Add debug modal for output cells

See the actual input that a model got for a specific cell. The formatting isn't great right now; should probably iterate on that.

package.json

@@ -25,8 +25,10 @@
     "@apidevtools/json-schema-ref-parser": "^10.1.0",
     "@babel/preset-typescript": "^7.22.5",
     "@babel/standalone": "^7.22.9",
+    "@chakra-ui/anatomy": "^2.2.0",
     "@chakra-ui/next-js": "^2.1.4",
     "@chakra-ui/react": "^2.7.1",
+    "@chakra-ui/styled-system": "^2.9.1",
     "@emotion/react": "^11.11.1",
     "@emotion/server": "^11.11.0",
     "@emotion/styled": "^11.11.0",
@@ -63,7 +65,7 @@
     "next-auth": "^4.22.1",
     "next-query-params": "^4.2.3",
     "nextjs-routes": "^2.0.1",
-    "openai": "4.0.0-beta.2",
+    "openai": "4.0.0-beta.7",
     "pluralize": "^8.0.0",
     "posthog-js": "^1.68.4",
     "prettier": "^3.0.0",
@@ -72,6 +74,7 @@
     "react-diff-viewer": "^3.1.1",
     "react-dom": "18.2.0",
     "react-icons": "^4.10.1",
+    "react-json-tree": "^0.18.0",
     "react-select": "^5.7.4",
     "react-syntax-highlighter": "^15.5.0",
     "react-textarea-autosize": "^8.5.0",

pnpm-lock.yaml

@@ -17,12 +17,18 @@ dependencies:
   '@babel/standalone':
     specifier: ^7.22.9
     version: 7.22.9
+  '@chakra-ui/anatomy':
+    specifier: ^2.2.0
+    version: 2.2.0
   '@chakra-ui/next-js':
     specifier: ^2.1.4
     version: 2.1.4(@chakra-ui/react@2.7.1)(@emotion/react@11.11.1)(next@13.4.2)(react@18.2.0)
   '@chakra-ui/react':
     specifier: ^2.7.1
     version: 2.7.1(@emotion/react@11.11.1)(@emotion/styled@11.11.0)(@types/react@18.2.6)(framer-motion@10.12.17)(react-dom@18.2.0)(react@18.2.0)
+  '@chakra-ui/styled-system':
+    specifier: ^2.9.1
+    version: 2.9.1
   '@emotion/react':
     specifier: ^11.11.1
     version: 11.11.1(@types/react@18.2.6)(react@18.2.0)
@@ -132,8 +138,8 @@ dependencies:
     specifier: ^2.0.1
     version: 2.0.1(next@13.4.2)
   openai:
-    specifier: 4.0.0-beta.2
+    specifier: 4.0.0-beta.7
-    version: 4.0.0-beta.2
+    version: 4.0.0-beta.7
   pluralize:
     specifier: ^8.0.0
     version: 8.0.0
@@ -158,6 +164,9 @@ dependencies:
   react-icons:
     specifier: ^4.10.1
     version: 4.10.1(react@18.2.0)
+  react-json-tree:
+    specifier: ^0.18.0
+    version: 0.18.0(@types/react@18.2.6)(react@18.2.0)
   react-select:
     specifier: ^5.7.4
     version: 5.7.4(@types/react@18.2.6)(react-dom@18.2.0)(react@18.2.0)
@@ -686,6 +695,10 @@ packages:
     resolution: {integrity: sha512-pKfOS/mztc4sUXHNc8ypJ1gPWSolWT770jrgVRfolVbYlki8y5Y+As996zMF6k5lewTu6j9DQequ7Cc9a69IVQ==}
     dev: false
 
+  /@chakra-ui/anatomy@2.2.0:
+    resolution: {integrity: sha512-cD8Ms5C8+dFda0LrORMdxiFhAZwOIY1BSlCadz6/mHUIgNdQy13AHPrXiq6qWdMslqVHq10k5zH7xMPLt6kjFg==}
+    dev: false
+
   /@chakra-ui/avatar@2.2.11(@chakra-ui/system@2.5.8)(react@18.2.0):
     resolution: {integrity: sha512-CJFkoWvlCTDJTUBrKA/aVyG5Zz6TBEIVmmsJtqC6VcQuVDTxkWod8ruXnjb0LT2DUveL7xR5qZM9a5IXcsH3zg==}
     peerDependencies:
@@ -2831,6 +2844,10 @@ packages:
       '@babel/types': 7.22.5
     dev: true
 
+  /@types/base16@1.0.2:
+    resolution: {integrity: sha512-oYO/U4VD1DavwrKuCSQWdLG+5K22SLPem2OQaHmFcQuwHoVeGC+JGVRji2MUqZUAIQZHEonOeVfAX09hYiLsdg==}
+    dev: false
+
   /@types/body-parser@1.19.2:
     resolution: {integrity: sha512-ALYone6pm6QmwZoAgeyNksccT9Q4AWZQ6PvfwR37GT6r6FWUPguq6sUmNGSMV2Wr761oQoBxwGGa6DR5o1DC9g==}
     dependencies:
@@ -3013,6 +3030,7 @@ packages:
 
   /@types/qs@6.9.7:
     resolution: {integrity: sha512-FGa1F62FT09qcrueBA6qYTrJPVDzah9a+493+o2PCXsesWHIn27G98TsSMs3WPNbZIEj4+VJf6saSFpvD+3Zsw==}
+    dev: true
 
   /@types/range-parser@1.2.4:
     resolution: {integrity: sha512-EEhsLsD6UsDM1yFhAvy0Cjr6VwmpMWqFBCb9w07wVugF7w9nfajxLuVmngTIpgS6svCnm6Vaw+MZhoDCKnOfsw==}
@@ -3658,6 +3676,10 @@ packages:
     resolution: {integrity: sha512-Y5gU45svrR5tI2Vt/X9GPd3L0HNIKzGu202EjxrXMpuc2V2CiKgemAbUUsqYmZJvPtCXoUKjNZwBJzsNScUbXA==}
     dev: false
 
+  /base16@1.0.0:
+    resolution: {integrity: sha512-pNdYkNPiJUnEhnfXV56+sQy8+AaPcG3POZAUnwr4EeqCUZFz4u2PePbo3e5Gj4ziYPCWGUZT9RHisvJKnwFuBQ==}
+    dev: false
+
   /base64-js@0.0.8:
     resolution: {integrity: sha512-3XSA2cR/h/73EzlXXdU6YNycmYI7+kicTxks4eJg2g39biHR84slg2+des+p7iHYhbRg/udIS4TD53WabcOUkw==}
     engines: {node: '>= 0.4'}
@@ -3916,10 +3938,24 @@ packages:
   /color-name@1.1.4:
     resolution: {integrity: sha512-dOy+3AuW3a2wNbZHIuMZpTcgjGuLU/uBL/ubcZF9OXbDo8ff4O8yVp5Bf0efS8uEoYo5q4Fx7dY9OgQGXgAsQA==}
 
+  /color-string@1.9.1:
+    resolution: {integrity: sha512-shrVawQFojnZv6xM40anx4CkoDP+fZsw/ZerEMsW/pyzsRbElpsL/DBVW7q3ExxwusdNXI3lXpuhEZkzs8p5Eg==}
+    dependencies:
+      color-name: 1.1.4
+      simple-swizzle: 0.2.2
+    dev: false
+
   /color2k@2.0.2:
     resolution: {integrity: sha512-kJhwH5nAwb34tmyuqq/lgjEKzlFXn1U99NlnB6Ws4qVaERcRUYeYP1cBw6BJ4vxaWStAUEef4WMr7WjOCnBt8w==}
     dev: false
 
+  /color@3.2.1:
+    resolution: {integrity: sha512-aBl7dZI9ENN6fUGC7mWpMTPNHmWUSNan9tuWN6ahh5ZLNk9baLJOnSMlrQkHcrfFgz2/RigjUVAjdx36VcemKA==}
+    dependencies:
+      color-convert: 1.9.3
+      color-string: 1.9.1
+    dev: false
+
   /combined-stream@1.0.8:
     resolution: {integrity: sha512-FQN4MRfuJeHf7cBbBMJFXhKSDq+2kAArBlmRBvcvFE5BB1HZKXtSFASDhdlz9zOYwxh8lDdnvmMOe/+5cdoEdg==}
     engines: {node: '>= 0.8'}
@@ -5536,6 +5572,10 @@ packages:
     resolution: {integrity: sha512-zz06S8t0ozoDXMG+ube26zeCTNXcKIPJZJi8hBrF4idCLms4CG9QtK7qBl1boi5ODzFpjswb5JPmHCbMpjaYzg==}
     dev: false
 
+  /is-arrayish@0.3.2:
+    resolution: {integrity: sha512-eVRqCvVlZbuw3GrM63ovNSNAeA1K16kaR/LRY/92w0zxQ5/1YzwblUX652i4Xs9RwAGjW9d9y6X88t8OaAJfWQ==}
+    dev: false
+
   /is-bigint@1.0.4:
     resolution: {integrity: sha512-zB9CruMamjym81i2JZ3UMn54PKGsQzsJeo6xvN3HJJ4CAsQNB6iRutp2To77OfCNuoxspsIhzaPoO1zyCEhFOg==}
     dependencies:
@@ -5910,6 +5950,10 @@ packages:
     resolution: {integrity: sha512-H5ZhCF25riFd9uB5UCkVKo61m3S/xZk1x4wA6yp/L3RFP6Z/eHH1ymQcGLo7J3GMPfm0V/7m1tryHuGVxpqEBQ==}
     dev: false
 
+  /lodash.curry@4.1.1:
+    resolution: {integrity: sha512-/u14pXGviLaweY5JI0IUzgzF2J6Ne8INyzAZjImcryjgkZ+ebruBxy2/JaOOkTqScddcYtakjhSaeemV8lR0tA==}
+    dev: false
+
   /lodash.merge@4.6.2:
     resolution: {integrity: sha512-0KpjqXRVvrYyCsX1swR/XTK0va6VQkQM6MNo7PqW77ByjAhoARA8EfrP1N4+KlKj8YS0ZUCtRT/YUuhyYDujIQ==}
     dev: true
@@ -6387,19 +6431,17 @@ packages:
       is-wsl: 2.2.0
     dev: true
 
-  /openai@4.0.0-beta.2:
+  /openai@4.0.0-beta.7:
-    resolution: {integrity: sha512-zTuAxBFe5nSO7LngbV+/O0udtgHWfXb2lFei8/sDY4GB5cOdnrRoSOtiyUfV65ANdvlI4F75oYZX7w067cxj3w==}
+    resolution: {integrity: sha512-jHjwvpMuGkNxiQ3erwLZsOvPEhcVrMtwtfNeYmGCjhbdB+oStVw/7pIhIPkualu8rlhLwgMR7awknIaN3IQcOA==}
     dependencies:
       '@types/node': 18.16.0
       '@types/node-fetch': 2.6.4
-      '@types/qs': 6.9.7
       abort-controller: 3.0.0
       agentkeepalive: 4.3.0
       digest-fetch: 1.3.0
       form-data-encoder: 1.7.2
       formdata-node: 4.4.1
       node-fetch: 2.6.12
-      qs: 6.11.2
     transitivePeerDependencies:
       - encoding
       - supports-color
@@ -6832,13 +6874,6 @@ packages:
       side-channel: 1.0.4
     dev: false
 
-  /qs@6.11.2:
-    resolution: {integrity: sha512-tDNIz22aBzCDxLtVH++VnTfzxlfeK5CbqohpSqpJgj1Wg/cQbStNAz3NuqCs5vV+pjBsK4x4pN9HlVh7rcYRiA==}
-    engines: {node: '>=0.6'}
-    dependencies:
-      side-channel: 1.0.4
-    dev: false
-
   /queue-microtask@1.2.3:
     resolution: {integrity: sha512-NuaNSa6flKT5JaSYQzJok04JzTL1CA6aGhv5rfLW3PgqA+M2ChpZQnAC8h8i4ZFkBS8X5RqkDBHA7r4hej3K9A==}
     dev: true
@@ -6875,6 +6910,18 @@ packages:
       webpack: 5.88.2
     dev: true
 
+  /react-base16-styling@0.9.1:
+    resolution: {integrity: sha512-1s0CY1zRBOQ5M3T61wetEpvQmsYSNtWEcdYzyZNxKa8t7oDvaOn9d21xrGezGAHFWLM7SHcktPuPTrvoqxSfKw==}
+    dependencies:
+      '@babel/runtime': 7.22.6
+      '@types/base16': 1.0.2
+      '@types/lodash': 4.14.195
+      base16: 1.0.0
+      color: 3.2.1
+      csstype: 3.1.2
+      lodash.curry: 4.1.1
+    dev: false
+
   /react-clientside-effect@1.2.6(react@18.2.0):
     resolution: {integrity: sha512-XGGGRQAKY+q25Lz9a/4EPqom7WRjz3z9R2k4jhVKA/puQFH/5Nt27vFZYql4m4NVNdUvX8PS3O7r/Zzm7cjUlg==}
     peerDependencies:
@@ -6949,6 +6996,19 @@ packages:
     resolution: {integrity: sha512-xWGDIW6x921xtzPkhiULtthJHoJvBbF3q26fzloPCK0hsvxtPVelvftw3zjbHWSkR2km9Z+4uxbDDK/6Zw9B8w==}
     dev: true
 
+  /react-json-tree@0.18.0(@types/react@18.2.6)(react@18.2.0):
+    resolution: {integrity: sha512-Qe6HKSXrr++n9Y31nkRJ3XvQMATISpqigH1vEKhLwB56+nk5thTP0ITThpjxY6ZG/ubpVq/aEHIcyLP/OPHxeA==}
+    peerDependencies:
+      '@types/react': ^16.8.0 || ^17.0.0 || ^18.0.0
+      react: ^16.8.0 || ^17.0.0 || ^18.0.0
+    dependencies:
+      '@babel/runtime': 7.22.6
+      '@types/lodash': 4.14.195
+      '@types/react': 18.2.6
+      react: 18.2.0
+      react-base16-styling: 0.9.1
+    dev: false
+
   /react-remove-scroll-bar@2.3.4(@types/react@18.2.6)(react@18.2.0):
     resolution: {integrity: sha512-63C4YQBUt0m6ALadE9XV56hV8BgJWDmmTPY758iIJjfQKt2nYwoUrPk0LXRXcB/yIj82T1/Ixfdpdk68LwIB0A==}
     engines: {node: '>=10'}
@@ -7360,6 +7420,12 @@ packages:
     resolution: {integrity: sha512-wnD2ZE+l+SPC/uoS0vXeE9L1+0wuaMqKlfz9AMUo38JsyLSBWSFcHR1Rri62LZc12vLr1gb3jl7iwQhgwpAbGQ==}
     dev: true
 
+  /simple-swizzle@0.2.2:
+    resolution: {integrity: sha512-JA//kQgZtbuY83m+xT+tXJkmJncGMTFT+C+g2h2R9uxkYIrE2yy9sgmcLhCnw57/WSD+Eh3J97FPEDFnbXnDUg==}
+    dependencies:
+      is-arrayish: 0.3.2
+    dev: false
+
   /slash@3.0.0:
     resolution: {integrity: sha512-g9Q1haeby36OSStwb4ntCGGGaKsaVSjQ68fBxoQcutl5fS1vuY18H3wSt3jFyFtrkx+Kz0V1G85A4MyAdDMi2Q==}
     engines: {node: '>=8'}

src/components/ChangeModelModal/ModelSearch.tsx

@@ -22,8 +22,8 @@ export const ModelSearch = (props: {
   const [containerRef, containerDimensions] = useElementDimensions();
 
   return (
-    <VStack ref={containerRef as LegacyRef<HTMLDivElement>} w="full">
+    <VStack ref={containerRef as LegacyRef<HTMLDivElement>} w="full" fontFamily="inconsolata">
-      <Text>Browse Models</Text>
+      <Text fontWeight="bold">Browse Models</Text>
       <Select<ProviderModel>
         styles={{ control: (provided) => ({ ...provided, width: containerDimensions?.width }) }}
         getOptionLabel={(data) => modelLabel(data.provider, data.model)}

src/components/ChangeModelModal/ModelStatsCard.tsx

@@ -23,16 +23,24 @@ export const ModelStatsCard = ({
         {label}
       </Text>
 
-      <VStack w="full" spacing={6} bgColor="gray.100" p={4} borderRadius={4}>
+      <VStack
+        w="full"
+        spacing={6}
+        borderWidth={1}
+        borderColor="gray.300"
+        p={4}
+        borderRadius={8}
+        fontFamily="inconsolata"
+      >
         <HStack w="full" align="flex-start">
-          <Text flex={1} fontSize="lg">
+          <VStack flex={1} fontSize="lg" alignItems="flex-start">
-            <Text as="span" color="gray.600">
-              {model.provider} /{" "}
-            </Text>
             <Text as="span" fontWeight="bold" color="gray.900">
               {model.name}
             </Text>
-          </Text>
+            <Text as="span" color="gray.600" fontSize="sm">
+              Provider: {model.provider}
+            </Text>
+          </VStack>
           <Link
             href={model.learnMoreUrl}
             isExternal

src/components/OutputsTable/OutputCell/CellContent.tsx

@@ -1,19 +0,0 @@
-import { type StackProps, VStack } from "@chakra-ui/react";
-import { CellOptions } from "./CellOptions";
-
-export const CellContent = ({
-  hardRefetch,
-  hardRefetching,
-  children,
-  ...props
-}: {
-  hardRefetch: () => void;
-  hardRefetching: boolean;
-} & StackProps) => (
-  <VStack w="full" alignItems="flex-start" {...props}>
-    <CellOptions refetchingOutput={hardRefetching} refetchOutput={hardRefetch} />
-    <VStack w="full" alignItems="flex-start" maxH={500} overflowY="auto">
-      {children}
-    </VStack>
-  </VStack>
-);

src/components/OutputsTable/OutputCell/CellOptions.tsx

@@ -1,37 +0,0 @@
-import { Button, HStack, Icon, Spinner, Tooltip } from "@chakra-ui/react";
-import { BsArrowClockwise } from "react-icons/bs";
-import { useExperimentAccess } from "~/utils/hooks";
-
-export const CellOptions = ({
-  refetchingOutput,
-  refetchOutput,
-}: {
-  refetchingOutput: boolean;
-  refetchOutput: () => void;
-}) => {
-  const { canModify } = useExperimentAccess();
-  return (
-    <HStack justifyContent="flex-end" w="full">
-      {canModify && (
-        <Tooltip label="Refetch output" aria-label="refetch output">
-          <Button
-            size="xs"
-            w={4}
-            h={4}
-            py={4}
-            px={4}
-            minW={0}
-            borderRadius={8}
-            color="gray.500"
-            variant="ghost"
-            cursor="pointer"
-            onClick={refetchOutput}
-            aria-label="refetch output"
-          >
-            <Icon as={refetchingOutput ? Spinner : BsArrowClockwise} boxSize={4} />
-          </Button>
-        </Tooltip>
-      )}
-    </HStack>
-  );
-};

src/components/OutputsTable/OutputCell/OutputCell.tsx

@@ -1,17 +1,17 @@
 import { api } from "~/utils/api";
 import { type PromptVariant, type Scenario } from "../types";
-import { Text, VStack } from "@chakra-ui/react";
+import { type StackProps, Text, VStack } from "@chakra-ui/react";
 import { useExperiment, useHandledAsyncCallback } from "~/utils/hooks";
 import SyntaxHighlighter from "react-syntax-highlighter";
 import { docco } from "react-syntax-highlighter/dist/cjs/styles/hljs";
 import stringify from "json-stringify-pretty-compact";
-import { type ReactElement, useState, useEffect, Fragment } from "react";
+import { type ReactElement, useState, useEffect, Fragment, useCallback } from "react";
 import useSocket from "~/utils/useSocket";
 import { OutputStats } from "./OutputStats";
 import { RetryCountdown } from "./RetryCountdown";
 import frontendModelProviders from "~/modelProviders/frontendModelProviders";
 import { ResponseLog } from "./ResponseLog";
-import { CellContent } from "./CellContent";
+import { CellOptions } from "./TopActions";
 
 const WAITING_MESSAGE_INTERVAL = 20000;
 
@@ -72,37 +72,49 @@ export default function OutputCell({
   // TODO: disconnect from socket if we're not streaming anymore
   const streamedMessage = useSocket<OutputSchema>(cell?.id);
 
+  const mostRecentResponse = cell?.modelResponses[cell.modelResponses.length - 1];
+
+  const CellWrapper = useCallback(
+    ({ children, ...props }: StackProps) => (
+      <VStack w="full" alignItems="flex-start" {...props} px={2} py={2} h="100%">
+        {cell && (
+          <CellOptions refetchingOutput={hardRefetching} refetchOutput={hardRefetch} cell={cell} />
+        )}
+        <VStack w="full" alignItems="flex-start" maxH={500} overflowY="auto" flex={1}>
+          {children}
+        </VStack>
+        {mostRecentResponse && (
+          <OutputStats modelResponse={mostRecentResponse} scenario={scenario} />
+        )}
+      </VStack>
+    ),
+    [hardRefetching, hardRefetch, mostRecentResponse, scenario],
+  );
+
   if (!vars) return null;
 
   if (!cell && !fetchingOutput)
     return (
-      <CellContent hardRefetching={hardRefetching} hardRefetch={hardRefetch}>
+      <CellWrapper>
         <Text color="gray.500">Error retrieving output</Text>
-      </CellContent>
+      </CellWrapper>
     );
 
   if (cell && cell.errorMessage) {
     return (
-      <CellContent hardRefetching={hardRefetching} hardRefetch={hardRefetch}>
+      <CellWrapper>
         <Text color="red.500">{cell.errorMessage}</Text>
-      </CellContent>
+      </CellWrapper>
     );
   }
 
   if (disabledReason) return <Text color="gray.500">{disabledReason}</Text>;
 
-  const mostRecentResponse = cell?.modelResponses[cell.modelResponses.length - 1];
   const showLogs = !streamedMessage && !mostRecentResponse?.output;
 
   if (showLogs)
     return (
-      <CellContent
+      <CellWrapper alignItems="flex-start" fontFamily="inconsolata, monospace" spacing={0}>
-        hardRefetching={hardRefetching}
-        hardRefetch={hardRefetch}
-        alignItems="flex-start"
-        fontFamily="inconsolata, monospace"
-        spacing={0}
-      >
         {cell?.jobQueuedAt && <ResponseLog time={cell.jobQueuedAt} title="Job queued" />}
         {cell?.jobStartedAt && <ResponseLog time={cell.jobStartedAt} title="Job started" />}
         {cell?.modelResponses?.map((response) => {
@@ -124,9 +136,13 @@ export default function OutputCell({
                 Array.from({ length: numWaitingMessages }, (_, i) => (
                   <ResponseLog
                     key={`waiting-${i}`}
-                    // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+                    time={
-                    time={new Date(response.requestedAt!.getTime() + i * WAITING_MESSAGE_INTERVAL)}
+                      new Date(
-                    title="Waiting for response"
+                        (response.requestedAt?.getTime?.() ?? 0) +
+                          (i + 1) * WAITING_MESSAGE_INTERVAL,
+                      )
+                    }
+                    title="Waiting for response..."
                   />
                 ))}
               {response.receivedAt && (
@@ -144,7 +160,7 @@ export default function OutputCell({
         {mostRecentResponse?.retryTime && (
           <RetryCountdown retryTime={mostRecentResponse.retryTime} />
         )}
-      </CellContent>
+      </CellWrapper>
     );
 
   const normalizedOutput = mostRecentResponse?.output
@@ -155,50 +171,27 @@ export default function OutputCell({
 
   if (mostRecentResponse?.output && normalizedOutput?.type === "json") {
     return (
-      <VStack
+      <CellWrapper>
-        w="100%"
+        <SyntaxHighlighter
-        h="100%"
+          customStyle={{ overflowX: "unset", width: "100%", flex: 1 }}
-        fontSize="xs"
+          language="json"
-        flexWrap="wrap"
+          style={docco}
-        overflowX="hidden"
+          lineProps={{
-        justifyContent="space-between"
+            style: { wordBreak: "break-all", whiteSpace: "pre-wrap" },
-      >
+          }}
-        <CellContent
+          wrapLines
-          hardRefetching={hardRefetching}
-          hardRefetch={hardRefetch}
-          w="full"
-          flex={1}
-          spacing={0}
         >
-          <SyntaxHighlighter
+          {stringify(normalizedOutput.value, { maxLength: 40 })}
-            customStyle={{ overflowX: "unset", width: "100%", flex: 1 }}
+        </SyntaxHighlighter>
-            language="json"
+      </CellWrapper>
-            style={docco}
-            lineProps={{
-              style: { wordBreak: "break-all", whiteSpace: "pre-wrap" },
-            }}
-            wrapLines
-          >
-            {stringify(normalizedOutput.value, { maxLength: 40 })}
-          </SyntaxHighlighter>
-        </CellContent>
-        <OutputStats modelResponse={mostRecentResponse} scenario={scenario} />
-      </VStack>
     );
   }
 
   const contentToDisplay = (normalizedOutput?.type === "text" && normalizedOutput.value) || "";
 
   return (
-    <VStack w="100%" h="100%" justifyContent="space-between" whiteSpace="pre-wrap">
+    <CellWrapper>
-      <VStack w="full" alignItems="flex-start" spacing={0}>
+      <Text>{contentToDisplay}</Text>
-        <CellContent hardRefetching={hardRefetching} hardRefetch={hardRefetch}>
+    </CellWrapper>
-          <Text>{contentToDisplay}</Text>
-        </CellContent>
-      </VStack>
-      {mostRecentResponse?.output && (
-        <OutputStats modelResponse={mostRecentResponse} scenario={scenario} />
-      )}
-    </VStack>
   );
 }

src/components/OutputsTable/OutputCell/PromptModal.tsx

@@ -0,0 +1,36 @@
+import {
+  Modal,
+  ModalBody,
+  ModalCloseButton,
+  ModalContent,
+  ModalHeader,
+  ModalOverlay,
+  type UseDisclosureReturn,
+} from "@chakra-ui/react";
+import { type RouterOutputs } from "~/utils/api";
+import { JSONTree } from "react-json-tree";
+
+export default function ExpandedModal(props: {
+  cell: NonNullable<RouterOutputs["scenarioVariantCells"]["get"]>;
+  disclosure: UseDisclosureReturn;
+}) {
+  return (
+    <Modal isOpen={props.disclosure.isOpen} onClose={props.disclosure.onClose} size="2xl">
+      <ModalOverlay />
+      <ModalContent>
+        <ModalHeader>Prompt</ModalHeader>
+        <ModalCloseButton />
+        <ModalBody>
+          <JSONTree
+            data={props.cell.prompt}
+            invertTheme={true}
+            theme="chalk"
+            shouldExpandNodeInitially={() => true}
+            getItemString={() => ""}
+            hideRoot
+          />
+        </ModalBody>
+      </ModalContent>
+    </Modal>
+  );
+}

src/components/OutputsTable/OutputCell/TopActions.tsx

@@ -0,0 +1,53 @@
+import { HStack, Icon, IconButton, Spinner, Tooltip, useDisclosure } from "@chakra-ui/react";
+import { BsArrowClockwise, BsInfoCircle } from "react-icons/bs";
+import { useExperimentAccess } from "~/utils/hooks";
+import ExpandedModal from "./PromptModal";
+import { type RouterOutputs } from "~/utils/api";
+
+export const CellOptions = ({
+  cell,
+  refetchingOutput,
+  refetchOutput,
+}: {
+  cell: RouterOutputs["scenarioVariantCells"]["get"];
+  refetchingOutput: boolean;
+  refetchOutput: () => void;
+}) => {
+  const { canModify } = useExperimentAccess();
+
+  const modalDisclosure = useDisclosure();
+
+  return (
+    <HStack justifyContent="flex-end" w="full">
+      {cell && (
+        <>
+          <Tooltip label="See Prompt">
+            <IconButton
+              aria-label="See Prompt"
+              icon={<Icon as={BsInfoCircle} boxSize={4} />}
+              onClick={modalDisclosure.onOpen}
+              size="xs"
+              colorScheme="gray"
+              color="gray.500"
+              variant="ghost"
+            />
+          </Tooltip>
+          <ExpandedModal cell={cell} disclosure={modalDisclosure} />
+        </>
+      )}
+      {canModify && (
+        <Tooltip label="Refetch output">
+          <IconButton
+            size="xs"
+            color="gray.500"
+            variant="ghost"
+            cursor="pointer"
+            onClick={refetchOutput}
+            aria-label="refetch output"
+            icon={<Icon as={refetchingOutput ? Spinner : BsArrowClockwise} boxSize={4} />}
+          />
+        </Tooltip>
+      )}
+    </HStack>
+  );
+};

src/components/OutputsTable/ScenarioEditor.tsx

@@ -1,9 +1,8 @@
-import { useEffect, type DragEvent } from "react";
-import { api } from "~/utils/api";
 import { isEqual } from "lodash-es";
-import { type Scenario } from "./types";
+import { useEffect, useState, type DragEvent } from "react";
+import { api } from "~/utils/api";
 import { useExperiment, useExperimentAccess, useHandledAsyncCallback } from "~/utils/hooks";
-import { useState } from "react";
+import { type Scenario } from "./types";
 
 import {
   Box,
@@ -12,14 +11,12 @@ import {
   Icon,
   IconButton,
   Spinner,
-  Stack,
+  Text,
   Tooltip,
   VStack,
-  Text,
 } from "@chakra-ui/react";
-import { cellPadding } from "../constants";
 import { BsArrowsAngleExpand, BsX } from "react-icons/bs";
-import { RiDraggable } from "react-icons/ri";
+import { cellPadding } from "../constants";
 import { FloatingLabelInput } from "./FloatingLabelInput";
 import { ScenarioEditorModal } from "./ScenarioEditorModal";
 
@@ -115,60 +112,44 @@ export default function ScenarioEditor({
         onDrop={onReorder}
         backgroundColor={isDragTarget ? "gray.100" : "transparent"}
       >
-        {canModify && props.canHide && (
-          <Stack
-            alignSelf="flex-start"
-            opacity={props.hovered ? 1 : 0}
-            spacing={0}
-            ml={-cellPadding.x}
-          >
-            <Tooltip label="Hide scenario" hasArrow>
-              {/* for some reason the tooltip can't position itself properly relative to the icon without the wrapping box */}
-              <Button
-                variant="unstyled"
-                color="gray.400"
-                height="unset"
-                width="unset"
-                minW="unset"
-                onClick={onHide}
-                _hover={{
-                  color: "gray.800",
-                  cursor: "pointer",
-                }}
-              >
-                <Icon as={hidingInProgress ? Spinner : BsX} boxSize={hidingInProgress ? 4 : 6} />
-              </Button>
-            </Tooltip>
-            <Icon
-              as={RiDraggable}
-              boxSize={6}
-              color="gray.400"
-              _hover={{ color: "gray.800", cursor: "pointer" }}
-            />
-          </Stack>
-        )}
-
         {variableLabels.length === 0 ? (
           <Box color="gray.500">
             {vars.data ? "No scenario variables configured" : "Loading..."}
           </Box>
         ) : (
           <VStack spacing={4} flex={1} py={2}>
-            <HStack justifyContent="space-between" w="100%">
+            <HStack justifyContent="space-between" w="100%" align="center" spacing={0}>
-              <Text color="gray.500">Scenario</Text>
+              <Text flex={1}>Scenario</Text>
-              <IconButton
+              <Tooltip label="Expand" hasArrow>
-                className="fullscreen-toggle"
+                <IconButton
-                aria-label="Maximize"
+                  aria-label="Expand"
-                icon={<BsArrowsAngleExpand />}
+                  icon={<Icon as={BsArrowsAngleExpand} boxSize={3} />}
-                onClick={() => setScenarioEditorModalOpen(true)}
+                  onClick={() => setScenarioEditorModalOpen(true)}
-                boxSize={6}
+                  size="xs"
-                borderRadius={4}
+                  colorScheme="gray"
-                p={1.5}
+                  color="gray.500"
-                minW={0}
+                  variant="ghost"
-                colorScheme="gray"
+                />
-                color="gray.500"
+              </Tooltip>
-                variant="ghost"
+              {canModify && props.canHide && (
-              />
+                <Tooltip label="Delete" hasArrow>
+                  <IconButton
+                    aria-label="Delete"
+                    icon={
+                      <Icon
+                        as={hidingInProgress ? Spinner : BsX}
+                        boxSize={hidingInProgress ? 4 : 6}
+                      />
+                    }
+                    onClick={onHide}
+                    size="xs"
+                    display="flex"
+                    colorScheme="gray"
+                    color="gray.500"
+                    variant="ghost"
+                  />
+                </Tooltip>
+              )}
             </HStack>
             {variableLabels.map((key) => {
               const value = values[key] ?? "";

src/components/OutputsTable/ScenarioEditorModal.tsx

@@ -1,7 +1,6 @@
 import {
   Button,
   HStack,
-  Icon,
   Modal,
   ModalBody,
   ModalCloseButton,
@@ -14,7 +13,6 @@ import {
   VStack,
 } from "@chakra-ui/react";
 import { useEffect, useState } from "react";
-import { BsFileTextFill } from "react-icons/bs";
 import { isEqual } from "lodash-es";
 
 import { api } from "~/utils/api";
@@ -60,8 +58,6 @@ export const ScenarioEditorModal = ({
     await utils.scenarios.list.invalidate();
   }, [mutation, values]);
 
-  console.log("scenario", scenario);
-
   const vars = api.templateVars.list.useQuery({ experimentId: experiment.data?.id ?? "" });
   const variableLabels = vars.data?.map((v) => v.label) ?? [];
 
@@ -73,12 +69,7 @@ export const ScenarioEditorModal = ({
     >
       <ModalOverlay />
       <ModalContent w={1200}>
-        <ModalHeader>
+        <ModalHeader />
-          <HStack>
-            <Icon as={BsFileTextFill} />
-            <Text>Scenario</Text>
-          </HStack>
-        </ModalHeader>
         <ModalCloseButton />
         <ModalBody maxW="unset">
           <VStack spacing={8}>

src/components/OutputsTable/ScenarioRow.tsx

@@ -1,6 +1,5 @@
-import { Box, GridItem } from "@chakra-ui/react";
+import { GridItem } from "@chakra-ui/react";
 import React, { useState } from "react";
-import { cellPadding } from "../constants";
 import OutputCell from "./OutputCell/OutputCell";
 import ScenarioEditor from "./ScenarioEditor";
 import type { PromptVariant, Scenario } from "./types";
@@ -39,9 +38,7 @@ const ScenarioRow = (props: {
           colStart={i + 2}
           {...borders}
         >
-          <Box h="100%" w="100%" px={cellPadding.x} py={cellPadding.y}>
+          <OutputCell key={variant.id} scenario={props.scenario} variant={variant} />
-            <OutputCell key={variant.id} scenario={props.scenario} variant={variant} />
-          </Box>
         </GridItem>
       ))}
     </>

src/modelProviders/openai-ChatCompletion/getCompletion.ts

@@ -6,7 +6,7 @@ import {
 } from "openai/resources/chat";
 import { countOpenAIChatTokens } from "~/utils/countTokens";
 import { type CompletionResponse } from "../types";
-import { omit } from "lodash-es";
+import { isArray, isString, omit } from "lodash-es";
 import { openai } from "~/server/utils/openai";
 import { truthyFilter } from "~/utils/utils";
 import { APIError } from "openai";
@@ -40,6 +40,8 @@ const mergeStreamedChunks = (
           ((choice.delta.function_call.arguments as string) ?? "");
       }
     } else {
+      // @ts-expect-error the types are correctly telling us that finish_reason
+      // could be null, but don't want to fix it right now.
       choices.push({ ...omit(choice, "delta"), message: { role: "assistant", ...choice.delta } });
     }
   }
@@ -64,6 +66,7 @@ export async function getCompletion(
 
   try {
     if (onStream) {
+      console.log("got started");
       const resp = await openai.chat.completions.create(
         { ...input, stream: true },
         {
@@ -71,9 +74,11 @@ export async function getCompletion(
         },
       );
       for await (const part of resp) {
+        console.log("got part", part);
         finalCompletion = mergeStreamedChunks(finalCompletion, part);
         onStream(finalCompletion);
       }
+      console.log("got final", finalCompletion);
       if (!finalCompletion) {
         return {
           type: "error",
@@ -121,9 +126,17 @@ export async function getCompletion(
     };
   } catch (error: unknown) {
     if (error instanceof APIError) {
+      // The types from the sdk are wrong
+      const rawMessage = error.message as string | string[];
+      // If the message is not a string, stringify it
+      const message = isString(rawMessage)
+        ? rawMessage
+        : isArray(rawMessage)
+        ? rawMessage.map((m) => m.toString()).join("\n")
+        : (rawMessage as any).toString();
       return {
         type: "error",
-        message: error.message,
+        message,
         autoRetry: error.status === 429 || error.status === 503,
         statusCode: error.status,
       };

src/modelProviders/replicate-llama2/getCompletion.ts

@@ -8,8 +8,8 @@ const replicate = new Replicate({
 });
 
 const modelIds: Record<ReplicateLlama2Input["model"], string> = {
-  "7b-chat": "5ec5fdadd80ace49f5a2b2178cceeb9f2f77c493b85b1131002c26e6b2b13184",
+  "7b-chat": "058333670f2a6e88cf1b29b8183405b17bb997767282f790b82137df8c090c1f",
-  "13b-chat": "6b4da803a2382c08868c5af10a523892f38e2de1aafb2ee55b020d9efef2fdb8",
+  "13b-chat": "d5da4236b006f967ceb7da037be9cfc3924b20d21fed88e1e94f19d56e2d3111",
   "70b-chat": "2c1608e18606fad2812020dc541930f2d0495ce32eee50074220b87300bc16e1",
 };
 

src/modelProviders/replicate-llama2/index.ts

@@ -54,7 +54,7 @@ const modelProvider: ReplicateLlama2Provider = {
       temperature: {
         type: "number",
         description:
-          "Adjusts randomness of outputs, greater than 1 is random and 0 is deterministic, 0.75 is a good starting value. (minimum: 0.01; maximum: 5)",
+          "Adjusts randomness of outputs, 0.1 is a good starting value. (minimum: 0.01; maximum: 5)",
       },
       top_p: {
         type: "number",

src/pages/api/experiments/og-image.tsx

@@ -9,7 +9,7 @@ const inconsolataRegularFontP = fetch(
   new URL("../../../../public/fonts/Inconsolata_SemiExpanded-Medium.ttf", import.meta.url),
 ).then((res) => res.arrayBuffer());
 
-const OgImage = async (req: NextApiRequest, res: NextApiResponse) => {
+const OgImage = async (req: NextApiRequest, _res: NextApiResponse) => {
   // @ts-expect-error - nextUrl is not defined on NextApiRequest for some reason
   const searchParams = req.nextUrl?.searchParams as URLSearchParams;
   const experimentLabel = searchParams.get("experimentLabel");

src/pages/world-champs/signup.tsx

@@ -17,6 +17,7 @@ import {
   Tr,
   VStack,
   useInterval,
+  Image,
 } from "@chakra-ui/react";
 import { signIn, useSession } from "next-auth/react";
 import Head from "next/head";
@@ -27,9 +28,23 @@ import { api } from "~/utils/api";
 import dayjs from "~/utils/dayjs";
 import { useHandledAsyncCallback } from "~/utils/hooks";
 
+const TopNavbar = () => (
+  <DarkMode>
+    <GlobalStyle />
+    <HStack px={4} py={2}>
+      <HStack as={Link} href="/" _hover={{ textDecoration: "none" }} spacing={0} py={2} pr={16}>
+        <Image src="/logo.svg" alt="" boxSize={6} mr={4} />
+        <Heading size="md" fontFamily="inconsolata, monospace">
+          OpenPipe
+        </Heading>
+      </HStack>
+    </HStack>
+  </DarkMode>
+);
+
 // Shows how long until the competition starts. Refreshes every second
 function CountdownTimer(props: { date: Date } & TextProps) {
-  const [now, setNow] = useState(dayjs(0));
+  const [now, setNow] = useState(dayjs());
 
   useInterval(() => {
     setNow(dayjs());
@@ -37,7 +52,7 @@ function CountdownTimer(props: { date: Date } & TextProps) {
 
   const { date, ...rest } = props;
 
-  const kickoff = dayjs(props.date);
+  const kickoff = dayjs(date);
   const diff = kickoff.diff(now, "second");
   const days = Math.floor(diff / 86400);
   const hours = Math.floor((diff % 86400) / 3600);
@@ -45,7 +60,7 @@ function CountdownTimer(props: { date: Date } & TextProps) {
   const seconds = Math.floor(diff % 60);
 
   return (
-    <Text {...rest}>
+    <Text {...rest} suppressHydrationWarning>
       <Text as="span" fontWeight="bold">
         Kickoff in
       </Text>{" "}
@@ -128,7 +143,8 @@ export default function Signup() {
         />
       </Head>
 
-      <Box bgColor="gray.900" color="gray.200" minH="100vh" w="full">
+      <Box color="gray.200" minH="100vh" w="full">
+        <TopNavbar />
         <VStack mx="auto" py={24} maxW="2xl" align="start" fontSize="lg">
           <Heading size="lg">🏆 Prompt Engineering World Championships</Heading>
           <CountdownTimer

src/server/api/routers/worldChamps.router.ts

@@ -3,7 +3,7 @@ import { prisma } from "~/server/db";
 import { requireNothing } from "~/utils/accessControl";
 
 export const worldChampsRouter = createTRPCRouter({
-  userStatus: publicProcedure.query(async ({ input, ctx }) => {
+  userStatus: publicProcedure.query(async ({ ctx }) => {
     const userId = ctx.session?.user.id;
 
     if (!userId) {

src/server/utils/deriveNewContructFn.ts

@@ -3,7 +3,7 @@ import ivm from "isolated-vm";
 import dedent from "dedent";
 import { openai } from "./openai";
 import { isObject } from "lodash-es";
-import { type CompletionCreateParams } from "openai/resources/chat/completions";
+import type { CreateChatCompletionRequestMessage } from "openai/resources/chat/completions";
 import formatPromptConstructor from "~/utils/formatPromptConstructor";
 import { type SupportedProvider, type Model } from "~/modelProviders/types";
 import modelProviders from "~/modelProviders/modelProviders";
@@ -44,14 +44,14 @@ const requestUpdatedPromptFunction = async (
   let newContructionFn = "";
   for (let i = 0; i < NUM_RETRIES; i++) {
     try {
-      const messages: CompletionCreateParams.CreateChatCompletionRequestNonStreaming.Message[] = [
+      const messages: CreateChatCompletionRequestMessage[] = [
         {
           role: "system",
           content: `Your job is to update prompt constructor functions. Here is the api shape for the current model:\n---\n${JSON.stringify(
             originalModelProvider.inputSchema,
             null,
             2,
-          )}\n\nDo not add any assistant messages.`,
+          )}\n\nDo not add any assistant messages. Do not add any extra fields to the schema. Try to keep temperature consistent.`,
         },
         {
           role: "user",

src/theme/ChakraThemeProvider.tsx

@@ -1,11 +1,31 @@
 import { extendTheme } from "@chakra-ui/react";
 import "@fontsource/inconsolata";
 import { ChakraProvider } from "@chakra-ui/react";
+import { modalAnatomy } from "@chakra-ui/anatomy";
+import { createMultiStyleConfigHelpers } from "@chakra-ui/styled-system";
 
 const systemFont =
   'ui-sans-serif, -apple-system, "system-ui", "Segoe UI", Helvetica, "Apple Color Emoji", Arial, sans-serif, "Segoe UI Emoji", "Segoe UI Symbol"';
 
+/* eslint-disable @typescript-eslint/unbound-method */
+const { definePartsStyle, defineMultiStyleConfig } = createMultiStyleConfigHelpers(
+  modalAnatomy.keys,
+);
+
+const modalTheme = defineMultiStyleConfig({
+  baseStyle: definePartsStyle({
+    dialog: { borderRadius: "sm" },
+  }),
+});
+
 const theme = extendTheme({
+  styles: {
+    global: (props: { colorMode: "dark" | "light" }) => ({
+      "html, body": {
+        backgroundColor: props.colorMode === "dark" ? "gray.900" : "white",
+      },
+    }),
+  },
   fonts: {
     heading: systemFont,
     body: systemFont,
@@ -32,6 +52,7 @@ const theme = extendTheme({
         },
       },
     },
+    Modal: modalTheme,
   },
 });