Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
24 changes: 24 additions & 0 deletions src/app/components/chat-panel/chat-panel.component.html
Original file line number Diff line number Diff line change
Expand Up @@ -216,6 +216,30 @@
}
</div>
}
@if (message.metricResults?.length) {
<div class="message-metric-results">
@for (metricResult of message.metricResults; track metricResult.metricName) {
<span
class="metric-result-chip"
[ngClass]="{
'metric-pass': metricResult.evalStatus === 1 || metricResult.evalStatus === 'PASSED',
'metric-fail': metricResult.evalStatus === 2 || metricResult.evalStatus === 'FAILED',
'metric-neutral': metricResult.evalStatus !== 1 && metricResult.evalStatus !== 2 && metricResult.evalStatus !== 'PASSED' && metricResult.evalStatus !== 'FAILED'
}"
>
<span class="metric-result-name">{{ metricResult.metricName }}</span>
@if (metricResult.score !== undefined) {
<span class="metric-result-score">Score: {{ metricResult.score }}</span>
}
@if (metricResult.threshold !== undefined) {
<span class="metric-result-threshold"
>Threshold: {{ metricResult.threshold }}</span
>
}
</span>
}
</div>
}
</mat-card>
} @if (message.functionCall) {
<button
Expand Down
33 changes: 33 additions & 0 deletions src/app/components/chat-panel/chat-panel.component.scss
Original file line number Diff line number Diff line change
Expand Up @@ -151,6 +151,39 @@
font-weight: 600;
}

.message-metric-results {
display: flex;
flex-wrap: wrap;
gap: 6px;
margin-top: 10px;
}

.metric-result-chip {
display: inline-flex;
align-items: center;
gap: 6px;
border-radius: 12px;
padding: 4px 8px;
background-color: var(
--chat-panel-metric-result-chip-background,
rgba(255, 255, 255, 0.1)
);
font-size: 12px;
color: var(--chat-panel-metric-result-chip-color, #e8eaed);

&.metric-pass {
color: var(--chat-panel-metric-result-chip-pass-color, #1e8e3e);
}

&.metric-fail {
color: var(--chat-panel-metric-result-chip-fail-color, #d93025);
}

&.metric-neutral {
color: var(--chat-panel-metric-result-chip-neutral-color, #9aa0a6);
}
}

.eval-response-header {
padding-bottom: 5px;
border-bottom: 2px solid var(--chat-panel-eval-response-header-border-bottom-color);
Expand Down
71 changes: 67 additions & 4 deletions src/app/components/eval-tab/eval-tab.component.html
Original file line number Diff line number Diff line change
Expand Up @@ -63,6 +63,36 @@
<div>
@if (!showEvalHistory()) {
<div>
<div class="metric-selection">
<mat-form-field appearance="outline" class="metric-select-field">
<mat-label>Evaluation metrics</mat-label>
<mat-select multiple [value]="selectedMetricNames" (valueChange)="onMetricSelectionChange($event)">
@for (metric of metricOptions; track metric.metricName) {
<mat-option [value]="metric.metricName">
<span>{{ metric.metricName }}</span>
@if (metric.description) {
<span class="metric-option-description"> — {{ metric.description }}</span>
}
</mat-option>
}
</mat-select>
</mat-form-field>
@if (selectedMetricNames.length === 0) {
<div class="metric-selection-helper">
Select at least one metric before running an evaluation.
</div>
} @else {
<div class="selected-metric-summary">
@for (metric of metricOptions; track metric.metricName) {
@if (metric.selected) {
<span class="selected-metric-chip">
{{ metric.metricName }} · threshold: {{ metric.threshold }}
</span>
}
}
</div>
}
</div>
<div class="evaluation-tab-header">
<button class="run-eval-btn" (click)="openEvalConfigDialog()">{{ i18n.runEvaluationButton }}</button>
<mat-icon class="evaluation-history-icon" (click)="toggleEvalHistoryButton()" [matTooltip]="i18n.viewEvalRunHistoryTooltip">history</mat-icon>
Expand Down Expand Up @@ -128,9 +158,15 @@
</div>
@if (getEvalMetrics(evalResult)) {
<div class="status-card__metrics">
@for (evalMetric of getEvalMetrics(evalResult); track evalMetric) {
<span class="status-card__metric"> {{ evalMetric.metricName }}:
{{ evalMetric.threshold }}
@for (evalMetric of getEvalMetrics(evalResult); track evalMetric.metricName) {
<span class="status-card__metric">
{{ evalMetric.metricName }} · Threshold: {{ evalMetric.threshold }}
@if (evalMetric.score !== undefined) {
· Score: {{ evalMetric.score }}
}
@if (evalMetric.evalStatus) {
· Status: {{ evalMetric.evalStatus }}
}
</span>
}
</div>
Expand All @@ -155,6 +191,34 @@
<div style="padding-top: 4px;">{{ evalResult.finalEvalStatus == 1 ? i18n.passStatusCaps : i18n.failStatusCaps}} </div>
</button>
</div>
@if (evalResult.overallEvalMetricResults?.length) {
<div class="status-card__metric-results">
@for (metricResult of evalResult.overallEvalMetricResults; track metricResult.metricName) {
<div
class="status-card__metric-result"
[ngClass]="{
'status-card__metric-result--pass': metricResult.evalStatus === 'PASSED' || metricResult.evalStatus === 1,
'status-card__metric-result--fail': metricResult.evalStatus === 'FAILED' || metricResult.evalStatus === 2,
'status-card__metric-result--neutral': metricResult.evalStatus !== 'PASSED' && metricResult.evalStatus !== 'FAILED' && metricResult.evalStatus !== 1 && metricResult.evalStatus !== 2
}"
>
<span class="status-card__metric-result-name">
{{ metricResult.metricName }}
</span>
@if (metricResult.score !== undefined) {
<span class="status-card__metric-result-score">
Score: {{ metricResult.score }}
</span>
}
@if (metricResult.threshold !== undefined) {
<span class="status-card__metric-result-threshold">
Threshold: {{ metricResult.threshold }}
</span>
}
</div>
}
</div>
}
</div>
}
</div>
Expand Down Expand Up @@ -184,4 +248,3 @@
</div>
}
</div>

72 changes: 72 additions & 0 deletions src/app/components/eval-tab/eval-tab.component.scss
Original file line number Diff line number Diff line change
Expand Up @@ -165,6 +165,45 @@
width: 100%;
}

.metric-selection {
display: flex;
flex-direction: column;
margin-top: 12px;
gap: 8px;
}

.metric-select-field {
width: 100%;
}

.metric-option-description {
color: var(--eval-tab-metric-option-description-color, #9aa0a6);
font-size: 12px;
margin-left: 4px;
}

.metric-selection-helper {
color: var(--eval-tab-metric-selection-helper-color, #9aa0a6);
font-size: 12px;
}

.selected-metric-summary {
display: flex;
flex-wrap: wrap;
gap: 6px;
font-size: 12px;
color: var(--eval-tab-selected-metric-summary-color, #9aa0a6);
}

.selected-metric-chip {
background-color: var(
--eval-tab-selected-metric-chip-background,
rgba(255, 255, 255, 0.08)
);
border-radius: 12px;
padding: 4px 8px;
}

.evaluation-history-icon {
cursor: pointer;
margin-top: 4px;
Expand Down Expand Up @@ -271,6 +310,39 @@
width: 100%;
margin-top: 15px;
}

&__metric-results {
display: flex;
flex-wrap: wrap;
gap: 6px;
margin-top: 8px;
}

&__metric-result {
display: inline-flex;
align-items: center;
gap: 6px;
border-radius: 12px;
padding: 4px 8px;
background-color: var(
--eval-tab-status-card-metric-result-background,
rgba(255, 255, 255, 0.08)
);
font-size: 12px;
color: var(--eval-tab-status-card-metric-result-color, #e8eaed);

&--pass {
color: var(--eval-tab-status-card-metric-result-pass-color, #1e8e3e);
}

&--fail {
color: var(--eval-tab-status-card-metric-result-fail-color, #d93025);
}

&--neutral {
color: var(--eval-tab-status-card-metric-result-neutral-color, #9aa0a6);
}
}
}

.eval-spinner {
Expand Down
Loading