): string {
const hwKey = String(pts[0].hwKey);
- const cfg = getHardwareConfig(hwKey);
+ const cfg = getHardwareConfig(hwKey, pts[0].model);
const hwLabel = cfg ? getDisplayLabel(cfg) : hwKey;
return `${hwLabel} • ${comparisonEntryLabel(String(pts[0].date), numbering)}`;
}
@@ -266,7 +266,7 @@ const GPUGraph = React.memo(
const knownIssueAnnotations = useMemo(
(): KnownIssueAnnotation[] =>
matchKnownConfigIssues(modelLabel, filteredData).map((issue) => {
- const cfg = getHardwareConfig(issue.hwKey);
+ const cfg = getHardwareConfig(issue.hwKey, modelLabel);
const colorEntry = allGraphs.find(
(entry) => entry.hwKey === issue.hwKey && activeDates.has(entry.id),
);
@@ -835,7 +835,7 @@ const GPUGraph = React.memo(
hw: id,
label: comparisonEntryLabel(date, runNumbering),
color,
- title: getDisplayLabel(getHardwareConfig(hwKey)),
+ title: getDisplayLabel(getHardwareConfig(hwKey, modelLabel)),
isActive: activeDates.has(id),
onClick: () => {
toggleActiveDate(id);
diff --git a/packages/app/src/components/inference/ui/InferenceTable.tsx b/packages/app/src/components/inference/ui/InferenceTable.tsx
index c300e60d..5bdb713c 100644
--- a/packages/app/src/components/inference/ui/InferenceTable.tsx
+++ b/packages/app/src/components/inference/ui/InferenceTable.tsx
@@ -51,8 +51,8 @@ export default function InferenceTable({
() => [
{
header: 'GPU',
- cell: (row) => getDisplayLabel(getHardwareConfig(row.hwKey)),
- sortValue: (row) => getDisplayLabel(getHardwareConfig(row.hwKey)),
+ cell: (row) => getDisplayLabel(getHardwareConfig(row.hwKey, row.model)),
+ sortValue: (row) => getDisplayLabel(getHardwareConfig(row.hwKey, row.model)),
className: 'font-medium whitespace-nowrap',
},
{
diff --git a/packages/app/src/components/inference/ui/ScatterGraph.tsx b/packages/app/src/components/inference/ui/ScatterGraph.tsx
index fec5be80..68e9cc14 100644
--- a/packages/app/src/components/inference/ui/ScatterGraph.tsx
+++ b/packages/app/src/components/inference/ui/ScatterGraph.tsx
@@ -102,17 +102,23 @@ const formatChangelogDescription = (desc: string | string[]): React.JSX.Element
const CHART_MARGIN = { top: 24, right: 10, bottom: 60, left: 60 };
-// Derive a readable label from a hwKey using the HARDWARE_CONFIG source of truth
-const parseHwKeyToLabel = (hwKey: string): { name: string; label: string } => {
- const config = getHardwareConfig(hwKey);
+// Derive a readable label from a hwKey using the HARDWARE_CONFIG source of truth.
+// `model` (display name) enables per-model suffix overrides (e.g. M3 MTP → EAGLE).
+const parseHwKeyToLabel = (hwKey: string, model?: string): { name: string; label: string } => {
+ const config = getHardwareConfig(hwKey, model);
return { name: config.label, label: getDisplayLabel(config) };
};
// Line-label text for a curve. When more than one precision is shown, each curve
// is its own line, so append the precision (e.g. "B200 (vLLM) FP8") to keep the
// FP4 and FP8 curves of the same hardware distinguishable.
-const lineLabelText = (hwKey: string, precision: string, includePrecision: boolean): string => {
- const base = parseHwKeyToLabel(hwKey).label;
+const lineLabelText = (
+ hwKey: string,
+ precision: string,
+ includePrecision: boolean,
+ model?: string,
+): string => {
+ const base = parseHwKeyToLabel(hwKey, model).label;
return includePrecision ? `${base} ${getPrecisionLabel(precision as Precision)}` : base;
};
@@ -366,7 +372,7 @@ const ScatterGraph = React.memo(
const visiblePoints = [...filteredData, ...visibleOverlayPoints];
return matchKnownConfigIssues(modelLabel, visiblePoints).map((issue) => ({
issue,
- label: parseHwKeyToLabel(issue.hwKey).label,
+ label: parseHwKeyToLabel(issue.hwKey, modelLabel).label,
color: getCssColor(resolveColor(issue.hwKey)),
points: visiblePoints
.filter((p) => pointMatchesIssue(issue, p))
@@ -1065,7 +1071,7 @@ const ScatterGraph = React.memo(
placeLabel(
entry.key,
entry.hw,
- lineLabelText(entry.hw, entry.precision, multiPrecision),
+ lineLabelText(entry.hw, entry.precision, multiPrecision, modelLabel),
getCssColor(resolveColor(entry.hw)),
entry.points,
);
@@ -1079,7 +1085,7 @@ const ScatterGraph = React.memo(
lineLabels.push({
key: entry.key,
hw: entry.hw,
- label: lineLabelText(entry.hw, entry.precision, multiPrecision),
+ label: lineLabelText(entry.hw, entry.precision, multiPrecision, modelLabel),
color: getCssColor(resolveColor(entry.hw)),
x: xScale(entry.points[0].x),
y: yScale(entry.points[0].y),
@@ -1101,7 +1107,7 @@ const ScatterGraph = React.memo(
const info = unofficialRunInfos[runIndex];
const base = info
? `✕ ${info.branch || `run ${info.id}`}`
- : parseHwKeyToLabel(hwKey).label;
+ : parseHwKeyToLabel(hwKey, modelLabel).label;
return multiPrecision
? `${base} ${getPrecisionLabel(precision as Precision)}`
: base;
@@ -1144,7 +1150,7 @@ const ScatterGraph = React.memo(
lineLabels.push({
key: entry.key,
hw: entry.hw,
- label: lineLabelText(entry.hw, entry.precision, multiPrecision),
+ label: lineLabelText(entry.hw, entry.precision, multiPrecision, modelLabel),
color: getCssColor(resolveColor(entry.hw)),
x: xScale(pt.x),
y: yScale(pt.y),
@@ -1158,7 +1164,7 @@ const ScatterGraph = React.memo(
const info = unofficialRunInfos[group.runIndex];
const branchOrHw = info
? `✕ ${info.branch || `run ${info.id}`}`
- : parseHwKeyToLabel(group.hwKey).label;
+ : parseHwKeyToLabel(group.hwKey, modelLabel).label;
const labelText = multiPrecision
? `${branchOrHw} ${getPrecisionLabel((group.points[0]?.precision ?? '') as Precision)}`
: branchOrHw;
diff --git a/packages/app/src/components/inference/ui/UnofficialChartDisplay.tsx b/packages/app/src/components/inference/ui/UnofficialChartDisplay.tsx
index f9b1b3c8..799854d7 100644
--- a/packages/app/src/components/inference/ui/UnofficialChartDisplay.tsx
+++ b/packages/app/src/components/inference/ui/UnofficialChartDisplay.tsx
@@ -106,7 +106,7 @@ export function UnofficialChartDisplay() {
hardwareConfig: Object.fromEntries(
Object.entries(dataForChart.gpus || {}).map(([k, v]) => [
k,
- { ...getHardwareConfig(k), ...v },
+ { ...getHardwareConfig(k, selectedModel), ...v },
]),
),
};
diff --git a/packages/app/src/components/inference/utils/changelogFormatters.test.ts b/packages/app/src/components/inference/utils/changelogFormatters.test.ts
index 79910096..905e8c89 100644
--- a/packages/app/src/components/inference/utils/changelogFormatters.test.ts
+++ b/packages/app/src/components/inference/utils/changelogFormatters.test.ts
@@ -17,6 +17,13 @@ describe('formatConfigKeys', () => {
expect(result).toContain('FP8');
});
+ it('renders M3 mtp as EAGLE (not MTP)', () => {
+ const result = formatConfigKeys('minimaxm3-fp8-h100-vllm-mtp');
+ expect(result).toContain('H100');
+ expect(result).toContain('EAGLE');
+ expect(result).not.toContain('MTP');
+ });
+
it('formats compound framework names', () => {
const result = formatConfigKeys('gptoss-fp4-b200-dynamo-sglang');
expect(result).toContain('B200');
diff --git a/packages/app/src/components/inference/utils/changelogFormatters.tsx b/packages/app/src/components/inference/utils/changelogFormatters.tsx
index 7574f252..aae20a0b 100644
--- a/packages/app/src/components/inference/utils/changelogFormatters.tsx
+++ b/packages/app/src/components/inference/utils/changelogFormatters.tsx
@@ -1,4 +1,7 @@
-import { resolveFrameworkAliasesInString } from '@semianalysisai/inferencex-constants';
+import {
+ resolveFrameworkAliasesInString,
+ resolveFrameworkPartLabel,
+} from '@semianalysisai/inferencex-constants';
import { type Precision, MODEL_PREFIX_MAPPING, getPrecisionLabel } from '@/lib/data-mappings';
import { getFrameworkLabel } from '@/lib/utils';
@@ -45,6 +48,8 @@ export function formatConfigKeys(key: string) {
const isMtp = framework.endsWith('-mtp');
const baseFramework = isMtp ? framework.slice(0, -4) : framework;
const baseLabel = getFrameworkLabel(baseFramework);
- const frameworkLabel = isMtp ? `${baseLabel}, MTP` : baseLabel;
+ // M3's `mtp` spec token renders as "EAGLE"; every other model keeps "MTP".
+ const mtpLabel = resolveFrameworkPartLabel(MODEL_PREFIX_MAPPING[model], 'mtp');
+ const frameworkLabel = isMtp ? `${baseLabel}, ${mtpLabel}` : baseLabel;
return `${gpu.toUpperCase()} (${frameworkLabel}) ${MODEL_PREFIX_MAPPING[model]} ${getPrecisionLabel(precision as Precision)}`;
}
diff --git a/packages/app/src/components/submissions/SubmissionsTable.tsx b/packages/app/src/components/submissions/SubmissionsTable.tsx
index d7845e30..5dd56f47 100644
--- a/packages/app/src/components/submissions/SubmissionsTable.tsx
+++ b/packages/app/src/components/submissions/SubmissionsTable.tsx
@@ -3,6 +3,11 @@
import { ChevronDown, ChevronRight, GitCompare, Info } from 'lucide-react';
import { useCallback, useEffect, useMemo, useState } from 'react';
+import {
+ DB_MODEL_TO_DISPLAY,
+ resolveFrameworkPartLabel,
+} from '@semianalysisai/inferencex-constants';
+
import { track } from '@/lib/analytics';
import { MODEL_PREFIX_MAPPING, getModelLabel } from '@/lib/data-mappings';
import type { SubmissionSummaryRow } from '@/lib/submissions-types';
@@ -302,7 +307,7 @@ function SubmissionRow({
| {row.precision} |
{row.spec_method && row.spec_method !== 'none' ? (
- row.spec_method
+ resolveFrameworkPartLabel(DB_MODEL_TO_DISPLAY[row.model], row.spec_method)
) : (
—
)}
@@ -368,7 +373,9 @@ function SubmissionRow({
label="Spec Method:"
tip="Speculative decoding method (e.g. MTP, Eagle)"
>
- {row.spec_method || 'none'}
+ {row.spec_method && row.spec_method !== 'none'
+ ? resolveFrameworkPartLabel(DB_MODEL_TO_DISPLAY[row.model], row.spec_method)
+ : 'none'}
{
if (v <= 0 || !isFinite(mins[i]) || maxs[i] === mins[i]) return null;
const norm = (v - mins[i]) / (maxs[i] - mins[i]);
diff --git a/packages/app/src/lib/benchmark-transform.test.ts b/packages/app/src/lib/benchmark-transform.test.ts
index b49fae39..8f27cc8f 100644
--- a/packages/app/src/lib/benchmark-transform.test.ts
+++ b/packages/app/src/lib/benchmark-transform.test.ts
@@ -302,6 +302,26 @@ describe('transformBenchmarkRows', () => {
const hwKeys = Object.keys(hardwareConfig);
expect(hwKeys.length).toBeGreaterThanOrEqual(2);
});
+
+ it('labels M3 mtp configs with the "M3 EAGLE" suffix', () => {
+ const rows = [
+ makeRow({ model: 'minimaxm3', hardware: 'h100', framework: 'vllm', spec_method: 'mtp' }),
+ ];
+ const { hardwareConfig } = transformBenchmarkRows(rows);
+ const entry = hardwareConfig['h100_vllm_mtp'];
+ expect(entry).toBeDefined();
+ expect(entry.suffix).toBe('(vLLM, M3 EAGLE)');
+ });
+
+ it('keeps the generic MTP suffix for non-M3 mtp configs', () => {
+ const rows = [
+ makeRow({ model: 'dsr1', hardware: 'h200', framework: 'sglang', spec_method: 'mtp' }),
+ ];
+ const { hardwareConfig } = transformBenchmarkRows(rows);
+ const entry = hardwareConfig['h200_sglang_mtp'];
+ expect(entry).toBeDefined();
+ expect(entry.suffix).toBe('(SGLang, MTP)');
+ });
});
// ---------------------------------------------------------------------------
diff --git a/packages/app/src/lib/benchmark-transform.ts b/packages/app/src/lib/benchmark-transform.ts
index 87b48558..ac806b79 100644
--- a/packages/app/src/lib/benchmark-transform.ts
+++ b/packages/app/src/lib/benchmark-transform.ts
@@ -122,7 +122,7 @@ export function transformBenchmarkRows(rows: BenchmarkRow[]): {
entry.hwKey = hwKey;
if (!hwConfigCache.has(hwKey)) {
- const hwConfig = getHardwareConfig(hwKey);
+ const hwConfig = getHardwareConfig(hwKey, entry.model);
hwConfigCache.set(hwKey, hwConfig);
if (hwConfig) gpuConfig[hwKey] = { ...hwConfig, name: hwKey };
}
diff --git a/packages/app/src/lib/constants.test.ts b/packages/app/src/lib/constants.test.ts
index 24fb371c..dd849319 100644
--- a/packages/app/src/lib/constants.test.ts
+++ b/packages/app/src/lib/constants.test.ts
@@ -154,6 +154,27 @@ describe('getHardwareConfig', () => {
warnSpy.mockRestore();
});
+ it('renders M3 mtp configs with the "M3 EAGLE" suffix when model is passed', () => {
+ const config = getHardwareConfig('h100_vllm_mtp', 'MiniMax-M3');
+ expect(config.suffix).toBe('(vLLM, M3 EAGLE)');
+ expect(config.gpu).toContain('M3 EAGLE');
+ });
+
+ it('keeps the generic MTP suffix for non-M3 models', () => {
+ expect(getHardwareConfig('h100_sglang_mtp', 'DeepSeek-R1-0528').suffix).toBe('(SGLang, MTP)');
+ });
+
+ it('keeps the generic MTP suffix when no model is passed (backward compatible)', () => {
+ expect(getHardwareConfig('h100_vllm_mtp').suffix).toBe('(vLLM, MTP)');
+ });
+
+ it('does not let the model-scoped cache leak into the model-less lookup', () => {
+ // Prime the cache with the M3 (EAGLE) variant first, then ensure the
+ // model-less call still returns the generic label.
+ getHardwareConfig('b200_vllm_mtp', 'MiniMax-M3');
+ expect(getHardwareConfig('b200_vllm_mtp').suffix).toBe('(vLLM, MTP)');
+ });
+
it('HW_REGISTRY has non-zero power for all entries', () => {
for (const entry of Object.values(HW_REGISTRY)) {
expect(entry.power).toBeGreaterThan(0);
diff --git a/packages/app/src/lib/constants.ts b/packages/app/src/lib/constants.ts
index a3987de3..a720077d 100644
--- a/packages/app/src/lib/constants.ts
+++ b/packages/app/src/lib/constants.ts
@@ -1,4 +1,4 @@
-import { FRAMEWORK_LABELS, HW_REGISTRY } from '@semianalysisai/inferencex-constants';
+import { HW_REGISTRY, resolveFrameworkPartLabel } from '@semianalysisai/inferencex-constants';
/** d3.schemeTableau10 — 10-color categorical palette for tracked configs. */
export const TABLEAU_10 = [
@@ -59,8 +59,11 @@ const UNKNOWN_HARDWARE: HardwareEntry = {
/**
* Build a hardware config entry from a key like "h100_dynamo-trt_mtp".
* Derives all display fields from GPU_KEYS/GPU_VENDORS + FRAMEWORK_LABELS.
+ *
+ * `model` (frontend display name) enables per-model suffix overrides — e.g. M3's
+ * `mtp` token renders as "EAGLE" rather than the generic "MTP".
*/
-function buildHardwareEntry(hwKey: string): HardwareEntry | null {
+function buildHardwareEntry(hwKey: string, model?: string): HardwareEntry | null {
const base = hwKey.split('_')[0];
const reg = HW_REGISTRY[base];
if (!reg) return null;
@@ -68,7 +71,7 @@ function buildHardwareEntry(hwKey: string): HardwareEntry | null {
const parts = hwKey.split('_').slice(1);
const label = reg.label;
const gpuName = base.toUpperCase(); // always raw uppercase for gpu string
- const partLabels = parts.map((p) => FRAMEWORK_LABELS[p] ?? p.toUpperCase());
+ const partLabels = parts.map((p) => resolveFrameworkPartLabel(model, p));
return {
name: hwKey.replaceAll('_', '-'),
@@ -132,14 +135,18 @@ const hwCache = new Map();
/**
* Get hardware config for a GPU key, building it dynamically from shared GPU constants + FRAMEWORK_LABELS.
* Returns UNKNOWN_HARDWARE for unrecognized base GPUs.
+ *
+ * Pass `model` (frontend display name) to apply per-model suffix overrides
+ * (e.g. M3 `mtp` → "EAGLE"). Omitting it preserves the generic labels.
*/
-export function getHardwareConfig(hwKey: string): HardwareEntry {
- const cached = hwCache.get(hwKey);
+export function getHardwareConfig(hwKey: string, model?: string): HardwareEntry {
+ const cacheKey = model ? `${model}|${hwKey}` : hwKey;
+ const cached = hwCache.get(cacheKey);
if (cached) return cached;
- const entry = buildHardwareEntry(hwKey);
+ const entry = buildHardwareEntry(hwKey, model);
if (entry) {
- hwCache.set(hwKey, entry);
+ hwCache.set(cacheKey, entry);
return entry;
}
diff --git a/packages/constants/src/framework-aliases.test.ts b/packages/constants/src/framework-aliases.test.ts
index 5f819312..a28bea34 100644
--- a/packages/constants/src/framework-aliases.test.ts
+++ b/packages/constants/src/framework-aliases.test.ts
@@ -3,8 +3,10 @@ import { describe, expect, it } from 'vitest';
import {
FRAMEWORK_ALIASES,
FRAMEWORK_LABELS,
+ MODEL_SPEC_METHOD_LABELS,
resolveFrameworkAlias,
resolveFrameworkAliasesInString,
+ resolveFrameworkPartLabel,
} from './framework-aliases';
describe('FRAMEWORK_LABELS', () => {
@@ -17,6 +19,34 @@ describe('FRAMEWORK_LABELS', () => {
});
});
+describe('MODEL_SPEC_METHOD_LABELS', () => {
+ it('maps MiniMax-M3 mtp to "M3 EAGLE"', () => {
+ expect(MODEL_SPEC_METHOD_LABELS['MiniMax-M3']?.mtp).toBe('M3 EAGLE');
+ });
+});
+
+describe('resolveFrameworkPartLabel', () => {
+ it('renders M3 mtp as "M3 EAGLE"', () => {
+ expect(resolveFrameworkPartLabel('MiniMax-M3', 'mtp')).toBe('M3 EAGLE');
+ });
+
+ it('keeps the generic MTP label for other models', () => {
+ expect(resolveFrameworkPartLabel('DeepSeek-R1-0528', 'mtp')).toBe('MTP');
+ });
+
+ it('keeps the generic MTP label when no model is provided', () => {
+ expect(resolveFrameworkPartLabel(undefined, 'mtp')).toBe('MTP');
+ });
+
+ it('falls back to FRAMEWORK_LABELS for non-overridden parts even for M3', () => {
+ expect(resolveFrameworkPartLabel('MiniMax-M3', 'vllm')).toBe('vLLM');
+ });
+
+ it('uppercases unknown tokens', () => {
+ expect(resolveFrameworkPartLabel('MiniMax-M3', 'foo')).toBe('FOO');
+ });
+});
+
describe('FRAMEWORK_ALIASES', () => {
it('maps sglang-disagg to mori-sglang with disagg=true', () => {
expect(FRAMEWORK_ALIASES['sglang-disagg']).toEqual({ canonical: 'mori-sglang', disagg: true });
diff --git a/packages/constants/src/framework-aliases.ts b/packages/constants/src/framework-aliases.ts
index 6794069a..208fbd57 100644
--- a/packages/constants/src/framework-aliases.ts
+++ b/packages/constants/src/framework-aliases.ts
@@ -48,6 +48,31 @@ export const FRAMEWORK_LABELS: Record = {
mtp: 'MTP',
};
+/**
+ * Per-model display overrides for hwKey suffix parts (framework / spec_method
+ * tokens), keyed by frontend display model name → token → label.
+ *
+ * M3's speculative-decoding runs are ingested under the generic `mtp` token but
+ * actually use EAGLE, so for MiniMax-M3 the suffix reads "EAGLE" while every
+ * other model keeps the generic "MTP" label.
+ */
+export const MODEL_SPEC_METHOD_LABELS: Record> = {
+ 'MiniMax-M3': { mtp: 'M3 EAGLE' },
+};
+
+/**
+ * Resolve a single hwKey suffix part (framework or spec_method token) to its
+ * display label, applying any per-model override before the generic
+ * FRAMEWORK_LABELS map. `model` is the frontend display model name.
+ */
+export function resolveFrameworkPartLabel(model: string | undefined, part: string): string {
+ return (
+ (model ? MODEL_SPEC_METHOD_LABELS[model]?.[part] : undefined) ??
+ FRAMEWORK_LABELS[part] ??
+ part.toUpperCase()
+ );
+}
+
/**
* Resolve a framework name to its canonical form.
* Returns the input lowercased if no alias exists.
|