push up hopefully working build to test with guidellm after the preview is live in gh-pages branch

Author: DaltheCow

.github/workflows/development.yml

@@ -29,6 +29,11 @@ jobs:
       - name: Check out code
         uses: actions/checkout@v3
 
+      - name: Set up Node.js 22
+        uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+
       - name: Install dependencies
         run: npm ci
 
@@ -59,6 +64,11 @@ jobs:
       - name: Check out code
         uses: actions/checkout@v3
 
+      - name: Set up Node.js 22
+        uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+
       - name: Install dependencies
         run: npm ci
 
@@ -89,6 +99,11 @@ jobs:
       - name: Check out code
         uses: actions/checkout@v3
 
+      - name: Set up Node.js 22
+        uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+
       - name: Install dependencies
         run: npm ci
 
@@ -119,6 +134,11 @@ jobs:
       - name: Check out code
         uses: actions/checkout@v3
 
+      - name: Set up Node.js 22
+        uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+
       - name: Install dependencies
         run: npm ci
 
@@ -149,6 +169,11 @@ jobs:
       - name: Check out code
         uses: actions/checkout@v3
 
+      - name: Set up Node.js 22
+        uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+
       - name: Install dependencies
         run: npm ci
 
@@ -219,6 +244,11 @@ jobs:
         with:
           fetch-depth: 0
 
+      - name: Set up Node.js 22
+        uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+
       - name: Check if UI-related files changed
         id: check-changes
         run: |
@@ -250,7 +280,8 @@ jobs:
 
           # Set asset prefix and base path with PR number
           ASSET_PREFIX=https://neuralmagic.github.io/guidellm/ui/pr/${PR_NUMBER}
-          USE_MOCK_DATA=true
+          # temporarily setting to false to test if this build works with guidellm
+          USE_MOCK_DATA=false
           BASE_PATH=/ui/pr/${PR_NUMBER}
           GIT_SHA=${{ github.sha }}
           export ASSET_PREFIX=${ASSET_PREFIX}

.github/workflows/main.yml

@@ -30,6 +30,11 @@ jobs:
       - name: Check out code
         uses: actions/checkout@v3
 
+      - name: Set up Node.js 22
+        uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+
       - name: Install dependencies
         run: npm ci
 
@@ -60,6 +65,11 @@ jobs:
       - name: Check out code
         uses: actions/checkout@v3
 
+      - name: Set up Node.js 22
+        uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+
       - name: Install dependencies
         run: npm ci
 
@@ -90,6 +100,11 @@ jobs:
       - name: Check out code
         uses: actions/checkout@v3
 
+      - name: Set up Node.js 22
+        uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+
       - name: Install dependencies
         run: npm ci
 
@@ -120,6 +135,11 @@ jobs:
       - name: Check out code
         uses: actions/checkout@v3
 
+      - name: Set up Node.js 22
+        uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+
       - name: Install dependencies
         run: npm ci
 
@@ -150,6 +170,11 @@ jobs:
       - name: Check out code
         uses: actions/checkout@v3
 
+      - name: Set up Node.js 22
+        uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+
       - name: Install dependencies
         run: npm ci
 
@@ -165,6 +190,11 @@ jobs:
       - name: Check out code
         uses: actions/checkout@v3
 
+      - name: Set up Node.js 22
+        uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+
       - name: Install dependencies
         run: npm ci
 

src/guidellm/objects/statistics.py

@@ -37,6 +37,9 @@ class Percentiles(StandardBaseModel):
     p25: float = Field(
         description="The 25th percentile of the distribution.",
     )
+    p50: float = Field(
+        description="The 50th percentile of the distribution.",
+    )
     p75: float = Field(
         description="The 75th percentile of the distribution.",
     )
@@ -159,6 +162,7 @@ def from_distribution_function(
                     p05=cdf[np.argmax(cdf[:, 1] >= 0.05), 0].item(),  # noqa: PLR2004
                     p10=cdf[np.argmax(cdf[:, 1] >= 0.1), 0].item(),  # noqa: PLR2004
                     p25=cdf[np.argmax(cdf[:, 1] >= 0.25), 0].item(),  # noqa: PLR2004
+                    p50=cdf[np.argmax(cdf[:, 1] >= 0.50), 0].item(),  # noqa: PLR2004
                     p75=cdf[np.argmax(cdf[:, 1] >= 0.75), 0].item(),  # noqa: PLR2004
                     p90=cdf[np.argmax(cdf[:, 1] >= 0.9), 0].item(),  # noqa: PLR2004
                     p95=cdf[np.argmax(cdf[:, 1] >= 0.95), 0].item(),  # noqa: PLR2004
@@ -172,6 +176,7 @@ def from_distribution_function(
                     p05=0,
                     p10=0,
                     p25=0,
+                    p50=0,
                     p75=0,
                     p90=0,
                     p95=0,

src/guidellm/presentation/__init__.py

@@ -1,5 +1,5 @@
 from .builder import UIDataBuilder
-from .data_models import (Bucket, Model, Dataset, RunInfo, TokenDistribution, TokenDetails, Server, WorkloadDetails, BenchmarkDatum)
+from .data_models import (Bucket, Model, Dataset, RunInfo, Distribution, TokenDetails, Server, WorkloadDetails, BenchmarkDatum)
 from .injector import (create_report, inject_data)
 
 __all__ = [
@@ -8,7 +8,7 @@
     "Model",
     "Dataset",
     "RunInfo",
-    "TokenDistribution",
+    "Distribution",
     "TokenDetails",
     "Server",
     "WorkloadDetails",

src/guidellm/presentation/data_models.py

@@ -1,13 +1,13 @@
 from collections import defaultdict
 from math import ceil
-from pydantic import BaseModel
+from pydantic import computed_field, BaseModel
 import random
 from typing import List, Optional, Tuple
 
 from guidellm.benchmark.benchmark import GenerativeBenchmark
 from guidellm.objects.statistics import DistributionSummary
 
-__all__ = ["Bucket", "Model", "Dataset", "RunInfo", "TokenDistribution", "TokenDetails", "Server", "WorkloadDetails", "BenchmarkDatum"]
+__all__ = ["Bucket", "Model", "Dataset", "RunInfo", "Distribution", "TokenDetails", "Server", "WorkloadDetails", "BenchmarkDatum"]
 
 class Bucket(BaseModel):
   value: float
@@ -69,22 +69,22 @@ def from_benchmarks(cls, benchmarks: list[GenerativeBenchmark]):
       dataset=Dataset(name="N/A")
     )
 
-class TokenDistribution(BaseModel):
+class Distribution(BaseModel):
   statistics: Optional[DistributionSummary] = None
   buckets: list[Bucket]
   bucket_width: float
 
 
 class TokenDetails(BaseModel):
   samples: list[str]
-  token_distributions: TokenDistribution
+  token_distributions: Distribution
 
 class Server(BaseModel):
   target: str
 
 class RequestOverTime(BaseModel):
    num_benchmarks: int
-   requests_over_time: TokenDistribution
+   requests_over_time: Distribution
 
 class WorkloadDetails(BaseModel):
   prompts: TokenDetails
@@ -109,8 +109,8 @@ def from_benchmarks(cls, benchmarks: list[GenerativeBenchmark]):
 
     prompt_token_stats = DistributionSummary.from_values(prompt_tokens)
     output_token_stats = DistributionSummary.from_values(output_tokens)
-    prompt_token_distributions = TokenDistribution(statistics=prompt_token_stats, buckets=prompt_token_buckets, bucket_width=1)
-    output_token_distributions = TokenDistribution(statistics=output_token_stats, buckets=output_token_buckets, bucket_width=1)
+    prompt_token_distributions = Distribution(statistics=prompt_token_stats, buckets=prompt_token_buckets, bucket_width=1)
+    output_token_distributions = Distribution(statistics=output_token_stats, buckets=output_token_buckets, bucket_width=1)
 
     min_start_time = benchmarks[0].run_stats.start_time
 
@@ -122,7 +122,7 @@ def from_benchmarks(cls, benchmarks: list[GenerativeBenchmark]):
     ]
     number_of_buckets = len(benchmarks)
     request_over_time_buckets, bucket_width = Bucket.from_data(all_req_times, None, number_of_buckets)
-    request_over_time_distribution = TokenDistribution(buckets=request_over_time_buckets, bucket_width=bucket_width)
+    request_over_time_distribution = Distribution(buckets=request_over_time_buckets, bucket_width=bucket_width)
     return cls(
        prompts=TokenDetails(samples=sample_prompts, token_distributions=prompt_token_distributions),
        generations=TokenDetails(samples=sample_outputs, token_distributions=output_token_distributions),
@@ -131,19 +131,39 @@ def from_benchmarks(cls, benchmarks: list[GenerativeBenchmark]):
        server=Server(target=target)
     )
 
+class TabularDistributionSummary(DistributionSummary):
+    """
+    Same fields as `DistributionSummary`, but adds a ready-to-serialize/iterate
+    `percentile_rows` helper.
+    """
+
+    @computed_field
+    @property
+    def percentile_rows(self) -> list[dict[str, float]]:
+        return [
+            {"percentile": name, "value": value}
+            for name, value in self.percentiles.model_dump().items()
+        ]
+
+    @classmethod
+    def from_distribution_summary(
+        cls, distribution: DistributionSummary
+    ) -> "TabularDistributionSummary":
+        return cls(**distribution.model_dump())
+
 class BenchmarkDatum(BaseModel):
   requests_per_second: float
-  tpot: DistributionSummary
-  ttft: DistributionSummary
-  throughput: DistributionSummary
-  time_per_request: DistributionSummary
+  tpot: TabularDistributionSummary
+  ttft: TabularDistributionSummary
+  throughput: TabularDistributionSummary
+  time_per_request: TabularDistributionSummary
 
   @classmethod
   def from_benchmark(cls, bm: GenerativeBenchmark):
     return cls(
        requests_per_second=bm.metrics.requests_per_second.successful.mean,
-       tpot=bm.metrics.inter_token_latency_ms.successful,
-       ttft=bm.metrics.time_to_first_token_ms.successful,
-       throughput=bm.metrics.output_tokens_per_second.successful,
-       time_per_request=bm.metrics.request_latency.successful,
+       tpot=TabularDistributionSummary.from_distribution_summary(bm.metrics.inter_token_latency_ms.successful),
+       ttft=TabularDistributionSummary.from_distribution_summary(bm.metrics.time_to_first_token_ms.successful),
+       throughput=TabularDistributionSummary.from_distribution_summary(bm.metrics.output_tokens_per_second.successful),
+       time_per_request=TabularDistributionSummary.from_distribution_summary(bm.metrics.request_latency.successful),
     )

src/ui/lib/store/slices/benchmarks/benchmarks.api.ts

@@ -1,7 +1,7 @@
 import { ThunkDispatch, UnknownAction } from '@reduxjs/toolkit';
 import { createApi, fetchBaseQuery } from '@reduxjs/toolkit/query/react';
 
-import { Benchmarks, MetricData } from './benchmarks.interfaces';
+import { Benchmarks, Statistics } from './benchmarks.interfaces';
 import { formatNumber } from '../../../utils/helpers';
 import { defaultPercentile } from '../slo/slo.constants';
 import { setSloData } from '../slo/slo.slice';
@@ -13,14 +13,14 @@ const fetchBenchmarks = () => {
 };
 
 const getAverageValueForPercentile = (
-  firstMetric: MetricData,
-  lastMetric: MetricData,
+  firstMetric: Statistics,
+  lastMetric: Statistics,
   percentile: string
 ) => {
-  const firstPercentile = firstMetric.percentiles.find(
+  const firstPercentile = firstMetric.percentileRows.find(
     (p) => p.percentile === percentile
   );
-  const lastPercentile = lastMetric.percentiles.find(
+  const lastPercentile = lastMetric.percentileRows.find(
     (p) => p.percentile === percentile
   );
   return ((firstPercentile?.value ?? 0) + (lastPercentile?.value ?? 0)) / 2;

src/ui/lib/store/slices/benchmarks/benchmarks.interfaces.ts

@@ -1,44 +1,32 @@
 export type Name = 'benchmarks';
 
-interface Statistics {
+export interface Statistics {
   total: number;
   mean: number;
   std: number;
   median: number;
   min: number;
   max: number;
+  percentileRows: Percentile[];
+  percentiles: Record<PercentileValues, number>;
 }
 
 export type PercentileValues = 'p50' | 'p90' | 'p95' | 'p99';
 
 interface Percentile {
-  percentile: string;
+  percentile: PercentileValues;
   value: number;
 }
 
-interface Bucket {
-  value: number;
-  count: number;
-}
-
-export interface MetricData {
-  statistics: Statistics;
-  percentiles: Percentile[];
-  buckets: Bucket[];
-  bucketWidth: number;
-}
-
 export interface BenchmarkMetrics {
-  ttft: MetricData;
-  tpot: MetricData;
-  timePerRequest: MetricData;
-  throughput: MetricData;
+  ttft: Statistics;
+  tpot: Statistics;
+  timePerRequest: Statistics;
+  throughput: Statistics;
 }
 
 export interface Benchmark extends BenchmarkMetrics {
   requestsPerSecond: number;
 }
 
-export type Benchmarks = {
-  benchmarks: Benchmark[];
-};
+export type Benchmarks = Benchmark[];