Input Validation, Tensorboard Tweaks + Drag-n-drop support by O-J1 · Pull Request #908 · Nerogar/OneTrainer

O-J1 · 2025-07-04T17:43:06Z

High level summary of what this PR does:

Drag-n-drop support for all path_entry widgets
Drag-n-drop support for the concepts tab. Wont duplicate
Merges file_entry and dir_entry into a single function
Adds more extensive input validation, including to path_entry and especially model output destination
Adds debouncing to all fields derived from entry including time_entry which will partially prevent the division by 0 backup bug, combined with prior fix this will make it impossible
Fixes the tensorboard workspace spam when typing a new one and always on
Streamlines Tensorboard management down to a single input and makes it work with cloud in Always On.

Initial checklist

Validation specific checklist

User enters without extenstion, gets auto-added
User sets a prefix, is used for autocomplete
User can choose to use friendlywords instead of datetime
Tensorboard will use model output name.
Overwrite prevention works and is toggeable

Tensorboard has now changed from enable + toggle to just a dropdown

Drag and drop demo

drag-n-drop-pr.mp4

for more information, see https://pre-commit.ci

… utils

O-J1 · 2026-01-02T03:01:00Z

@dxqb

is this now used for everything? Saves, tensorboard, samples, ...?

Tensorboard
Configs saved to workspace folder
sample filenames
saves
backups

Input validation - what is detected, what is rejected?

Every field has basic validation on its data type, reverting if an invalid input is provided (with a debounced delay). The only field with anything really fancy is Output model destination. That has a whole bunch of logic around ensuring the entered output matches the output format and if it will overwrite. This would be a lot easier if Nero let us split the fields.

Do you think we still need the automatic revert then? That always seemed unnatural to me (no other UI does that) but it was better than silently ignoring invalid values.

I strongly believe we still need, it will be toggeable though!. If we had more gatekeepers to getting started with OT or it had remained power user only software then yes I would agree, however that is not our reality

This reverts commit 71cde89.

O-J1 · 2026-02-04T09:16:25Z

I have now disabled DnD on linux after extensive conversations with the maintainer of TKInterdnd2 and another dev, who noted DnD does not even work with Qt on all linux distros (the gold standard).

Didnt want to do this but, no choice :(, tested again on Linux and Windows.

…ith invalid values. Logic tweaks.

O-J1 · 2026-02-04T12:40:26Z

After discussing further with Dxq, value reversion has been removed. In its place I have added

Undo/Redo: Ctrl + Z/Shift+Ctrl + Z or Crtl + Y to CTKEntry fields.
Alert dialog pops up if you try to start training with invalid values (havent changed logic around whats invalid)
Added logic to prevent edge case prefix/suffix recursion.

I think this is ready for further reviews again.

dxqb

it still reverts, for example when I enter a string into Learning Rate. Is that intended? [see Discord - as long as it's a strict improvement it isn't a stopper, even though I prefer just the red marking with no reverting]
minor: "prevent overwrite" is enabled. I select a preset, "Model Output Destination" goes to default "model/model.safetensors". That file exists, so "Prevent Overwrite" is adding "_1" to the file. But the orange marking stays, even though there isn't a problem anymore
it complains that "File does not exist" when Base Model is a huggingface path like lodestones/Chroma1-HD. Not initially when selected a preset, but after I manually entered it
the "File does not exist" tooltip when it appears and then I switch to another tab like training, it blinks out of nowhere into that tab even though the error is on the model tab
I'm a slightly confused by all the functionality and switches affecting the overwrite-files behaviour (Auto-correct input, Prevent Overwrite, Use Friendly Names). It basically only affects the Model Output Destination, right? The only other output path_entrys I could find are directories (workspace, debug, cache), where this functionality mostly doesn't apply.
The "run name" is used in backups, for example Creating Backup workspace/run/backup/2026-02-06_13-17-53-**model_1**-backup-20-0-20. I think using any type of prefix/suffix in backups breaks the backup discovery when you enable "Continue from last backup". At least, I remember this to be the reason why we haven't used the "Save Prefix" for backups previously.
the "run name" is consistently used for saves, tensorboard and samples. However, the "Save Filename Prefix" is ignored in most cases. The "run name" is the output file name. I think this breaks how the prefix is used currently by many users, including myself: You do 3 differents runs with different parameters, and use the save prefix so you can later identify the runs in tensorboard, samples and saves. For example, you might use "LR1e-3" as a prefix. Now you only get "model_1", "model_2", and "model_3" instead, unless you anticipate that and modify the output filename accordingly.
review: test functionality that might affect it with cloud training

dxqb · 2026-02-06T11:46:40Z

            is_full_width=False,
-            show_toggle_button=True
+            show_toggle_button=True,
+            placeholder_text="You have no concepts defined. Either click the 'Add Concept' button or drag and drop a directory here to add a new concept."


please add a check for DnD being available, so we don't tell the user to do something that doesn't work

dxqb · 2026-02-06T12:14:08Z

+        data.append(("validation_auto_correct", True, bool, False))
+        data.append(("validation_show_tooltips", True, bool, False))
+        data.append(("use_friendly_names", False, bool, False))
+        data.append(("prevent_overwrite", True, bool, False))


should this be the default?
I don't like it personally, because it'll accumulate output files even though I never use them. I only use saves.

dxqb · 2026-02-06T12:14:40Z


+        # validation settings
+        data.append(("validation_auto_correct", True, bool, False))
+        data.append(("validation_show_tooltips", True, bool, False))


why is this a setting? Isn't this always good? Why would I want to disable this?

dxqb · 2026-02-06T12:27:40Z

        self.bucket_ax.cla()
-        aspects = [str(x) for x in list(aspect_buckets.keys())]
-        aspect_ratios = [self.decimal_to_aspect_ratio(x) for x in list(aspect_buckets.keys())]
+        aspect_ratios = [self.decimal_to_aspect_ratio(x) for x in aspect_buckets]


what are these code changes to aspect statistics?
Can't find anything about this in the PR description

Just for self note: It was because of testing around the matplotlib TkAgg issue with DnD and I noted we were doing it inefficently

dxqb · 2026-02-06T12:31:46Z

+            name = path.stem
+
+        # fall back to prefix if no name could be extracted
+        if not name:


see comment in main. save prefix is ignored in most cases.

dxqb · 2026-02-06T12:36:23Z

@@ -1,56 +1,170 @@
+from __future__ import annotations


not reviewing this file - please double check before merge

dxqb · 2026-02-06T12:36:56Z

 from typing import Any

 from customtkinter import CTk, CTkToplevel
+from tkinterdnd2 import DND_FILES


always import, even if Linux? See the other file for conditional import

dxqb · 2026-02-06T12:38:05Z

@@ -0,0 +1,482 @@
+"""Input validation utilities for UI components."""


please reconsider if we need all this code for one entry field (see main comment)

dxqb · 2026-02-06T12:40:08Z


 if sys.version_info < deprecated_ver:
-    sys.stderr.write("Warning: Deprecated Python version found. Update to %s or newer\n" % (tuple_to_str(deprecated_ver)))
+    sys.stderr.write("Warning: Deprecated Python version found. Its strongly recommended to update to %s or newer\n" % (tuple_to_str(deprecated_ver)))


just for my information: why?

Because we have had 8+ users coming in going "OMGWTF how do I update to python 3.11?!" so instead of saying you must update, instead strongly recomend. Python 3.10 support will likely be dropped by the end of this year for PT (speculation)

dxqb · 2026-02-06T12:40:36Z

 safetensors==0.5.3
-tensorboard==2.19.0
+tensorboard==2.20.0
+StrEnum==0.4.15; python_version < "3.11" # For tensoboardMode on 3.10


O-J1 · 2026-02-07T19:16:40Z

Closing as spliting up.

O-J1 added 5 commits July 4, 2025 22:56

Move trainer cleanup under finally

fad971a

Tweak UI layout and labels

13efc33

Centralise tensorboard management + mutually exclusive toggles

e5026c8

Create tensorboard_util that both BaseTrainer and TrainUI can import

84c9f89

Move to option_kv to reduce clutter and make UI logic easier.

6c7a7bc

This comment was marked as outdated.

Sign in to view

Made the default value always on

aef07ad

This comment was marked as outdated.

Sign in to view

Potential fix for cloud always-on conflict.

7f7796c

dxqb self-requested a review August 20, 2025 11:10

dxqb self-assigned this Aug 20, 2025

O-J1 marked this pull request as ready for review August 29, 2025 07:37

Merge branch 'master' into tensorboard-on-tweaks

e84e9e4

This comment was marked as outdated.

Sign in to view

[pre-commit.ci] auto fixes from pre-commit.com hooks

15256d8

for more information, see https://pre-commit.ci

O-J1 added the Effort: Medium label Sep 4, 2025

Merge remote-tracking branch 'origin/master' into tensorboard-on-tweaks

1a8ef47

This comment was marked as outdated.

Sign in to view

O-J1 added Effort: Low and removed Effort: Medium labels Oct 4, 2025

dxqb reviewed Oct 4, 2025

View reviewed changes

Comment thread modules/util/config/TrainConfig.py

O-J1 added 8 commits October 13, 2025 03:14

Adds initial drag-n-drop fuctionality

272e8d4

Add dnd to conceptsTab and small cleanup

15bad8b

Add tkinterdnd2 to requirements

3ca7973

Allow multi dnd for concepts and cleanup validation further + move to…

edd7001

… utils

Adds XL Lora output correction to saving + bugfix

c216f19

Merge remote-tracking branch 'origin/master' into tensorboard-on-tweaks

e2c728f

Move the migration to migration 8

f759a8f

Attempt at fixing debounced workspace bug + cleanup

784fb3d

O-J1 added 7 commits January 2, 2026 09:57

Fix accidental row reuse

69aca0c

Make model output dest dragNdrop reject invalid filepaths i.e txt

0f9cb0d

Modify scalene env var name to OT_PROFILE

2a128cf

Remove redundant checks

1e6923c

Tweak gitignore.

66653dd

Fix CLI tensorboard oversight

30b156d

Make saves and backups use run name

d8ebd21

dxqb reviewed Jan 6, 2026

View reviewed changes

Comment thread modules/ui/ConfigList.py Outdated

O-J1 added 13 commits January 11, 2026 15:09

Attempts at debugging Xubuntu issues with dnd

ff8edb9

Make the placeholder text implementation better

b2b44f2

Bump tensorboard

ab07e7b

Remove debug, bump av to allow fresh install

1fe3257

Destroy fig and canvas when window is closed

7e09675

Allow HF repo style inputs to skip validation

2e93765

Add two HF rules I wasnt aware of.

6105204

Allow HF HTTP urls to also skip validation

2e60c7b

Experimental test to see if FigureCanvasTkAgg is the issue

71cde89

Revert "Experimental test to see if FigureCanvasTkAgg is the issue"

c260031

This reverts commit 71cde89.

Move statistics to default Agg backend instead of interactive one

f434a76

Skip scanning for invalid concepts

a145528

Disable DnD on linux due to unstable behaviour

79498d8

Add undo/redo. Remove revert behaviour. Prevent training startining w…

935dc0e

…ith invalid values. Logic tweaks.

dxqb requested changes Feb 6, 2026

View reviewed changes

O-J1 mentioned this pull request Feb 7, 2026

Phase 1 - Input validation #1296

Merged

O-J1 closed this Feb 7, 2026

O-J1 removed a link to an issue Apr 26, 2026

[Bug]: Always-On Tensorboard create unwanted workspace directories #1028

Open

		@@ -0,0 +1,482 @@
		"""Input validation utilities for UI components."""

Uh oh!

Conversation

O-J1 commented Jul 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Validation specific checklist

Tensorboard has now changed from enable + toggle to just a dropdown

Drag and drop demo

Uh oh!

This comment was marked as outdated.

This comment was marked as outdated.

This comment was marked as outdated.

This comment was marked as outdated.

This comment was marked as outdated.

Uh oh!

O-J1 commented Jan 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

O-J1 commented Feb 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

O-J1 commented Feb 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dxqb left a comment • edited by O-J1 Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

O-J1 Feb 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

O-J1 commented Feb 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

O-J1 commented Jul 4, 2025 •

edited

Loading

O-J1 commented Jan 2, 2026 •

edited

Loading

O-J1 commented Feb 4, 2026 •

edited

Loading

O-J1 commented Feb 4, 2026 •

edited

Loading

dxqb left a comment •

edited by O-J1

Loading

O-J1 Feb 6, 2026 •

edited

Loading