add Model, ISL, OSL columns to inference CSV export

Author: adibarra

packages/app/src/components/inference/ui/ChartDisplay.tsx

@@ -268,7 +268,11 @@ export default function ChartDisplay() {
                   const visibleData = graph.data.filter((d) =>
                     activeHwTypes.has(d.hwKey as string),
                   );
-                  const { headers, rows } = inferenceChartToCsv(visibleData);
+                  const { headers, rows } = inferenceChartToCsv(
+                    visibleData,
+                    graph.model,
+                    graph.sequence,
+                  );
                   exportToCsv(`chart-${graphIndex}-${Date.now()}`, headers, rows);
                 }}
               />

packages/app/src/lib/csv-export-helpers.test.ts

@@ -57,7 +57,7 @@ describe('inferenceChartToCsv', () => {
 
   it('exports all raw benchmark fields', () => {
     const data = [makePoint()];
-    const { headers, rows } = inferenceChartToCsv(data);
+    const { headers, rows } = inferenceChartToCsv(data, 'llama-3.1-405b', '1k/1k');
 
     // Should have all metric columns
     expect(headers).toContain('Throughput/GPU (tok/s)');
@@ -71,9 +71,22 @@ describe('inferenceChartToCsv', () => {
     expect(rows).toHaveLength(1);
   });
 
+  it('includes Model, ISL, and OSL columns from model and sequence', () => {
+    const data = [makePoint()];
+    const { headers, rows } = inferenceChartToCsv(data, 'llama-3.1-405b', '1k/8k');
+    const row = rows[0];
+
+    expect(headers[0]).toBe('Model');
+    expect(headers[1]).toBe('ISL');
+    expect(headers[2]).toBe('OSL');
+    expect(row[0]).toBe('llama-3.1-405b');
+    expect(row[1]).toBe(1024);
+    expect(row[2]).toBe(8192);
+  });
+
   it('includes throughput and latency values in correct columns', () => {
     const data = [makePoint()];
-    const { headers, rows } = inferenceChartToCsv(data);
+    const { headers, rows } = inferenceChartToCsv(data, 'llama-3.1-405b', '1k/1k');
     const row = rows[0];
 
     const tputIdx = headers.indexOf('Throughput/GPU (tok/s)');
@@ -88,13 +101,13 @@ describe('inferenceChartToCsv', () => {
 
   it('filters out hidden data points', () => {
     const data = [makePoint(), makePoint({ hidden: true })];
-    const { rows } = inferenceChartToCsv(data);
+    const { rows } = inferenceChartToCsv(data, 'llama-3.1-405b', '1k/1k');
     expect(rows).toHaveLength(1);
   });
 
   it('includes disaggregated and parallelism fields', () => {
     const data = [makePoint({ disagg: true, num_prefill_gpu: 2, num_decode_gpu: 6, ep: 4 })];
-    const { headers, rows } = inferenceChartToCsv(data);
+    const { headers, rows } = inferenceChartToCsv(data, 'llama-3.1-405b', '1k/1k');
     const row = rows[0];
 
     expect(row[headers.indexOf('Disaggregated')]).toBe(true);
@@ -104,7 +117,7 @@ describe('inferenceChartToCsv', () => {
   });
 
   it('handles empty data', () => {
-    const { rows } = inferenceChartToCsv([]);
+    const { rows } = inferenceChartToCsv([], 'llama-3.1-405b', '1k/1k');
     expect(rows).toHaveLength(0);
   });
 
@@ -129,7 +142,7 @@ describe('inferenceChartToCsv', () => {
         costri: { y: 0, roof: false },
       } as InferenceData,
     ];
-    const { headers, rows } = inferenceChartToCsv(data);
+    const { headers, rows } = inferenceChartToCsv(data, 'llama-3.1-405b', '1k/1k');
     const row = rows[0];
 
     // Missing optional fields should be ''

packages/app/src/lib/csv-export-helpers.ts

@@ -9,6 +9,8 @@
 
 import type { InferenceData, TrendDataPoint } from '@/components/inference/types';
 
+import { sequenceToIslOsl } from '@semianalysisai/inferencex-constants';
+
 interface CsvData {
   headers: string[];
   rows: (string | number | boolean | null | undefined)[][];
@@ -19,8 +21,16 @@ interface CsvData {
  * Exports all raw benchmark metrics so the user gets a full data dump,
  * regardless of which axes are currently plotted.
  */
-export function inferenceChartToCsv(data: InferenceData[]): CsvData {
+export function inferenceChartToCsv(
+  data: InferenceData[],
+  model: string,
+  sequence: string,
+): CsvData {
+  const islOsl = sequenceToIslOsl(sequence);
   const headers = [
+    'Model',
+    'ISL',
+    'OSL',
     'Hardware',
     'Hardware Key',
     'Framework',
@@ -71,6 +81,9 @@ export function inferenceChartToCsv(data: InferenceData[]): CsvData {
   const rows = data
     .filter((d) => !d.hidden)
     .map((d) => [
+      model,
+      islOsl?.isl ?? '',
+      islOsl?.osl ?? '',
       d.hw ?? '',
       d.hwKey,
       d.framework ?? '',