Add gemma2:2b test and ollama pull format as cli

packages/cli/src/nodehost.ts

@@ -70,7 +70,7 @@ class ModelManager implements ModelService {
         if (provider === MODEL_PROVIDER_OLLAMA) {
             if (this.pulled.includes(modelid)) return { ok: true }
 
-            logVerbose(`ollama: pulling ${modelid}...`)
+            logVerbose(`ollama pull ${model}`)
             const conn = await this.getModelToken(modelid)
             const res = await fetch(`${conn.base}/api/pull`, {
                 method: "POST",

packages/core/src/models.test.ts

@@ -11,29 +11,37 @@ import {
 // generate unit tests for parseModelIdentifier
 describe("parseModelIdentifier", () => {
     test("aici:gpt-3.5:en", () => {
-        const { provider, model, tag, modelId } =
+        const { provider, model, tag, family } =
             parseModelIdentifier("aici:gpt-3.5:en")
         assert(provider === MODEL_PROVIDER_AICI)
-        assert(model === "gpt-3.5")
+        assert(family === "gpt-3.5")
         assert(tag === "en")
-        assert(modelId === "gpt-3.5:en")
+        assert(model === "gpt-3.5:en")
     })
     test("ollama:phi3", () => {
-        const { provider, model, tag, modelId } =
+        const { provider, model, tag, family } =
             parseModelIdentifier("ollama:phi3")
         assert(provider === MODEL_PROVIDER_OLLAMA)
         assert(model === "phi3")
-        assert(modelId === "phi3")
+        assert(family === "phi3")
+    })
+    test("ollama:gemma2:2b", () => {
+        const { provider, model, tag, family } =
+            parseModelIdentifier("ollama:gemma2:2b")
+        assert(provider === MODEL_PROVIDER_OLLAMA)
+        assert(model === "gemma2:2b")
+        assert(family === "gemma2")
     })
     test("llamafile", () => {
-        const { provider, model } = parseModelIdentifier("llamafile")
+        const { provider, model, family } = parseModelIdentifier("llamafile")
         assert(provider === MODEL_PROVIDER_LLAMAFILE)
-        assert(model === "*")
+        assert(family === "*")
+        assert(model === "llamafile")
     })
     test("gpt4", () => {
-        const { provider, model, modelId } = parseModelIdentifier("gpt4")
+        const { provider, model, family } = parseModelIdentifier("gpt4")
         assert(provider === MODEL_PROVIDER_OPENAI)
         assert(model === "gpt4")
-        assert(modelId === "gpt4")
+        assert(family === "gpt4")
     })
 })

packages/core/src/models.ts

@@ -9,22 +9,27 @@ import { assert } from "./util"
  * provider:model
  * provider:model:tag where modelId model:tag
  */
-export function parseModelIdentifier(id: string) {
+export function parseModelIdentifier(id: string): {
+    provider: string
+    family: string
+    model: string
+    tag?: string
+} {
     assert(!!id)
     id = id.replace("-35-", "-3.5-")
     const parts = id.split(":")
     if (parts.length >= 3)
         return {
             provider: parts[0],
-            model: parts[1],
+            family: parts[1],
             tag: parts.slice(2).join(":"),
-            modelId: parts.slice(1).join(":"),
+            model: parts.slice(1).join(":"),
         }
     else if (parts.length === 2)
-        return { provider: parts[0], model: parts[1], modelId: parts[1] }
+        return { provider: parts[0], family: parts[1], model: parts[1] }
     else if (id === MODEL_PROVIDER_LLAMAFILE)
-        return { provider: MODEL_PROVIDER_LLAMAFILE, model: "*", modelId: id }
-    else return { provider: MODEL_PROVIDER_OPENAI, model: id, modelId: id }
+        return { provider: MODEL_PROVIDER_LLAMAFILE, family: "*", model: id }
+    else return { provider: MODEL_PROVIDER_OPENAI, family: id, model: id }
 }
 
 export interface ModelConnectionInfo

packages/core/src/ollama.ts

@@ -16,23 +16,23 @@ export const OllamaCompletion: ChatCompletionHandler = async (
         return await OpenAIChatCompletion(req, cfg, options, trace)
     } catch (e) {
         if (isRequestError(e)) {
-            const { modelId } = parseModelIdentifier(req.model)
+            const { model } = parseModelIdentifier(req.model)
             if (
                 e.status === 404 &&
                 e.body?.type === "api_error" &&
-                e.body?.message?.includes(`model '${modelId}' not found`)
+                e.body?.message?.includes(`model '${model}' not found`)
             ) {
-                trace.log(`model ${modelId} not found, trying to pull it`)
+                trace.log(`model ${model} not found, trying to pull it`)
                 // model not installed locally
                 // trim v1
                 const fetch = await createFetch({ trace })
                 const res = await fetch(cfg.base.replace("/v1", "/api/pull"), {
                     method: "POST",
-                    body: JSON.stringify({ name: modelId, stream: false }),
+                    body: JSON.stringify({ name: model, stream: false }),
                 })
                 if (!res.ok) {
                     throw new Error(
-                        `Failed to pull model ${modelId}: ${res.status} ${res.statusText}`
+                        `Failed to pull model ${model}: ${res.status} ${res.statusText}`
                     )
                 }
                 trace.log(`model pulled`)

packages/sample/genaisrc/summarize-ollama-gemma2.genai.js

@@ -0,0 +1,15 @@
+script({
+    model: "ollama:gemma2:2b",
+    title: "summarize with ollama gemma 2 2b",
+    system: [],
+    files: "src/rag/markdown.md",
+    tests: {
+        files: "src/rag/markdown.md",
+        keywords: "markdown",
+    },
+})
+
+const file = def("FILE", env.files)
+
+$`Summarize ${file} in a sentence. Make it short.
+`