Merge pull request #61 from unslothai/feature/training-frontend-integration
training frontend integration + backend sync v2
This commit is contained in:
Wasim Yousef Said
GitHub
commit
a7c6432ffd
43 changed files with 2150 additions and 470 deletions
|
|
@ -1,6 +1,8 @@
|
|||
"""
|
||||
Datasets API routes
|
||||
"""
|
||||
import base64
|
||||
import io
|
||||
import sys
|
||||
from pathlib import Path
|
||||
from fastapi import APIRouter, HTTPException
|
||||
|
|
@ -30,6 +32,42 @@ if not logger.handlers:
|
|||
from models.datasets import CheckFormatRequest, CheckFormatResponse
|
||||
|
||||
|
||||
def _serialize_preview_value(value):
|
||||
"""make it json safe for client preview ⊂(◉‿◉)つ"""
|
||||
if value is None or isinstance(value, (str, int, float, bool)):
|
||||
return value
|
||||
|
||||
try:
|
||||
from PIL.Image import Image as PILImage
|
||||
if isinstance(value, PILImage):
|
||||
buffer = io.BytesIO()
|
||||
value.convert("RGB").save(buffer, format="JPEG", quality=85)
|
||||
return {
|
||||
"type": "image",
|
||||
"mime": "image/jpeg",
|
||||
"width": value.width,
|
||||
"height": value.height,
|
||||
"data": base64.b64encode(buffer.getvalue()).decode("ascii"),
|
||||
}
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
if isinstance(value, dict):
|
||||
return {str(key): _serialize_preview_value(item) for key, item in value.items()}
|
||||
|
||||
if isinstance(value, (list, tuple)):
|
||||
return [_serialize_preview_value(item) for item in value]
|
||||
|
||||
return str(value)
|
||||
|
||||
|
||||
def _serialize_preview_rows(rows):
|
||||
return [
|
||||
{str(key): _serialize_preview_value(value) for key, value in dict(row).items()}
|
||||
for row in rows
|
||||
]
|
||||
|
||||
|
||||
# --- Endpoints ---
|
||||
|
||||
@router.post("/check-format", response_model=CheckFormatResponse)
|
||||
|
|
@ -92,15 +130,15 @@ async def check_format(request: CheckFormatRequest):
|
|||
custom_format_mapping=result.get("suggested_mapping"),
|
||||
)
|
||||
processed = format_result["dataset"]
|
||||
preview_samples = [dict(row) for row in processed]
|
||||
preview_samples = _serialize_preview_rows(processed)
|
||||
except Exception as e:
|
||||
logger.warning(f"Processed preview generation failed (non-fatal): {e}")
|
||||
# Fall back to raw samples so frontend still has something
|
||||
preview_samples = [dict(row) for row in preview_slice]
|
||||
preview_samples = _serialize_preview_rows(preview_slice)
|
||||
else:
|
||||
# Format detection failed — return raw samples so user can
|
||||
# see actual data and map columns in the frontend
|
||||
preview_samples = [dict(row) for row in preview_slice]
|
||||
preview_samples = _serialize_preview_rows(preview_slice)
|
||||
|
||||
return CheckFormatResponse(
|
||||
requires_manual_mapping=result["requires_manual_mapping"],
|
||||
|
|
|
|||
|
|
@ -180,14 +180,19 @@ async def start_training(
|
|||
except Exception as e:
|
||||
logger.error(f"Error updating progress: {e}")
|
||||
|
||||
# Consume the generator - this actually runs the training
|
||||
update_count = 0
|
||||
for _update_tuple in backend.start_training(**training_kwargs):
|
||||
update_count += 1
|
||||
if update_count % 10 == 0:
|
||||
logger.info(f"Training progress update #{update_count}")
|
||||
# start_training returns bool (not generator)
|
||||
run_result = backend.start_training(**training_kwargs)
|
||||
logger.info(
|
||||
"Training job %s backend.start_training returned type=%s value=%r",
|
||||
job_id,
|
||||
type(run_result).__name__,
|
||||
run_result,
|
||||
)
|
||||
if not run_result:
|
||||
progress_error = backend.trainer.training_progress.error
|
||||
raise RuntimeError(progress_error or "Training failed to start")
|
||||
|
||||
logger.info(f"Training job {job_id} completed successfully")
|
||||
logger.info(f"Training job {job_id} started successfully")
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"Training error in job {job_id}: {e}", exc_info=True)
|
||||
|
|
@ -653,4 +658,3 @@ async def stream_training_progress(
|
|||
"X-Accel-Buffering": "no",
|
||||
}
|
||||
)
|
||||
|
||||
|
|
|
|||
|
|
@ -17,6 +17,8 @@
|
|||
"@assistant-ui/react-markdown": "^0.12.1",
|
||||
"@assistant-ui/react-streamdown": "^0.1.0",
|
||||
"@base-ui/react": "^1.1.0",
|
||||
"@dagrejs/dagre": "^2.0.4",
|
||||
"@dagrejs/graphlib": "^3.0.4",
|
||||
"@fontsource-variable/figtree": "^5.2.10",
|
||||
"@fontsource-variable/inter": "^5.2.8",
|
||||
"@fontsource-variable/space-grotesk": "^5.2.10",
|
||||
|
|
|
|||
227
studio/frontend/src/components/ui/terminal.tsx
Normal file
227
studio/frontend/src/components/ui/terminal.tsx
Normal file
|
|
@ -0,0 +1,227 @@
|
|||
import { cn } from "@/lib/utils"
|
||||
import {
|
||||
Children,
|
||||
cloneElement,
|
||||
isValidElement,
|
||||
useEffect,
|
||||
useRef,
|
||||
useState,
|
||||
} from "react"
|
||||
import type { ElementType, ReactElement, ReactNode } from "react"
|
||||
|
||||
type TerminalProps = {
|
||||
children: ReactNode
|
||||
className?: string
|
||||
sequence?: boolean
|
||||
startOnView?: boolean
|
||||
}
|
||||
|
||||
type InternalLineProps = {
|
||||
__isActive?: boolean
|
||||
__onDone?: () => void
|
||||
__sequence?: boolean
|
||||
}
|
||||
|
||||
function useStartOnView(enabled: boolean): {
|
||||
ref: React.RefObject<HTMLDivElement | null>
|
||||
started: boolean
|
||||
} {
|
||||
const ref = useRef<HTMLDivElement | null>(null)
|
||||
const [isInView, setIsInView] = useState(false)
|
||||
const started = !enabled || isInView
|
||||
|
||||
useEffect(() => {
|
||||
if (!enabled) {
|
||||
return
|
||||
}
|
||||
|
||||
const node = ref.current
|
||||
if (!node) {
|
||||
return
|
||||
}
|
||||
|
||||
const observer = new IntersectionObserver(
|
||||
([entry]) => {
|
||||
if (entry?.isIntersecting) {
|
||||
setIsInView(true)
|
||||
observer.disconnect()
|
||||
}
|
||||
},
|
||||
{ threshold: 0.2 }
|
||||
)
|
||||
|
||||
observer.observe(node)
|
||||
return () => observer.disconnect()
|
||||
}, [enabled])
|
||||
|
||||
return { ref, started }
|
||||
}
|
||||
|
||||
export function Terminal({
|
||||
children,
|
||||
className,
|
||||
sequence = true,
|
||||
startOnView = true,
|
||||
}: TerminalProps): ReactElement {
|
||||
const { ref, started } = useStartOnView(startOnView)
|
||||
const childElements = Children.toArray(children).filter(isValidElement)
|
||||
const [activeIndex, setActiveIndex] = useState(0)
|
||||
const visibleIndex = sequence
|
||||
? started
|
||||
? activeIndex
|
||||
: -1
|
||||
: Number.MAX_SAFE_INTEGER
|
||||
|
||||
function handleLineDone(index: number): void {
|
||||
if (!sequence) {
|
||||
return
|
||||
}
|
||||
|
||||
setActiveIndex((prev) => {
|
||||
if (prev !== index) {
|
||||
return prev
|
||||
}
|
||||
return Math.min(index + 1, childElements.length)
|
||||
})
|
||||
}
|
||||
|
||||
return (
|
||||
<div
|
||||
ref={ref}
|
||||
className={cn(
|
||||
"w-full rounded-2xl border border-border bg-card px-6 py-5 font-mono text-sm text-foreground shadow-2xl",
|
||||
className
|
||||
)}
|
||||
>
|
||||
{childElements.map((child, index) =>
|
||||
cloneElement(child, {
|
||||
__sequence: sequence,
|
||||
__isActive: !sequence || visibleIndex >= index,
|
||||
__onDone: () => handleLineDone(index),
|
||||
key: child.key ?? index,
|
||||
} as InternalLineProps)
|
||||
)}
|
||||
</div>
|
||||
)
|
||||
}
|
||||
|
||||
type AnimatedSpanProps = InternalLineProps & {
|
||||
children: ReactNode
|
||||
className?: string
|
||||
delay?: number
|
||||
startOnView?: boolean
|
||||
}
|
||||
|
||||
export function AnimatedSpan({
|
||||
children,
|
||||
className,
|
||||
delay = 0,
|
||||
startOnView = false,
|
||||
__isActive,
|
||||
__sequence,
|
||||
__onDone,
|
||||
}: AnimatedSpanProps): ReactElement {
|
||||
const { ref, started } = useStartOnView(startOnView)
|
||||
const [visible, setVisible] = useState(false)
|
||||
const doneRef = useRef(false)
|
||||
const onDoneRef = useRef(__onDone)
|
||||
const shouldStart = __sequence ? __isActive : started
|
||||
|
||||
useEffect(() => {
|
||||
onDoneRef.current = __onDone
|
||||
}, [__onDone])
|
||||
|
||||
useEffect(() => {
|
||||
if (!shouldStart || doneRef.current) {
|
||||
return
|
||||
}
|
||||
|
||||
const timeout = window.setTimeout(() => {
|
||||
setVisible(true)
|
||||
doneRef.current = true
|
||||
onDoneRef.current?.()
|
||||
}, delay)
|
||||
|
||||
return () => window.clearTimeout(timeout)
|
||||
}, [delay, shouldStart])
|
||||
|
||||
return (
|
||||
<div
|
||||
ref={ref}
|
||||
className={cn(
|
||||
"min-h-5 transition-opacity duration-300",
|
||||
visible ? "opacity-100" : "opacity-0",
|
||||
className
|
||||
)}
|
||||
>
|
||||
{children}
|
||||
</div>
|
||||
)
|
||||
}
|
||||
|
||||
type TypingAnimationProps = InternalLineProps & {
|
||||
children: string
|
||||
className?: string
|
||||
duration?: number
|
||||
delay?: number
|
||||
as?: ElementType
|
||||
startOnView?: boolean
|
||||
}
|
||||
|
||||
export function TypingAnimation({
|
||||
children,
|
||||
className,
|
||||
duration = 60,
|
||||
delay = 0,
|
||||
as: Component = "span",
|
||||
startOnView = true,
|
||||
__isActive,
|
||||
__sequence,
|
||||
__onDone,
|
||||
}: TypingAnimationProps): ReactElement {
|
||||
const { ref, started } = useStartOnView(startOnView)
|
||||
const [typed, setTyped] = useState("")
|
||||
const doneRef = useRef(false)
|
||||
const onDoneRef = useRef(__onDone)
|
||||
const shouldStart = __sequence ? __isActive : started
|
||||
|
||||
useEffect(() => {
|
||||
onDoneRef.current = __onDone
|
||||
}, [__onDone])
|
||||
|
||||
useEffect(() => {
|
||||
if (!shouldStart || doneRef.current) {
|
||||
return
|
||||
}
|
||||
|
||||
let index = 0
|
||||
let intervalId: number | null = null
|
||||
const startTimer = window.setTimeout(() => {
|
||||
intervalId = window.setInterval(() => {
|
||||
index += 1
|
||||
setTyped(children.slice(0, index))
|
||||
|
||||
if (index >= children.length) {
|
||||
if (intervalId) {
|
||||
window.clearInterval(intervalId)
|
||||
}
|
||||
doneRef.current = true
|
||||
onDoneRef.current?.()
|
||||
}
|
||||
}, duration)
|
||||
}, delay)
|
||||
|
||||
return () => {
|
||||
window.clearTimeout(startTimer)
|
||||
if (intervalId) {
|
||||
window.clearInterval(intervalId)
|
||||
}
|
||||
}
|
||||
}, [children, delay, duration, shouldStart])
|
||||
|
||||
return (
|
||||
<div ref={ref} className="min-h-5">
|
||||
<Component className={cn("whitespace-pre-wrap", className)}>{typed}</Component>
|
||||
</div>
|
||||
)
|
||||
}
|
||||
|
|
@ -1,6 +1,6 @@
|
|||
export { LoginPage } from "./login-page";
|
||||
export { SignupPage } from "./signup-page";
|
||||
export { refreshSession } from "./api";
|
||||
export { authFetch, refreshSession } from "./api";
|
||||
export {
|
||||
getPostAuthRoute,
|
||||
hasAuthToken,
|
||||
|
|
|
|||
|
|
@ -13,7 +13,8 @@ import {
|
|||
TooltipContent,
|
||||
TooltipTrigger,
|
||||
} from "@/components/ui/tooltip";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingRuntimeStore } from "@/features/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import { isAdapterMethod } from "@/types/training";
|
||||
import { InformationCircleIcon, PackageIcon } from "@hugeicons/core-free-icons";
|
||||
import { HugeiconsIcon } from "@hugeicons/react";
|
||||
|
|
@ -36,36 +37,31 @@ export function ExportPage() {
|
|||
trainingMethod,
|
||||
selectedModel,
|
||||
saveSteps,
|
||||
trainingMetrics,
|
||||
epochs,
|
||||
loraRank,
|
||||
hfToken,
|
||||
setHfToken,
|
||||
} = useWizardStore(
|
||||
} = useTrainingConfigStore(
|
||||
useShallow((s) => ({
|
||||
trainingMethod: s.trainingMethod,
|
||||
selectedModel: s.selectedModel,
|
||||
saveSteps: s.saveSteps,
|
||||
trainingMetrics: s.trainingMetrics,
|
||||
epochs: s.epochs,
|
||||
loraRank: s.loraRank,
|
||||
hfToken: s.hfToken,
|
||||
setHfToken: s.setHfToken,
|
||||
})),
|
||||
);
|
||||
const totalSteps = useTrainingRuntimeStore((state) => state.totalSteps);
|
||||
const isAdapter = isAdapterMethod(trainingMethod);
|
||||
|
||||
const checkpoints = useMemo(() => {
|
||||
if (isAdapter) {
|
||||
const interval = saveSteps > 0 ? saveSteps : 100;
|
||||
const total = trainingMetrics?.totalSteps ?? 500;
|
||||
const total = totalSteps > 0 ? totalSteps : 500;
|
||||
const entries: { value: string; label: string; detail: string }[] = [];
|
||||
for (let step = interval; step <= total; step += interval) {
|
||||
const loss = (
|
||||
1.5 -
|
||||
(step / total) * 0.7 +
|
||||
Math.random() * 0.05
|
||||
).toFixed(2);
|
||||
const loss = (1.5 - (step / total) * 0.7).toFixed(2);
|
||||
entries.push({
|
||||
value: `checkpoint-${step}`,
|
||||
label: `checkpoint-${step}`,
|
||||
|
|
@ -81,7 +77,7 @@ export function ExportPage() {
|
|||
detail: "Full fine-tuned weights",
|
||||
},
|
||||
];
|
||||
}, [isAdapter, saveSteps, trainingMetrics?.totalSteps]);
|
||||
}, [isAdapter, saveSteps, totalSteps]);
|
||||
|
||||
const [checkpoint, setCheckpoint] = useState<string | null>(null);
|
||||
const [exportMethod, setExportMethod] = useState<ExportMethod | null>(null);
|
||||
|
|
|
|||
|
|
@ -38,7 +38,7 @@ import {
|
|||
useInfiniteScroll,
|
||||
} from "@/hooks";
|
||||
import { cn, formatCompact } from "@/lib/utils";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import type { DatasetFormat } from "@/types/training";
|
||||
import {
|
||||
InformationCircleIcon,
|
||||
|
|
@ -69,7 +69,7 @@ export function DatasetStep() {
|
|||
setDataset,
|
||||
uploadedFile,
|
||||
setUploadedFile,
|
||||
} = useWizardStore(
|
||||
} = useTrainingConfigStore(
|
||||
useShallow((s) => ({
|
||||
hfToken: s.hfToken,
|
||||
setHfToken: s.setHfToken,
|
||||
|
|
|
|||
|
|
@ -20,7 +20,7 @@ import {
|
|||
TooltipTrigger,
|
||||
} from "@/components/ui/tooltip";
|
||||
import { CONTEXT_LENGTHS } from "@/config/training";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import { InformationCircleIcon } from "@hugeicons/core-free-icons";
|
||||
import { HugeiconsIcon } from "@hugeicons/react";
|
||||
import { useShallow } from "zustand/react/shallow";
|
||||
|
|
@ -40,7 +40,7 @@ export function HyperparametersStep() {
|
|||
setLoraAlpha,
|
||||
loraDropout,
|
||||
setLoraDropout,
|
||||
} = useWizardStore(
|
||||
} = useTrainingConfigStore(
|
||||
useShallow((s) => ({
|
||||
trainingMethod: s.trainingMethod,
|
||||
epochs: s.epochs,
|
||||
|
|
|
|||
|
|
@ -37,7 +37,7 @@ import {
|
|||
useInfiniteScroll,
|
||||
} from "@/hooks";
|
||||
import { formatCompact } from "@/lib/utils";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import type { TrainingMethod } from "@/types/training";
|
||||
import {
|
||||
InformationCircleIcon,
|
||||
|
|
@ -57,7 +57,7 @@ export function ModelSelectionStep() {
|
|||
setTrainingMethod,
|
||||
hfToken,
|
||||
setHfToken,
|
||||
} = useWizardStore(
|
||||
} = useTrainingConfigStore(
|
||||
useShallow((s) => ({
|
||||
modelType: s.modelType,
|
||||
selectedModel: s.selectedModel,
|
||||
|
|
|
|||
|
|
@ -8,7 +8,7 @@ import {
|
|||
} from "@/components/ui/tooltip";
|
||||
import { MODEL_TYPES } from "@/config/training";
|
||||
import { cn } from "@/lib/utils";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import type { ModelType } from "@/types/training";
|
||||
import {
|
||||
Database02Icon,
|
||||
|
|
@ -38,7 +38,7 @@ const TYPE_TOOLTIPS: Record<ModelType, string> = {
|
|||
const COMING_SOON: ModelType[] = ["tts", "embeddings"];
|
||||
|
||||
export function ModelTypeStep(): ReactElement {
|
||||
const { modelType, setModelType } = useWizardStore(
|
||||
const { modelType, setModelType } = useTrainingConfigStore(
|
||||
useShallow((s) => ({
|
||||
modelType: s.modelType,
|
||||
setModelType: s.setModelType,
|
||||
|
|
|
|||
|
|
@ -1,7 +1,7 @@
|
|||
import { Badge } from "@/components/ui/badge";
|
||||
import { Card, CardContent, CardHeader, CardTitle } from "@/components/ui/card";
|
||||
import { Separator } from "@/components/ui/separator";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import { isAdapterMethod } from "@/types/training";
|
||||
import { GpuIcon } from "@hugeicons/core-free-icons";
|
||||
import { HugeiconsIcon } from "@hugeicons/react";
|
||||
|
|
@ -30,7 +30,7 @@ export function SummaryStep() {
|
|||
loraRank,
|
||||
loraAlpha,
|
||||
loraDropout,
|
||||
} = useWizardStore(
|
||||
} = useTrainingConfigStore(
|
||||
useShallow(
|
||||
({
|
||||
modelType,
|
||||
|
|
|
|||
|
|
@ -1,5 +1,5 @@
|
|||
import { STEPS } from "@/config/training";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import type { StepNumber } from "@/types/training";
|
||||
import { DatasetStep } from "./steps/dataset-step";
|
||||
import { HyperparametersStep } from "./steps/hyperparameters-step";
|
||||
|
|
@ -24,7 +24,7 @@ const STEP_MASCOTS: Record<StepNumber, string> = {
|
|||
};
|
||||
|
||||
export function WizardContent() {
|
||||
const currentStep = useWizardStore((s) => s.currentStep);
|
||||
const currentStep = useTrainingConfigStore((s) => s.currentStep);
|
||||
const stepConfig = STEPS[currentStep - 1];
|
||||
const StepComponent = STEP_COMPONENTS[currentStep];
|
||||
const mascotSrc = STEP_MASCOTS[currentStep];
|
||||
|
|
|
|||
|
|
@ -1,14 +1,14 @@
|
|||
import { Button } from "@/components/ui/button";
|
||||
import { STEPS } from "@/config/training";
|
||||
import { markOnboardingDone } from "@/features/auth";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import { ArrowLeft02Icon, ArrowRight02Icon } from "@hugeicons/core-free-icons";
|
||||
import { HugeiconsIcon } from "@hugeicons/react";
|
||||
import { useNavigate } from "@tanstack/react-router";
|
||||
import { useShallow } from "zustand/react/shallow";
|
||||
|
||||
export function WizardFooter() {
|
||||
const { currentStep, prevStep, nextStep, canProceed } = useWizardStore(
|
||||
const { currentStep, prevStep, nextStep, canProceed } = useTrainingConfigStore(
|
||||
useShallow((s) => ({
|
||||
currentStep: s.currentStep,
|
||||
prevStep: s.prevStep,
|
||||
|
|
|
|||
|
|
@ -6,7 +6,7 @@ import { Suspense, lazy, useEffect, useRef, useState } from "react";
|
|||
import type { ConfettiRef } from "@/components/ui/confetti";
|
||||
import { STEPS } from "@/config/training";
|
||||
import { isOnboardingDone, markOnboardingDone } from "@/features/auth";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import { SplashScreen } from "./splash-screen";
|
||||
import { WizardContent } from "./wizard-content";
|
||||
import { WizardFooter } from "./wizard-footer";
|
||||
|
|
@ -19,7 +19,7 @@ const Confetti = lazy(() =>
|
|||
export function WizardLayout() {
|
||||
const navigate = useNavigate();
|
||||
const [showSplash, setShowSplash] = useState(true);
|
||||
const currentStep = useWizardStore((s) => s.currentStep);
|
||||
const currentStep = useTrainingConfigStore((s) => s.currentStep);
|
||||
const confettiRef = useRef<ConfettiRef>(null);
|
||||
const hasFiredRef = useRef(false);
|
||||
const isFinalStep = currentStep === STEPS.length;
|
||||
|
|
|
|||
|
|
@ -1,10 +1,10 @@
|
|||
import { Progress } from "@/components/ui/progress";
|
||||
import { STEPS } from "@/config/training";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import { WizardStepItem } from "./wizard-step-item";
|
||||
|
||||
export function WizardSidebar() {
|
||||
const currentStep = useWizardStore((s) => s.currentStep);
|
||||
const currentStep = useTrainingConfigStore((s) => s.currentStep);
|
||||
const progress = ((currentStep - 1) / (STEPS.length - 1)) * 100;
|
||||
|
||||
return (
|
||||
|
|
|
|||
|
|
@ -1,5 +1,5 @@
|
|||
import { cn } from "@/lib/utils";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import type { StepConfig, StepNumber } from "@/types/training";
|
||||
import { useShallow } from "zustand/react/shallow";
|
||||
|
||||
|
|
@ -8,7 +8,7 @@ interface WizardStepItemProps {
|
|||
}
|
||||
|
||||
export function WizardStepItem({ step }: WizardStepItemProps) {
|
||||
const { currentStep, setStep } = useWizardStore(
|
||||
const { currentStep, setStep } = useTrainingConfigStore(
|
||||
useShallow((s) => ({ currentStep: s.currentStep, setStep: s.setStep })),
|
||||
);
|
||||
const isActive = currentStep === step.number;
|
||||
|
|
|
|||
|
|
@ -23,7 +23,6 @@ import {
|
|||
} from "@/components/ui/dropdown-menu";
|
||||
import { Label } from "@/components/ui/label";
|
||||
import { Slider } from "@/components/ui/slider";
|
||||
import type { TrainingMetrics } from "@/types/training";
|
||||
import { ChartAverageIcon, Settings02Icon } from "@hugeicons/core-free-icons";
|
||||
import { HugeiconsIcon } from "@hugeicons/react";
|
||||
import { type ReactElement, useMemo, useState } from "react";
|
||||
|
|
@ -44,9 +43,11 @@ const lossConfig = {
|
|||
const lrConfig = {
|
||||
lr: { label: "LR", color: "#8b5cf6" },
|
||||
} satisfies ChartConfig;
|
||||
|
||||
const gradNormConfig = {
|
||||
gradNorm: { label: "Grad Norm", color: "#f97316" },
|
||||
} satisfies ChartConfig;
|
||||
|
||||
const evalLossConfig = {
|
||||
loss: { label: "Eval Loss", color: "#ef4444" },
|
||||
} satisfies ChartConfig;
|
||||
|
|
@ -62,10 +63,81 @@ const placeholderEvalData = [
|
|||
type LossHistoryItem = { step: number; loss: number };
|
||||
type SmoothedLossItem = LossHistoryItem & { smoothed: number };
|
||||
|
||||
interface TrainingChartSeries {
|
||||
lossHistory: LossHistoryItem[];
|
||||
lrHistory: { step: number; lr: number }[];
|
||||
gradNormHistory: { step: number; gradNorm: number }[];
|
||||
}
|
||||
|
||||
const CHART_SYNC_ID = "train-metrics-sync";
|
||||
const MAX_RENDER_POINTS = 800;
|
||||
const DEFAULT_VISIBLE_POINTS = 160;
|
||||
|
||||
function formatStepTick(value: number): string {
|
||||
if (value >= 1_000_000) {
|
||||
return `${(value / 1_000_000).toFixed(1)}M`;
|
||||
}
|
||||
if (value >= 1_000) {
|
||||
return `${(value / 1_000).toFixed(1)}k`;
|
||||
}
|
||||
return String(Math.round(value));
|
||||
}
|
||||
|
||||
function compressSeries<T>(data: T[], maxPoints: number): T[] {
|
||||
if (data.length <= maxPoints) {
|
||||
return data;
|
||||
}
|
||||
|
||||
const stride = Math.ceil(data.length / maxPoints);
|
||||
return data.filter(
|
||||
(_item, index) => index % stride === 0 || index === data.length - 1,
|
||||
);
|
||||
}
|
||||
|
||||
function buildStepTicks(min: number, max: number, targetCount = 6): number[] {
|
||||
if (!Number.isFinite(min) || !Number.isFinite(max)) {
|
||||
return [0, 1];
|
||||
}
|
||||
if (max <= min) {
|
||||
return [min, max];
|
||||
}
|
||||
|
||||
const stepSize = Math.max(1, Math.ceil((max - min) / (targetCount - 1)));
|
||||
const ticks: number[] = [];
|
||||
let current = min;
|
||||
|
||||
while (current < max) {
|
||||
ticks.push(current);
|
||||
current += stepSize;
|
||||
}
|
||||
|
||||
ticks.push(max);
|
||||
return Array.from(new Set(ticks));
|
||||
}
|
||||
|
||||
function buildYDomain(values: number[]): [number, number] {
|
||||
if (values.length === 0) {
|
||||
return [0, 1];
|
||||
}
|
||||
|
||||
const min = Math.min(...values);
|
||||
const max = Math.max(...values);
|
||||
|
||||
if (min === max) {
|
||||
const base = Math.abs(min);
|
||||
const pad = base > 0 ? base * 0.08 : 0.1;
|
||||
return [min - pad, max + pad];
|
||||
}
|
||||
|
||||
const pad = (max - min) * 0.12;
|
||||
return [min - pad, max + pad];
|
||||
}
|
||||
|
||||
function ema(data: LossHistoryItem[], alpha: number): SmoothedLossItem[] {
|
||||
if (data.length === 0) {
|
||||
return [];
|
||||
}
|
||||
|
||||
let s = data[0].loss;
|
||||
return data.map((d) => {
|
||||
s = alpha * d.loss + (1 - alpha) * s;
|
||||
|
|
@ -75,18 +147,113 @@ function ema(data: LossHistoryItem[], alpha: number): SmoothedLossItem[] {
|
|||
|
||||
export function ChartsContent({
|
||||
metrics,
|
||||
}: { metrics: TrainingMetrics }): ReactElement {
|
||||
const [smoothing, setSmoothing] = useState(0.6);
|
||||
}: { metrics: TrainingChartSeries }): ReactElement {
|
||||
const [smoothing, setSmoothing] = useState(0.75);
|
||||
const [showRaw, setShowRaw] = useState(true);
|
||||
const [showSmoothed, setShowSmoothed] = useState(true);
|
||||
const [showAvgLine, setShowAvgLine] = useState(true);
|
||||
|
||||
const lossHistory = metrics.lossHistory;
|
||||
const smoothedData = useMemo(
|
||||
() => (lossHistory ? ema(lossHistory, 1 - smoothing) : []),
|
||||
() => (lossHistory.length > 0 ? ema(lossHistory, 1 - smoothing) : []),
|
||||
[lossHistory, smoothing],
|
||||
);
|
||||
|
||||
const reducedLossData = useMemo(
|
||||
() => compressSeries(smoothedData, MAX_RENDER_POINTS),
|
||||
[smoothedData],
|
||||
);
|
||||
|
||||
const reducedGradNormData = useMemo(
|
||||
() => compressSeries(metrics.gradNormHistory, MAX_RENDER_POINTS),
|
||||
[metrics.gradNormHistory],
|
||||
);
|
||||
|
||||
const reducedLrData = useMemo(
|
||||
() => compressSeries(metrics.lrHistory, MAX_RENDER_POINTS),
|
||||
[metrics.lrHistory],
|
||||
);
|
||||
|
||||
const visibleStepDomain = useMemo<[number, number]>(() => {
|
||||
const allSteps = [
|
||||
...reducedLossData.map((point) => point.step),
|
||||
...reducedGradNormData.map((point) => point.step),
|
||||
...reducedLrData.map((point) => point.step),
|
||||
].sort((a, b) => a - b);
|
||||
|
||||
if (allSteps.length === 0) {
|
||||
return [0, 1];
|
||||
}
|
||||
|
||||
const minStep = allSteps[0] ?? 0;
|
||||
const endStep = allSteps[allSteps.length - 1] ?? 1;
|
||||
const startIndex = Math.max(0, allSteps.length - DEFAULT_VISIBLE_POINTS);
|
||||
const startStep = allSteps[startIndex] ?? minStep;
|
||||
if (startStep === endStep) {
|
||||
return [startStep, startStep + 4];
|
||||
}
|
||||
if (endStep - startStep < 6) {
|
||||
return [Math.max(minStep, endStep - 6), endStep];
|
||||
}
|
||||
return [startStep, endStep];
|
||||
}, [reducedGradNormData, reducedLossData, reducedLrData]);
|
||||
|
||||
const xAxisTicks = useMemo(
|
||||
() => buildStepTicks(visibleStepDomain[0], visibleStepDomain[1]),
|
||||
[visibleStepDomain],
|
||||
);
|
||||
|
||||
const visibleLossValues = useMemo(
|
||||
() =>
|
||||
reducedLossData
|
||||
.filter(
|
||||
(point) =>
|
||||
point.step >= visibleStepDomain[0] && point.step <= visibleStepDomain[1],
|
||||
)
|
||||
.map((point) => point.loss),
|
||||
[reducedLossData, visibleStepDomain],
|
||||
);
|
||||
|
||||
const visibleSmoothValues = useMemo(
|
||||
() =>
|
||||
reducedLossData
|
||||
.filter(
|
||||
(point) =>
|
||||
point.step >= visibleStepDomain[0] && point.step <= visibleStepDomain[1],
|
||||
)
|
||||
.map((point) => point.smoothed),
|
||||
[reducedLossData, visibleStepDomain],
|
||||
);
|
||||
|
||||
const visibleGradValues = useMemo(
|
||||
() =>
|
||||
reducedGradNormData
|
||||
.filter(
|
||||
(point) =>
|
||||
point.step >= visibleStepDomain[0] && point.step <= visibleStepDomain[1],
|
||||
)
|
||||
.map((point) => point.gradNorm),
|
||||
[reducedGradNormData, visibleStepDomain],
|
||||
);
|
||||
|
||||
const visibleLrValues = useMemo(
|
||||
() =>
|
||||
reducedLrData
|
||||
.filter(
|
||||
(point) =>
|
||||
point.step >= visibleStepDomain[0] && point.step <= visibleStepDomain[1],
|
||||
)
|
||||
.map((point) => point.lr),
|
||||
[reducedLrData, visibleStepDomain],
|
||||
);
|
||||
|
||||
const lossDomain = useMemo(
|
||||
() => buildYDomain([...visibleLossValues, ...visibleSmoothValues]),
|
||||
[visibleLossValues, visibleSmoothValues],
|
||||
);
|
||||
const gradDomain = useMemo(() => buildYDomain(visibleGradValues), [visibleGradValues]);
|
||||
const lrDomain = useMemo(() => buildYDomain(visibleLrValues), [visibleLrValues]);
|
||||
|
||||
const avg =
|
||||
metrics.lossHistory.length > 0
|
||||
? +(
|
||||
|
|
@ -96,8 +263,7 @@ export function ChartsContent({
|
|||
: 0;
|
||||
|
||||
return (
|
||||
<div className="grid grid-cols-1 lg:grid-cols-2 gap-6">
|
||||
{/* Training Loss */}
|
||||
<div className="grid grid-cols-1 gap-6 lg:grid-cols-2">
|
||||
<Card size="sm">
|
||||
<CardHeader>
|
||||
<CardTitle className="text-sm">Training Loss</CardTitle>
|
||||
|
|
@ -106,7 +272,7 @@ export function ChartsContent({
|
|||
<DropdownMenuTrigger asChild={true}>
|
||||
<button
|
||||
type="button"
|
||||
className="rounded-md p-1 text-muted-foreground hover:bg-muted hover:text-foreground transition-colors cursor-pointer"
|
||||
className="cursor-pointer rounded-md p-1 text-muted-foreground transition-colors hover:bg-muted hover:text-foreground"
|
||||
>
|
||||
<HugeiconsIcon icon={Settings02Icon} className="size-3.5" />
|
||||
</button>
|
||||
|
|
@ -155,12 +321,10 @@ export function ChartsContent({
|
|||
</CardAction>
|
||||
</CardHeader>
|
||||
<CardContent>
|
||||
<ChartContainer
|
||||
config={lossConfig}
|
||||
className="h-[200px] w-full -ml-3"
|
||||
>
|
||||
<ChartContainer config={lossConfig} className="-ml-3 h-[220px] w-full">
|
||||
<LineChart
|
||||
data={smoothedData}
|
||||
data={reducedLossData}
|
||||
syncId={CHART_SYNC_ID}
|
||||
accessibilityLayer={true}
|
||||
margin={{ left: 0, right: 8 }}
|
||||
>
|
||||
|
|
@ -168,19 +332,27 @@ export function ChartsContent({
|
|||
<XAxis
|
||||
dataKey="step"
|
||||
type="number"
|
||||
domain={["dataMin", "dataMax"]}
|
||||
domain={visibleStepDomain}
|
||||
ticks={xAxisTicks}
|
||||
allowDataOverflow={true}
|
||||
allowDecimals={false}
|
||||
minTickGap={28}
|
||||
tickLine={false}
|
||||
axisLine={false}
|
||||
tickMargin={8}
|
||||
fontSize={10}
|
||||
tickFormatter={(value) => formatStepTick(Number(value))}
|
||||
interval="preserveStartEnd"
|
||||
/>
|
||||
<YAxis
|
||||
domain={lossDomain}
|
||||
allowDataOverflow={true}
|
||||
tickLine={false}
|
||||
axisLine={false}
|
||||
tickMargin={4}
|
||||
fontSize={10}
|
||||
width={40}
|
||||
width={52}
|
||||
tickFormatter={(value) => Number(value).toFixed(2)}
|
||||
/>
|
||||
<ChartTooltip
|
||||
content={
|
||||
|
|
@ -207,22 +379,30 @@ export function ChartsContent({
|
|||
)}
|
||||
{showRaw && (
|
||||
<Line
|
||||
type="monotone"
|
||||
type="monotoneX"
|
||||
dataKey="loss"
|
||||
stroke="var(--color-loss)"
|
||||
strokeWidth={1.5}
|
||||
strokeOpacity={showSmoothed ? 0.3 : 1}
|
||||
strokeWidth={1.2}
|
||||
strokeOpacity={showSmoothed ? 0.35 : 1}
|
||||
dot={false}
|
||||
activeDot={{ r: 3, strokeWidth: 0 }}
|
||||
connectNulls={true}
|
||||
strokeLinecap="round"
|
||||
strokeLinejoin="round"
|
||||
isAnimationActive={false}
|
||||
/>
|
||||
)}
|
||||
{showSmoothed && (
|
||||
<Line
|
||||
type="monotone"
|
||||
type="monotoneX"
|
||||
dataKey="smoothed"
|
||||
stroke="var(--color-smoothed)"
|
||||
strokeWidth={2}
|
||||
strokeWidth={2.2}
|
||||
dot={false}
|
||||
activeDot={{ r: 3, strokeWidth: 0 }}
|
||||
connectNulls={true}
|
||||
strokeLinecap="round"
|
||||
strokeLinejoin="round"
|
||||
isAnimationActive={false}
|
||||
/>
|
||||
)}
|
||||
|
|
@ -232,7 +412,6 @@ export function ChartsContent({
|
|||
</CardContent>
|
||||
</Card>
|
||||
|
||||
{/* Grad Norm */}
|
||||
<Card size="sm">
|
||||
<CardHeader>
|
||||
<CardTitle className="text-sm">Gradient Norm</CardTitle>
|
||||
|
|
@ -240,10 +419,11 @@ export function ChartsContent({
|
|||
<CardContent>
|
||||
<ChartContainer
|
||||
config={gradNormConfig}
|
||||
className="h-[200px] w-full -ml-3"
|
||||
className="-ml-3 h-[220px] w-full"
|
||||
>
|
||||
<LineChart
|
||||
data={metrics.gradNormHistory}
|
||||
data={reducedGradNormData}
|
||||
syncId={CHART_SYNC_ID}
|
||||
accessibilityLayer={true}
|
||||
margin={{ left: 0, right: 8 }}
|
||||
>
|
||||
|
|
@ -251,19 +431,27 @@ export function ChartsContent({
|
|||
<XAxis
|
||||
dataKey="step"
|
||||
type="number"
|
||||
domain={["dataMin", "dataMax"]}
|
||||
domain={visibleStepDomain}
|
||||
ticks={xAxisTicks}
|
||||
allowDataOverflow={true}
|
||||
allowDecimals={false}
|
||||
minTickGap={28}
|
||||
tickLine={false}
|
||||
axisLine={false}
|
||||
tickMargin={8}
|
||||
fontSize={10}
|
||||
tickFormatter={(value) => formatStepTick(Number(value))}
|
||||
interval="preserveStartEnd"
|
||||
/>
|
||||
<YAxis
|
||||
domain={gradDomain}
|
||||
allowDataOverflow={true}
|
||||
tickLine={false}
|
||||
axisLine={false}
|
||||
tickMargin={4}
|
||||
fontSize={10}
|
||||
width={40}
|
||||
width={52}
|
||||
tickFormatter={(value) => Number(value).toFixed(2)}
|
||||
/>
|
||||
<ChartTooltip
|
||||
content={
|
||||
|
|
@ -275,11 +463,15 @@ export function ChartsContent({
|
|||
}
|
||||
/>
|
||||
<Line
|
||||
type="monotone"
|
||||
type="monotoneX"
|
||||
dataKey="gradNorm"
|
||||
stroke="var(--color-gradNorm)"
|
||||
strokeWidth={2}
|
||||
dot={false}
|
||||
activeDot={{ r: 3, strokeWidth: 0 }}
|
||||
connectNulls={true}
|
||||
strokeLinecap="round"
|
||||
strokeLinejoin="round"
|
||||
isAnimationActive={false}
|
||||
/>
|
||||
<ChartLegend content={<ChartLegendContent />} />
|
||||
|
|
@ -288,18 +480,15 @@ export function ChartsContent({
|
|||
</CardContent>
|
||||
</Card>
|
||||
|
||||
{/* Learning Rate */}
|
||||
<Card size="sm">
|
||||
<CardHeader>
|
||||
<CardTitle className="text-sm">Learning Rate</CardTitle>
|
||||
</CardHeader>
|
||||
<CardContent>
|
||||
<ChartContainer
|
||||
config={lrConfig}
|
||||
className="h-[200px] w-full -ml-1.5"
|
||||
>
|
||||
<ChartContainer config={lrConfig} className="-ml-1.5 h-[220px] w-full">
|
||||
<LineChart
|
||||
data={metrics.lrHistory}
|
||||
data={reducedLrData}
|
||||
syncId={CHART_SYNC_ID}
|
||||
accessibilityLayer={true}
|
||||
margin={{ left: 0, right: 8 }}
|
||||
>
|
||||
|
|
@ -307,20 +496,27 @@ export function ChartsContent({
|
|||
<XAxis
|
||||
dataKey="step"
|
||||
type="number"
|
||||
domain={["dataMin", "dataMax"]}
|
||||
domain={visibleStepDomain}
|
||||
ticks={xAxisTicks}
|
||||
allowDataOverflow={true}
|
||||
allowDecimals={false}
|
||||
minTickGap={28}
|
||||
tickLine={false}
|
||||
axisLine={false}
|
||||
tickMargin={8}
|
||||
fontSize={10}
|
||||
tickFormatter={(value) => formatStepTick(Number(value))}
|
||||
interval="preserveStartEnd"
|
||||
/>
|
||||
<YAxis
|
||||
domain={lrDomain}
|
||||
allowDataOverflow={true}
|
||||
tickLine={false}
|
||||
axisLine={false}
|
||||
tickMargin={4}
|
||||
fontSize={10}
|
||||
width={40}
|
||||
tickFormatter={(v) => v.toExponential(0)}
|
||||
width={52}
|
||||
tickFormatter={(value) => Number(value).toExponential(0)}
|
||||
/>
|
||||
<ChartTooltip
|
||||
content={
|
||||
|
|
@ -328,19 +524,20 @@ export function ChartsContent({
|
|||
labelFormatter={(_value, payload) =>
|
||||
`Step ${payload?.[0]?.payload?.step ?? ""}`
|
||||
}
|
||||
formatter={(value) => [
|
||||
Number(value).toExponential(3),
|
||||
"LR",
|
||||
]}
|
||||
formatter={(value) => [Number(value).toExponential(3), "LR"]}
|
||||
/>
|
||||
}
|
||||
/>
|
||||
<Line
|
||||
type="monotone"
|
||||
type="monotoneX"
|
||||
dataKey="lr"
|
||||
stroke="var(--color-lr)"
|
||||
strokeWidth={2}
|
||||
dot={false}
|
||||
activeDot={{ r: 3, strokeWidth: 0 }}
|
||||
connectNulls={true}
|
||||
strokeLinecap="round"
|
||||
strokeLinejoin="round"
|
||||
isAnimationActive={false}
|
||||
/>
|
||||
<ChartLegend content={<ChartLegendContent />} />
|
||||
|
|
@ -349,7 +546,6 @@ export function ChartsContent({
|
|||
</CardContent>
|
||||
</Card>
|
||||
|
||||
{/* Eval Loss (disabled/blurred) */}
|
||||
<Card size="sm">
|
||||
<CardHeader>
|
||||
<CardTitle className="text-sm text-muted-foreground">
|
||||
|
|
@ -360,7 +556,7 @@ export function ChartsContent({
|
|||
<div className="relative">
|
||||
<ChartContainer
|
||||
config={evalLossConfig}
|
||||
className="h-[200px] w-full -ml-3 blur"
|
||||
className="-ml-3 h-[220px] w-full blur"
|
||||
>
|
||||
<LineChart
|
||||
data={placeholderEvalData}
|
||||
|
|
|
|||
|
|
@ -1,5 +1,5 @@
|
|||
import { useWizardStore } from "@/stores/training";
|
||||
import { type ReactElement, Suspense, lazy } from "react";
|
||||
import { useTrainingRuntimeStore } from "@/features/training";
|
||||
import { type ReactElement, Suspense, lazy, useMemo } from "react";
|
||||
|
||||
const ChartsContent = lazy(() =>
|
||||
import("./charts-content").then((module) => ({
|
||||
|
|
@ -14,9 +14,39 @@ const SKELETON_KEYS = [
|
|||
];
|
||||
|
||||
export function ChartsSection(): ReactElement | null {
|
||||
const metrics = useWizardStore((s) => s.trainingMetrics);
|
||||
const currentStep = useTrainingRuntimeStore((state) => state.currentStep);
|
||||
const totalSteps = useTrainingRuntimeStore((state) => state.totalSteps);
|
||||
const lossHistoryRaw = useTrainingRuntimeStore((state) => state.lossHistory);
|
||||
const lrHistoryRaw = useTrainingRuntimeStore((state) => state.lrHistory);
|
||||
const gradNormHistoryRaw = useTrainingRuntimeStore(
|
||||
(state) => state.gradNormHistory,
|
||||
);
|
||||
|
||||
if (!metrics) {
|
||||
const series = useMemo(
|
||||
() => ({
|
||||
currentStep,
|
||||
totalSteps,
|
||||
lossHistory: lossHistoryRaw.map((point) => ({
|
||||
step: point.step,
|
||||
loss: point.value,
|
||||
})),
|
||||
lrHistory: lrHistoryRaw.map((point) => ({
|
||||
step: point.step,
|
||||
lr: point.value,
|
||||
})),
|
||||
gradNormHistory: gradNormHistoryRaw.map((point) => ({
|
||||
step: point.step,
|
||||
gradNorm: point.value,
|
||||
})),
|
||||
}),
|
||||
[currentStep, gradNormHistoryRaw, lossHistoryRaw, lrHistoryRaw, totalSteps],
|
||||
);
|
||||
|
||||
if (
|
||||
series.lossHistory.length === 0 &&
|
||||
series.lrHistory.length === 0 &&
|
||||
series.gradNormHistory.length === 0
|
||||
) {
|
||||
return null;
|
||||
}
|
||||
|
||||
|
|
@ -33,7 +63,7 @@ export function ChartsSection(): ReactElement | null {
|
|||
</div>
|
||||
}
|
||||
>
|
||||
<ChartsContent metrics={metrics} />
|
||||
<ChartsContent metrics={series} />
|
||||
</Suspense>
|
||||
);
|
||||
}
|
||||
|
|
|
|||
|
|
@ -27,6 +27,14 @@ type CheckFormatResponse = {
|
|||
total_rows?: number | null;
|
||||
};
|
||||
|
||||
type PreviewImagePayload = {
|
||||
type: "image";
|
||||
mime?: string;
|
||||
width?: number;
|
||||
height?: number;
|
||||
data?: string;
|
||||
};
|
||||
|
||||
type DatasetPreviewDialogProps = {
|
||||
open: boolean;
|
||||
onOpenChange: (open: boolean) => void;
|
||||
|
|
@ -123,6 +131,36 @@ export function DatasetPreviewDialog({
|
|||
),
|
||||
cell: ({ getValue }: { getValue: () => unknown }) => {
|
||||
const value = getValue();
|
||||
const images = collectPreviewImages(value);
|
||||
if (images.length > 0) {
|
||||
return (
|
||||
<div className="flex flex-wrap gap-2">
|
||||
{images.slice(0, 4).map((image, index) => {
|
||||
const mime = image.mime || "image/jpeg";
|
||||
const src = image.data ? `data:${mime};base64,${image.data}` : "";
|
||||
const width = image.width ?? 128;
|
||||
const height = image.height ?? 128;
|
||||
return (
|
||||
<img
|
||||
key={`${colName}-img-${index}`}
|
||||
src={src}
|
||||
alt={`preview-${index}`}
|
||||
className="h-16 w-auto max-w-40 rounded-md border object-contain bg-muted"
|
||||
width={width}
|
||||
height={height}
|
||||
loading="lazy"
|
||||
/>
|
||||
);
|
||||
})}
|
||||
{images.length > 4 && (
|
||||
<span className="text-xs text-muted-foreground self-end">
|
||||
+{images.length - 4} more
|
||||
</span>
|
||||
)}
|
||||
</div>
|
||||
);
|
||||
}
|
||||
|
||||
const text = formatCell(value);
|
||||
if (!text) {
|
||||
return (
|
||||
|
|
@ -285,3 +323,41 @@ function formatCell(value: unknown): string {
|
|||
return JSON.stringify(value).slice(0, 500);
|
||||
return String(value);
|
||||
}
|
||||
|
||||
function isPreviewImagePayload(value: unknown): value is PreviewImagePayload {
|
||||
if (!value || typeof value !== "object") return false;
|
||||
const record = value as Record<string, unknown>;
|
||||
return (
|
||||
record.type === "image" &&
|
||||
typeof record.data === "string" &&
|
||||
record.data.length > 0
|
||||
);
|
||||
}
|
||||
|
||||
function collectPreviewImages(value: unknown): PreviewImagePayload[] {
|
||||
const images: PreviewImagePayload[] = [];
|
||||
const stack: unknown[] = [value];
|
||||
let steps = 0;
|
||||
|
||||
while (stack.length > 0 && steps < 200) {
|
||||
steps += 1;
|
||||
const current = stack.pop();
|
||||
if (isPreviewImagePayload(current)) {
|
||||
images.push(current);
|
||||
continue;
|
||||
}
|
||||
|
||||
if (Array.isArray(current)) {
|
||||
for (const item of current) stack.push(item);
|
||||
continue;
|
||||
}
|
||||
|
||||
if (current && typeof current === "object") {
|
||||
for (const nested of Object.values(current as Record<string, unknown>)) {
|
||||
stack.push(nested);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return images;
|
||||
}
|
||||
|
|
|
|||
|
|
@ -28,7 +28,7 @@ import {
|
|||
useInfiniteScroll,
|
||||
} from "@/hooks";
|
||||
import { formatCompact } from "@/lib/utils";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import {
|
||||
CloudUploadIcon,
|
||||
Database02Icon,
|
||||
|
|
@ -44,7 +44,7 @@ import { DatasetPreviewDialog } from "./dataset-preview-dialog";
|
|||
|
||||
export function DatasetSection() {
|
||||
const { dataset, setDataset, datasetFormat, setDatasetFormat, hfToken } =
|
||||
useWizardStore(
|
||||
useTrainingConfigStore(
|
||||
useShallow(
|
||||
({
|
||||
dataset,
|
||||
|
|
|
|||
|
|
@ -32,7 +32,7 @@ import {
|
|||
useInfiniteScroll,
|
||||
} from "@/hooks";
|
||||
import { formatCompact } from "@/lib/utils";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import type { TrainingMethod } from "@/types/training";
|
||||
import {
|
||||
ChipIcon,
|
||||
|
|
@ -65,7 +65,7 @@ export function ModelSection() {
|
|||
setTrainingMethod,
|
||||
hfToken,
|
||||
setHfToken,
|
||||
} = useWizardStore(
|
||||
} = useTrainingConfigStore(
|
||||
useShallow(
|
||||
({
|
||||
modelType,
|
||||
|
|
|
|||
|
|
@ -21,7 +21,7 @@ import {
|
|||
TooltipTrigger,
|
||||
} from "@/components/ui/tooltip";
|
||||
import { CONTEXT_LENGTHS, TARGET_MODULES } from "@/config/training";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import type { GradientCheckpointing } from "@/types/training";
|
||||
import {
|
||||
ArrowDown01Icon,
|
||||
|
|
@ -107,7 +107,7 @@ function SliderRow({
|
|||
}
|
||||
|
||||
export function ParamsSection(): ReactElement {
|
||||
const store = useWizardStore();
|
||||
const store = useTrainingConfigStore();
|
||||
const isLora = store.trainingMethod !== "full";
|
||||
const isVision = store.modelType === "vision";
|
||||
const [loraOpen, setLoraOpen] = useState(false);
|
||||
|
|
|
|||
|
|
@ -5,7 +5,12 @@ import {
|
|||
PopoverContent,
|
||||
PopoverTrigger,
|
||||
} from "@/components/ui/popover";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import {
|
||||
useTrainingConfigStore,
|
||||
useTrainingActions,
|
||||
useTrainingRuntimeStore,
|
||||
type TrainingPhase,
|
||||
} from "@/features/training";
|
||||
import {
|
||||
ChartAverageIcon,
|
||||
DashboardSpeed01Icon,
|
||||
|
|
@ -16,66 +21,160 @@ import {
|
|||
ZapIcon,
|
||||
} from "@hugeicons/core-free-icons";
|
||||
import { HugeiconsIcon } from "@hugeicons/react";
|
||||
import type { ReactElement, ReactNode } from "react";
|
||||
import { useEffect, useRef, useState, type ReactElement, type ReactNode } from "react";
|
||||
import { useShallow } from "zustand/react/shallow";
|
||||
|
||||
export function ProgressSection(): ReactElement | null {
|
||||
const store = useWizardStore();
|
||||
const metrics = store.trainingMetrics;
|
||||
if (!metrics) {
|
||||
return null;
|
||||
const phaseLabel: Record<TrainingPhase, string> = {
|
||||
idle: "Idle",
|
||||
loading_model: "Loading model",
|
||||
loading_dataset: "Loading dataset",
|
||||
configuring: "Configuring",
|
||||
training: "Training",
|
||||
completed: "Completed",
|
||||
error: "Error",
|
||||
stopped: "Stopped",
|
||||
};
|
||||
|
||||
const phaseColors: Record<TrainingPhase, string> = {
|
||||
idle: "bg-muted text-muted-foreground",
|
||||
loading_model: "bg-amber-100 text-amber-700 dark:bg-amber-900 dark:text-amber-300",
|
||||
loading_dataset:
|
||||
"bg-amber-100 text-amber-700 dark:bg-amber-900 dark:text-amber-300",
|
||||
configuring: "bg-blue-100 text-blue-700 dark:bg-blue-900 dark:text-blue-300",
|
||||
training:
|
||||
"bg-emerald-100 text-emerald-700 dark:bg-emerald-900 dark:text-emerald-300",
|
||||
completed:
|
||||
"bg-emerald-100 text-emerald-700 dark:bg-emerald-900 dark:text-emerald-300",
|
||||
error: "bg-red-100 text-red-700 dark:bg-red-900 dark:text-red-300",
|
||||
stopped: "bg-muted text-muted-foreground",
|
||||
};
|
||||
|
||||
function formatDuration(seconds: number | null): string {
|
||||
if (seconds == null || seconds < 0) {
|
||||
return "--";
|
||||
}
|
||||
const total = Math.floor(seconds);
|
||||
const min = Math.floor(total / 60);
|
||||
const sec = total % 60;
|
||||
return `${min}m ${sec}s`;
|
||||
}
|
||||
|
||||
const pct = Math.round((metrics.currentStep / metrics.totalSteps) * 100);
|
||||
const etaSec =
|
||||
metrics.totalSteps > 0
|
||||
? Math.round(
|
||||
((metrics.totalSteps - metrics.currentStep) /
|
||||
Math.max(metrics.currentStep, 1)) *
|
||||
metrics.elapsed,
|
||||
function formatNumber(value: number | null | undefined, digits: number): string {
|
||||
if (value == null || !Number.isFinite(value)) {
|
||||
return "--";
|
||||
}
|
||||
return value.toFixed(digits);
|
||||
}
|
||||
|
||||
export function ProgressSection(): ReactElement {
|
||||
const runtime = useTrainingRuntimeStore(
|
||||
useShallow((state) => ({
|
||||
phase: state.phase,
|
||||
message: state.message,
|
||||
error: state.error,
|
||||
currentStep: state.currentStep,
|
||||
totalSteps: state.totalSteps,
|
||||
currentEpoch: state.currentEpoch,
|
||||
currentLoss: state.currentLoss,
|
||||
currentLearningRate: state.currentLearningRate,
|
||||
currentGradNorm: state.currentGradNorm,
|
||||
progressPercent: state.progressPercent,
|
||||
elapsedSeconds: state.elapsedSeconds,
|
||||
etaSeconds: state.etaSeconds,
|
||||
currentNumTokens: state.currentNumTokens,
|
||||
isTrainingRunning: state.isTrainingRunning,
|
||||
})),
|
||||
);
|
||||
|
||||
const config = useTrainingConfigStore(
|
||||
useShallow((state) => ({
|
||||
selectedModel: state.selectedModel,
|
||||
trainingMethod: state.trainingMethod,
|
||||
epochs: state.epochs,
|
||||
batchSize: state.batchSize,
|
||||
learningRate: state.learningRate,
|
||||
maxSteps: state.maxSteps,
|
||||
contextLength: state.contextLength,
|
||||
warmupSteps: state.warmupSteps,
|
||||
loraRank: state.loraRank,
|
||||
loraAlpha: state.loraAlpha,
|
||||
loraDropout: state.loraDropout,
|
||||
loraVariant: state.loraVariant,
|
||||
})),
|
||||
);
|
||||
|
||||
const { stopTrainingRun } = useTrainingActions();
|
||||
const localStartAtRef = useRef<number | null>(null);
|
||||
const [, setLocalTick] = useState(0);
|
||||
|
||||
const pct =
|
||||
runtime.totalSteps > 0
|
||||
? Math.min(
|
||||
100,
|
||||
Math.max(
|
||||
0,
|
||||
Math.round((runtime.currentStep / runtime.totalSteps) * 100),
|
||||
),
|
||||
)
|
||||
: 0;
|
||||
const fmtTime = (s: number) => {
|
||||
const m = Math.floor(s / 60);
|
||||
const sec = s % 60;
|
||||
return `${m}m ${sec}s`;
|
||||
};
|
||||
: Math.round(runtime.progressPercent);
|
||||
|
||||
const statusColors = {
|
||||
training:
|
||||
"bg-emerald-100 text-emerald-700 dark:bg-emerald-900 dark:text-emerald-300",
|
||||
warmup: "bg-amber-100 text-amber-700 dark:bg-amber-900 dark:text-amber-300",
|
||||
saving:
|
||||
"bg-emerald-100 text-emerald-700 dark:bg-emerald-900 dark:text-emerald-300",
|
||||
};
|
||||
const statusLabels = {
|
||||
training: "Training",
|
||||
warmup: "Warming up",
|
||||
saving: "Saving checkpoint",
|
||||
};
|
||||
useEffect(() => {
|
||||
if (runtime.elapsedSeconds != null && runtime.elapsedSeconds >= 0) {
|
||||
localStartAtRef.current = Date.now() - runtime.elapsedSeconds * 1000;
|
||||
return;
|
||||
}
|
||||
if (runtime.currentStep > 0 && localStartAtRef.current == null) {
|
||||
localStartAtRef.current = Date.now();
|
||||
}
|
||||
}, [runtime.currentStep, runtime.elapsedSeconds]);
|
||||
|
||||
const modelName = store.selectedModel ?? "—";
|
||||
useEffect(() => {
|
||||
if (!runtime.isTrainingRunning) {
|
||||
return;
|
||||
}
|
||||
const timer = window.setInterval(() => {
|
||||
setLocalTick((prev) => prev + 1);
|
||||
}, 1000);
|
||||
return () => window.clearInterval(timer);
|
||||
}, [runtime.isTrainingRunning]);
|
||||
|
||||
const elapsed =
|
||||
runtime.elapsedSeconds ??
|
||||
(localStartAtRef.current == null
|
||||
? null
|
||||
: Math.max(0, Math.floor((Date.now() - localStartAtRef.current) / 1000)));
|
||||
const derivedEta =
|
||||
elapsed != null && pct > 0
|
||||
? Math.round((elapsed * (100 - pct)) / Math.max(pct, 1))
|
||||
: null;
|
||||
const eta = runtime.etaSeconds ?? derivedEta;
|
||||
|
||||
const stepsPerSecond =
|
||||
elapsed != null && elapsed > 0
|
||||
? runtime.currentStep / elapsed
|
||||
: null;
|
||||
|
||||
const configItems = [
|
||||
{
|
||||
section: "Hyperparams",
|
||||
rows: [
|
||||
["Epochs", store.epochs],
|
||||
["Batch size", store.batchSize],
|
||||
["Learning rate", store.learningRate],
|
||||
["Max steps", store.maxSteps],
|
||||
["Context length", store.contextLength],
|
||||
["Warmup steps", store.warmupSteps],
|
||||
["Epochs", config.epochs],
|
||||
["Batch size", config.batchSize],
|
||||
["Learning rate", config.learningRate],
|
||||
["Max steps", config.maxSteps],
|
||||
["Context length", config.contextLength],
|
||||
["Warmup steps", config.warmupSteps],
|
||||
],
|
||||
},
|
||||
...(store.trainingMethod !== "full"
|
||||
...(config.trainingMethod !== "full"
|
||||
? [
|
||||
{
|
||||
section: "LoRA",
|
||||
rows: [
|
||||
["Rank", store.loraRank],
|
||||
["Alpha", store.loraAlpha],
|
||||
["Dropout", store.loraDropout],
|
||||
["Variant", store.loraVariant],
|
||||
["Rank", config.loraRank],
|
||||
["Alpha", config.loraAlpha],
|
||||
["Dropout", config.loraDropout],
|
||||
["Variant", config.loraVariant],
|
||||
],
|
||||
},
|
||||
]
|
||||
|
|
@ -86,7 +185,7 @@ export function ProgressSection(): ReactElement | null {
|
|||
<SectionCard
|
||||
icon={<HugeiconsIcon icon={ChartAverageIcon} className="size-5" />}
|
||||
title="Training Progress"
|
||||
description="Live training metrics"
|
||||
description={runtime.message || "Live training metrics"}
|
||||
accent="emerald"
|
||||
className="shadow-border ring-1 ring-border"
|
||||
headerAction={
|
||||
|
|
@ -130,7 +229,8 @@ export function ProgressSection(): ReactElement | null {
|
|||
variant="destructive"
|
||||
size="sm"
|
||||
className="h-7 cursor-pointer px-3 text-xs"
|
||||
onClick={() => store.setIsTraining(false)}
|
||||
onClick={() => void stopTrainingRun()}
|
||||
disabled={!runtime.isTrainingRunning}
|
||||
>
|
||||
<HugeiconsIcon icon={StopIcon} className="size-3" /> Stop
|
||||
</Button>
|
||||
|
|
@ -138,24 +238,22 @@ export function ProgressSection(): ReactElement | null {
|
|||
}
|
||||
>
|
||||
<div className="grid grid-cols-1 gap-6 md:grid-cols-2">
|
||||
{/* Left: Progress */}
|
||||
<div className="flex flex-col gap-4">
|
||||
<div className="flex items-center gap-2">
|
||||
<span
|
||||
className={`rounded-full px-2.5 py-0.5 text-[10px] font-semibold ${statusColors[metrics.status]}`}
|
||||
className={`rounded-full px-2.5 py-0.5 text-[10px] font-semibold ${phaseColors[runtime.phase]}`}
|
||||
>
|
||||
{statusLabels[metrics.status]}
|
||||
{phaseLabel[runtime.phase]}
|
||||
</span>
|
||||
<span className="text-[10px] tabular-nums text-muted-foreground">
|
||||
Epoch {metrics.currentEpoch.toFixed(2)} / {metrics.totalEpochs}
|
||||
Epoch {runtime.currentEpoch.toFixed(2)}
|
||||
</span>
|
||||
</div>
|
||||
|
||||
{/* Progress bar */}
|
||||
<div className="flex flex-col gap-1.5">
|
||||
<div className="flex justify-between text-xs text-muted-foreground">
|
||||
<span>
|
||||
Step {metrics.currentStep} / {metrics.totalSteps}
|
||||
Step {runtime.currentStep} / {runtime.totalSteps || "--"}
|
||||
</span>
|
||||
<span>{pct}%</span>
|
||||
</div>
|
||||
|
|
@ -167,51 +265,59 @@ export function ProgressSection(): ReactElement | null {
|
|||
</div>
|
||||
</div>
|
||||
|
||||
{/* Metrics */}
|
||||
<div className="flex items-baseline gap-4">
|
||||
{runtime.error && (
|
||||
<p className="text-xs text-red-500 leading-relaxed">{runtime.error}</p>
|
||||
)}
|
||||
|
||||
<div className="flex flex-wrap items-baseline gap-4">
|
||||
<div>
|
||||
<p className="text-xs text-muted-foreground">Loss</p>
|
||||
<p className="text-3xl font-bold tabular-nums tracking-tight">
|
||||
{metrics.currentLoss.toFixed(4)}
|
||||
{runtime.currentLoss.toFixed(4)}
|
||||
</p>
|
||||
</div>
|
||||
<div>
|
||||
<p className="text-xs text-muted-foreground">LR</p>
|
||||
<p className="text-lg font-semibold tabular-nums">
|
||||
{metrics.currentLR.toExponential(2)}
|
||||
{runtime.currentLearningRate.toExponential(2)}
|
||||
</p>
|
||||
</div>
|
||||
<div>
|
||||
<p className="text-xs text-muted-foreground">Grad Norm</p>
|
||||
<p className="text-lg font-semibold tabular-nums">
|
||||
{metrics.gradNorm.toFixed(3)}
|
||||
{formatNumber(runtime.currentGradNorm, 3)}
|
||||
</p>
|
||||
</div>
|
||||
<div>
|
||||
<p className="text-xs text-muted-foreground">Model</p>
|
||||
<p className="text-lg font-semibold truncate max-w-[140px]">
|
||||
{modelName}
|
||||
{config.selectedModel ?? "--"}
|
||||
</p>
|
||||
</div>
|
||||
<div>
|
||||
<p className="text-xs text-muted-foreground">Method</p>
|
||||
<p className="text-lg font-semibold">{store.trainingMethod}</p>
|
||||
<p className="text-lg font-semibold">
|
||||
{config.trainingMethod.toUpperCase()}
|
||||
</p>
|
||||
</div>
|
||||
</div>
|
||||
|
||||
{/* Timings */}
|
||||
<div className="flex gap-4 text-xs text-muted-foreground">
|
||||
<span>Elapsed: {fmtTime(metrics.elapsed)}</span>
|
||||
<span>ETA: {fmtTime(etaSec)}</span>
|
||||
<span>{metrics.samplesPerSecond} samples/s</span>
|
||||
<div className="flex flex-wrap gap-4 text-xs text-muted-foreground">
|
||||
<span>Elapsed: {formatDuration(elapsed)}</span>
|
||||
<span>ETA: {formatDuration(eta)}</span>
|
||||
<span>
|
||||
{stepsPerSecond == null
|
||||
? "-- steps/s"
|
||||
: `${stepsPerSecond.toFixed(2)} steps/s`}
|
||||
</span>
|
||||
{runtime.currentNumTokens != null && (
|
||||
<span>Tokens: {runtime.currentNumTokens}</span>
|
||||
)}
|
||||
</div>
|
||||
</div>
|
||||
|
||||
{/* Right: GPU */}
|
||||
<div className="flex flex-col gap-3">
|
||||
<p className="text-xs font-medium text-muted-foreground">
|
||||
GPU Monitor
|
||||
</p>
|
||||
<p className="text-xs font-medium text-muted-foreground">GPU Monitor</p>
|
||||
<div className="grid grid-cols-2 gap-3">
|
||||
<GpuStat
|
||||
label="Utilization"
|
||||
|
|
@ -221,29 +327,27 @@ export function ProgressSection(): ReactElement | null {
|
|||
className="size-3.5"
|
||||
/>
|
||||
}
|
||||
value={`${metrics.gpuUtil}%`}
|
||||
pct={metrics.gpuUtil}
|
||||
value="--"
|
||||
pct={0}
|
||||
/>
|
||||
<GpuStat
|
||||
label="Temperature"
|
||||
icon={
|
||||
<HugeiconsIcon icon={TemperatureIcon} className="size-3.5" />
|
||||
}
|
||||
value={`${metrics.gpuTemp}°C`}
|
||||
pct={metrics.gpuTemp}
|
||||
icon={<HugeiconsIcon icon={TemperatureIcon} className="size-3.5" />}
|
||||
value="--"
|
||||
pct={0}
|
||||
max={100}
|
||||
/>
|
||||
<GpuStat
|
||||
label="VRAM"
|
||||
icon={<HugeiconsIcon icon={RamMemoryIcon} className="size-3.5" />}
|
||||
value={`${metrics.gpuVramUsed.toFixed(1)} / ${metrics.gpuVramTotal}GB`}
|
||||
pct={(metrics.gpuVramUsed / metrics.gpuVramTotal) * 100}
|
||||
value="--"
|
||||
pct={0}
|
||||
/>
|
||||
<GpuStat
|
||||
label="Power"
|
||||
icon={<HugeiconsIcon icon={ZapIcon} className="size-3.5" />}
|
||||
value={`${metrics.gpuPower}W`}
|
||||
pct={(metrics.gpuPower / 350) * 100}
|
||||
value="--"
|
||||
pct={0}
|
||||
/>
|
||||
</div>
|
||||
</div>
|
||||
|
|
@ -272,6 +376,7 @@ function GpuStat({
|
|||
} else if (clamped < 95) {
|
||||
barColor = "bg-amber-500";
|
||||
}
|
||||
|
||||
return (
|
||||
<div className="flex flex-col gap-1.5 rounded-xl bg-muted/50 p-3">
|
||||
<div className="flex items-center justify-between text-xs">
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ import {
|
|||
CollapsibleTrigger,
|
||||
} from "@/components/ui/collapsible";
|
||||
import { Input } from "@/components/ui/input";
|
||||
import { useWizardStore } from "@/stores/training";
|
||||
import { useTrainingActions, useTrainingConfigStore } from "@/features/training";
|
||||
import {
|
||||
Archive04Icon,
|
||||
ArrowDown01Icon,
|
||||
|
|
@ -35,7 +35,8 @@ const placeholderData = [
|
|||
];
|
||||
|
||||
export function TrainingSection() {
|
||||
const store = useWizardStore();
|
||||
const store = useTrainingConfigStore();
|
||||
const { isStarting, startError, startTrainingRun } = useTrainingActions();
|
||||
const [logOpen, setLogOpen] = useState(false);
|
||||
|
||||
return (
|
||||
|
|
@ -94,11 +95,15 @@ export function TrainingSection() {
|
|||
{/* Start/Stop */}
|
||||
<Button
|
||||
className="w-full cursor-pointer bg-gradient-to-r from-emerald-500 to-teal-500 text-white hover:from-emerald-600 hover:to-teal-600"
|
||||
onClick={() => store.setIsTraining(true)}
|
||||
onClick={() => void startTrainingRun()}
|
||||
disabled={isStarting}
|
||||
>
|
||||
<HugeiconsIcon icon={Rocket01Icon} className="size-4" /> Start
|
||||
Training
|
||||
<HugeiconsIcon icon={Rocket01Icon} className="size-4" />
|
||||
{isStarting ? "Starting..." : "Start Training"}
|
||||
</Button>
|
||||
{startError && (
|
||||
<p className="text-xs text-red-500 leading-relaxed">{startError}</p>
|
||||
)}
|
||||
|
||||
{/* Save / Clear */}
|
||||
<div className="grid grid-cols-2 gap-2">
|
||||
|
|
|
|||
|
|
@ -1,4 +1,8 @@
|
|||
import { useWizardStore } from "@/stores/training";
|
||||
import {
|
||||
shouldShowTrainingView,
|
||||
useTrainingRuntimeLifecycle,
|
||||
useTrainingRuntimeStore,
|
||||
} from "@/features/training";
|
||||
import type { ReactElement } from "react";
|
||||
import { DatasetSection } from "./sections/dataset-section";
|
||||
import { ModelSection } from "./sections/model-section";
|
||||
|
|
@ -7,7 +11,11 @@ import { TrainingSection } from "./sections/training-section";
|
|||
import { TrainingView } from "./training-view";
|
||||
|
||||
export function StudioPage(): ReactElement {
|
||||
const isTraining = useWizardStore((s) => s.isTraining);
|
||||
useTrainingRuntimeLifecycle();
|
||||
const showTrainingView = useTrainingRuntimeStore(shouldShowTrainingView);
|
||||
const runtimeMessage = useTrainingRuntimeStore((state) => state.message);
|
||||
const isHydratingRuntime = useTrainingRuntimeStore((state) => state.isHydrating);
|
||||
const hasHydratedRuntime = useTrainingRuntimeStore((state) => state.hasHydrated);
|
||||
|
||||
return (
|
||||
<div className="min-h-screen bg-background">
|
||||
|
|
@ -18,13 +26,17 @@ export function StudioPage(): ReactElement {
|
|||
Fine-tuning Studio
|
||||
</h1>
|
||||
<p className="text-sm text-muted-foreground">
|
||||
{isTraining
|
||||
? "Training in progress"
|
||||
{showTrainingView
|
||||
? runtimeMessage || "Training in progress"
|
||||
: "Configure and start training"}
|
||||
</p>
|
||||
</div>
|
||||
|
||||
{isTraining ? (
|
||||
{!hasHydratedRuntime && isHydratingRuntime ? (
|
||||
<div className="rounded-xl border bg-card p-8 text-sm text-muted-foreground">
|
||||
Loading training runtime...
|
||||
</div>
|
||||
) : showTrainingView ? (
|
||||
<TrainingView />
|
||||
) : (
|
||||
<div className="grid grid-cols-12 items-start gap-6">
|
||||
|
|
|
|||
|
|
@ -0,0 +1,58 @@
|
|||
import {
|
||||
AnimatedSpan,
|
||||
Terminal,
|
||||
TypingAnimation,
|
||||
} from "@/components/ui/terminal"
|
||||
import type { ReactElement } from "react"
|
||||
|
||||
type TrainingStartOverlayProps = {
|
||||
message: string
|
||||
currentStep: number
|
||||
}
|
||||
|
||||
export function TrainingStartOverlay({
|
||||
message,
|
||||
currentStep,
|
||||
}: TrainingStartOverlayProps): ReactElement {
|
||||
return (
|
||||
<div className="pointer-events-none absolute inset-0 z-30 flex items-center justify-center rounded-2xl bg-background/45 backdrop-blur-[1px]">
|
||||
<div className="flex w-[860px] max-w-[calc(100%-2rem)] flex-col items-center gap-4">
|
||||
<img
|
||||
src="/Sloth emojis/large sloth wave.png"
|
||||
alt="Unsloth mascot"
|
||||
className="size-24 animate-bounce object-contain"
|
||||
/>
|
||||
<Terminal
|
||||
className="w-full min-h-[390px] rounded-2xl px-7 py-6 text-left"
|
||||
startOnView={false}
|
||||
>
|
||||
<TypingAnimation
|
||||
duration={36}
|
||||
className="bg-gradient-to-r from-emerald-300 via-lime-300 to-teal-300 bg-clip-text font-semibold text-transparent"
|
||||
>
|
||||
{"> unsloth training starts..."}
|
||||
</TypingAnimation>
|
||||
<AnimatedSpan className="my-2">
|
||||
<pre className="whitespace-pre text-left text-muted-foreground">{`==((====))==
|
||||
\\\\ /|
|
||||
O^O/ \\_/ \\
|
||||
\\ /
|
||||
"-____-"`}</pre>
|
||||
</AnimatedSpan>
|
||||
<TypingAnimation duration={44}>
|
||||
{"> Preparing model and dataset..."}
|
||||
</TypingAnimation>
|
||||
<TypingAnimation duration={44}>
|
||||
{"> We are getting everything ready for your run..."}
|
||||
</TypingAnimation>
|
||||
<TypingAnimation duration={44}>
|
||||
{"> Did you know, Mugi is actually short for \"Mugiwara\" xd"}
|
||||
</TypingAnimation>
|
||||
<AnimatedSpan className="mt-2 text-muted-foreground">
|
||||
{`> ${message || "starting training..."} | waiting for first step... (${currentStep})`}
|
||||
</AnimatedSpan>
|
||||
</Terminal>
|
||||
</div>
|
||||
</div>
|
||||
)
|
||||
}
|
||||
|
|
@ -1,164 +1,47 @@
|
|||
import { useWizardStore } from "@/stores/training";
|
||||
import type { TrainingMetrics } from "@/types/training";
|
||||
import { type ReactElement, useEffect, useRef } from "react";
|
||||
import { cn } from "@/lib/utils";
|
||||
import { useTrainingRuntimeStore } from "@/features/training";
|
||||
import type { ReactElement } from "react";
|
||||
import { useShallow } from "zustand/react/shallow";
|
||||
import { ChartsSection } from "./sections/charts-section";
|
||||
import { ProgressSection } from "./sections/progress-section";
|
||||
|
||||
function createInitialMetrics(
|
||||
totalSteps: number,
|
||||
totalEpochs: number,
|
||||
lr: number,
|
||||
): TrainingMetrics {
|
||||
return {
|
||||
currentStep: 0,
|
||||
totalSteps,
|
||||
currentEpoch: 0,
|
||||
totalEpochs,
|
||||
currentLoss: 2.5,
|
||||
currentLR: lr * 0.1,
|
||||
gradNorm: 0,
|
||||
samplesPerSecond: 0,
|
||||
lossHistory: [],
|
||||
lrHistory: [],
|
||||
gradNormHistory: [],
|
||||
gpuUtil: 0,
|
||||
gpuTemp: 45,
|
||||
gpuVramUsed: 0,
|
||||
gpuVramTotal: 24,
|
||||
gpuPower: 50,
|
||||
elapsed: 0,
|
||||
status: "warmup",
|
||||
};
|
||||
}
|
||||
import { TrainingStartOverlay } from "./training-start-overlay";
|
||||
|
||||
export function TrainingView(): ReactElement {
|
||||
const { maxSteps, epochs, learningRate, warmupSteps, setTrainingMetrics } =
|
||||
useWizardStore();
|
||||
const metricsRef = useRef<ReturnType<typeof setInterval> | null>(null);
|
||||
const chartsRef = useRef<ReturnType<typeof setInterval> | null>(null);
|
||||
const runtime = useTrainingRuntimeStore(
|
||||
useShallow((state) => ({
|
||||
phase: state.phase,
|
||||
message: state.message,
|
||||
currentStep: state.currentStep,
|
||||
firstStepReceived: state.firstStepReceived,
|
||||
isStarting: state.isStarting,
|
||||
})),
|
||||
);
|
||||
|
||||
useEffect(() => {
|
||||
const totalSteps = maxSteps || 500;
|
||||
const totalEpochs = epochs || 3;
|
||||
const peakLR = learningRate;
|
||||
const warmup = warmupSteps || 20;
|
||||
|
||||
setTrainingMetrics(createInitialMetrics(totalSteps, totalEpochs, peakLR));
|
||||
|
||||
let step = 0;
|
||||
let elapsed = 0;
|
||||
|
||||
const computeStep = () => {
|
||||
step++;
|
||||
if (step > totalSteps) {
|
||||
return null;
|
||||
}
|
||||
elapsed++;
|
||||
|
||||
let lr: number;
|
||||
if (step < warmup) {
|
||||
lr = peakLR * (step / warmup);
|
||||
} else {
|
||||
const progress = (step - warmup) / (totalSteps - warmup);
|
||||
lr = peakLR * 0.5 * (1 + Math.cos(Math.PI * progress));
|
||||
}
|
||||
|
||||
const baseLoss = 2.5 * Math.exp((-3 * step) / totalSteps) + 0.3;
|
||||
const noise = (Math.random() - 0.5) * 0.08;
|
||||
const loss = Math.max(0.1, baseLoss + noise);
|
||||
const status =
|
||||
step < warmup ? "warmup" : step % 100 === 0 ? "saving" : "training";
|
||||
const gradNorm = +(
|
||||
1.2 * Math.exp(-step / totalSteps) +
|
||||
0.1 +
|
||||
(Math.random() - 0.5) * 0.05
|
||||
).toFixed(3);
|
||||
|
||||
return {
|
||||
step,
|
||||
elapsed,
|
||||
lr,
|
||||
loss: +loss.toFixed(4),
|
||||
status: status as TrainingMetrics["status"],
|
||||
gradNorm,
|
||||
};
|
||||
};
|
||||
|
||||
// Top card values — update every 1s
|
||||
metricsRef.current = setInterval(() => {
|
||||
const s = computeStep();
|
||||
if (!s) {
|
||||
if (metricsRef.current) {
|
||||
clearInterval(metricsRef.current);
|
||||
}
|
||||
if (chartsRef.current) {
|
||||
clearInterval(chartsRef.current);
|
||||
}
|
||||
return;
|
||||
}
|
||||
|
||||
const prev = useWizardStore.getState().trainingMetrics;
|
||||
setTrainingMetrics({
|
||||
currentStep: s.step,
|
||||
totalSteps,
|
||||
currentEpoch:
|
||||
Math.floor((s.step / totalSteps) * totalEpochs * 100) / 100,
|
||||
totalEpochs,
|
||||
currentLoss: s.loss,
|
||||
currentLR: s.lr,
|
||||
gradNorm: s.gradNorm,
|
||||
samplesPerSecond: +(12 + (Math.random() - 0.5) * 2).toFixed(1),
|
||||
lossHistory: prev?.lossHistory ?? [],
|
||||
lrHistory: prev?.lrHistory ?? [],
|
||||
gradNormHistory: prev?.gradNormHistory ?? [],
|
||||
gpuUtil: Math.min(99, 85 + Math.round((Math.random() - 0.5) * 10)),
|
||||
gpuTemp: Math.min(89, 68 + Math.round((Math.random() - 0.5) * 6)),
|
||||
gpuVramUsed: +(18.2 + (Math.random() - 0.5) * 0.4).toFixed(1),
|
||||
gpuVramTotal: 24,
|
||||
gpuPower: Math.round(280 + (Math.random() - 0.5) * 30),
|
||||
elapsed: s.elapsed,
|
||||
status: s.status,
|
||||
});
|
||||
}, 1000);
|
||||
|
||||
// Chart history — update every 5s
|
||||
chartsRef.current = setInterval(() => {
|
||||
const prev = useWizardStore.getState().trainingMetrics;
|
||||
if (!prev || prev.currentStep === 0) {
|
||||
return;
|
||||
}
|
||||
|
||||
setTrainingMetrics({
|
||||
...prev,
|
||||
lossHistory: [
|
||||
...prev.lossHistory,
|
||||
{ step: prev.currentStep, loss: prev.currentLoss },
|
||||
],
|
||||
lrHistory: [
|
||||
...prev.lrHistory,
|
||||
{ step: prev.currentStep, lr: prev.currentLR },
|
||||
],
|
||||
gradNormHistory: [
|
||||
...prev.gradNormHistory,
|
||||
{ step: prev.currentStep, gradNorm: prev.gradNorm },
|
||||
],
|
||||
});
|
||||
}, 5000);
|
||||
|
||||
return () => {
|
||||
if (metricsRef.current) {
|
||||
clearInterval(metricsRef.current);
|
||||
}
|
||||
if (chartsRef.current) {
|
||||
clearInterval(chartsRef.current);
|
||||
}
|
||||
};
|
||||
}, [epochs, learningRate, maxSteps, setTrainingMetrics, warmupSteps]);
|
||||
const isPreparingPhase =
|
||||
runtime.phase === "loading_model" ||
|
||||
runtime.phase === "loading_dataset" ||
|
||||
runtime.phase === "configuring";
|
||||
const isWaitingForFirstStep =
|
||||
runtime.phase === "training" && !runtime.firstStepReceived;
|
||||
const showOverlay =
|
||||
runtime.isStarting ||
|
||||
isPreparingPhase ||
|
||||
(isWaitingForFirstStep && runtime.currentStep <= 0);
|
||||
|
||||
return (
|
||||
<div className="flex flex-col gap-6">
|
||||
<ProgressSection />
|
||||
<ChartsSection />
|
||||
<div className={cn("relative", showOverlay && "min-h-[72vh]")}>
|
||||
<div
|
||||
className={cn("flex flex-col gap-6 transition-[filter]", showOverlay && "blur")}
|
||||
>
|
||||
<ProgressSection />
|
||||
<ChartsSection />
|
||||
</div>
|
||||
{showOverlay ? (
|
||||
<TrainingStartOverlay
|
||||
message={runtime.message}
|
||||
currentStep={runtime.currentStep}
|
||||
/>
|
||||
) : null}
|
||||
</div>
|
||||
);
|
||||
}
|
||||
|
|
|
|||
63
studio/frontend/src/features/training/api/mappers.ts
Normal file
63
studio/frontend/src/features/training/api/mappers.ts
Normal file
|
|
@ -0,0 +1,63 @@
|
|||
import type { TrainingConfigState } from "../types/config";
|
||||
import type { TrainingStartRequest } from "../types/api";
|
||||
|
||||
const BACKEND_LORA_TYPE = "LoRA/QLoRA";
|
||||
const BACKEND_FULL_TYPE = "Full Finetuning";
|
||||
|
||||
export function toBackendTrainingType(trainingMethod: string): string {
|
||||
return trainingMethod === "full" ? BACKEND_FULL_TYPE : BACKEND_LORA_TYPE;
|
||||
}
|
||||
|
||||
export function buildTrainingStartPayload(
|
||||
config: TrainingConfigState,
|
||||
): TrainingStartRequest {
|
||||
const adapterMethod = config.trainingMethod !== "full";
|
||||
const isQlorMethod = config.trainingMethod === "qlora";
|
||||
const hfDataset = config.datasetSource === "huggingface" ? config.dataset : null;
|
||||
|
||||
return {
|
||||
model_name: config.selectedModel ?? "",
|
||||
training_type: toBackendTrainingType(config.trainingMethod),
|
||||
hf_token: config.hfToken.trim() || null,
|
||||
load_in_4bit: adapterMethod ? isQlorMethod : false,
|
||||
max_seq_length: config.contextLength,
|
||||
hf_dataset: hfDataset,
|
||||
local_datasets: [],
|
||||
format_type: config.datasetFormat,
|
||||
num_epochs: config.epochs,
|
||||
learning_rate: String(config.learningRate),
|
||||
batch_size: config.batchSize,
|
||||
gradient_accumulation_steps: config.gradientAccumulation,
|
||||
warmup_steps: config.warmupSteps,
|
||||
warmup_ratio: null,
|
||||
max_steps: config.maxSteps,
|
||||
save_steps: config.saveSteps,
|
||||
weight_decay: config.weightDecay,
|
||||
random_seed: config.randomSeed,
|
||||
packing: config.packing,
|
||||
optim: "adamw_8bit",
|
||||
lr_scheduler_type: "linear",
|
||||
use_lora: adapterMethod,
|
||||
lora_r: config.loraRank,
|
||||
lora_alpha: config.loraAlpha,
|
||||
lora_dropout: config.loraDropout,
|
||||
target_modules: adapterMethod ? config.targetModules : [],
|
||||
gradient_checkpointing: config.gradientCheckpointing,
|
||||
use_rslora: config.loraVariant === "rslora",
|
||||
use_loftq: config.loraVariant === "loftq",
|
||||
train_on_completions: config.trainOnCompletions,
|
||||
finetune_vision_layers: config.finetuneVisionLayers,
|
||||
finetune_language_layers: config.finetuneLanguageLayers,
|
||||
finetune_attention_modules: config.finetuneAttentionModules,
|
||||
finetune_mlp_modules: config.finetuneMLPModules,
|
||||
enable_wandb: config.enableWandb,
|
||||
wandb_token: config.enableWandb ? config.wandbToken.trim() || null : null,
|
||||
wandb_project: config.enableWandb
|
||||
? config.wandbProject.trim() || null
|
||||
: null,
|
||||
enable_tensorboard: config.enableTensorboard,
|
||||
tensorboard_dir: config.enableTensorboard
|
||||
? config.tensorboardDir.trim() || null
|
||||
: null,
|
||||
};
|
||||
}
|
||||
173
studio/frontend/src/features/training/api/train-api.ts
Normal file
173
studio/frontend/src/features/training/api/train-api.ts
Normal file
|
|
@ -0,0 +1,173 @@
|
|||
import { authFetch } from "@/features/auth";
|
||||
import type {
|
||||
TrainingStartRequest,
|
||||
TrainingStartResponse,
|
||||
TrainingStopResponse,
|
||||
} from "../types/api";
|
||||
import type {
|
||||
TrainingMetricsResponse,
|
||||
TrainingProgressPayload,
|
||||
TrainingStatusResponse,
|
||||
} from "../types/runtime";
|
||||
|
||||
function isAbortError(error: unknown): boolean {
|
||||
return error instanceof DOMException && error.name === "AbortError";
|
||||
}
|
||||
|
||||
async function readError(response: Response): Promise<string> {
|
||||
try {
|
||||
const payload = (await response.json()) as { detail?: string; message?: string };
|
||||
return payload.detail || payload.message || `Request failed (${response.status})`;
|
||||
} catch {
|
||||
return `Request failed (${response.status})`;
|
||||
}
|
||||
}
|
||||
|
||||
async function parseJson<T>(response: Response): Promise<T> {
|
||||
if (!response.ok) {
|
||||
throw new Error(await readError(response));
|
||||
}
|
||||
return (await response.json()) as T;
|
||||
}
|
||||
|
||||
export async function startTraining(
|
||||
payload: TrainingStartRequest,
|
||||
): Promise<TrainingStartResponse> {
|
||||
const response = await authFetch("/api/train/start", {
|
||||
method: "POST",
|
||||
headers: { "Content-Type": "application/json" },
|
||||
body: JSON.stringify(payload),
|
||||
});
|
||||
return parseJson<TrainingStartResponse>(response);
|
||||
}
|
||||
|
||||
export async function stopTraining(): Promise<TrainingStopResponse> {
|
||||
const response = await authFetch("/api/train/stop", { method: "POST" });
|
||||
return parseJson<TrainingStopResponse>(response);
|
||||
}
|
||||
|
||||
export async function getTrainingStatus(): Promise<TrainingStatusResponse> {
|
||||
const response = await authFetch("/api/train/status");
|
||||
return parseJson<TrainingStatusResponse>(response);
|
||||
}
|
||||
|
||||
export async function getTrainingMetrics(): Promise<TrainingMetricsResponse> {
|
||||
const response = await authFetch("/api/train/metrics");
|
||||
return parseJson<TrainingMetricsResponse>(response);
|
||||
}
|
||||
|
||||
type ProgressEventName = "progress" | "heartbeat" | "complete" | "error";
|
||||
|
||||
interface ParsedSseEvent {
|
||||
event: ProgressEventName;
|
||||
payload: TrainingProgressPayload;
|
||||
id: number | null;
|
||||
}
|
||||
|
||||
function parseSseEvent(rawEvent: string): ParsedSseEvent | null {
|
||||
const lines = rawEvent.split(/\r?\n/);
|
||||
let eventName: ProgressEventName = "progress";
|
||||
let id: number | null = null;
|
||||
const dataLines: string[] = [];
|
||||
|
||||
for (const line of lines) {
|
||||
if (!line) {
|
||||
continue;
|
||||
}
|
||||
if (line.startsWith("event:")) {
|
||||
const value = line.slice(6).trim();
|
||||
if (
|
||||
value === "progress" ||
|
||||
value === "heartbeat" ||
|
||||
value === "complete" ||
|
||||
value === "error"
|
||||
) {
|
||||
eventName = value;
|
||||
}
|
||||
continue;
|
||||
}
|
||||
if (line.startsWith("id:")) {
|
||||
const value = Number(line.slice(3).trim());
|
||||
id = Number.isFinite(value) ? value : null;
|
||||
continue;
|
||||
}
|
||||
if (line.startsWith("data:")) {
|
||||
dataLines.push(line.slice(5).trimStart());
|
||||
}
|
||||
}
|
||||
|
||||
if (dataLines.length === 0) {
|
||||
return null;
|
||||
}
|
||||
|
||||
const parsed = JSON.parse(dataLines.join("\n")) as TrainingProgressPayload;
|
||||
return { event: eventName, payload: parsed, id };
|
||||
}
|
||||
|
||||
export async function streamTrainingProgress(options: {
|
||||
signal: AbortSignal;
|
||||
lastEventId?: number | null;
|
||||
onOpen?: () => void;
|
||||
onEvent: (event: ParsedSseEvent) => void;
|
||||
}): Promise<void> {
|
||||
const headers = new Headers();
|
||||
if (typeof options.lastEventId === "number") {
|
||||
headers.set("Last-Event-ID", String(options.lastEventId));
|
||||
}
|
||||
|
||||
const response = await authFetch("/api/train/progress", {
|
||||
method: "GET",
|
||||
headers,
|
||||
signal: options.signal,
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
throw new Error(await readError(response));
|
||||
}
|
||||
|
||||
if (!response.body) {
|
||||
throw new Error("Progress stream unavailable");
|
||||
}
|
||||
|
||||
options.onOpen?.();
|
||||
|
||||
const reader = response.body.getReader();
|
||||
const decoder = new TextDecoder();
|
||||
let buffer = "";
|
||||
|
||||
while (true) {
|
||||
const { value, done } = await reader.read();
|
||||
if (done) {
|
||||
break;
|
||||
}
|
||||
|
||||
buffer += decoder.decode(value, { stream: true });
|
||||
|
||||
let separatorIndex = buffer.search(/\r?\n\r?\n/);
|
||||
while (separatorIndex >= 0) {
|
||||
const rawEvent = buffer.slice(0, separatorIndex);
|
||||
const separatorLength = buffer[separatorIndex] === "\r" ? 4 : 2;
|
||||
buffer = buffer.slice(separatorIndex + separatorLength);
|
||||
|
||||
if (rawEvent.startsWith("retry:")) {
|
||||
separatorIndex = buffer.search(/\r?\n\r?\n/);
|
||||
continue;
|
||||
}
|
||||
|
||||
try {
|
||||
const event = parseSseEvent(rawEvent);
|
||||
if (event) {
|
||||
options.onEvent(event);
|
||||
}
|
||||
} catch (error) {
|
||||
if (!isAbortError(error)) {
|
||||
throw error;
|
||||
}
|
||||
}
|
||||
|
||||
separatorIndex = buffer.search(/\r?\n\r?\n/);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
export { isAbortError };
|
||||
|
|
@ -0,0 +1,70 @@
|
|||
import { useCallback } from "react";
|
||||
import { useTrainingConfigStore } from "../stores/training-config-store";
|
||||
import { useTrainingRuntimeStore } from "../stores/training-runtime-store";
|
||||
import { startTraining, stopTraining } from "../api/train-api";
|
||||
import { buildTrainingStartPayload } from "../api/mappers";
|
||||
import { syncTrainingRuntimeFromBackend } from "../lib/sync-runtime";
|
||||
import { validateTrainingConfig } from "../lib/validation";
|
||||
|
||||
export function useTrainingActions() {
|
||||
const isStarting = useTrainingRuntimeStore((state) => state.isStarting);
|
||||
const startError = useTrainingRuntimeStore((state) => state.startError);
|
||||
|
||||
const startTrainingRun = useCallback(async (): Promise<boolean> => {
|
||||
const config = useTrainingConfigStore.getState();
|
||||
const runtimeStore = useTrainingRuntimeStore.getState();
|
||||
|
||||
runtimeStore.setStartError(null);
|
||||
const validation = validateTrainingConfig(config);
|
||||
if (!validation.ok) {
|
||||
runtimeStore.setStartError(validation.message);
|
||||
return false;
|
||||
}
|
||||
|
||||
runtimeStore.setStarting(true);
|
||||
|
||||
try {
|
||||
const payload = buildTrainingStartPayload(config);
|
||||
const response = await startTraining(payload);
|
||||
|
||||
if (response.status === "error") {
|
||||
runtimeStore.setStartError(response.error || response.message);
|
||||
runtimeStore.setStarting(false);
|
||||
return false;
|
||||
}
|
||||
|
||||
runtimeStore.setStartQueued(response.job_id, response.message);
|
||||
await syncTrainingRuntimeFromBackend();
|
||||
return true;
|
||||
} catch (error) {
|
||||
const message =
|
||||
error instanceof Error ? error.message : "Failed to start training";
|
||||
runtimeStore.setStartError(message);
|
||||
runtimeStore.setStarting(false);
|
||||
return false;
|
||||
}
|
||||
}, []);
|
||||
|
||||
const stopTrainingRun = useCallback(async (): Promise<boolean> => {
|
||||
const runtimeStore = useTrainingRuntimeStore.getState();
|
||||
runtimeStore.setStartError(null);
|
||||
|
||||
try {
|
||||
await stopTraining();
|
||||
await syncTrainingRuntimeFromBackend();
|
||||
return true;
|
||||
} catch (error) {
|
||||
const message =
|
||||
error instanceof Error ? error.message : "Failed to stop training";
|
||||
runtimeStore.setRuntimeError(message);
|
||||
return false;
|
||||
}
|
||||
}, []);
|
||||
|
||||
return {
|
||||
isStarting,
|
||||
startError,
|
||||
startTrainingRun,
|
||||
stopTrainingRun,
|
||||
};
|
||||
}
|
||||
|
|
@ -0,0 +1,183 @@
|
|||
import { useEffect } from "react";
|
||||
import {
|
||||
getTrainingMetrics,
|
||||
getTrainingStatus,
|
||||
isAbortError,
|
||||
streamTrainingProgress,
|
||||
} from "../api/train-api";
|
||||
import { useTrainingRuntimeStore } from "../stores/training-runtime-store";
|
||||
import type { TrainingRuntimeStore } from "../types/runtime";
|
||||
|
||||
const STATUS_POLL_INTERVAL_MS = 3000;
|
||||
const METRICS_POLL_INTERVAL_MS = 5000;
|
||||
const STREAM_RECONNECT_DELAY_MS = 1500;
|
||||
|
||||
function shouldUseLiveSync(state: TrainingRuntimeStore): boolean {
|
||||
return (
|
||||
state.isTrainingRunning ||
|
||||
state.phase === "loading_model" ||
|
||||
state.phase === "loading_dataset" ||
|
||||
state.phase === "configuring" ||
|
||||
state.phase === "training"
|
||||
);
|
||||
}
|
||||
|
||||
export function useTrainingRuntimeLifecycle(): void {
|
||||
useEffect(() => {
|
||||
let disposed = false;
|
||||
let openingStream = false;
|
||||
let streamController: AbortController | null = null;
|
||||
let reconnectTimer: ReturnType<typeof setTimeout> | null = null;
|
||||
|
||||
const runtimeStore = useTrainingRuntimeStore;
|
||||
|
||||
const clearReconnect = () => {
|
||||
if (reconnectTimer) {
|
||||
clearTimeout(reconnectTimer);
|
||||
reconnectTimer = null;
|
||||
}
|
||||
};
|
||||
|
||||
const stopStream = () => {
|
||||
clearReconnect();
|
||||
if (streamController) {
|
||||
streamController.abort();
|
||||
streamController = null;
|
||||
}
|
||||
runtimeStore.getState().setSseConnected(false);
|
||||
};
|
||||
|
||||
const pollMetrics = async () => {
|
||||
try {
|
||||
const metrics = await getTrainingMetrics();
|
||||
if (disposed) {
|
||||
return;
|
||||
}
|
||||
runtimeStore.getState().applyMetrics(metrics);
|
||||
} catch (error) {
|
||||
if (!isAbortError(error) && !disposed) {
|
||||
runtimeStore.getState().setSseConnected(false);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
const pollStatus = async () => {
|
||||
try {
|
||||
const status = await getTrainingStatus();
|
||||
if (disposed) {
|
||||
return;
|
||||
}
|
||||
|
||||
runtimeStore.getState().applyStatus(status);
|
||||
|
||||
const nextState = runtimeStore.getState();
|
||||
if (shouldUseLiveSync(nextState)) {
|
||||
void ensureStream();
|
||||
} else {
|
||||
stopStream();
|
||||
}
|
||||
} catch (error) {
|
||||
if (!isAbortError(error) && !disposed) {
|
||||
runtimeStore.getState().setSseConnected(false);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
const ensureStream = async () => {
|
||||
const state = runtimeStore.getState();
|
||||
if (
|
||||
disposed ||
|
||||
openingStream ||
|
||||
streamController ||
|
||||
!shouldUseLiveSync(state)
|
||||
) {
|
||||
return;
|
||||
}
|
||||
|
||||
clearReconnect();
|
||||
openingStream = true;
|
||||
const controller = new AbortController();
|
||||
streamController = controller;
|
||||
|
||||
try {
|
||||
await streamTrainingProgress({
|
||||
signal: controller.signal,
|
||||
lastEventId: state.lastEventId,
|
||||
onOpen: () => {
|
||||
runtimeStore.getState().setSseConnected(true);
|
||||
},
|
||||
onEvent: (event) => {
|
||||
const liveStore = runtimeStore.getState();
|
||||
if (typeof event.id === "number") {
|
||||
liveStore.setLastEventId(event.id);
|
||||
}
|
||||
|
||||
liveStore.applyProgress(event.payload, event.id ?? undefined);
|
||||
|
||||
if (event.event === "complete") {
|
||||
void pollStatus();
|
||||
void pollMetrics();
|
||||
stopStream();
|
||||
}
|
||||
|
||||
if (event.event === "error") {
|
||||
liveStore.setRuntimeError("Training stream error");
|
||||
stopStream();
|
||||
}
|
||||
},
|
||||
});
|
||||
} catch (error) {
|
||||
if (!disposed && !controller.signal.aborted && !isAbortError(error)) {
|
||||
runtimeStore.getState().setSseConnected(false);
|
||||
}
|
||||
} finally {
|
||||
openingStream = false;
|
||||
if (streamController === controller) {
|
||||
streamController = null;
|
||||
}
|
||||
runtimeStore.getState().setSseConnected(false);
|
||||
|
||||
if (!disposed && !controller.signal.aborted) {
|
||||
const liveState = runtimeStore.getState();
|
||||
if (shouldUseLiveSync(liveState)) {
|
||||
reconnectTimer = setTimeout(() => {
|
||||
void ensureStream();
|
||||
}, STREAM_RECONNECT_DELAY_MS);
|
||||
}
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
const hydrate = async () => {
|
||||
runtimeStore.getState().setHydrating(true);
|
||||
try {
|
||||
await Promise.all([pollStatus(), pollMetrics()]);
|
||||
} finally {
|
||||
if (!disposed) {
|
||||
runtimeStore.getState().setHydrating(false);
|
||||
runtimeStore.getState().setHasHydrated(true);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
void hydrate();
|
||||
|
||||
const statusTimer = setInterval(() => {
|
||||
void pollStatus();
|
||||
}, STATUS_POLL_INTERVAL_MS);
|
||||
|
||||
const metricsTimer = setInterval(() => {
|
||||
const state = runtimeStore.getState();
|
||||
if (shouldUseLiveSync(state) || state.currentStep > 0) {
|
||||
void pollMetrics();
|
||||
}
|
||||
}, METRICS_POLL_INTERVAL_MS);
|
||||
|
||||
return () => {
|
||||
disposed = true;
|
||||
clearInterval(statusTimer);
|
||||
clearInterval(metricsTimer);
|
||||
stopStream();
|
||||
};
|
||||
}, []);
|
||||
}
|
||||
8
studio/frontend/src/features/training/index.ts
Normal file
8
studio/frontend/src/features/training/index.ts
Normal file
|
|
@ -0,0 +1,8 @@
|
|||
export { useTrainingConfigStore } from "./stores/training-config-store";
|
||||
export {
|
||||
shouldShowTrainingView,
|
||||
useTrainingRuntimeStore,
|
||||
} from "./stores/training-runtime-store";
|
||||
export { useTrainingActions } from "./hooks/use-training-actions";
|
||||
export { useTrainingRuntimeLifecycle } from "./hooks/use-training-runtime-lifecycle";
|
||||
export type { TrainingPhase } from "./types/runtime";
|
||||
19
studio/frontend/src/features/training/lib/sync-runtime.ts
Normal file
19
studio/frontend/src/features/training/lib/sync-runtime.ts
Normal file
|
|
@ -0,0 +1,19 @@
|
|||
import {
|
||||
getTrainingMetrics,
|
||||
getTrainingStatus,
|
||||
} from "../api/train-api";
|
||||
import { useTrainingRuntimeStore } from "../stores/training-runtime-store";
|
||||
import type { TrainingStatusResponse } from "../types/runtime";
|
||||
|
||||
export async function syncTrainingRuntimeFromBackend(): Promise<TrainingStatusResponse> {
|
||||
const [status, metrics] = await Promise.all([
|
||||
getTrainingStatus(),
|
||||
getTrainingMetrics(),
|
||||
]);
|
||||
|
||||
const runtimeStore = useTrainingRuntimeStore.getState();
|
||||
runtimeStore.applyStatus(status);
|
||||
runtimeStore.applyMetrics(metrics);
|
||||
|
||||
return status;
|
||||
}
|
||||
27
studio/frontend/src/features/training/lib/validation.ts
Normal file
27
studio/frontend/src/features/training/lib/validation.ts
Normal file
|
|
@ -0,0 +1,27 @@
|
|||
import type { TrainingConfigState } from "../types/config";
|
||||
|
||||
export interface StartValidationResult {
|
||||
ok: boolean;
|
||||
message: string | null;
|
||||
}
|
||||
|
||||
export function validateTrainingConfig(
|
||||
config: TrainingConfigState,
|
||||
): StartValidationResult {
|
||||
if (!config.selectedModel) {
|
||||
return { ok: false, message: "Select a base model first." };
|
||||
}
|
||||
|
||||
if (config.datasetSource !== "huggingface") {
|
||||
return {
|
||||
ok: false,
|
||||
message: "Only Hugging Face dataset source is enabled right now.",
|
||||
};
|
||||
}
|
||||
|
||||
if (!config.dataset) {
|
||||
return { ok: false, message: "Select a Hugging Face dataset first." };
|
||||
}
|
||||
|
||||
return { ok: true, message: null };
|
||||
}
|
||||
|
|
@ -0,0 +1,105 @@
|
|||
import { DEFAULT_HYPERPARAMS, STEPS } from "@/config/training";
|
||||
import type { StepNumber } from "@/types/training";
|
||||
import { create } from "zustand";
|
||||
import { persist } from "zustand/middleware";
|
||||
import type { TrainingConfigState, TrainingConfigStore } from "../types/config";
|
||||
|
||||
const MIN_STEP: StepNumber = 1;
|
||||
const MAX_STEP: StepNumber = STEPS.length as StepNumber;
|
||||
|
||||
const initialState: TrainingConfigState = {
|
||||
currentStep: MIN_STEP,
|
||||
modelType: null,
|
||||
selectedModel: null,
|
||||
trainingMethod: "qlora",
|
||||
hfToken: "",
|
||||
datasetSource: "huggingface",
|
||||
datasetFormat: "auto",
|
||||
dataset: null,
|
||||
uploadedFile: null,
|
||||
...DEFAULT_HYPERPARAMS,
|
||||
};
|
||||
|
||||
function clampStep(step: number): StepNumber {
|
||||
return Math.min(MAX_STEP, Math.max(MIN_STEP, step)) as StepNumber;
|
||||
}
|
||||
|
||||
function canProceedForStep(state: TrainingConfigState): boolean {
|
||||
switch (state.currentStep) {
|
||||
case 1:
|
||||
return state.modelType !== null;
|
||||
case 2:
|
||||
return state.selectedModel !== null;
|
||||
case 3:
|
||||
return state.datasetSource === "upload"
|
||||
? state.uploadedFile !== null
|
||||
: state.dataset !== null;
|
||||
case 4:
|
||||
case 5:
|
||||
return true;
|
||||
default:
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
export const useTrainingConfigStore = create<TrainingConfigStore>()(
|
||||
persist(
|
||||
(set, get) => ({
|
||||
...initialState,
|
||||
setStep: (step) => set({ currentStep: step }),
|
||||
nextStep: () => set({ currentStep: clampStep(get().currentStep + 1) }),
|
||||
prevStep: () => set({ currentStep: clampStep(get().currentStep - 1) }),
|
||||
setModelType: (modelType) => set({ modelType, selectedModel: null }),
|
||||
setSelectedModel: (selectedModel) => set({ selectedModel }),
|
||||
setTrainingMethod: (trainingMethod) => set({ trainingMethod }),
|
||||
setHfToken: (hfToken) => set({ hfToken }),
|
||||
setDatasetSource: (datasetSource) => set({ datasetSource }),
|
||||
setDatasetFormat: (datasetFormat) => set({ datasetFormat }),
|
||||
setDataset: (dataset) => set({ dataset }),
|
||||
setUploadedFile: (uploadedFile) => set({ uploadedFile }),
|
||||
setEpochs: (epochs) => set({ epochs }),
|
||||
setContextLength: (contextLength) => set({ contextLength }),
|
||||
setLearningRate: (learningRate) => set({ learningRate }),
|
||||
setLoraRank: (loraRank) => set({ loraRank }),
|
||||
setLoraAlpha: (loraAlpha) => set({ loraAlpha }),
|
||||
setLoraDropout: (loraDropout) => set({ loraDropout }),
|
||||
setLoraVariant: (loraVariant) => set({ loraVariant }),
|
||||
setBatchSize: (batchSize) => set({ batchSize }),
|
||||
setGradientAccumulation: (gradientAccumulation) =>
|
||||
set({ gradientAccumulation }),
|
||||
setWeightDecay: (weightDecay) => set({ weightDecay }),
|
||||
setWarmupSteps: (warmupSteps) => set({ warmupSteps }),
|
||||
setMaxSteps: (maxSteps) => set({ maxSteps }),
|
||||
setSaveSteps: (saveSteps) => set({ saveSteps }),
|
||||
setPacking: (packing) => set({ packing }),
|
||||
setTrainOnCompletions: (trainOnCompletions) =>
|
||||
set({ trainOnCompletions }),
|
||||
setGradientCheckpointing: (gradientCheckpointing) =>
|
||||
set({ gradientCheckpointing }),
|
||||
setRandomSeed: (randomSeed) => set({ randomSeed }),
|
||||
setEnableWandb: (enableWandb) => set({ enableWandb }),
|
||||
setWandbToken: (wandbToken) => set({ wandbToken }),
|
||||
setWandbProject: (wandbProject) => set({ wandbProject }),
|
||||
setEnableTensorboard: (enableTensorboard) => set({ enableTensorboard }),
|
||||
setTensorboardDir: (tensorboardDir) => set({ tensorboardDir }),
|
||||
setLogFrequency: (logFrequency) => set({ logFrequency }),
|
||||
setFinetuneVisionLayers: (finetuneVisionLayers) =>
|
||||
set({ finetuneVisionLayers }),
|
||||
setFinetuneLanguageLayers: (finetuneLanguageLayers) =>
|
||||
set({ finetuneLanguageLayers }),
|
||||
setFinetuneAttentionModules: (finetuneAttentionModules) =>
|
||||
set({ finetuneAttentionModules }),
|
||||
setFinetuneMLPModules: (finetuneMLPModules) => set({ finetuneMLPModules }),
|
||||
setTargetModules: (targetModules) => set({ targetModules }),
|
||||
canProceed: () => canProceedForStep(get()),
|
||||
reset: () => set(initialState),
|
||||
}),
|
||||
{
|
||||
name: "unsloth_training_config_v1",
|
||||
partialize: (state) => {
|
||||
const { modelType, ...rest } = state;
|
||||
return rest;
|
||||
},
|
||||
},
|
||||
),
|
||||
);
|
||||
|
|
@ -0,0 +1,229 @@
|
|||
import { create } from "zustand";
|
||||
import type {
|
||||
TrainingMetricsResponse,
|
||||
TrainingProgressPayload,
|
||||
TrainingRuntimeState,
|
||||
TrainingRuntimeStore,
|
||||
TrainingSeriesPoint,
|
||||
TrainingStatusResponse,
|
||||
} from "../types/runtime";
|
||||
|
||||
const initialState: TrainingRuntimeState = {
|
||||
jobId: null,
|
||||
phase: "idle",
|
||||
isTrainingRunning: false,
|
||||
message: "Ready to train",
|
||||
error: null,
|
||||
isHydrating: false,
|
||||
hasHydrated: false,
|
||||
isStarting: false,
|
||||
startError: null,
|
||||
sseConnected: false,
|
||||
firstStepReceived: false,
|
||||
lastEventId: null,
|
||||
currentStep: 0,
|
||||
totalSteps: 0,
|
||||
currentEpoch: 0,
|
||||
currentLoss: 0,
|
||||
currentLearningRate: 0,
|
||||
progressPercent: 0,
|
||||
elapsedSeconds: null,
|
||||
etaSeconds: null,
|
||||
currentGradNorm: null,
|
||||
currentNumTokens: null,
|
||||
lossHistory: [],
|
||||
lrHistory: [],
|
||||
gradNormHistory: [],
|
||||
};
|
||||
|
||||
function sortSeries(points: TrainingSeriesPoint[]): TrainingSeriesPoint[] {
|
||||
return [...points].sort((a, b) => a.step - b.step);
|
||||
}
|
||||
|
||||
function toSeries(steps: number[], values: number[]): TrainingSeriesPoint[] {
|
||||
const points: TrainingSeriesPoint[] = [];
|
||||
for (let i = 0; i < steps.length; i += 1) {
|
||||
const step = steps[i];
|
||||
const value = values[i];
|
||||
if (!Number.isFinite(step) || !Number.isFinite(value)) {
|
||||
continue;
|
||||
}
|
||||
points.push({ step, value });
|
||||
}
|
||||
return sortSeries(points);
|
||||
}
|
||||
|
||||
function upsertPoint(
|
||||
points: TrainingSeriesPoint[],
|
||||
step: number,
|
||||
value: number,
|
||||
): TrainingSeriesPoint[] {
|
||||
const next = points.slice();
|
||||
const index = next.findIndex((point) => point.step === step);
|
||||
if (index >= 0) {
|
||||
next[index] = { step, value };
|
||||
return next;
|
||||
}
|
||||
next.push({ step, value });
|
||||
return sortSeries(next);
|
||||
}
|
||||
|
||||
function applyMetricHistoryFromStatus(payload: TrainingStatusResponse): {
|
||||
lossHistory: TrainingSeriesPoint[] | null;
|
||||
lrHistory: TrainingSeriesPoint[] | null;
|
||||
} {
|
||||
const history = payload.metric_history;
|
||||
if (!history || !history.steps?.length) {
|
||||
return { lossHistory: null, lrHistory: null };
|
||||
}
|
||||
|
||||
const steps = history.steps;
|
||||
const lossHistory = history.loss ? toSeries(steps, history.loss) : null;
|
||||
const lrHistory = history.lr ? toSeries(steps, history.lr) : null;
|
||||
|
||||
return { lossHistory, lrHistory };
|
||||
}
|
||||
|
||||
export const useTrainingRuntimeStore = create<TrainingRuntimeStore>()((set) => ({
|
||||
...initialState,
|
||||
|
||||
setHydrating: (value) => set({ isHydrating: value }),
|
||||
setHasHydrated: (value) => set({ hasHydrated: value }),
|
||||
setStarting: (value) => set({ isStarting: value }),
|
||||
setStartError: (value) => set({ startError: value }),
|
||||
setSseConnected: (value) => set({ sseConnected: value }),
|
||||
setLastEventId: (value) => set({ lastEventId: value }),
|
||||
|
||||
resetRuntime: () =>
|
||||
set({
|
||||
...initialState,
|
||||
lossHistory: [],
|
||||
lrHistory: [],
|
||||
gradNormHistory: [],
|
||||
}),
|
||||
|
||||
setStartQueued: (jobId, message) =>
|
||||
set({
|
||||
jobId,
|
||||
message,
|
||||
error: null,
|
||||
startError: null,
|
||||
phase: "configuring",
|
||||
isStarting: false,
|
||||
}),
|
||||
|
||||
setRuntimeError: (message) =>
|
||||
set({
|
||||
error: message,
|
||||
phase: "error",
|
||||
isStarting: false,
|
||||
startError: null,
|
||||
sseConnected: false,
|
||||
}),
|
||||
|
||||
applyStatus: (payload) =>
|
||||
set((state) => {
|
||||
const metricHistory = applyMetricHistoryFromStatus(payload);
|
||||
const detailStep = payload.details?.step;
|
||||
const detailTotal = payload.details?.total_steps;
|
||||
const detailLoss = payload.details?.loss;
|
||||
const detailLr = payload.details?.learning_rate;
|
||||
const detailEpoch = payload.details?.epoch;
|
||||
|
||||
return {
|
||||
...state,
|
||||
jobId: payload.job_id || state.jobId,
|
||||
phase: payload.phase,
|
||||
isTrainingRunning: payload.is_training_running,
|
||||
message: payload.message,
|
||||
error: payload.error,
|
||||
startError: null,
|
||||
currentStep:
|
||||
typeof detailStep === "number" ? Math.max(detailStep, 0) : state.currentStep,
|
||||
totalSteps:
|
||||
typeof detailTotal === "number"
|
||||
? Math.max(detailTotal, 0)
|
||||
: state.totalSteps,
|
||||
currentLoss:
|
||||
typeof detailLoss === "number" ? detailLoss : state.currentLoss,
|
||||
currentLearningRate:
|
||||
typeof detailLr === "number" ? detailLr : state.currentLearningRate,
|
||||
currentEpoch:
|
||||
typeof detailEpoch === "number" ? detailEpoch : state.currentEpoch,
|
||||
lossHistory: metricHistory.lossHistory ?? state.lossHistory,
|
||||
lrHistory: metricHistory.lrHistory ?? state.lrHistory,
|
||||
};
|
||||
}),
|
||||
|
||||
applyMetrics: (payload: TrainingMetricsResponse) =>
|
||||
set((state) => {
|
||||
const lossHistory = toSeries(payload.step_history, payload.loss_history);
|
||||
const lrHistory = toSeries(payload.step_history, payload.lr_history);
|
||||
const latestStep =
|
||||
payload.current_step ??
|
||||
(payload.step_history.length > 0
|
||||
? payload.step_history[payload.step_history.length - 1]
|
||||
: null);
|
||||
|
||||
return {
|
||||
...state,
|
||||
lossHistory: lossHistory.length > 0 ? lossHistory : state.lossHistory,
|
||||
lrHistory: lrHistory.length > 0 ? lrHistory : state.lrHistory,
|
||||
currentStep:
|
||||
typeof latestStep === "number"
|
||||
? Math.max(latestStep, state.currentStep)
|
||||
: state.currentStep,
|
||||
currentLoss:
|
||||
typeof payload.current_loss === "number"
|
||||
? payload.current_loss
|
||||
: state.currentLoss,
|
||||
currentLearningRate:
|
||||
typeof payload.current_lr === "number"
|
||||
? payload.current_lr
|
||||
: state.currentLearningRate,
|
||||
};
|
||||
}),
|
||||
|
||||
applyProgress: (payload: TrainingProgressPayload, eventId?: number) =>
|
||||
set((state) => {
|
||||
const step = Math.max(payload.step, 0);
|
||||
return {
|
||||
...state,
|
||||
jobId: payload.job_id || state.jobId,
|
||||
currentStep: step,
|
||||
totalSteps: Math.max(payload.total_steps, state.totalSteps),
|
||||
currentLoss: payload.loss,
|
||||
currentLearningRate: payload.learning_rate,
|
||||
progressPercent: payload.progress_percent,
|
||||
currentEpoch: payload.epoch ?? state.currentEpoch,
|
||||
elapsedSeconds: payload.elapsed_seconds,
|
||||
etaSeconds: payload.eta_seconds,
|
||||
currentGradNorm: payload.grad_norm,
|
||||
currentNumTokens: payload.num_tokens,
|
||||
firstStepReceived: state.firstStepReceived || step > 0,
|
||||
lastEventId: typeof eventId === "number" ? eventId : state.lastEventId,
|
||||
lossHistory:
|
||||
step > 0
|
||||
? upsertPoint(state.lossHistory, step, payload.loss)
|
||||
: state.lossHistory,
|
||||
lrHistory:
|
||||
step > 0
|
||||
? upsertPoint(state.lrHistory, step, payload.learning_rate)
|
||||
: state.lrHistory,
|
||||
gradNormHistory:
|
||||
step > 0 && typeof payload.grad_norm === "number"
|
||||
? upsertPoint(state.gradNormHistory, step, payload.grad_norm)
|
||||
: state.gradNormHistory,
|
||||
};
|
||||
}),
|
||||
}));
|
||||
|
||||
export function shouldShowTrainingView(state: TrainingRuntimeStore): boolean {
|
||||
return (
|
||||
state.phase !== "idle" ||
|
||||
state.isTrainingRunning ||
|
||||
state.isStarting ||
|
||||
state.lossHistory.length > 0 ||
|
||||
state.currentStep > 0
|
||||
);
|
||||
}
|
||||
53
studio/frontend/src/features/training/types/api.ts
Normal file
53
studio/frontend/src/features/training/types/api.ts
Normal file
|
|
@ -0,0 +1,53 @@
|
|||
export interface TrainingStartRequest {
|
||||
model_name: string;
|
||||
training_type: string;
|
||||
hf_token: string | null;
|
||||
load_in_4bit: boolean;
|
||||
max_seq_length: number;
|
||||
hf_dataset: string | null;
|
||||
local_datasets: string[];
|
||||
format_type: string;
|
||||
num_epochs: number;
|
||||
learning_rate: string;
|
||||
batch_size: number;
|
||||
gradient_accumulation_steps: number;
|
||||
warmup_steps: number | null;
|
||||
warmup_ratio: number | null;
|
||||
max_steps: number | null;
|
||||
save_steps: number;
|
||||
weight_decay: number;
|
||||
random_seed: number;
|
||||
packing: boolean;
|
||||
optim: string;
|
||||
lr_scheduler_type: string;
|
||||
use_lora: boolean;
|
||||
lora_r: number;
|
||||
lora_alpha: number;
|
||||
lora_dropout: number;
|
||||
target_modules: string[];
|
||||
gradient_checkpointing: string;
|
||||
use_rslora: boolean;
|
||||
use_loftq: boolean;
|
||||
train_on_completions: boolean;
|
||||
finetune_vision_layers: boolean;
|
||||
finetune_language_layers: boolean;
|
||||
finetune_attention_modules: boolean;
|
||||
finetune_mlp_modules: boolean;
|
||||
enable_wandb: boolean;
|
||||
wandb_token: string | null;
|
||||
wandb_project: string | null;
|
||||
enable_tensorboard: boolean;
|
||||
tensorboard_dir: string | null;
|
||||
}
|
||||
|
||||
export interface TrainingStartResponse {
|
||||
job_id: string;
|
||||
status: "queued" | "error";
|
||||
message: string;
|
||||
error: string | null;
|
||||
}
|
||||
|
||||
export interface TrainingStopResponse {
|
||||
status: "stopped" | "idle";
|
||||
message: string;
|
||||
}
|
||||
96
studio/frontend/src/features/training/types/config.ts
Normal file
96
studio/frontend/src/features/training/types/config.ts
Normal file
|
|
@ -0,0 +1,96 @@
|
|||
import type {
|
||||
DatasetFormat,
|
||||
DatasetSource,
|
||||
GradientCheckpointing,
|
||||
ModelType,
|
||||
StepNumber,
|
||||
TrainingMethod,
|
||||
} from "@/types/training";
|
||||
|
||||
export type LoraVariant = "lora" | "rslora" | "loftq";
|
||||
|
||||
export interface TrainingConfigState {
|
||||
currentStep: StepNumber;
|
||||
modelType: ModelType | null;
|
||||
selectedModel: string | null;
|
||||
trainingMethod: TrainingMethod;
|
||||
hfToken: string;
|
||||
datasetSource: DatasetSource;
|
||||
datasetFormat: DatasetFormat;
|
||||
dataset: string | null;
|
||||
uploadedFile: string | null;
|
||||
epochs: number;
|
||||
contextLength: number;
|
||||
learningRate: number;
|
||||
loraRank: number;
|
||||
loraAlpha: number;
|
||||
loraDropout: number;
|
||||
loraVariant: LoraVariant;
|
||||
batchSize: number;
|
||||
gradientAccumulation: number;
|
||||
weightDecay: number;
|
||||
warmupSteps: number;
|
||||
maxSteps: number;
|
||||
saveSteps: number;
|
||||
packing: boolean;
|
||||
trainOnCompletions: boolean;
|
||||
gradientCheckpointing: GradientCheckpointing;
|
||||
randomSeed: number;
|
||||
enableWandb: boolean;
|
||||
wandbToken: string;
|
||||
wandbProject: string;
|
||||
enableTensorboard: boolean;
|
||||
tensorboardDir: string;
|
||||
logFrequency: number;
|
||||
finetuneVisionLayers: boolean;
|
||||
finetuneLanguageLayers: boolean;
|
||||
finetuneAttentionModules: boolean;
|
||||
finetuneMLPModules: boolean;
|
||||
targetModules: string[];
|
||||
}
|
||||
|
||||
export interface TrainingConfigActions {
|
||||
setStep: (step: StepNumber) => void;
|
||||
nextStep: () => void;
|
||||
prevStep: () => void;
|
||||
setModelType: (type: ModelType) => void;
|
||||
setSelectedModel: (model: string | null) => void;
|
||||
setTrainingMethod: (method: TrainingMethod) => void;
|
||||
setHfToken: (token: string) => void;
|
||||
setDatasetSource: (source: DatasetSource) => void;
|
||||
setDatasetFormat: (format: DatasetFormat) => void;
|
||||
setDataset: (dataset: string | null) => void;
|
||||
setUploadedFile: (file: string | null) => void;
|
||||
setEpochs: (epochs: number) => void;
|
||||
setContextLength: (length: number) => void;
|
||||
setLearningRate: (rate: number) => void;
|
||||
setLoraRank: (rank: number) => void;
|
||||
setLoraAlpha: (alpha: number) => void;
|
||||
setLoraDropout: (dropout: number) => void;
|
||||
setLoraVariant: (variant: LoraVariant) => void;
|
||||
setBatchSize: (value: number) => void;
|
||||
setGradientAccumulation: (value: number) => void;
|
||||
setWeightDecay: (value: number) => void;
|
||||
setWarmupSteps: (value: number) => void;
|
||||
setMaxSteps: (value: number) => void;
|
||||
setSaveSteps: (value: number) => void;
|
||||
setPacking: (value: boolean) => void;
|
||||
setTrainOnCompletions: (value: boolean) => void;
|
||||
setGradientCheckpointing: (value: GradientCheckpointing) => void;
|
||||
setRandomSeed: (value: number) => void;
|
||||
setEnableWandb: (value: boolean) => void;
|
||||
setWandbToken: (value: string) => void;
|
||||
setWandbProject: (value: string) => void;
|
||||
setEnableTensorboard: (value: boolean) => void;
|
||||
setTensorboardDir: (value: string) => void;
|
||||
setLogFrequency: (value: number) => void;
|
||||
setFinetuneVisionLayers: (value: boolean) => void;
|
||||
setFinetuneLanguageLayers: (value: boolean) => void;
|
||||
setFinetuneAttentionModules: (value: boolean) => void;
|
||||
setFinetuneMLPModules: (value: boolean) => void;
|
||||
setTargetModules: (value: string[]) => void;
|
||||
canProceed: () => boolean;
|
||||
reset: () => void;
|
||||
}
|
||||
|
||||
export type TrainingConfigStore = TrainingConfigState & TrainingConfigActions;
|
||||
102
studio/frontend/src/features/training/types/runtime.ts
Normal file
102
studio/frontend/src/features/training/types/runtime.ts
Normal file
|
|
@ -0,0 +1,102 @@
|
|||
export type TrainingPhase =
|
||||
| "idle"
|
||||
| "loading_model"
|
||||
| "loading_dataset"
|
||||
| "configuring"
|
||||
| "training"
|
||||
| "completed"
|
||||
| "error"
|
||||
| "stopped";
|
||||
|
||||
export interface TrainingStatusResponse {
|
||||
job_id: string;
|
||||
phase: TrainingPhase;
|
||||
is_training_running: boolean;
|
||||
message: string;
|
||||
error: string | null;
|
||||
details?: {
|
||||
epoch?: number;
|
||||
step?: number;
|
||||
total_steps?: number;
|
||||
loss?: number;
|
||||
learning_rate?: number;
|
||||
} | null;
|
||||
metric_history?: {
|
||||
steps?: number[];
|
||||
loss?: number[];
|
||||
lr?: number[];
|
||||
} | null;
|
||||
}
|
||||
|
||||
export interface TrainingMetricsResponse {
|
||||
loss_history: number[];
|
||||
lr_history: number[];
|
||||
step_history: number[];
|
||||
current_loss: number | null;
|
||||
current_lr: number | null;
|
||||
current_step: number | null;
|
||||
}
|
||||
|
||||
export interface TrainingProgressPayload {
|
||||
job_id: string;
|
||||
step: number;
|
||||
total_steps: number;
|
||||
loss: number;
|
||||
learning_rate: number;
|
||||
progress_percent: number;
|
||||
epoch: number | null;
|
||||
elapsed_seconds: number | null;
|
||||
eta_seconds: number | null;
|
||||
grad_norm: number | null;
|
||||
num_tokens: number | null;
|
||||
}
|
||||
|
||||
export interface TrainingSeriesPoint {
|
||||
step: number;
|
||||
value: number;
|
||||
}
|
||||
|
||||
export interface TrainingRuntimeState {
|
||||
jobId: string | null;
|
||||
phase: TrainingPhase;
|
||||
isTrainingRunning: boolean;
|
||||
message: string;
|
||||
error: string | null;
|
||||
isHydrating: boolean;
|
||||
hasHydrated: boolean;
|
||||
isStarting: boolean;
|
||||
startError: string | null;
|
||||
sseConnected: boolean;
|
||||
firstStepReceived: boolean;
|
||||
lastEventId: number | null;
|
||||
currentStep: number;
|
||||
totalSteps: number;
|
||||
currentEpoch: number;
|
||||
currentLoss: number;
|
||||
currentLearningRate: number;
|
||||
progressPercent: number;
|
||||
elapsedSeconds: number | null;
|
||||
etaSeconds: number | null;
|
||||
currentGradNorm: number | null;
|
||||
currentNumTokens: number | null;
|
||||
lossHistory: TrainingSeriesPoint[];
|
||||
lrHistory: TrainingSeriesPoint[];
|
||||
gradNormHistory: TrainingSeriesPoint[];
|
||||
}
|
||||
|
||||
export interface TrainingRuntimeActions {
|
||||
setHydrating: (value: boolean) => void;
|
||||
setHasHydrated: (value: boolean) => void;
|
||||
setStarting: (value: boolean) => void;
|
||||
setStartError: (value: string | null) => void;
|
||||
setSseConnected: (value: boolean) => void;
|
||||
setLastEventId: (value: number | null) => void;
|
||||
resetRuntime: () => void;
|
||||
applyStatus: (payload: TrainingStatusResponse) => void;
|
||||
applyMetrics: (payload: TrainingMetricsResponse) => void;
|
||||
applyProgress: (payload: TrainingProgressPayload, eventId?: number) => void;
|
||||
setStartQueued: (jobId: string, message: string) => void;
|
||||
setRuntimeError: (message: string) => void;
|
||||
}
|
||||
|
||||
export type TrainingRuntimeStore = TrainingRuntimeState & TrainingRuntimeActions;
|
||||
|
|
@ -23,6 +23,28 @@ const EXCLUDED_TAGS = new Set([
|
|||
"ctranslate2",
|
||||
]);
|
||||
|
||||
function withPopularitySort(
|
||||
input: Parameters<typeof fetch>[0],
|
||||
init?: Parameters<typeof fetch>[1],
|
||||
): ReturnType<typeof fetch> {
|
||||
const rawUrl =
|
||||
typeof input === "string"
|
||||
? input
|
||||
: input instanceof URL
|
||||
? input.toString()
|
||||
: input.url;
|
||||
const url = new URL(rawUrl);
|
||||
|
||||
if (!url.searchParams.has("sort")) {
|
||||
url.searchParams.set("sort", "downloads");
|
||||
}
|
||||
if (!url.searchParams.has("direction")) {
|
||||
url.searchParams.set("direction", "-1");
|
||||
}
|
||||
|
||||
return fetch(url, init);
|
||||
}
|
||||
|
||||
function mapModel(raw: unknown): HfModelResult | null {
|
||||
const m = raw as {
|
||||
name: string;
|
||||
|
|
@ -53,10 +75,10 @@ export function useHfModelSearch(
|
|||
listModels({
|
||||
search: {
|
||||
...(query.trim() ? { query } : { owner: "unsloth" }),
|
||||
tags: ["transformers"],
|
||||
...(task ? { task } : {}),
|
||||
},
|
||||
additionalFields: ["safetensors", "tags"],
|
||||
fetch: withPopularitySort,
|
||||
...(accessToken ? { credentials: { accessToken } } : {}),
|
||||
}) as AsyncGenerator<unknown>,
|
||||
[query, task, accessToken],
|
||||
|
|
|
|||
|
|
@ -1,106 +1,4 @@
|
|||
import { DEFAULT_HYPERPARAMS } from "@/config/training";
|
||||
import type { StepNumber, WizardActions, WizardState } from "@/types/training";
|
||||
import { create } from "zustand";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
|
||||
const MIN_STEP: StepNumber = 1;
|
||||
const MAX_STEP: StepNumber = 5;
|
||||
|
||||
const initialState: WizardState = {
|
||||
isTraining: false,
|
||||
trainingMetrics: null,
|
||||
currentStep: MIN_STEP,
|
||||
modelType: null,
|
||||
selectedModel: null,
|
||||
trainingMethod: "qlora",
|
||||
hfToken: "",
|
||||
datasetSource: "huggingface",
|
||||
datasetFormat: "auto",
|
||||
dataset: null,
|
||||
uploadedFile: null,
|
||||
...DEFAULT_HYPERPARAMS,
|
||||
};
|
||||
|
||||
function clampStep(step: number): StepNumber {
|
||||
return Math.min(MAX_STEP, Math.max(MIN_STEP, step)) as StepNumber;
|
||||
}
|
||||
|
||||
function canProceedForStep(state: WizardState): boolean {
|
||||
switch (state.currentStep) {
|
||||
case 1:
|
||||
return state.modelType !== null;
|
||||
case 2:
|
||||
return state.selectedModel !== null;
|
||||
case 3: {
|
||||
if (state.datasetSource === "upload") {
|
||||
return state.uploadedFile !== null;
|
||||
}
|
||||
return state.dataset !== null;
|
||||
}
|
||||
case 4:
|
||||
case 5:
|
||||
return true;
|
||||
default:
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
export const useWizardStore = create<WizardState & WizardActions>(
|
||||
(set, get) => ({
|
||||
...initialState,
|
||||
|
||||
setStep: (step) => set({ currentStep: step }),
|
||||
|
||||
nextStep: () => {
|
||||
const { currentStep } = get();
|
||||
set({ currentStep: clampStep(currentStep + 1) });
|
||||
},
|
||||
|
||||
prevStep: () => {
|
||||
const { currentStep } = get();
|
||||
set({ currentStep: clampStep(currentStep - 1) });
|
||||
},
|
||||
|
||||
setModelType: (type) => set({ modelType: type, selectedModel: null }),
|
||||
setSelectedModel: (model) => set({ selectedModel: model }),
|
||||
setTrainingMethod: (method) => set({ trainingMethod: method }),
|
||||
setHfToken: (token) => set({ hfToken: token }),
|
||||
setDatasetSource: (source) => set({ datasetSource: source }),
|
||||
setDatasetFormat: (format) => set({ datasetFormat: format }),
|
||||
setDataset: (dataset) => set({ dataset }),
|
||||
setUploadedFile: (file) => set({ uploadedFile: file }),
|
||||
setEpochs: (epochs) => set({ epochs }),
|
||||
setContextLength: (length) => set({ contextLength: length }),
|
||||
setLearningRate: (rate) => set({ learningRate: rate }),
|
||||
setLoraRank: (rank) => set({ loraRank: rank }),
|
||||
setLoraAlpha: (alpha) => set({ loraAlpha: alpha }),
|
||||
setLoraDropout: (dropout) => set({ loraDropout: dropout }),
|
||||
setLoraVariant: (v) => set({ loraVariant: v }),
|
||||
setBatchSize: (v) => set({ batchSize: v }),
|
||||
setGradientAccumulation: (v) => set({ gradientAccumulation: v }),
|
||||
setWeightDecay: (v) => set({ weightDecay: v }),
|
||||
setWarmupSteps: (v) => set({ warmupSteps: v }),
|
||||
setMaxSteps: (v) => set({ maxSteps: v }),
|
||||
setSaveSteps: (v) => set({ saveSteps: v }),
|
||||
setPacking: (v) => set({ packing: v }),
|
||||
setTrainOnCompletions: (v) => set({ trainOnCompletions: v }),
|
||||
setGradientCheckpointing: (v) => set({ gradientCheckpointing: v }),
|
||||
setRandomSeed: (v) => set({ randomSeed: v }),
|
||||
setEnableWandb: (v) => set({ enableWandb: v }),
|
||||
setWandbToken: (v) => set({ wandbToken: v }),
|
||||
setWandbProject: (v) => set({ wandbProject: v }),
|
||||
setEnableTensorboard: (v) => set({ enableTensorboard: v }),
|
||||
setTensorboardDir: (v) => set({ tensorboardDir: v }),
|
||||
setLogFrequency: (v) => set({ logFrequency: v }),
|
||||
setFinetuneVisionLayers: (v) => set({ finetuneVisionLayers: v }),
|
||||
setFinetuneLanguageLayers: (v) => set({ finetuneLanguageLayers: v }),
|
||||
setFinetuneAttentionModules: (v) => set({ finetuneAttentionModules: v }),
|
||||
setFinetuneMLPModules: (v) => set({ finetuneMLPModules: v }),
|
||||
setTargetModules: (v) => set({ targetModules: v }),
|
||||
setIsTraining: (v) => set({ isTraining: v }),
|
||||
setTrainingMetrics: (v) => set({ trainingMetrics: v }),
|
||||
|
||||
canProceed: () => canProceedForStep(get()),
|
||||
|
||||
reset: () => set(initialState),
|
||||
}),
|
||||
);
|
||||
export const useWizardStore = useTrainingConfigStore;
|
||||
export { useTrainingConfigStore };
|
||||
|
|
|
|||
|
|
@ -9,30 +9,7 @@ export type DatasetSource = "huggingface" | "upload";
|
|||
export type DatasetFormat = "auto" | "alpaca" | "chatml" | "sharegpt";
|
||||
export type GradientCheckpointing = "none" | "true" | "unsloth";
|
||||
|
||||
export interface TrainingMetrics {
|
||||
currentStep: number;
|
||||
totalSteps: number;
|
||||
currentEpoch: number;
|
||||
totalEpochs: number;
|
||||
currentLoss: number;
|
||||
currentLR: number;
|
||||
gradNorm: number;
|
||||
samplesPerSecond: number;
|
||||
lossHistory: { step: number; loss: number }[];
|
||||
lrHistory: { step: number; lr: number }[];
|
||||
gradNormHistory: { step: number; gradNorm: number }[];
|
||||
gpuUtil: number;
|
||||
gpuTemp: number;
|
||||
gpuVramUsed: number;
|
||||
gpuVramTotal: number;
|
||||
gpuPower: number;
|
||||
elapsed: number;
|
||||
status: "training" | "warmup" | "saving";
|
||||
}
|
||||
|
||||
export interface WizardState {
|
||||
isTraining: boolean;
|
||||
trainingMetrics: TrainingMetrics | null;
|
||||
currentStep: StepNumber;
|
||||
modelType: ModelType | null;
|
||||
selectedModel: string | null;
|
||||
|
|
@ -112,8 +89,6 @@ export interface WizardActions {
|
|||
setFinetuneAttentionModules: (v: boolean) => void;
|
||||
setFinetuneMLPModules: (v: boolean) => void;
|
||||
setTargetModules: (v: string[]) => void;
|
||||
setIsTraining: (v: boolean) => void;
|
||||
setTrainingMetrics: (v: TrainingMetrics | null) => void;
|
||||
canProceed: () => boolean;
|
||||
reset: () => void;
|
||||
}
|
||||
|
|
|
|||
|
|
@ -6,18 +6,43 @@ import { defineConfig } from "vite";
|
|||
// https://vite.dev/config/
|
||||
export default defineConfig({
|
||||
plugins: [react(), tailwindcss()],
|
||||
optimizeDeps: {
|
||||
include: ["@dagrejs/dagre", "@dagrejs/graphlib"],
|
||||
},
|
||||
server: {
|
||||
allowedHosts: ["playground.wasimhub.dev"],
|
||||
host: "0.0.0.0",
|
||||
allowedHosts: true,
|
||||
proxy: {
|
||||
"/api": {
|
||||
target: "http://127.0.0.1:8000",
|
||||
changeOrigin: true,
|
||||
},
|
||||
"/preview": {
|
||||
target: "http://127.0.0.1:8004",
|
||||
changeOrigin: true,
|
||||
},
|
||||
"/validate": {
|
||||
target: "http://127.0.0.1:8004",
|
||||
changeOrigin: true,
|
||||
},
|
||||
"/tools": {
|
||||
target: "http://127.0.0.1:8004",
|
||||
changeOrigin: true,
|
||||
},
|
||||
},
|
||||
},
|
||||
resolve: {
|
||||
alias: {
|
||||
"@": path.resolve(__dirname, "./src"),
|
||||
"@dagrejs/dagre": path.resolve(
|
||||
__dirname,
|
||||
"./node_modules/@dagrejs/dagre/dist/dagre.cjs.js",
|
||||
),
|
||||
},
|
||||
},
|
||||
build: {
|
||||
commonjsOptions: {
|
||||
include: [/node_modules/, /@dagrejs\/dagre/, /@dagrejs\/graphlib/],
|
||||
},
|
||||
},
|
||||
});
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue