diff --git a/studio/backend/core/inference/inference.py b/studio/backend/core/inference/inference.py index 1147c281b7..329f5d944b 100644 --- a/studio/backend/core/inference/inference.py +++ b/studio/backend/core/inference/inference.py @@ -268,47 +268,6 @@ class InferenceBackend: logger.error(traceback.format_exc()) return False, None - def load_adapter(self, base_model_name: str, adapter_path: str, adapter_name: str = None) -> bool: - """ - Load a LoRA adapter onto the base model if it's not already registered. - This method is idempotent. - """ - if base_model_name not in self.models: - logger.error(f"Base model {base_model_name} not loaded") - return False - - model = self.models[base_model_name].get("model") - if model is None: - logger.error(f"Model object for {base_model_name} is None.") - return False - - if adapter_name is None: - adapter_name = adapter_path.split("/")[-1].replace(".", "_") - - # If we've loaded this adapter before, we don't need to do anything. - if adapter_name in self.models[base_model_name].get("loaded_adapters", {}): - logger.info(f"Adapter '{adapter_name}' is already registered. Skipping.") - return True - - try: - logger.info(f"Loading new adapter '{adapter_name}' from '{adapter_path}' onto {base_model_name}") - - # Unsloth modifies the model in-place and returns None. Do NOT re-assign. - model.load_adapter(adapter_path, adapter_name=adapter_name) - - # Update our internal registry so we don't load it again. - self.models[base_model_name]["loaded_adapters"][adapter_name] = adapter_path - - total_adapters = len(getattr(model, 'peft_config', {})) - logger.info(f"Adapter '{adapter_name}' loaded successfully. (Total adapters on model: {total_adapters})") - return True - except Exception as e: - logger.error(f"Failed to load adapter '{adapter_name}': {e}") - import traceback - logger.error(traceback.format_exc()) - return False - pass - def enable_adapter(self, base_model_name: str, adapter_name: str) -> bool: """Enable specific adapter (for generation)""" if base_model_name not in self.models: @@ -341,55 +300,6 @@ class InferenceBackend: logger.error(f"Failed to disable adapters: {e}") return False - # In backend/inference.py - - def load_for_eval(self, lora_path: str, max_seq_length: int = 2048, - dtype = None, load_in_4bit: bool = True, - hf_token: Optional[str] = None) -> Tuple[bool, Optional[str], Optional[str]]: - """ - Prepare for eval: ensure base model and the specified adapter are loaded. - """ - try: - from utils.models import ModelConfig - lora_config = ModelConfig.from_lora_path(lora_path, hf_token) - if not lora_config: - return False, None, None - - base_model_name = lora_config.base_model - - # 1. Load the base model if it's not already in memory (this logic is correct) - if base_model_name not in self.models or not self.models[base_model_name].get("model"): - logger.info(f"Base model '{base_model_name}' not loaded, loading now.") - base_config = ModelConfig.from_ui_selection(base_model_name, None, is_lora=False) - if not self.load_model(base_config, max_seq_length, dtype, load_in_4bit, hf_token): - return False, None, None - else: - logger.info(f"Base model '{base_model_name}' is already in memory.") - - self.active_model_name = base_model_name - - # 2. Delegate to our now-idempotent load_adapter function. - # It will handle all cases: first adapter, or subsequent adapters. - adapter_name = lora_path.split("/")[-1].replace(".", "_") - adapter_success = self.load_adapter( - base_model_name=base_model_name, - adapter_path=lora_path, - adapter_name=adapter_name - ) - - if not adapter_success: - return False, base_model_name, None - - return True, base_model_name, adapter_name - - except Exception as e: - logger.error(f"Error during load_for_eval: {e}") - import traceback - logger.error(traceback.format_exc()) - return False, None, None - pass - - def load_for_eval(self, lora_path: str, max_seq_length: int = 2048, dtype = None, load_in_4bit: bool = True, hf_token: Optional[str] = None) -> Tuple[bool, Optional[str], Optional[str]]: @@ -1272,47 +1182,6 @@ class InferenceBackend: """Get name of currently loading model""" return next(iter(self.loading_models)) if self.loading_models else None - def load_model_simple(self, - model_path: str, - hf_token: Optional[str] = None, - max_seq_length: int = 2048, - load_in_4bit: bool = True) -> bool: - """ - Simple model loading wrapper for chat interface. - Accepts model path as string and handles ModelConfig creation internally. - - Args: - model_path: Model name or path (e.g., "unsloth/llama-3-8b") - hf_token: HuggingFace token for gated models - max_seq_length: Maximum sequence length - load_in_4bit: Whether to use 4-bit quantization - - Returns: - bool: True if successful, False otherwise - """ - try: - # Create config from string path - config = ModelConfig.from_ui_selection( - model_path, - lora_path=None, # No LoRA for chat - is_lora=False - ) - - # Call existing load_model with config - return self.load_model( - config=config, - max_seq_length=max_seq_length, - dtype=None, # Auto-detect - load_in_4bit=load_in_4bit, - hf_token=hf_token - ) - - except Exception as e: - logger.error(f"Error in load_model_simple: {e}") - return False - - - def load_model_simple(self, model_path: str, hf_token: Optional[str] = None, diff --git a/studio/backend/core/training/trainer.py b/studio/backend/core/training/trainer.py index 9987f0a260..0ff317ecf1 100644 --- a/studio/backend/core/training/trainer.py +++ b/studio/backend/core/training/trainer.py @@ -354,7 +354,9 @@ class UnslothTrainer: subset: str = None, train_split: str = "train", eval_split: str = None, - eval_steps: float = 0.00) -> Optional[tuple]: + eval_steps: float = 0.00, + dataset_slice_start: int = None, + dataset_slice_end: int = None) -> Optional[tuple]: """ Load and prepare dataset for training. @@ -466,6 +468,18 @@ class UnslothTrainer: if dataset is None: raise ValueError("No dataset provided") + # Apply index range slicing if requested (inclusive on both ends) + if dataset_slice_start is not None or dataset_slice_end is not None: + total_rows = len(dataset) + start = dataset_slice_start if dataset_slice_start is not None else 0 + end = dataset_slice_end if dataset_slice_end is not None else total_rows - 1 + # Clamp to valid range + start = max(0, min(start, total_rows - 1)) + end = max(start, min(end, total_rows - 1)) + dataset = dataset.select(range(start, end + 1)) + print(f"Sliced dataset to rows [{start}, {end}]: {len(dataset)} of {total_rows} rows\n") + self._update_progress(status_message=f"Sliced dataset to {len(dataset)} rows (indices {start}-{end})") + # Check if stopped before applying template if self.should_stop: print("Stopped before applying chat template\n") @@ -482,6 +496,7 @@ class UnslothTrainer: format_type=format_type, dataset_name=dataset_source, custom_format_mapping=custom_format_mapping, + progress_callback=self._update_progress, ) # Check if stopped during formatting @@ -489,6 +504,14 @@ class UnslothTrainer: print("Stopped during dataset formatting\n") return None + # Abort if dataset formatting/conversion failed + if not dataset_info.get("success", True): + errors = dataset_info.get("errors", []) + error_msg = "; ".join(errors) if errors else "Dataset formatting failed" + logger.error(f"Dataset conversion failed: {error_msg}") + self._update_progress(error=error_msg) + return None + self._update_progress(status_message=f"Dataset formatted and ready for training") print(f"Dataset formatted successfully\n") diff --git a/studio/backend/core/training/training.py b/studio/backend/core/training/training.py index 9123d36b39..153f4335e3 100644 --- a/studio/backend/core/training/training.py +++ b/studio/backend/core/training/training.py @@ -116,7 +116,9 @@ class TrainingBackend: train_split: str = "train", eval_split: str = None, eval_steps: float = 0.00, - is_dataset_multimodal: bool = False) -> bool: + is_dataset_multimodal: bool = False, + dataset_slice_start: int = None, + dataset_slice_end: int = None) -> bool: """ Start training. @@ -224,6 +226,8 @@ class TrainingBackend: train_split=train_split, eval_split=eval_split, eval_steps=eval_steps, + dataset_slice_start=dataset_slice_start, + dataset_slice_end=dataset_slice_end, ) # Unpack: load_and_format_dataset returns (dataset, eval_dataset) diff --git a/studio/backend/models/datasets.py b/studio/backend/models/datasets.py index 9e0643415b..e615180b7f 100644 --- a/studio/backend/models/datasets.py +++ b/studio/backend/models/datasets.py @@ -35,6 +35,7 @@ class CheckFormatResponse(BaseModel): detected_text_column: Optional[str] = None preview_samples: Optional[List[Dict]] = None total_rows: Optional[int] = None + warning: Optional[str] = None class LocalDatasetItem(BaseModel): diff --git a/studio/backend/models/training.py b/studio/backend/models/training.py index 54de974100..b6b30989bd 100644 --- a/studio/backend/models/training.py +++ b/studio/backend/models/training.py @@ -22,6 +22,8 @@ class TrainingStartRequest(BaseModel): train_split: Optional[str] = Field("train", description="Training split name") eval_split: Optional[str] = Field(None, description="Eval split name. None = auto-detect") eval_steps: float = Field(0.00, description="Fraction of total steps between evals (0-1)") + dataset_slice_start: Optional[int] = Field(None, description="Inclusive start row index for dataset slicing") + dataset_slice_end: Optional[int] = Field(None, description="Inclusive end row index for dataset slicing") @model_validator(mode="before") @classmethod diff --git a/studio/backend/routes/datasets.py b/studio/backend/routes/datasets.py index 61f1bf94da..19d70d654f 100644 --- a/studio/backend/routes/datasets.py +++ b/studio/backend/routes/datasets.py @@ -6,7 +6,7 @@ import io import json import sys from pathlib import Path -from fastapi import APIRouter, HTTPException +from fastapi import APIRouter, Depends, HTTPException import logging # Add backend directory to path @@ -16,6 +16,7 @@ if str(backend_path) not in sys.path: # Import dataset utilities from utils.datasets import check_dataset_format +from auth.authentication import get_current_subject router = APIRouter() logger = logging.getLogger(__name__) @@ -257,7 +258,10 @@ def list_local_datasets() -> LocalDatasetsResponse: @router.post("/check-format", response_model=CheckFormatResponse) -def check_format(request: CheckFormatRequest): +def check_format( + request: CheckFormatRequest, + current_subject: str = Depends(get_current_subject), +): """ Check if a dataset requires manual column mapping. @@ -373,6 +377,21 @@ def check_format(request: CheckFormatRequest): else: preview_samples = _serialize_preview_rows(preview_slice) + # Lightweight URL-based image detection for VLM datasets + warning = None + image_col = result.get("detected_image_column") + if image_col and image_col in (result.get("columns") or []): + try: + sample_val = preview_slice[0][image_col] + if isinstance(sample_val, str) and sample_val.startswith(("http://", "https://")): + warning = ( + "This dataset contains image URLs instead of embedded images. " + "Images will be downloaded during training, which may be slow for large datasets." + ) + logger.info(f"URL-based image column detected: {image_col}") + except Exception: + pass + return CheckFormatResponse( requires_manual_mapping=result["requires_manual_mapping"], detected_format=result["detected_format"], @@ -384,6 +403,7 @@ def check_format(request: CheckFormatRequest): detected_text_column=result.get("detected_text_column"), preview_samples=preview_samples, total_rows=total_rows, + warning=warning, ) except HTTPException: diff --git a/studio/backend/routes/training.py b/studio/backend/routes/training.py index f8de2f639f..497daaedd3 100644 --- a/studio/backend/routes/training.py +++ b/studio/backend/routes/training.py @@ -149,6 +149,8 @@ async def start_training( "train_split": request.train_split, "eval_split": request.eval_split, "eval_steps": request.eval_steps, + "dataset_slice_start": request.dataset_slice_start, + "dataset_slice_end": request.dataset_slice_end, "custom_format_mapping": request.custom_format_mapping, "num_epochs": request.num_epochs, "learning_rate": request.learning_rate, diff --git a/studio/backend/utils/datasets/dataset_utils.py b/studio/backend/utils/datasets/dataset_utils.py index 3a4d54f93f..9e1f54a75c 100644 --- a/studio/backend/utils/datasets/dataset_utils.py +++ b/studio/backend/utils/datasets/dataset_utils.py @@ -593,6 +593,7 @@ def format_and_template_dataset( aliases_for_assistant=["gpt", "assistant", "output",], batch_size=1000, num_proc=None, + progress_callback=None, ): """ Convenience function that combines format_dataset and apply_chat_template_to_dataset. @@ -638,6 +639,7 @@ def format_and_template_dataset( text_column=user_vlm_text_column, image_column=user_vlm_image_column, dataset_name=dataset_name, + progress_callback=progress_callback, ) warnings.append(f"Applied user VLM mapping: image='{user_vlm_image_column}', text='{user_vlm_text_column}'") @@ -734,6 +736,7 @@ def format_and_template_dataset( text_column=vlm_text_column, image_column=vlm_image_column, dataset_name=dataset_name, + progress_callback=progress_callback, ) if vlm_instruction: diff --git a/studio/backend/utils/datasets/format_conversion.py b/studio/backend/utils/datasets/format_conversion.py index 6436e7a82a..41a9617857 100644 --- a/studio/backend/utils/datasets/format_conversion.py +++ b/studio/backend/utils/datasets/format_conversion.py @@ -238,24 +238,51 @@ def convert_alpaca_to_chatml(dataset, batch_size=1000, num_proc=None): return dataset.map(_convert, **dataset_map_kwargs) +def _format_eta(seconds): + """Format seconds into a human-readable ETA string.""" + if seconds < 60: + return f"{seconds:.0f}s" + elif seconds < 3600: + m, s = divmod(int(seconds), 60) + return f"{m}m {s}s" + else: + h, remainder = divmod(int(seconds), 3600) + m, _ = divmod(remainder, 60) + return f"{h}h {m}m" + + def convert_to_vlm_format( dataset, instruction=None, text_column="text", image_column="image", dataset_name=None, + progress_callback=None, ): """ Converts simple {image, text} format to VLM messages format. Returns a LIST, not a HuggingFace Dataset (to preserve PIL Images). + For URL-based image datasets, runs a 200-sample parallel probe first to + estimate download speed and failure rate, then reports time estimate or + warning through progress_callback before proceeding with the full conversion. + + Args: + progress_callback: Optional callable(status_message=str) to report + progress to the training overlay. + Returns: list: List of dicts with 'messages' field """ from PIL import Image from .vlm_processing import generate_smart_vlm_instruction + def _notify(msg): + """Send status update to the training overlay if callback is available.""" + if progress_callback: + progress_callback(status_message=msg) + # Generate smart instruction if not provided if instruction is None: instruction_info = generate_smart_vlm_instruction( @@ -281,12 +308,17 @@ def convert_to_vlm_format( def _convert_single_sample(sample): """Convert a single sample to VLM format.""" - # Get image (might be PIL Image or path) + # Get image (might be PIL Image, local path, or URL) image_data = sample[image_column] - # Handle image paths if isinstance(image_data, str): - image_data = Image.open(image_data).convert("RGB") + if image_data.startswith(("http://", "https://")): + import fsspec + from io import BytesIO + with fsspec.open(image_data, "rb", expand=True) as f: + image_data = Image.open(BytesIO(f.read())).convert("RGB") + else: + image_data = Image.open(image_data).convert("RGB") # Get text text_data = sample[text_column] @@ -317,11 +349,143 @@ def convert_to_vlm_format( # Return dict with messages return {"messages": messages} - # Use list comprehension and return the LIST directly - print(f"πŸ”„ Converting {len(dataset)} samples to VLM format...") - converted_list = [_convert_single_sample(sample) for sample in dataset] + total = len(dataset) + first_image = next(iter(dataset))[image_column] + has_urls = isinstance(first_image, str) and first_image.startswith(("http://", "https://")) - print(f"βœ… Converted {len(converted_list)} samples") + # ── URL probe: 200 samples with parallel workers to estimate speed + failure rate ── + PROBE_SIZE = 200 + MAX_FAIL_RATE = 0.3 + + if has_urls and total > PROBE_SIZE: + import time + from concurrent.futures import ThreadPoolExecutor, as_completed + from utils.hardware import safe_num_proc + + num_workers = safe_num_proc() + _notify(f"Probing {PROBE_SIZE} image URLs with {num_workers} workers...") + print(f"πŸ” Probing {PROBE_SIZE}/{total} image URLs with {num_workers} workers...") + + probe_samples = [dataset[i] for i in range(PROBE_SIZE)] + probe_ok = 0 + probe_fail = 0 + probe_start = time.time() + + with ThreadPoolExecutor(max_workers=num_workers) as executor: + futures = {executor.submit(_convert_single_sample, s): s for s in probe_samples} + for future in as_completed(futures): + try: + future.result() + probe_ok += 1 + except Exception: + probe_fail += 1 + + probe_elapsed = time.time() - probe_start + probe_total = probe_ok + probe_fail + fail_rate = probe_fail / probe_total if probe_total > 0 else 0 + throughput = probe_total / probe_elapsed if probe_elapsed > 0 else 0 + + if fail_rate >= MAX_FAIL_RATE: + msg = ( + f"⚠️ {fail_rate:.0%} of the first {PROBE_SIZE} images failed to download " + f"({probe_fail}/{probe_total}). " + "This dataset has too many broken or unreachable image URLs. " + "Consider using a dataset with embedded images instead." + ) + print(msg) + _notify(msg) + raise ValueError(msg) + + # Estimate total time for remaining samples + remaining = total - PROBE_SIZE + estimated_seconds = remaining / throughput if throughput > 0 else 0 + eta_str = _format_eta(estimated_seconds) + + info_msg = ( + f"Downloading {total:,} images ({num_workers} workers, ~{throughput:.1f} img/s). " + f"Estimated time: ~{eta_str}" + ) + if probe_fail > 0: + info_msg += f" | {fail_rate:.0%} broken URLs will be skipped" + + print(f"βœ… Probe passed: {probe_ok}/{probe_total} ok, {probe_fail} failed ({fail_rate:.0%}), {throughput:.1f} img/s") + print(f"⏱️ Estimated time for {total:,} samples: ~{eta_str}") + _notify(info_msg) + + # ── Full conversion with progress ── + from tqdm import tqdm + + print(f"πŸ”„ Converting {total} samples to VLM format...") + converted_list = [] + failed_count = 0 + + if has_urls: + # Parallel conversion for URL-based datasets + import time + from concurrent.futures import ThreadPoolExecutor, as_completed + from utils.hardware import safe_num_proc + + num_workers = safe_num_proc() + batch_size = 500 + start_time = time.time() + + for batch_start in range(0, total, batch_size): + batch_end = min(batch_start + batch_size, total) + batch_samples = [dataset[i] for i in range(batch_start, batch_end)] + + with ThreadPoolExecutor(max_workers=num_workers) as executor: + futures = {executor.submit(_convert_single_sample, s): i for i, s in enumerate(batch_samples)} + batch_results = [None] * len(batch_samples) + for future in as_completed(futures): + idx = futures[future] + try: + batch_results[idx] = future.result() + except Exception: + failed_count += 1 + + converted_list.extend(r for r in batch_results if r is not None) + + # Progress update every batch + elapsed = time.time() - start_time + done = batch_end + rate = done / elapsed if elapsed > 0 else 0 + remaining_time = (total - done) / rate if rate > 0 else 0 + eta_str = _format_eta(remaining_time) + progress_msg = f"Downloading images: {done:,}/{total:,} ({done*100//total}%) | ~{eta_str} remaining | {failed_count} skipped" + print(f" [{done}/{total}] {rate:.1f} img/s, {failed_count} failed, ETA {eta_str}") + _notify(progress_msg) + else: + # Sequential conversion for local/embedded images (fast, no I/O bottleneck) + pbar = tqdm(dataset, total=total, desc="Converting VLM samples", unit="sample") + for sample in pbar: + try: + converted_list.append(_convert_single_sample(sample)) + except Exception: + failed_count += 1 + pbar.set_postfix(ok=len(converted_list), failed=failed_count, refresh=False) + pbar.close() + + if failed_count > 0: + fail_rate = failed_count / total + print(f"⚠️ Skipped {failed_count}/{total} ({fail_rate:.0%}) samples with broken/unreachable images") + # For datasets that skipped the probe (small URL datasets), check fail rate now + if has_urls and fail_rate >= MAX_FAIL_RATE: + msg = ( + f"⚠️ {fail_rate:.0%} of images failed to download ({failed_count}/{total}). " + "This dataset has too many broken or unreachable image URLs. " + "Consider using a dataset with embedded images instead." + ) + _notify(msg) + raise ValueError(msg) + + if len(converted_list) == 0: + raise ValueError( + f"All {total} samples failed during VLM conversion β€” no usable images found. " + "This dataset may contain only image URLs that are no longer accessible." + ) + + print(f"βœ… Converted {len(converted_list)}/{total} samples") + _notify(f"Converted {len(converted_list):,}/{total:,} images successfully") # Return list, NOT Dataset return converted_list diff --git a/studio/frontend/src/components/ui/select.tsx b/studio/frontend/src/components/ui/select.tsx index 52fe4e5a75..4c500af08e 100644 --- a/studio/frontend/src/components/ui/select.tsx +++ b/studio/frontend/src/components/ui/select.tsx @@ -4,8 +4,8 @@ import { Select as SelectPrimitive } from "radix-ui"; import type * as React from "react"; import { createContext, useContext, useState } from "react"; -import { cn } from "@/lib/utils"; -import { useDialogPortalContainer } from "@/components/ui/dialog"; +import { cn } from "@/lib/utils"; +import { useDialogPortalContainer } from "@/components/ui/dialog"; import { ArrowDown01Icon, ArrowUp01Icon, @@ -92,22 +92,22 @@ function SelectTrigger({ ); } -function SelectContent({ - className, - children, - position = "item-aligned", - align = "center", - container, - ...props -}: React.ComponentProps & { - container?: HTMLElement | null; -}) { - const dialogContainer = useDialogPortalContainer(); - return ( - - & { + container?: HTMLElement | null; +}) { + const dialogContainer = useDialogPortalContainer(); + return ( + + + {data.warning && ( +
+ + {data.warning} +
+ )} + {mappingEnabled && ( ({ dataset: s.dataset, @@ -110,6 +115,10 @@ export function DatasetSection() { setUploadedFile: s.setUploadedFile, hfToken: s.hfToken, modelType: s.modelType, + datasetSliceStart: s.datasetSliceStart, + setDatasetSliceStart: s.setDatasetSliceStart, + datasetSliceEnd: s.datasetSliceEnd, + setDatasetSliceEnd: s.setDatasetSliceEnd, })), ); @@ -566,51 +575,118 @@ export function DatasetSection() { Advanced -
- - Target Format - - - - - - Format of your training data. Auto-detect works for most - datasets.{" "} - - Read more - - - - - +
+
+ + Target Format + + + + + + Format of your training data. Auto-detect works for most + datasets.{" "} + + Read more + + + + + +
+
+
+ + Train Split Start + + + + + + Only train on a subset of your training split by + specifying a start row index (inclusive, 0-based). + Leave empty to start from the first row. + + + + + setDatasetSliceStart(e.target.value || null) + } + /> +
+
+ + Train Split End + + + + + + Last row index to include from the training split + (inclusive, 0-based). For example, set Start to 0 and + End to 99 to train on the first 100 rows. Leave empty + to use all remaining rows. + + + + + setDatasetSliceEnd(e.target.value || null) + } + /> +
+
diff --git a/studio/frontend/src/features/training/api/datasets-api.ts b/studio/frontend/src/features/training/api/datasets-api.ts index 7149271c0e..d056d92189 100644 --- a/studio/frontend/src/features/training/api/datasets-api.ts +++ b/studio/frontend/src/features/training/api/datasets-api.ts @@ -2,6 +2,7 @@ import type { CheckFormatResponse, LocalDatasetsResponse, } from "../types/datasets"; +import { authFetch } from "@/features/auth"; type CheckDatasetFormatArgs = { datasetName: string; @@ -18,7 +19,7 @@ export async function checkDatasetFormat({ split, isVlm, }: CheckDatasetFormatArgs): Promise { - const res = await fetch("/api/datasets/check-format", { + const res = await authFetch("/api/datasets/check-format", { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify({ @@ -39,7 +40,7 @@ export async function checkDatasetFormat({ } export async function listLocalDatasets(): Promise { - const res = await fetch("/api/datasets/local"); + const res = await authFetch("/api/datasets/local"); if (!res.ok) { const body = await res.json().catch(() => null); throw new Error(body?.detail || `Request failed (${res.status})`); diff --git a/studio/frontend/src/features/training/api/mappers.ts b/studio/frontend/src/features/training/api/mappers.ts index 80a2b3aaf8..7fcb1592fe 100644 --- a/studio/frontend/src/features/training/api/mappers.ts +++ b/studio/frontend/src/features/training/api/mappers.ts @@ -4,6 +4,15 @@ import type { TrainingStartRequest } from "../types/api"; const BACKEND_LORA_TYPE = "LoRA/QLoRA"; const BACKEND_FULL_TYPE = "Full Finetuning"; +function parseSliceValue(value: string | null): number | null { + if (value == null) return null; + const trimmed = value.trim(); + if (!trimmed) return null; + const num = Number(trimmed); + if (!Number.isFinite(num) || !Number.isInteger(num)) return null; + return num; +} + export function toBackendTrainingType(trainingMethod: string): string { return trainingMethod === "full" ? BACKEND_FULL_TYPE : BACKEND_LORA_TYPE; } @@ -31,6 +40,8 @@ export function buildTrainingStartPayload( subset: hfDataset ? config.datasetSubset : null, train_split: hfDataset ? config.datasetSplit : null, eval_split: hfDataset ? config.datasetEvalSplit : null, + dataset_slice_start: parseSliceValue(config.datasetSliceStart), + dataset_slice_end: parseSliceValue(config.datasetSliceEnd), local_datasets: localDatasets, format_type: config.datasetFormat, custom_format_mapping: customFormatMapping, diff --git a/studio/frontend/src/features/training/components/hf-dataset-subset-split-selectors.tsx b/studio/frontend/src/features/training/components/hf-dataset-subset-split-selectors.tsx index d21fd55ff4..148341e4de 100644 --- a/studio/frontend/src/features/training/components/hf-dataset-subset-split-selectors.tsx +++ b/studio/frontend/src/features/training/components/hf-dataset-subset-split-selectors.tsx @@ -106,7 +106,7 @@ export function HfDatasetSubsetSplitSelectors({ : "rounded-lg border border-amber-200 bg-amber-50 px-3.5 py-2.5 text-xs text-amber-700 dark:border-amber-800 dark:bg-amber-950 dark:text-amber-400" } > - Could not fetch dataset splits: {error} + {error}
)} diff --git a/studio/frontend/src/features/training/hooks/use-training-actions.ts b/studio/frontend/src/features/training/hooks/use-training-actions.ts index bfe035cf4a..1f3db35ce5 100644 --- a/studio/frontend/src/features/training/hooks/use-training-actions.ts +++ b/studio/frontend/src/features/training/hooks/use-training-actions.ts @@ -16,6 +16,19 @@ const ROLE_REMAP: Record> = { sharegpt: { user: "human", assistant: "gpt", system: "system" }, }; +function normalizeTrainingStartError(message: string): string { + const normalized = message.toLowerCase(); + const isLegacyDatasetScriptError = + normalized.includes("failed to check dataset format") && + normalized.includes("dataset scripts are no longer supported"); + + if (isLegacyDatasetScriptError) { + return "This Hub dataset relies on a legacy custom script and isn’t supported in this training flow."; + } + + return message; +} + export function useTrainingActions() { const isStarting = useTrainingRuntimeStore((state) => state.isStarting); const startError = useTrainingRuntimeStore((state) => state.startError); @@ -79,7 +92,9 @@ export function useTrainingActions() { const response = await startTraining(payload); if (response.status === "error") { - runtimeStore.setStartError(response.error || response.message); + const rawMessage = response.error || response.message; + const safeMessage = normalizeTrainingStartError(rawMessage); + runtimeStore.setStartError(safeMessage); runtimeStore.setStarting(false); return false; } @@ -88,9 +103,10 @@ export function useTrainingActions() { await syncTrainingRuntimeFromBackend(); return true; } catch (error) { - const message = + const rawMessage = error instanceof Error ? error.message : "Failed to start training"; - runtimeStore.setStartError(message); + const safeMessage = normalizeTrainingStartError(rawMessage); + runtimeStore.setStartError(safeMessage); runtimeStore.setStarting(false); return false; } 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 22218e62a9..06808dd4c5 100644 --- a/studio/frontend/src/features/training/stores/training-config-store.ts +++ b/studio/frontend/src/features/training/stores/training-config-store.ts @@ -28,6 +28,8 @@ const initialState: TrainingConfigState = { datasetSplit: null, datasetEvalSplit: null, datasetManualMapping: emptyManualMapping(), + datasetSliceStart: null, + datasetSliceEnd: null, uploadedFile: null, isCheckingVision: false, isVisionModel: false, @@ -255,6 +257,8 @@ export const useTrainingConfigStore = create()( datasetSplit: null, datasetEvalSplit: null, datasetManualMapping: emptyManualMapping(), + datasetSliceStart: null, + datasetSliceEnd: null, isDatasetMultimodal: null, isCheckingDataset: false, }); @@ -311,6 +315,8 @@ export const useTrainingConfigStore = create()( }, setDatasetManualMapping: (datasetManualMapping) => set({ datasetManualMapping }), + setDatasetSliceStart: (datasetSliceStart) => set({ datasetSliceStart }), + setDatasetSliceEnd: (datasetSliceEnd) => set({ datasetSliceEnd }), setUploadedFile: (uploadedFile) => { _datasetCheckController?.abort(); _datasetCheckController = null; @@ -321,6 +327,8 @@ export const useTrainingConfigStore = create()( datasetSplit: null, datasetEvalSplit: null, datasetManualMapping: emptyManualMapping(), + datasetSliceStart: null, + datasetSliceEnd: null, isDatasetMultimodal: null, isCheckingDataset: false, }); @@ -381,7 +389,7 @@ export const useTrainingConfigStore = create()( }, { name: "unsloth_training_config_v1", - version: 6, + version: 7, migrate: (persisted, version) => { const s = persisted as Record; if (version < 2 && s.datasetSubset == null && s.datasetConfig != null) { @@ -400,6 +408,10 @@ export const useTrainingConfigStore = create()( if (version < 6 && s.datasetEvalSplit == null) { s.datasetEvalSplit = null; } + if (version < 7) { + s.datasetSliceStart ??= null; + s.datasetSliceEnd ??= null; + } return s as unknown as TrainingConfigStore; }, partialize: partializePersistedState, diff --git a/studio/frontend/src/features/training/stores/training-runtime-store.ts b/studio/frontend/src/features/training/stores/training-runtime-store.ts index 7d5c86549b..41b0fcca07 100644 --- a/studio/frontend/src/features/training/stores/training-runtime-store.ts +++ b/studio/frontend/src/features/training/stores/training-runtime-store.ts @@ -187,7 +187,6 @@ export const useTrainingRuntimeStore = create()((set) => ( evalEnabled: payload.eval_enabled ?? state.evalEnabled, message: payload.message, error: payload.error, - startError: null, currentStep: typeof detailStep === "number" ? Math.max(detailStep, 0) : state.currentStep, totalSteps: diff --git a/studio/frontend/src/features/training/types/api.ts b/studio/frontend/src/features/training/types/api.ts index 22f02e4331..e2fcc04ad4 100644 --- a/studio/frontend/src/features/training/types/api.ts +++ b/studio/frontend/src/features/training/types/api.ts @@ -8,6 +8,8 @@ export interface TrainingStartRequest { subset: string | null; train_split: string | null; eval_split: string | null; + dataset_slice_start: number | null; + dataset_slice_end: number | null; local_datasets: string[]; format_type: string; custom_format_mapping?: Record | null; diff --git a/studio/frontend/src/features/training/types/config.ts b/studio/frontend/src/features/training/types/config.ts index 6c2feec172..0e48d18861 100644 --- a/studio/frontend/src/features/training/types/config.ts +++ b/studio/frontend/src/features/training/types/config.ts @@ -26,6 +26,8 @@ export interface TrainingConfigState { datasetSplit: string | null; datasetEvalSplit: string | null; datasetManualMapping: DatasetManualMapping; + datasetSliceStart: string | null; + datasetSliceEnd: string | null; uploadedFile: string | null; epochs: number; contextLength: number; @@ -84,6 +86,8 @@ export interface TrainingConfigActions { setDatasetSplit: (split: string | null) => void; setDatasetEvalSplit: (split: string | null) => void; setDatasetManualMapping: (mapping: DatasetManualMapping) => void; + setDatasetSliceStart: (value: string | null) => void; + setDatasetSliceEnd: (value: string | null) => void; setUploadedFile: (file: string | null) => void; setEpochs: (epochs: number) => void; setContextLength: (length: number) => void; diff --git a/studio/frontend/src/features/training/types/datasets.ts b/studio/frontend/src/features/training/types/datasets.ts index 15f9a2d0b8..f81653735a 100644 --- a/studio/frontend/src/features/training/types/datasets.ts +++ b/studio/frontend/src/features/training/types/datasets.ts @@ -9,6 +9,7 @@ export type CheckFormatResponse = { total_rows?: number | null; is_multimodal?: boolean; multimodal_columns?: string[] | null; + warning?: string | null; }; export type LocalDatasetInfo = { diff --git a/studio/frontend/src/hooks/use-hf-dataset-splits.ts b/studio/frontend/src/hooks/use-hf-dataset-splits.ts index 7b8e4906ec..cda2fcbc34 100644 --- a/studio/frontend/src/hooks/use-hf-dataset-splits.ts +++ b/studio/frontend/src/hooks/use-hf-dataset-splits.ts @@ -35,6 +35,36 @@ export interface HfDatasetSplitsResult { const HF_SPLITS_API = "https://datasets-server.huggingface.co/splits"; +function normalizeDatasetSplitsError(message: string): string { + const normalized = message.toLowerCase(); + + // datasets-server returns technical script/runtime details for legacy datasets. + if ( + normalized.includes("dataset scripts are no longer supported") || + normalized.includes("runs arbitrary python code") + ) { + return "We can’t load subset/split options for this Hub dataset because it relies on a legacy custom script."; + } + + if ( + normalized.includes("unauthorized") || + normalized.includes("forbidden") || + normalized.includes("access token") || + normalized.includes("private") || + normalized.includes("gated") || + normalized.includes("401") || + normalized.includes("403") + ) { + return "Unable to load dataset splits. This dataset may be private or gated. Add a Hugging Face token with access and try again."; + } + + if (normalized.includes("not found") || normalized.includes("404")) { + return "Dataset not found. Check the dataset name and try again."; + } + + return "Unable to load dataset split options for this dataset."; +} + // --------------------------------------------------------------------------- // Hook // --------------------------------------------------------------------------- @@ -101,7 +131,18 @@ export function useHfDatasetSplits( }) .catch((err) => { if (!controller.signal.aborted) { - setError(err.message || "Failed to fetch dataset splits"); + const rawErrorMessage = + err instanceof Error + ? err.message + : typeof err === "string" + ? err + : "Failed to fetch dataset splits"; + console.warn("[useHfDatasetSplits] Failed to fetch dataset splits", { + datasetName, + message: rawErrorMessage, + error: err, + }); + setError(normalizeDatasetSplitsError(rawErrorMessage)); setEntries([]); } }) diff --git a/studio/frontend/src/index.css b/studio/frontend/src/index.css index 44c9df566c..f3f45f407f 100644 --- a/studio/frontend/src/index.css +++ b/studio/frontend/src/index.css @@ -270,6 +270,9 @@ @apply font-sans; scrollbar-gutter: stable; } + body[data-scroll-locked] { + margin-right: 0 !important; + } h1, h2, h3,