move rag tests to ollama

.github/workflows/ollama.yml

@@ -40,6 +40,8 @@ jobs:
             - name: download ollama docker
               run: docker run -d -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama
             - name: run summarize-ollama-phi3
-              run: yarn test:summarize --model ollama:phi3 --out ./temp
+              run: yarn test:summarize --model ollama:phi3 --out ./temp/summarize-ollama-phi3
+            - name: run rag
+              run: yarn run:script rag --out ./temp/rag
             - name: test summarize-links-phi3
-              run: yarn test:scripts summarize-link --models ollama:phi3 --out ./temp
+              run: yarn test:scripts summarize-link --models ollama:phi3 --out ./temp/summarize-link

.vscode/settings.json

@@ -23,6 +23,7 @@
         "markdownify",
         "memorystream",
         "millis",
+        "nomic",
         "ollama",
         "openai",
         "promptdom",

docs/src/content/docs/reference/scripts/system.mdx

@@ -710,6 +710,8 @@
         "Function to do a search using embeddings vector similarity distance.",
 })
 
+const embeddingsModel = env.var.embeddingsModel || undefined
+
 defTool(
     "retrieval_vector_search",
     "Search files using embeddings and similarity distance.",
@@ -737,7 +739,7 @@
         const res = await retrieval.vectorSearch(
             q,
             files.map((filename) => ({ filename }))
-        )
+            , { embeddingsModel })
         return YAML.stringify(res.map(({ filename }) => filename))
     }
 )

packages/core/src/genaisrc/system.retrieval_vector_search.genai.js

@@ -4,6 +4,8 @@ system({
         "Function to do a search using embeddings vector similarity distance.",
 })
 
+const embeddingsModel = env.var.embeddingsModel || undefined
+
 defTool(
     "retrieval_vector_search",
     "Search files using embeddings and similarity distance.",
@@ -31,7 +33,7 @@ defTool(
         const res = await retrieval.vectorSearch(
             q,
             files.map((filename) => ({ filename }))
-        )
+            , { embeddingsModel })
         return YAML.stringify(res.map(({ filename }) => filename))
     }
 )

packages/core/src/models.ts

@@ -100,8 +100,8 @@
         try {
             const configuration = await host.getLanguageModelConfiguration(
                 model,
                 {
-                    token: askToken,
+                    token: withToken,
                     signal,
                     trace,
                 }

packages/core/src/types/prompt_template.d.ts

@@ -187,7 +187,14 @@
     /**
      * LLM model to use for embeddings.
      */
-    embeddingsModel?: "openai:text-embedding-ada-002" | string
+    embeddingsModel?: OptionsOrString<
+        "openai:text-embedding-3-small",
+        "openai:text-embedding-3-large",
+        "openai:text-embedding-ada-002",
+        "github:text-embedding-3-small",
+        "github:text-embedding-3-large",
+        "ollama:nomic-embed-text"
+    >
 }
 
 interface EmbeddingsModelOptions extends EmbeddingsModelConnectionOptions {}

packages/core/src/vectorsearch.ts

@@ -8,7 +8,7 @@
 import { EmbeddingCreateParams, EmbeddingCreateResponse } from "./chattypes"
 import { LanguageModelConfiguration } from "./host"
 import { getConfigHeaders } from "./openai"
-import { trimTrailingSlash } from "./util"
+import { logVerbose, trimTrailingSlash } from "./util"
 import { TraceOptions } from "./trace"
 
 export interface EmbeddingsCacheKey {
@@ -72,10 +72,11 @@
             url = `${trimTrailingSlash(base)}/${model.replace(/\./g, "")}/embeddings?api-version=${AZURE_OPENAI_API_VERSION}`
             delete body.model
         } else {
-            url = `${base}/v1/embeddings`
+            url = `${base}/embeddings`
         }
         const fetch = await createFetch({ retryOn: [429] })
         if (trace) traceFetchPost(trace, url, headers, body)
+        logVerbose(`embedding ${model}`)
         const resp = await fetch(url, {
             method: "POST",
             headers,

packages/sample/genaisrc/rag.genai.js

@@ -1,34 +1,44 @@
 script({
     title: "rag",
-    model: "openai:gpt-3.5-turbo",
+    model: "ollama:phi3",
     files: "src/rag/*",
     tests: {
         files: "src/rag/*",
         keywords: ["lorem", "markdown", "word"],
     },
 })
 
-$`You are a helpful assistant. Summarize the files in MARKDOWN, PDF, WORD and ALL.`
+$`Summarize MARKDOWN, PDF, WORD and ALL. Use one short sentence.`
+
+const embeddingsModel = env.vars.embeddingsModel || "ollama:nomic-embed-text"
 
 def(
     "MARKDOWN",
     await retrieval.vectorSearch(
         "markdown",
-        env.files.filter((f) => f.filename.endsWith(".md"))
+        env.files.filter((f) => f.filename.endsWith(".md")),
+        { embeddingsModel }
     )
 )
 def(
     "PDF",
     await retrieval.vectorSearch(
         "lorem",
-        env.files.filter((f) => f.filename.endsWith(".pdf"))
+        env.files.filter((f) => f.filename.endsWith(".pdf")),
+        { embeddingsModel }
     )
 )
 def(
     "WORD",
     await retrieval.vectorSearch(
         "word",
-        env.files.filter((f) => f.filename.endsWith(".docx"))
+        env.files.filter((f) => f.filename.endsWith(".docx")),
+        { embeddingsModel }
     )
 )
-def("ALL", await retrieval.vectorSearch("lorem", env.files))
+def(
+    "ALL",
+    await retrieval.vectorSearch("lorem", env.files, {
+        embeddingsModel,
+    })
+)