rest,
     });
   }
 }

pnpm-lock.yaml

@@ -174,7 +174,7 @@ importers:
         specifier: ^4.0.0
         version: 4.0.0
       chromadb:
-        specifier: ^1.7.3
+        specifier: ~1.7.3
         version: 1.7.3(openai@4.20.1)
       file-type:
         specifier: ^18.7.0