Update Gemini models

README.md

@@ -68,7 +68,7 @@ Now you can open the browser of your choice and open `http://127.0.0.1:8000`.
 
 - OpenAI
 - Anthropic
-- Google (Gemini, PaLM2)
+- Google Gemini
 - DeepSeek
 - HuggingFace (Inference and Endpoints)
 - Together.ai

chainforge/react-server/package.json

@@ -13,8 +13,7 @@
     "@emoji-mart/data": "^1.1.2",
     "@emoji-mart/react": "^1.1.1",
     "@fontsource/geist-mono": "^5.0.1",
-    "@google-ai/generativelanguage": "^0.2.0",
-    "@google/generative-ai": "^0.21.0",
+    "@google/genai": "^1.19.0",
     "@mantine/core": "^6.0.9",
     "@mantine/dates": "^6.0.13",
     "@mantine/dropzone": "^6.0.19",
@@ -66,7 +65,7 @@
     "lz-string": "^1.5.0",
     "mantine-contextmenu": "^6.1.0",
     "mantine-react-table": "^1.3.4",
-    "markdown-it": "^13.0.1",
+    "markdown-it": "^13.0.2",
     "mathjs": "^11.8.2",
     "mdast-util-from-markdown": "^2.0.0",
     "net": "^1.0.2",
@@ -136,6 +135,7 @@
   "devDependencies": {
     "@craco/craco": "^7.1.0",
     "@types/lodash": "^4.17.0",
+    "@types/markdown-it": "^14.1.2",
     "@types/papaparse": "^5.3.14",
     "@types/react-beautiful-dnd": "^13.1.8",
     "@types/react-edit-text": "^5.0.4",

chainforge/react-server/src/App.tsx

@@ -106,26 +106,27 @@ import {
 // Device / Browser detection
 import {
   isMobile,
+  isTablet,
   isChrome,
   isFirefox,
   isEdgeChromium,
   isChromium,
+  isMobileSafari,
 } from "react-device-detect";
 import MultiEvalNode from "./MultiEvalNode";
 import FlowSidebar from "./FlowSidebar";
 import NestedMenu, { NestedMenuItemProps } from "./NestedMenu";
 import RequestClarificationModal, {
   RequestClarificationModalProps,
 } from "./RequestClarificationModal";
-import { xorBy } from "lodash";
 
 const IS_ACCEPTED_BROWSER =
   (isChrome ||
     isChromium ||
     isEdgeChromium ||
     isFirefox ||
     (navigator as any)?.brave !== undefined) &&
-  !isMobile;
+  (!isMobile || (isTablet && !isMobileSafari));
 
 // Whether we are running on localhost or not, and hence whether
 // we have access to the Flask backend for, e.g., Python code evaluation.
@@ -185,13 +186,13 @@ const INITIAL_LLM = () => {
     // Prefer OpenAI for majority of local users.
     const chatgpt = {
       key: uuid(),
-      name: "GPT3.5",
+      name: "GPT-4o-mini",
       emoji: "🤖",
-      model: "gpt-3.5-turbo",
-      base_model: "gpt-3.5-turbo",
+      model: "gpt-4o-mini",
+      base_model: "gpt-4",
       temp: 1.0,
-      settings: getDefaultModelSettings("gpt-3.5-turbo"),
-      formData: getDefaultModelFormData("gpt-3.5-turbo"),
+      settings: getDefaultModelSettings("gpt-4"),
+      formData: getDefaultModelFormData("gpt-4"),
     } satisfies LLMSpec;
     chatgpt.formData.shortname = chatgpt.name;
     chatgpt.formData.model = chatgpt.model;

chainforge/react-server/src/GlobalSettingsModal.tsx

@@ -537,7 +537,7 @@ const GlobalSettingsModal = forwardRef<GlobalSettingsModalRef, object>(
                 <br />
                 <TextInput
                   label="Google AI API Key (Gemini)"
-                  placeholder="Paste your Google Gemini/PaLM API key here"
+                  placeholder="Paste your Google Gemini API key here"
                   {...form.getInputProps("Google")}
                 />
                 <br />

chainforge/react-server/src/ModelSettingSchemas.tsx

@@ -91,7 +91,7 @@ const ChatGPTSettings: ModelSettingsDict = {
           "text-davinci-002",
           "code-davinci-002",
         ],
-        default: "gpt-3.5-turbo",
+        default: "gpt-4o-mini",
       },
       system_msg: {
         type: "string",
@@ -318,18 +318,18 @@ const GPT4Settings: ModelSettingsDict = {
         title: "Nickname",
         description:
           "Unique identifier to appear in ChainForge. Keep it short.",
-        default: "GPT-4o",
+        default: "GPT-4o-mini",
       },
       model: {
         ...ChatGPTSettings.schema.properties.model,
-        default: "gpt-4o",
+        default: "gpt-4o-mini",
       },
     },
   },
   uiSchema: {
     ...ChatGPTSettings.uiSchema,
     model: {
-      "ui:help": "Defaults to gpt-4o.",
+      "ui:help": "Defaults to gpt-4o-mini.",
       "ui:widget": "datalist",
     },
   },
@@ -833,7 +833,7 @@ const ClaudeSettings: ModelSettingsDict = {
   },
 };
 
-const PaLM2Settings: ModelSettingsDict = {
+const Gemini25Settings: ModelSettingsDict = {
   fullName: "Google AI Models (Gemini)",
   schema: {
     type: "object",
@@ -850,30 +850,29 @@ const PaLM2Settings: ModelSettingsDict = {
         type: "string",
         title: "Model",
         description:
-          "Select a PaLM model to query. For more details on the differences, see the Google PaLM API documentation.",
+          "Select a Gemini model to query. For more details on the differences, see the Google Gemini API documentation.",
         enum: [
-          "gemini-2.5-pro-preview-03-25",
+          "gemini-2.5-pro",
+          "gemini-2.5-flash",
+          "gemini-2.5-flash-lite",
           "gemini-2.0-flash",
           "gemini-2.0-flash-lite",
-          "gemini-1.5-flash",
-          "gemini-1.5-flash-8b",
-          "gemini-1.5-pro",
-          "gemini-1.0-pro",
-          "gemini-pro",
-          "text-bison-001",
-          "chat-bison-001",
+          "gemini-embedding-001",
+          "text-embedding-005",
+          "text-embedding-004",
+          "text-multilingual-embedding-002",
         ],
-        default: "gemini-1.5-flash",
+        default: "gemini-2.5-flash",
         shortname_map: {
-          "text-bison-001": "PaLM2-text",
-          "chat-bison-001": "PaLM2-chat",
-          "gemini-pro": "Gemini 1.0",
-          "gemini-2.5-pro-preview-03-25": "Gemini 2.5",
+          "gemini-2.5-pro": "Gemini 2.5 Pro",
+          "gemini-2.5-flash": "Gemini 2.5 Flash",
+          "gemini-2.5-flash-lite": "Gemini 2.5 Flash Lite",
           "gemini-2.0-flash": "Gemini 2.0 Flash",
-          "gemini-1.5-pro": "Gemini 1.5",
-          "gemini-1.0-pro": "Gemini 1.0",
-          "gemini-1.5-flash": "Gemini Flash",
-          "gemini-1.5-flash-8b": "Gemini Flash 8B",
+          "gemini-2.0-flash-lite": "Gemini 2.0 Flash Lite",
+          "gemini-embedding-001": "gemini-embedding-001",
+          "text-embedding-005": "text-embedding-005",
+          "text-embedding-004": "text-embedding-004",
+          "text-multilingual-embedding-002": "text-multilingual-embedding-002",
         },
       },
       system_msg: {
@@ -887,25 +886,25 @@ const PaLM2Settings: ModelSettingsDict = {
         type: "number",
         title: "temperature",
         description:
-          "Controls the randomness of the output. Must be positive. Typical values are in the range: [0.0, 1.0]. Higher values produce a more random and varied response. A temperature of zero will be deterministic. (ChainForge only allows a max 1.0 temperature for PaLM).",
-        default: 0.5,
+          "Controls the randomness of the output. Must be positive. Typical values are in the range: [0.0, 1.0]. Higher values produce a more random and varied response. A temperature of zero will be deterministic.",
+        default: 0.7,
         minimum: 0,
-        maximum: 1,
+        maximum: 2,
         multipleOf: 0.01,
       },
       top_k: {
         type: "integer",
         title: "top_k",
         description:
-          "Sets the maximum number of tokens to sample from on each step. (The PaLM API uses combined nucleus and top-k sampling.) Set to -1 to use the default value.",
+          "Sets the maximum number of tokens to sample from on each step. (The Gemini API uses combined nucleus and top-k sampling.) Set to -1 to use the default value.",
         minimum: -1,
         default: -1,
       },
       top_p: {
         type: "number",
         title: "top_p",
         description:
-          "Sets the maximum cumulative probability of tokens to sample from. (The PaLM API uses combined nucleus and top-k sampling.) Set to -1 to use the default value.",
+          "Sets the maximum cumulative probability of tokens to sample from. (The Gemini API uses combined nucleus and top-k sampling.) Set to -1 to use the default value.",
         default: -1,
         minimum: -1,
         maximum: 1,
@@ -935,14 +934,14 @@ const PaLM2Settings: ModelSettingsDict = {
       "ui:autofocus": true,
     },
     model: {
-      "ui:help": "Defaults to gemini-pro.",
+      "ui:help": "Defaults to gemini-2.5-flash.",
       "ui:widget": "datalist",
     },
     system_msg: {
       "ui:widget": "textarea",
     },
     temperature: {
-      "ui:help": "Defaults to 0.5.",
+      "ui:help": "Defaults to 0.7.",
       "ui:widget": "range",
     },
     max_output_tokens: {
@@ -2481,7 +2480,7 @@ export const ModelSettings: Dict<ModelSettingsDict> = {
   "dall-e": DalleSettings,
   "gpt-image-1": GPTImage1Settings,
   "claude-v1": ClaudeSettings,
-  "palm2-bison": PaLM2Settings,
+  "gemini-2.5": Gemini25Settings,
   "azure-openai": AzureOpenAISettings,
   hf: HuggingFaceTextInferenceSettings,
   "luminous-base": AlephAlphaLuminousSettings,
@@ -2508,7 +2507,7 @@ export function baseModelToProvider(base_model: string): LLMProvider {
     "dall-e": LLMProvider.OpenAI,
     "gpt-image-1": LLMProvider.OpenAI,
     "claude-v1": LLMProvider.Anthropic,
-    "palm2-bison": LLMProvider.Google,
+    "gemini-2.5": LLMProvider.Google,
     "azure-openai": LLMProvider.Azure_OpenAI,
     hf: LLMProvider.HuggingFace,
     "luminous-base": LLMProvider.Aleph_Alpha,
@@ -2537,7 +2536,7 @@ export function getSettingsSchemaForLLM(
   } = {
     [LLMProvider.OpenAI]: GPT4Settings,
     [LLMProvider.Anthropic]: ClaudeSettings,
-    [LLMProvider.Google]: PaLM2Settings,
+    [LLMProvider.Google]: Gemini25Settings,
     [LLMProvider.Azure_OpenAI]: AzureOpenAISettings,
     [LLMProvider.HuggingFace]: HuggingFaceTextInferenceSettings,
     [LLMProvider.Aleph_Alpha]: AlephAlphaLuminousSettings,

chainforge/react-server/src/backend/__test__/backend.test.ts

@@ -68,7 +68,7 @@ test("call three LLMs with a single prompt", async () => {
   const llms = [
     NativeLLM.OpenAI_ChatGPT,
     NativeLLM.Claude_v1,
-    NativeLLM.PaLM2_Chat_Bison,
+    NativeLLM.GEMINI_v2_5_flash,
   ];
   const n = 1;
   const progress_listener = (progress: { [key: symbol]: any }) => {

chainforge/react-server/src/backend/__test__/query.test.ts

@@ -94,6 +94,6 @@ test("basic prompt pipeline with anthropic", async () => {
   await prompt_model(NativeLLM.Claude_v1, LLMProvider.Anthropic);
 }, 40000);
 
-test("basic prompt pipeline with google palm2", async () => {
-  await prompt_model(NativeLLM.PaLM2_Chat_Bison, LLMProvider.Google);
+test("basic prompt pipeline with google gemini 2.5 flash", async () => {
+  await prompt_model(NativeLLM.GEMINI_v2_5_flash, LLMProvider.Google);
 }, 40000);

chainforge/react-server/src/backend/__test__/utils.test.ts

@@ -5,7 +5,6 @@ import {
   call_alephalpha,
   call_anthropic,
   call_chatgpt,
-  call_google_palm,
   extract_responses,
   merge_response_objs,
 } from "../utils";
@@ -116,48 +115,6 @@ test("anthropic models", async () => {
   expect(typeof resps[0]).toBe("string");
 }, 20000);
 
-test("google palm2 models", async () => {
-  // Call Google's PaLM Chat API with a basic question
-  let [query, response] = await call_google_palm(
-    "Who invented modern playing cards?",
-    NativeLLM.PaLM2_Chat_Bison,
-    3,
-    0.7,
-  );
-  expect(response.candidates).toHaveLength(3);
-  expect(query).toHaveProperty("candidateCount");
-
-  // Extract responses, check their type
-  let resps = extract_responses(
-    response,
-    NativeLLM.PaLM2_Chat_Bison,
-    LLMProvider.Google,
-  );
-  expect(resps).toHaveLength(3);
-  expect(typeof resps[0]).toBe("string");
-  console.log(JSON.stringify(resps));
-
-  // Call Google's PaLM Text Completions API with a basic question
-  [query, response] = await call_google_palm(
-    "Who invented modern playing cards? The answer ",
-    NativeLLM.PaLM2_Text_Bison,
-    3,
-    0.7,
-  );
-  expect(response.candidates).toHaveLength(3);
-  expect(query).toHaveProperty("maxOutputTokens");
-
-  // Extract responses, check their type
-  resps = extract_responses(
-    response,
-    NativeLLM.PaLM2_Chat_Bison,
-    LLMProvider.Google,
-  );
-  expect(resps).toHaveLength(3);
-  expect(typeof resps[0]).toBe("string");
-  console.log(JSON.stringify(resps));
-}, 40000);
-
 test("aleph alpha model", async () => {
   let [query, response] = await call_alephalpha(
     "Who invented modern playing cards?",
@@ -177,7 +134,6 @@ test("aleph alpha model", async () => {
   expect(typeof resps[0]).toBe("string");
   console.log(JSON.stringify(resps));
 
-  // Call Google's PaLM Text Completions API with a basic question
   // eslint-disable-next-line
   [query, response] = await call_alephalpha(
     "Who invented modern playing cards? The answer ",

chainforge/react-server/src/backend/models.ts

@@ -74,16 +74,14 @@ export enum NativeLLM {
   Claude_v1_instant = "claude-instant-v1",
 
   // Google models
-  PaLM2_Text_Bison = "text-bison-001", // it's really models/text-bison-001, but that's confusing
-  PaLM2_Chat_Bison = "chat-bison-001",
-  GEMINI_PRO = "gemini-pro",
-  GEMINI_v2_5_pro_prev = "gemini-2.5-pro-preview-03-25",
+  GEMINI_v2_5_pro = "gemini-2.5-pro",
+  GEMINI_v2_5_flash = "gemini-2.5-flash",
+  GEMINI_v2_5_flash_lite = "gemini-2.5-flash-lite",
   GEMINI_v2_flash = "gemini-2.0-flash",
   GEMINI_v2_flash_lite = "gemini-2.0-flash-lite",
   GEMINI_v1_5_flash = "gemini-1.5-flash",
   GEMINI_v1_5_flash_8B = "gemini-1.5-flash-8b",
   GEMINI_v1_5_pro = "gemini-1.5-pro",
-  GEMINI_v1_pro = "gemini-1.0-pro",
 
   // DeepSeek
   DeepSeek_Chat = "deepseek-chat",
@@ -253,8 +251,7 @@ export function getProvider(llm: LLM): LLMProvider | undefined {
   const llm_name = getEnumName(NativeLLM, llm.toString());
   if (llm_name?.startsWith("OpenAI")) return LLMProvider.OpenAI;
   else if (llm_name?.startsWith("Azure")) return LLMProvider.Azure_OpenAI;
-  else if (llm_name?.startsWith("PaLM2") || llm_name?.startsWith("GEMINI"))
-    return LLMProvider.Google;
+  else if (llm_name?.startsWith("GEMINI")) return LLMProvider.Google;
   else if (llm_name?.startsWith("HF_")) return LLMProvider.HuggingFace;
   else if (llm.toString().startsWith("claude")) return LLMProvider.Anthropic;
   else if (llm_name?.startsWith("Aleph_Alpha")) return LLMProvider.Aleph_Alpha;
@@ -290,9 +287,9 @@ export const RATE_LIMIT_BY_MODEL: { [key in LLM]?: number } = {
   [NativeLLM.OpenAI_DallE_2]: 10, // Should be 5 images per minute (1 img per every 10 seconds); here, we've been a bit lenient with it.
   [NativeLLM.OpenAI_DallE_3]: 10, // This differs per tier, see https://platform.openai.com/docs/guides/rate-limits/usage-tiers?context=tier-one
   [NativeLLM.Azure_OpenAI]: 500, // conservative
-  [NativeLLM.PaLM2_Text_Bison]: 60, // max 60 requests per minute as of Mar 2023
-  [NativeLLM.PaLM2_Chat_Bison]: 60,
-  [NativeLLM.GEMINI_PRO]: 60,
+  [NativeLLM.GEMINI_v2_5_pro]: 150,
+  [NativeLLM.GEMINI_v2_5_flash]: 1000,
+  [NativeLLM.GEMINI_v2_5_flash_lite]: 4000,
   [NativeLLM.Bedrock_Jurassic_Mid]: 400,
   [NativeLLM.Bedrock_Jurassic_Ultra]: 25,
   [NativeLLM.Bedrock_Titan_Light]: 800,