diff --git a/datasets/recipes/recipe_a28d33f9eab14660bba2437a6dfda290/builder_config.json b/datasets/recipes/recipe_a28d33f9eab14660bba2437a6dfda290/builder_config.json new file mode 100644 index 0000000000..c204eab0fa --- /dev/null +++ b/datasets/recipes/recipe_a28d33f9eab14660bba2437a6dfda290/builder_config.json @@ -0,0 +1,72 @@ +{ + "data_designer": { + "columns": [ + { + "name": "generated_instruction", + "drop": false, + "allow_resize": false, + "column_type": "llm-text", + "prompt": "Based on this target answer:\n{{ output }}\n\nWrite one high-quality plain text short and brief user instruction that this answer would satisfy.\nReturn only the instruction.", + "model_alias": "ministral", + "system_prompt": null, + "multi_modal_context": null, + "tool_alias": null, + "with_trace": "none", + "extract_reasoning_content": false + }, + { + "name": "instruction", + "drop": false, + "allow_resize": false, + "column_type": "seed-dataset" + }, + { + "name": "input", + "drop": false, + "allow_resize": false, + "column_type": "seed-dataset" + }, + { + "name": "output", + "drop": false, + "allow_resize": false, + "column_type": "seed-dataset" + } + ], + "model_configs": [ + { + "alias": "ministral", + "model": "mistralai/ministral-8b-2512", + "inference_parameters": { + "generation_type": "chat-completion", + "max_parallel_requests": 4, + "timeout": null, + "extra_body": null, + "temperature": 0.7, + "top_p": null, + "max_tokens": 1024 + }, + "provider": "openai_provider", + "skip_health_check": false + } + ], + "tool_configs": [], + "seed_config": { + "source": { + "seed_type": "hf", + "path": "datasets/unsloth/alpaca-cleaned/**/*.json", + "token": null, + "endpoint": "https://huggingface.co" + }, + "sampling_strategy": "ordered", + "selection_strategy": { + "start": 1, + "end": 100 + } + }, + "constraints": null, + "profilers": null, + "processors": null + }, + "library_version": "0.5.1" +} \ No newline at end of file diff --git a/datasets/recipes/recipe_a28d33f9eab14660bba2437a6dfda290/metadata.json b/datasets/recipes/recipe_a28d33f9eab14660bba2437a6dfda290/metadata.json new file mode 100644 index 0000000000..4b84f8d0f4 --- /dev/null +++ b/datasets/recipes/recipe_a28d33f9eab14660bba2437a6dfda290/metadata.json @@ -0,0 +1,36 @@ +{ + "actual_num_records": 50, + "buffer_size": 50, + "column_statistics": [ + { + "column_name": "generated_instruction", + "column_type": "llm-text", + "input_tokens_mean": 170.52, + "input_tokens_median": 132.5, + "input_tokens_stddev": 119.91, + "num_null": 0, + "num_records": 50, + "num_unique": 50, + "output_tokens_mean": 39.74, + "output_tokens_median": 31.0, + "output_tokens_stddev": 39.58, + "pyarrow_dtype": "string", + "simple_dtype": "string" + } + ], + "dataset_name": "recipe_a28d33f9eab14660bba2437a6dfda290", + "file_paths": { + "parquet-files": [ + "parquet-files/batch_00000.parquet" + ] + }, + "num_completed_batches": 1, + "schema": { + "generated_instruction": "string", + "input": "string", + "instruction": "string", + "output": "string" + }, + "target_num_records": 50, + "total_num_batches": 1 +} \ No newline at end of file diff --git a/datasets/recipes/recipe_a28d33f9eab14660bba2437a6dfda290/parquet-files/batch_00000.parquet b/datasets/recipes/recipe_a28d33f9eab14660bba2437a6dfda290/parquet-files/batch_00000.parquet new file mode 100644 index 0000000000..06eebafa26 Binary files /dev/null and b/datasets/recipes/recipe_a28d33f9eab14660bba2437a6dfda290/parquet-files/batch_00000.parquet differ diff --git a/studio/frontend/bun.lock b/studio/frontend/bun.lock index b4c447145a..7f1491f5fe 100644 --- a/studio/frontend/bun.lock +++ b/studio/frontend/bun.lock @@ -40,9 +40,16 @@ "clsx": "^2.1.1", "cmdk": "^1.1.1", "date-fns": "^4.1.0", +<<<<<<< HEAD "dexie": "^4.3.0", "framer-motion": "^11.18.2", "katex": "^0.16.28", +======= + "dexie": "^4.2.1", + "framer-motion": "^11.15.0", + "js-yaml": "^4.1.1", + "katex": "^0.16.22", +>>>>>>> origin/nightly "lucide-react": "^0.563.0", "mammoth": "^1.11.0", "motion": "^12.34.0", @@ -68,6 +75,7 @@ "devDependencies": { "@biomejs/biome": "^1.9.4", "@eslint/js": "^9.39.1", + "@types/js-yaml": "^4.0.9", "@types/node": "^24.10.1", "@types/react": "^19.2.5", "@types/react-dom": "^19.2.3", @@ -767,6 +775,8 @@ "@types/hast": ["@types/hast@3.0.4", "", { "dependencies": { "@types/unist": "*" } }, "sha512-WPs+bbQw5aCj+x6laNGWLH3wviHtoCv/P3+otBhbOhJgG8qtpdAMlTCxLtsTWA7LH1Oh/bFCHsBn0TPS5m30EQ=="], + "@types/js-yaml": ["@types/js-yaml@4.0.9", "", {}, "sha512-k4MGaQl5TGo/iipqb2UDG2UwjXziSWkh0uysQelTlJpX1qGlpUZYm8PnO4DxG1qBomtJUdYJ6qR6xdIah10JLg=="], + "@types/json-schema": ["@types/json-schema@7.0.15", "", {}, "sha512-5+fP8P8MFNC+AyZCDxrB2pkZFPGzqQWUzpSeuuVLvm8VMcorNYavBqoFcxK8bQz4Qsbn4oUEEem4wDLfcysGHA=="], "@types/katex": ["@types/katex@0.16.8", "", {}, "sha512-trgaNyfU+Xh2Tc+ABIb44a5AYUpicB3uwirOioeOkNPPbmgRNtcWyDeeFRzjPZENO9Vq8gvVqfhaaXWLlevVwg=="], @@ -841,7 +851,7 @@ "ansis": ["ansis@4.2.0", "", {}, "sha512-HqZ5rWlFjGiV0tDm3UxxgNRqsOTniqoKZu0pIAfh7TZQMGuZK+hH0drySty0si0QXj1ieop4+SkSfPZBPPkHig=="], - "argparse": ["argparse@1.0.10", "", { "dependencies": { "sprintf-js": "~1.0.2" } }, "sha512-o5Roy6tNG4SL/FOkCAN6RzjiakZS25RLYFrcMttJqbdd8BWrnA+fGz57iN5Pb06pvBGvl5gQ0B48dJlslXvoTg=="], + "argparse": ["argparse@2.0.1", "", {}, "sha512-8+9WqebbFzpX9OR+Wa6O29asIogeRMzcGtAINdpMHHyAg10f05aSFVBbcEqGf/PXw1EjAZ+q2/bEBg3DvurK3Q=="], "aria-hidden": ["aria-hidden@1.2.6", "", { "dependencies": { "tslib": "^2.0.0" } }, "sha512-ik3ZgC9dY/lYVVM++OISsaYDeg1tb0VtP5uL3ouh1koGOaUMDPpbFIei4JkFimWUFPn90sbMNMXQAIVOlnYKJA=="], @@ -2301,16 +2311,21 @@ "fast-glob/glob-parent": ["glob-parent@5.1.2", "", { "dependencies": { "is-glob": "^4.0.1" } }, "sha512-AOIgSQCepiJYwP3ARnGx+5VnTu2HBYdzbGP45eLw1vr3zB3vZLeyed1sC9hnbcOc9/SrMyM5RPQrkGz4aS9Zow=="], +<<<<<<< HEAD "js-yaml/argparse": ["argparse@2.0.1", "", {}, "sha512-8+9WqebbFzpX9OR+Wa6O29asIogeRMzcGtAINdpMHHyAg10f05aSFVBbcEqGf/PXw1EjAZ+q2/bEBg3DvurK3Q=="], "langsmith/chalk": ["chalk@5.6.2", "", {}, "sha512-7NzBL0rN6fMUW+f7A6Io4h40qQlG+xGmtMxfbnH/K7TAtt8JQWVQK+6g0UXKMeVJoyV5EkkNsErQ8pVD3bLHbA=="], "langsmith/semver": ["semver@7.7.4", "", { "bin": { "semver": "bin/semver.js" } }, "sha512-vFKC2IEtQnVhpT78h1Yp8wzwrf8CM+MzKMHGJZfBtzhZNycRFnXsHk6E5TxIkkMsgNS7mdX3AGB7x2QM2di4lA=="], +======= +>>>>>>> origin/nightly "log-symbols/chalk": ["chalk@5.6.2", "", {}, "sha512-7NzBL0rN6fMUW+f7A6Io4h40qQlG+xGmtMxfbnH/K7TAtt8JQWVQK+6g0UXKMeVJoyV5EkkNsErQ8pVD3bLHbA=="], "log-symbols/is-unicode-supported": ["is-unicode-supported@1.3.0", "", {}, "sha512-43r2mRvz+8JRIKnWJ+3j8JtjRKZ6GmjzfaE/qiBJnikNnYv/6bagRJ1kUhNk8R5EX/GkobD+r+sfxCPJsiKBLQ=="], + "mammoth/argparse": ["argparse@1.0.10", "", { "dependencies": { "sprintf-js": "~1.0.2" } }, "sha512-o5Roy6tNG4SL/FOkCAN6RzjiakZS25RLYFrcMttJqbdd8BWrnA+fGz57iN5Pb06pvBGvl5gQ0B48dJlslXvoTg=="], + "mdast-util-find-and-replace/escape-string-regexp": ["escape-string-regexp@5.0.0", "", {}, "sha512-/veY75JbMK4j1yjvuUxuVsiS/hr/4iHs9FTT6cgTexxdE0Ly/glccBAkloH/DofkjRbZU3bnoj38mOmhkZ0lHw=="], "mermaid/marked": ["marked@16.4.2", "", { "bin": { "marked": "bin/marked.js" } }, "sha512-TI3V8YYWvkVf3KJe1dRkpnjs68JUPyEa5vjKrp1XEEJUAOaQc+Qj+L1qWbPd0SJuAdQkFU0h73sXXqwDYxsiDA=="], diff --git a/studio/frontend/package.json b/studio/frontend/package.json index 4b9f8350bb..4ad2f3be60 100644 --- a/studio/frontend/package.json +++ b/studio/frontend/package.json @@ -50,6 +50,7 @@ "date-fns": "^4.1.0", "dexie": "^4.3.0", "framer-motion": "^11.18.2", + "js-yaml": "^4.1.1", "katex": "^0.16.28", "lucide-react": "^0.563.0", "mammoth": "^1.11.0", @@ -76,6 +77,7 @@ "devDependencies": { "@biomejs/biome": "^1.9.4", "@eslint/js": "^9.39.1", + "@types/js-yaml": "^4.0.9", "@types/node": "^24.10.1", "@types/react": "^19.2.5", "@types/react-dom": "^19.2.3", @@ -88,4 +90,4 @@ "typescript-eslint": "^8.55.0", "vite": "^7.3.1" } -} +} \ No newline at end of file diff --git a/studio/frontend/src/features/studio/sections/training-section.tsx b/studio/frontend/src/features/studio/sections/training-section.tsx index a8f8a9c475..3f342ad68c 100644 --- a/studio/frontend/src/features/studio/sections/training-section.tsx +++ b/studio/frontend/src/features/studio/sections/training-section.tsx @@ -2,23 +2,27 @@ import { SectionCard } from "@/components/section-card"; import { Button } from "@/components/ui/button"; import { ChartContainer } from "@/components/ui/chart"; import type { ChartConfig } from "@/components/ui/chart"; -import { Checkbox } from "@/components/ui/checkbox"; import { - Collapsible, - CollapsibleContent, - CollapsibleTrigger, -} from "@/components/ui/collapsible"; -import { Input } from "@/components/ui/input"; -import { useTrainingActions, useTrainingConfigStore } from "@/features/training"; + Tooltip, + TooltipContent, + TooltipTrigger, +} from "@/components/ui/tooltip"; +import { + parseYamlConfig, + serializeConfigToYaml, + useTrainingActions, + useTrainingConfigStore, +} from "@/features/training"; import { Archive04Icon, - ArrowDown01Icon, ChartAverageIcon, CleanIcon, + CloudUploadIcon, Rocket01Icon, } from "@hugeicons/core-free-icons"; import { HugeiconsIcon } from "@hugeicons/react"; -import { useState } from "react"; +import { useRef } from "react"; +import { toast } from "sonner"; import { CartesianGrid, Line, LineChart, XAxis, YAxis } from "recharts"; const chartConfig = { @@ -37,10 +41,55 @@ const placeholderData = [ export function TrainingSection() { const store = useTrainingConfigStore(); const { isStarting, startError, startTrainingRun } = useTrainingActions(); - const [logOpen, setLogOpen] = useState(false); const isIncompatible = !store.isVisionModel && store.isDatasetMultimodal === true; + const fileInputRef = useRef(null); + const handleFileUpload = (e: React.ChangeEvent) => { + const file = e.target.files?.[0]; + if (!file) return; + e.target.value = ""; + + const reader = new FileReader(); + reader.onload = () => { + try { + const config = parseYamlConfig(reader.result as string); + store.applyConfigPatch(config); + toast.success("Config loaded", { description: file.name }); + } catch (err) { + toast.error("Failed to load config", { + description: + err instanceof Error ? err.message : "Invalid YAML file", + }); + } + }; + reader.onerror = () => { + toast.error("Failed to read file"); + }; + reader.readAsText(file); + }; + + const handleSaveConfig = () => { + const yamlStr = serializeConfigToYaml(store, store.isVisionModel); + const blob = new Blob([yamlStr], { type: "text/yaml" }); + const url = URL.createObjectURL(blob); + const a = document.createElement("a"); + a.href = url; + + const model = (store.selectedModel ?? "model").split("/").pop(); + const method = store.trainingMethod ?? "qlora"; + const dataset = (store.dataset ?? "dataset").split("/").pop(); + const timestamp = new Date().toISOString().replace(/[:T]/g, "-").slice(0, 19); + a.download = `${model}_${method}_${dataset}_${timestamp}.yaml`; + + a.click(); + URL.revokeObjectURL(url); + }; + + const handleResetConfig = () => { + store.resetToModelDefaults(); + toast.success("Parameters reset to model defaults"); + }; return (
@@ -115,100 +164,61 @@ export function TrainingSection() {

)} - {/* Save / Clear */} -
- - + {/* Upload / Save / Reset */} +

Training Config

+
+ + + + + Load a saved YAML config + + + + + + Download current config as YAML + + + + + + Reset to model defaults +
- - {/* Logging */} - - - - Logging - - - {/* W&B */} -
- store.setEnableWandb(!!v)} - /> - -
- {store.enableWandb && ( -
- store.setWandbToken(e.target.value)} - /> - store.setWandbProject(e.target.value)} - /> -
- )} - - {/* TensorBoard */} -
- store.setEnableTensorboard(!!v)} - /> - -
- {store.enableTensorboard && ( -
- store.setTensorboardDir(e.target.value)} - /> -
- - Log frequency - - - store.setLogFrequency(Number(e.target.value)) - } - className="w-24" - /> -
-
- )} - - +
diff --git a/studio/frontend/src/features/studio/tour/steps/save.tsx b/studio/frontend/src/features/studio/tour/steps/save.tsx index d137aecf41..d93548fc3d 100644 --- a/studio/frontend/src/features/studio/tour/steps/save.tsx +++ b/studio/frontend/src/features/studio/tour/steps/save.tsx @@ -6,8 +6,8 @@ export const studioSaveStep: TourStep = { title: "Save config", body: ( <> - Save configs that worked. Re-running the same baseline makes it obvious - if a change helped (or if you just got lucky). + Save your training config as a YAML file. Re-running the same baseline + makes it obvious if a change helped (or if you just got lucky). ), }; diff --git a/studio/frontend/src/features/training/index.ts b/studio/frontend/src/features/training/index.ts index 24439f0c8a..d2fd4f2e43 100644 --- a/studio/frontend/src/features/training/index.ts +++ b/studio/frontend/src/features/training/index.ts @@ -10,3 +10,4 @@ export { useDatasetPreviewDialogStore } from "./stores/dataset-preview-dialog-st export { listLocalModels } from "./api/models-api"; export type { LocalModelInfo } from "./api/models-api"; export type { TrainingPhase } from "./types/runtime"; +export { parseYamlConfig, serializeConfigToYaml } from "./lib/yaml-config"; diff --git a/studio/frontend/src/features/training/lib/yaml-config.ts b/studio/frontend/src/features/training/lib/yaml-config.ts new file mode 100644 index 0000000000..3bc113f173 --- /dev/null +++ b/studio/frontend/src/features/training/lib/yaml-config.ts @@ -0,0 +1,81 @@ +import * as yaml from "js-yaml"; +import type { BackendModelConfig } from "../api/models-api"; +import type { TrainingConfigState } from "../types/config"; + +const EXPECTED_TOP_KEYS = new Set(["training", "lora", "logging", "inference"]); + +/** + * Parse a YAML string into a BackendModelConfig suitable for + * `mapBackendModelConfigToTrainingPatch`. Throws on invalid input. + */ +export function parseYamlConfig(text: string): BackendModelConfig { + const parsed = yaml.load(text); + if (parsed == null || typeof parsed !== "object" || Array.isArray(parsed)) { + throw new Error( + "Invalid config: expected a YAML mapping with training/lora/logging sections", + ); + } + + const raw = parsed as Record; + const unknownKeys = Object.keys(raw).filter( + (k) => !EXPECTED_TOP_KEYS.has(k), + ); + if (unknownKeys.length > 0) { + console.warn("Ignored unknown YAML keys:", unknownKeys.join(", ")); + } + + return { + training: (raw.training ?? undefined) as BackendModelConfig["training"], + lora: (raw.lora ?? undefined) as BackendModelConfig["lora"], + logging: (raw.logging ?? undefined) as BackendModelConfig["logging"], + }; +} + +/** + * Serialize the current training config state to a YAML string matching the + * backend model-defaults schema. + */ +export function serializeConfigToYaml( + state: TrainingConfigState, + includeVisionFields: boolean, +): string { + const lora: Record = { + lora_r: state.loraRank, + lora_alpha: state.loraAlpha, + lora_dropout: state.loraDropout, + target_modules: state.targetModules, + use_rslora: state.loraVariant === "rslora", + use_loftq: state.loraVariant === "loftq", + }; + + if (includeVisionFields) { + lora.finetune_vision_layers = state.finetuneVisionLayers; + lora.finetune_language_layers = state.finetuneLanguageLayers; + lora.finetune_attention_modules = state.finetuneAttentionModules; + lora.finetune_mlp_modules = state.finetuneMLPModules; + } + + const config = { + training: { + max_seq_length: state.contextLength, + num_epochs: state.epochs, + learning_rate: state.learningRate, + batch_size: state.batchSize, + gradient_accumulation_steps: state.gradientAccumulation, + warmup_steps: state.warmupSteps, + max_steps: state.maxSteps, + save_steps: state.saveSteps, + eval_steps: state.evalSteps, + weight_decay: state.weightDecay, + random_seed: state.randomSeed, + packing: state.packing, + train_on_completions: state.trainOnCompletions, + gradient_checkpointing: state.gradientCheckpointing, + optim: state.optimizerType, + lr_scheduler_type: state.lrSchedulerType, + }, + lora, + }; + + return yaml.dump(config, { lineWidth: -1, noRefs: true }); +} diff --git a/studio/frontend/src/features/training/stores/training-config-store.ts b/studio/frontend/src/features/training/stores/training-config-store.ts index b6f9c01c42..d124ffd8db 100644 --- a/studio/frontend/src/features/training/stores/training-config-store.ts +++ b/studio/frontend/src/features/training/stores/training-config-store.ts @@ -5,6 +5,7 @@ import { persist } from "zustand/middleware"; import { checkDatasetFormat } from "../api/datasets-api"; import { checkVisionModel, getModelConfig } from "../api/models-api"; import { mapBackendModelConfigToTrainingPatch } from "../lib/model-defaults"; +import type { BackendModelConfig } from "../api/models-api"; import type { TrainingConfigState, TrainingConfigStore } from "../types/config"; const MIN_STEP: StepNumber = 1; @@ -344,6 +345,16 @@ export const useTrainingConfigStore = create()( setTargetModules: (targetModules) => set({ targetModules }), canProceed: () => canProceedForStep(get()), reset: () => set(initialState), + resetToModelDefaults: () => { + const { selectedModel } = get(); + if (!selectedModel) return; + set({ modelDefaultsAppliedFor: null }); + loadAndApplyModelDefaults(selectedModel); + }, + applyConfigPatch: (config: BackendModelConfig) => { + const patch = mapBackendModelConfigToTrainingPatch(config); + set(patch); + }, }; }, { diff --git a/studio/frontend/src/features/training/types/config.ts b/studio/frontend/src/features/training/types/config.ts index b268a08b79..b0b74f1f5b 100644 --- a/studio/frontend/src/features/training/types/config.ts +++ b/studio/frontend/src/features/training/types/config.ts @@ -6,6 +6,7 @@ import type { StepNumber, TrainingMethod, } from "@/types/training"; +import type { BackendModelConfig } from "../api/models-api"; export type LoraVariant = "lora" | "rslora" | "loftq"; @@ -117,6 +118,8 @@ export interface TrainingConfigActions { setTargetModules: (value: string[]) => void; canProceed: () => boolean; reset: () => void; + resetToModelDefaults: () => void; + applyConfigPatch: (config: BackendModelConfig) => void; } export type TrainingConfigStore = TrainingConfigState & TrainingConfigActions;