docker: Colab-grade JupyterLab and Studio UX for the Unsloth image#6681
docker: Colab-grade JupyterLab and Studio UX for the Unsloth image#6681danielhanchen wants to merge 23 commits into
Conversation
Stacks a Colab-like JupyterLab and Studio experience on top of the existing Blackwell image. Additive only: the training stack, CUDA/torch pinning, and the Studio/JupyterLab/sshd service trio are unchanged. JupyterLab labextension (prebuilt in a throwaway builder stage, so the runtime image stays Node-free): - Unsloth Dark (Monokai) theme, adaptive light/dark by system preference - Colab-style ArrowDown/Up cell navigation - top-bar Unsloth logo (stock Jupyter logo disabled and locked) - #@title lines render as collapsible Heading-2 form bars - Ctrl+A in a cell output selects only that output, not the whole notebook (the old behaviour ran notebook:select-all and was laggy) - right activity bar hidden by default - overrides.json: per-cell run button without auto-advance, labeled Restart and Run All, windowing off so collapsing an output does not snap to the cell top, news/update prompts suppressed Studio and login branding: Unsloth favicon, page logo, and a dark Unsloth login page that rotates through the curated Studio sloth stickers (fail-soft to the logo). Notebook organization and Colab compatibility (base image): - categorized folder view built from relative symlinks mirroring the README sections, rebuilt each boot; real .ipynb files never moved, and the symlink tree is invisible to the sync state machine - AMD-* notebooks shown only on an AMD/HIP host (autodetected) - Docker-only strip of the Colab "Run all on Colab" intro sentence from unedited notebooks (upstream notebooks unchanged) - hoist %%capture above a leading #@title form so the cell runs - the per-cell transformers-sidecar log is silent unless UNSLOTH_ENABLE_LOGGING=1 Dependency pinning and naming: the curated notebook extras are pinned to their resolved versions for reproducible rebuilds; decord is split into its own fail-soft install (no aarch64 wheel). The lean base image is renamed from :base to :core. Adds tests/validate_studio_features.py, a static self-test for the labextension plugins, overrides keys, and branding wiring.
for more information, see https://pre-commit.ci
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces a custom, Monokai-themed JupyterLab extension (unsloth-jupyterlab) and several integration scripts to provide a Colab-like user experience within the Unsloth Docker environment. Key additions include custom cell navigation, Colab #@title form cell rendering, localized output selection, and automatic IPython magic hoisting for compatibility. It also adds automated scripts to strip Colab-specific introductory text, clean up stale widget outputs, and organize notebooks into a categorized folder view using relative symlinks. A critical bug was identified in the notebook view generation script (docker/unsloth_nb_view.py), where the directory removal helper (_rmtree) could follow symlinks to directories and accidentally delete the actual notebook files. A fix was suggested to check for symbolic links first before verifying directory status.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
|
|
||
| def _rmtree(path): | ||
| # Remove a previously built VIEW. Only unlinks symlinks + empty dirs we made, | ||
| # but a full rmtree is fine here because VIEW holds nothing but our symlinks. |
There was a problem hiding this comment.
In Python, os.path.isdir(path) returns True if path is a symbolic link pointing to a directory. Consequently, if _rmtree is called with a symlink to a directory, it will bypass the not os.path.isdir(path) check and proceed to os.walk(path). Since os.walk follows the top-level symlink, it will traverse the target directory and delete all files and subdirectories inside it, resulting in accidental data loss of the actual notebooks.
Checking os.path.islink(path) first ensures that if path is a symlink (even to a directory), it is safely unlinked/removed directly without traversing its target.
| def _rmtree(path): | |
| # Remove a previously built VIEW. Only unlinks symlinks + empty dirs we made, | |
| # but a full rmtree is fine here because VIEW holds nothing but our symlinks. | |
| if os.path.islink(path): | |
| os.remove(path) | |
| return | |
| if not os.path.isdir(path): | |
| return |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c9ff52ba04
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| !jupyter/overrides.json | ||
| !jupyter/favicon.ico | ||
| !jupyter/logo.png | ||
| !jupyter/login.html |
There was a problem hiding this comment.
When building Dockerfile.studio with docker/ as the context, this allowlist still excludes jupyter/install_sloth_stickers.py because the file is not whitelisted after the top-level ** ignore. The new COPY jupyter/install_sloth_stickers.py ... in the studio Dockerfile therefore fails with a missing source before the image can build; add an exception for that helper alongside the other jupyter/ assets.
Useful? React with 👍 / 👎.
| RUN apt-get update \ | ||
| && apt-get install -y --no-install-recommends nodejs npm git \ | ||
| && rm -rf /var/lib/apt/lists/* |
There was a problem hiding this comment.
On the Ubuntu 24.04 CUDA base, apt-get install nodejs resolves to Node 18, but the JupyterLab 4.6 JavaScript packages used by the next jlpm install declare a Node >=20 engine. In a clean studio build this makes the labextension builder fail before producing /opt/labext-src/unsloth-jupyterlab/labextension; install/pin a Node 20+ source instead of the distro nodejs package.
Useful? React with 👍 / 👎.
| # Rebuild VIEW from scratch. | ||
| _rmtree(view) | ||
| os.makedirs(view, exist_ok=True) |
There was a problem hiding this comment.
The categorized view is under /workspace by default and is where Jupyter now opens, so users can easily create or save files there (or point UNSLOTH_NOTEBOOKS_VIEW_DIR at an existing directory). Rebuilding the view calls _rmtree(view) unconditionally, and _rmtree removes every regular file it finds, so the next container start can silently delete user-created notebooks or other files in that folder instead of just replacing the symlinks this script owns.
Useful? React with 👍 / 👎.
| c.ServerApp.default_url = "/lab/tree/Unsloth Notebooks" | ||
| c.LabApp.default_url = "/lab/tree/Unsloth Notebooks" | ||
| c.ServerApp.preferred_dir = "/workspace/Unsloth Notebooks" |
There was a problem hiding this comment.
When users set UNSLOTH_NOTEBOOKS_VIEW_DIR or disable the view with UNSLOTH_SKIP_NOTEBOOK_VIEW=1, the sync script either builds the categorized tree somewhere else or does not build /workspace/Unsloth Notebooks at all, but this launcher still hard-codes Jupyter's landing URL and preferred directory to that default path. Those configurations therefore open JupyterLab on a missing or stale folder instead of the requested view/raw notebooks; derive these values from the same env settings or fall back when the view is skipped.
Useful? React with 👍 / 👎.
- unsloth_nb_view.py: rebuilding the categorized view no longer deletes user files. The view is also JupyterLab's landing dir, so a user may save real notebooks there; _clear_view now unlinks only the symlinks we own and removes only folders that end up empty, leaving regular files in place. It also tests islink before isdir, so a view that is itself a symlink to a directory is unlinked instead of being walked into (which would have wiped the symlink target). - studio_launch.sh: derive the landing URL and preferred_dir from UNSLOTH_NOTEBOOKS_VIEW_DIR / UNSLOTH_SKIP_NOTEBOOK_VIEW, the same env the sync script uses, instead of hard-coding /workspace/Unsloth Notebooks. A relocated or disabled view no longer opens JupyterLab on a missing folder; it falls back to the default /lab over /workspace. - Dockerfile.studio: the labext-builder stage now installs Node 20 from NodeSource. Ubuntu 24.04's distro nodejs is 18, below JupyterLab 4.6's declared Node >=20 engine. Node stays confined to the throwaway builder stage, so the runtime image is unchanged. - .dockerignore: explicitly allowlist jupyter/install_sloth_stickers.py alongside its sibling jupyter assets, rather than relying on the directory re-inclusion.
danielhanchen
changed the title
for more information, see https://pre-commit.ci
|
Thanks for the review. Addressed in e496095:
Hard-coded landing path in Node version for the labext builder (P1): confirmed the distro
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 50f7170be4
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| # in those cases JupyterLab just opens on its default (/lab) over /workspace. | ||
| _root_dir="/workspace" | ||
| _view_dir="${UNSLOTH_NOTEBOOKS_VIEW_DIR:-/workspace/Unsloth Notebooks}" | ||
| if [[ "${UNSLOTH_SKIP_NOTEBOOK_VIEW:-0}" != "1" && "${_view_dir}" == "${_root_dir}/"* ]]; then |
There was a problem hiding this comment.
The fresh case here is UNSLOTH_SKIP_NOTEBOOK_SYNC=1: the entrypoint exits before it can build /workspace/Unsloth Notebooks, but this condition still writes /lab/tree/Unsloth Notebooks because it only checks UNSLOTH_SKIP_NOTEBOOK_VIEW. Users who disable notebook sync therefore land on a missing or stale view instead of /workspace; include the sync skip or verify the view directory exists before setting default_url.
Useful? React with 👍 / 👎.
| && /opt/unsloth-venv/bin/python /tmp/unsloth-branding/install_sloth_stickers.py \ | ||
| --src "${UNSLOTH_STUDIO_HOME}/src/studio/frontend/public/Sloth emojis" \ | ||
| --dest "${JS}/static/sloth" \ | ||
| || echo ">> sloth stickers not installed (login falls back to the Unsloth logo)" \ |
There was a problem hiding this comment.
Because this || is attached to the entire preceding && chain, any required branding step above it, such as resolving JS or copying the favicon/logo/login template, is converted into the optional sticker warning and the build continues. If a pinned Jupyter Server layout changes, this would publish a partially branded or broken login page instead of failing the image build; group only the sticker installer with the fallback so earlier failures remain fatal.
Useful? React with 👍 / 👎.
| if os.path.islink(link) or os.path.exists(link): | ||
| os.remove(link) |
There was a problem hiding this comment.
Fresh evidence: _clear_view now preserves regular files, but this later os.remove(link) still deletes any regular file whose name collides with a generated notebook symlink. If a user uploads or saves a notebook with the same filename inside one of the categorized folders, the next view rebuild silently removes that file before creating the symlink; only unlink existing symlinks, and skip or rename around regular files.
Useful? React with 👍 / 👎.
| "@jupyterlab/apputils-extension:notification": { | ||
| "fetchNews": "false", | ||
| "checkForUpdates": false, | ||
| "doNotDisturbMode": true |
There was a problem hiding this comment.
In JupyterLab 4.6 this is the global do-not-disturb switch, so when kernels restart, connections drop, or extensions emit warnings, no toast is automatically displayed. The news/update prompt is already disabled by fetchNews and checkForUpdates, so this line hides unrelated operational feedback for every user opening the image.
Useful? React with 👍 / 👎.
| # so install it on its own and fail-soft: amd64 gets it; on arm64 the ERNIE-VL | ||
| # video path is skipped rather than breaking the whole image build. | ||
| RUN ${VENV}/bin/uv pip install --python ${VENV}/bin/python "decord==0.6.0" \ | ||
| || echo ">> decord skipped (no matching wheel for ${TARGETARCH:-amd64}); ERNIE-VL video decode unavailable" |
There was a problem hiding this comment.
This fallback is unconditional, so on amd64 a transient resolver failure or incompatible decord wheel still produces a successful image without the dependency, even though the ERNIE-VL video notebooks expect it and the comment says amd64 gets it. Restrict the fail-soft path to non-amd64 or add an amd64 import check so supported images fail instead of silently dropping video decode.
Useful? React with 👍 / 👎.
|
|
||
| IMAGE="${IMAGE:-unsloth/unsloth:latest}" | ||
| BASE_IMAGE="${BASE_IMAGE:-unsloth/unsloth:base}" | ||
| BASE_IMAGE="${BASE_IMAGE:-unsloth/unsloth:core}" |
There was a problem hiding this comment.
This default now pulls unsloth/unsloth:core, but the in-repo publish workflow still creates only :base and base-* tags (.github/workflows/docker-publish.yml lines 228-238 and 432-435) and never creates a :core tag. On fresh user machines the confirmation scripts therefore fail in the pull phase unless BASE_IMAGE is overridden; update the workflow tags too, or keep the scripts on the published tag.
Useful? React with 👍 / 👎.
Complete the base->core (and "studio as studio") tag rename so the publish workflow matches the user-facing helpers and the Dockerfile.studio header. - The lean training image now publishes as :core (core-<tag>, core-nightly, core-sha-*); run.sh / docker_confirm.* already told users to pull :core, but docker-publish.yml still tagged it :base, so that pull would have 404'd. The per-arch digest artifacts are renamed to match. - The full Studio image keeps :latest and gains a stable :studio alias, matching the Dockerfile.studio header. Both the merge and post-publish smoke-test metadata blocks are updated together. Internal "base image" wording (the layer Studio builds FROM) is left as-is.
danielhanchen
force-pushed
the
pr-jupyter-studio-ux
branch
from
45d76d5 to
b963403
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b963403b18
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if (!panel.node.contains(event.target as Node)) { | ||
| return; |
There was a problem hiding this comment.
When focus is inside an interactive output (for example an ipywidgets slider/dropdown/text box), the event target is still contained by the notebook panel, so this handler proceeds in command mode and later calls preventDefault() while moving the active cell. That swallows ArrowUp/ArrowDown that the focused widget needs for its own navigation/value changes, making those controls hard to use after a notebook cell creates them; add a guard for editable/output-widget targets before taking over the key.
Useful? React with 👍 / 👎.
| if (direction === 1 && line !== editor.lineCount - 1) { | ||
| return; | ||
| } | ||
| if (direction === -1 && line !== 0) { | ||
| return; |
There was a problem hiding this comment.
When a completion popup is open in an editor on the first/last logical line (a very common one-line setup cell), ArrowDown/ArrowUp should move through the suggestions, but this capture-phase handler prevents the event before CodeMirror can see it because it only checks the cursor line. That makes completions unusable at cell boundaries; skip the takeover while CodeMirror completion UI is active, or otherwise let the editor handle those arrows first.
Useful? React with 👍 / 👎.
- studio_launch.sh: also gate the categorized-view landing URL on
UNSLOTH_SKIP_NOTEBOOK_SYNC (the entrypoint skips building the view entirely in
that mode), not just UNSLOTH_SKIP_NOTEBOOK_VIEW, so a no-sync container does not
land on a missing folder.
- Dockerfile.studio: scope the sticker-install "|| echo" fallback to only the
sticker step via a { ...; } group. It was attached to the whole branding &&
chain, so a failure in a REQUIRED step (JS resolve, favicon/logo/login copy)
was swallowed and the build continued with broken branding.
- unsloth_nb_view.py: when creating the categorized symlinks, only replace our
own stale symlinks; if a real user file already occupies that name, keep it and
skip the link instead of os.remove-ing it.
- overrides.json: drop doNotDisturbMode (it silenced ALL JupyterLab toasts,
including kernel-restart / connection-drop feedback). The news/update prompts
are already off via fetchNews / checkForUpdates.
- Dockerfile: keep decord mandatory on amd64 (fail the build on a missing or
incompatible wheel) and only fail-soft on arm64/other arches that have no wheel.
- cellNav.ts: do not hijack ArrowUp/Down when focus is in an interactive output
widget / form control, or while a completion popup is open, so ipywidgets
controls and autocomplete at cell boundaries keep working.
for more information, see https://pre-commit.ci
|
Worked through the review feedback. The earlier (06-25) comments were re-reviews of the first commit and are already fixed in it (the
Labextension recompiles cleanly and the studio self-test passes. |
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 2eba8a4976
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if not cond: | ||
| _failures.append(name) |
There was a problem hiding this comment.
In the repo CPU workflow I inspected (studio-backend-ci.yml runs python -m pytest tests/), this new file is collected as six pytest tests because the functions are named test_*. Pytest never calls main(), so when a condition is false check() only appends to _failures and the individual test still returns normally, producing a false-green validation unless someone runs the script directly. Please have check() raise/assert, or wrap the script in a single pytest test that calls main().
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
pytest does not collect this file: it is named validate_studio_features.py, not test_*.py, so default discovery skips it. The checks are now run explicitly in the repo CPU job.
…jupyter-studio-ux
…tinuation
Move the sloth-sticker fail-soft explanation above the RUN so no comment line
sits between backslash-continued commands. BuildKit strips such comments, but
keeping the RUN body a plain && chain removes the ambiguity for non-BuildKit
builders and static linters. The { ...; } fail-soft scoping is unchanged.
|
@codex review |
|
Addressed the latest review pass:
Validation: |
|
Codex Review: Didn't find any major issues. What shall we delve into next? Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
…jupyter-studio-ux # Conflicts: # .github/workflows/docker-publish.yml
|
@codex review |
|
Codex Review: Didn't find any major issues. Delightful! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
…jupyter-studio-ux
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 2e4699c920
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| m = re.match(r"^###\s+(.*)$", line) | ||
| if m: | ||
| section = clean_section(m.group(1)) |
There was a problem hiding this comment.
When the image bakes unslothai/notebooks from its current default ref, that repo's README no longer exposes the categorized notebook list as ### sections around nb/*.ipynb links, so this condition never sets a section and parse_readme() returns no rows. The view builder then falls through to putting every notebook under Other Notebooks, which means the new Jupyter landing view loses the curated category folders it is meant to provide; either parse the current catalog source/heading format or ship a stable manifest instead of depending only on ### headings here.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
The current unslothai/notebooks README still exposes the categories as ### headers around nb/*.ipynb link tables, so parse_readme returns the full set (verified against main: 26 categorized folders).
…jupyter-studio-ux
for more information, see https://pre-commit.ci
|
reviewed on windows 11 / WSL2. quick stuff first: installed studio from this branch locally and it ran fine (GGUF chat worked), and the docker studio image is where it got interesting. building
pointing
the rollback hits the same error, so it gives up, deletes the working copy, and source-builds instead (no nvcc here, so the result is broken). basically |
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 6fd9744f31
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| seen_pairs = set() # (section, filename) already emitted | ||
| section = None | ||
| for line in text.splitlines(): | ||
| m = re.match(r"^###\s+(.*)$", line) |
There was a problem hiding this comment.
When the baked notebooks README reaches a non-### domain heading such as the current # AMD Notebooks / # Molab Notebooks, section remains whatever the previous ### set (currently Other Notebooks), so the subsequent all-notebooks links are emitted into that folder. Because build_view intentionally cross-lists per section, the default Jupyter view gets a huge duplicate Other Notebooks folder instead of just uncategorized notebooks; reset on any heading or handle #/## boundaries before collecting links.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed. parse_readme now resets the section on any heading level and strips a leading emoji run, so the AMD domain notebooks get their own folder instead of falling into Other Notebooks.
| checks the shipped JupyterLab config + labextension source, so a regression in | ||
| the notebook organisation, Colab compatibility, Colab-intro/widget stripping, | ||
| sidecar-log gating, the labextension plugins, the JupyterLab defaults, or the | ||
| login branding fails CI on every device. |
There was a problem hiding this comment.
This script is named validate_studio_features.py, so pytest's default test_*.py discovery will not collect it, and I don't see any .github workflow invoking it explicitly. That means the Docker/Jupyter checks described here never fail CI unless someone runs the script manually; either rename/wrap it as a pytest test or add an explicit workflow step.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Wired in: the repo CPU job now runs tests/validate_studio_features.py explicitly, so a regression fails CI.
…b image Make it obvious the image is built by Unsloth and hard to white-label out with a shallow find-and-replace, and surface the AGPLv3 license + copyright in the UI. Visible attribution (labextension): - Help > "About Unsloth Docker Studio" dialog (about.ts): Unsloth logo, the AGPLv3 notice, "Copyright 2026-Present the Unsloth team", and source/website/ license links. Added to the Help menu and the command palette. - The JupyterLab loading splash is replaced with a spinning Unsloth logo (splash.ts, provides ISplashScreen; honors prefers-reduced-motion). The stock @jupyterlab/apputils-extension:splash is disabled+locked at build time, like the stock logo. - AGPLv3 footer (license + copyright + links) on the branded login page. - Labextension relicensed AGPL-3.0-only; SPDX headers on every source file. Anti-tamper (no encoded/obfuscated strings -- plain readable text only; the one data URI is the logo image): - A canonical, plain-text attribution set lives in unsloth_branding.py with a TypeScript mirror (branding.ts) bundled verbatim into the labextension, so the phrase, copyright, links and plugin ids are spread across independent layers. - unsloth_branding.py verifies all of these across the installed files (AGPLv3 text, login footer, theme, labextension package + built bundle strings, logo, favicon) and fails loudly if any are missing. It runs at three layers: build time (fails the image build), the whole-container launcher (studio_launch.sh refuses to start), and as a jupyter_server extension (refuses to serve JupyterLab). - tests/studio/test_branding_guard.py: positive + per-marker negative coverage, plus a check that no base64/decoder obfuscation crept into the attribution.
for more information, see https://pre-commit.ci
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d1a5b58383
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| if stripped.startswith("%%") and i > 0: | ||
| return [line] + skipped + lines[i + 1 :] |
There was a problem hiding this comment.
When a Colab-form cell uses a non-Python/shell cell magic such as %%writefile, this branch moves the #@title/blank lines into the magic body. IPython then hands those preserved lines to the magic itself, so a titled %%writefile config.json cell writes the Colab metadata comment into the generated file and can corrupt helper/config outputs; only hoist for magics where the preserved comments are inert, or drop Colab form metadata instead of injecting it into every %% body.
Useful? React with 👍 / 👎.
| const selection = window.getSelection(); | ||
| let output = closestOutput(selection?.anchorNode ?? null); | ||
| if (!output) { | ||
| output = lastPointerOutput; |
There was a problem hiding this comment.
When the last browser selection remains inside an output, clicking back on a command-mode cell or the file browser does not necessarily move window.getSelection().anchorNode, so this branch still finds the old output and prevents Ctrl/Cmd+A outside the output. After a user selects or clicks output text once, normal select-all shortcuts in the notebook/file browser can keep selecting that stale output instead; require the current event target or recent pointer location to be inside an output before taking over.
Useful? React with 👍 / 👎.
| problems.append("labextension package.json is not valid JSON") | ||
|
|
||
| # 5. The built bundle still carries the visible attribution strings + plugins. | ||
| bundle = _bundle_text(paths["labext_static"]) |
There was a problem hiding this comment.
In a derived image or mounted Jupyter config, adding disabledExtensions for unsloth-jupyterlab (or its plugin ids) leaves the static bundle on disk, so this check still passes while JupyterLab loads without the Unsloth logo/About/splash attribution; Dockerfile.studio only disables/locks the stock logo/splash plugins, not this extension. Since the guard is meant to refuse stripped/altered attribution, also reject disabled page_config entries for these plugins or lock them.
Useful? React with 👍 / 👎.
…jupyter-studio-ux # Conflicts: # docker/Dockerfile
…anding guard) - unsloth_colab_compat.py: only hoist a leading `%%` cell magic above the Colab `#@title` form for magics whose body runs as code (capture/time/bash/python/ ...). Content magics (%%writefile, %%html, %%latex, ...) are left untouched so the form comment is never injected into the written file / rendered output. - outputSelect.ts: stop trusting the text selection anchor to decide ownership of Ctrl/Cmd+A. A stale selection inside an output survives a click onto a command-mode cell or the file browser, which made select-all keep re-selecting the old output. Gate on the keystroke target or the last pointer-down (reset to null on any click outside an output) instead. - unsloth_branding.py: also reject page_config.json that disables the Unsloth labextension or any of its plugin ids via disabledExtensions (dict or list form); that leaves the bundle on disk so the prior checks passed while the logo/About/splash attribution was stripped at load. Lock unsloth-jupyterlab in Dockerfile.studio as well (defense in depth), and add guard tests.
…t.ci formatting) # Conflicts: # docker/jupyter/unsloth_branding.py
for more information, see https://pre-commit.ci
|
Thanks for the review. Addressed all three in 2482719 (the llama.cpp EXDEV fix the WSL2 report below also hits landed in the base PR #5748).
|
JupyterLab UX + branding verified in the running imageCaptured from the published Docker image booted on Login page: Unsloth mascot, "Sign in to JupyterLab", and the AGPLv3 footer with the source link.
Help > About "Unsloth Docker Studio": full attribution (Built by Unsloth, Licensed under the GNU AGPLv3, Source / Website / License links, Copyright 2026-Present the Unsloth team).
Lab shell: Unsloth top-bar logo, the "Unsloth Notebooks" launcher, and the full bundled notebook set (01-26) in the file tree.
Spinning-logo splash on load:
Other checks that passed in the same run:
|
Pin the unsloth-jupyterlab npm deps to exact versions matching the baked jupyterlab==4.6.0 (builder stays 4.5.9, its newest release) instead of floating ^/~ ranges, so the same commit always builds the same labextension bundle. Also probe JupyterLab /login (not /api, which 403s behind a password hash) in the Windows confirmation script.
# Conflicts: # docker/Dockerfile
…on into CI unsloth_nb_view.parse_readme only reset the folder section on level-3 (###) headings. The notebooks README carries level-1 domain headers (# AMD Notebooks, # Kaggle Notebooks) with their own nb/*.ipynb link tables and no intervening ###, so those notebooks were mis-filed under the previous stale section (all 148 AMD notebooks landed in Other Notebooks on an --amd build). Reset on any heading level and strip a leading emoji/symbol run so the domain notebooks get their own clean folder. Also run tests/validate_studio_features.py explicitly in the repo CPU job. It is named validate_* (not test_*) so pytest never collected it, which meant a regression in the notebook view, Colab compat, strip, JupyterLab defaults or login branding failed CI only when run by hand.
|
Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits. |
The exact pins introduced earlier (@jupyterlab/* 4.6.0, @lumino/widgets 2.8.0, @jupyterlab/builder 4.5.9) break the Dockerfile.studio labext-builder stage. Exact-pinning the framework packages defeats jlpm's (yarn classic) hoisting: transitive @jupyterlab deps request caret ranges that resolve to newer patch releases (e.g. @jupyterlab/ notebook pulls @jupyterlab/cells ^4.6.0 -> a newer patch), so jlpm installs a second nested copy alongside the exact top-level one. Two copies of @jupyterlab/cells and @lumino/widgets in the tree produce TS2345 "not assignable" errors (protected-member/identity mismatch) and the build fails. Caret ranges let jlpm collapse every @jupyterlab and @Lumino package to a single hoisted copy, which is required for a JupyterLab prebuilt (federated) extension: at runtime those packages are shared singletons provided by the host JupyterLab, so the build-time versions only need to type-check against one consistent tree, not match an exact runtime patch. This is the version set the published image was built and validated with end to end. Verified by building the labext in isolation against the base image (Node 20 + bundled jlpm): caret ranges build clean (webpack compiled successfully); the exact pins fail with the duplicate-package TS errors.
|
Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits. |
3 participants
Summary
This stacks a Colab-grade JupyterLab and Studio experience on top of the Unsloth Docker image (the one that runs on any NVIDIA GPU host, not just Blackwell). It is additive on
docker-blackwell-build: no change to the training stack, the CUDA/torch pinning, or the existing service trio (Studio on 8000, JupyterLab on 8888, sshd on 22). The goal is that a user whodocker pulls the image and opens JupyterLab lands in something that looks and behaves like Colab, with the notebooks already organized and the rough edges (Ctrl+A selecting the whole notebook, Colab-only intro text,%%captureunder a#@titleform) fixed.JupyterLab extension (prebuilt, Node-free runtime)
A small prebuilt JupyterLab 4 labextension (
docker/jupyter/unsloth_labext) is compiled in a throwawaylabext-builderstage inDockerfile.studioand copied into the final image as static assets, so Node never ships in the runtime image. Plugins:#@titleColab forms: a#@title ...first line renders as a collapsible heading bar (Heading 2 sized text, large caret) instead of a bare comment.notebook:select-all, which selected every cell and was laggy on long notebooks.System defaults ship in
docker/jupyter/overrides.json: the theme, a per-cell run button that does not auto-advance, a labeled "Restart & Run All", windowing disabled so collapsing a long output does not snap the cell to the top, and the Jupyter news/update prompts suppressed.Studio and login branding
Dockerfile.studiobrands jupyter_server: the browser-tab favicon and page logo become the Unsloth logo, and the login screen uses a dark Unsloth-themedlogin.htmlthat rotates through the curated Studio sloth stickers (installed byinstall_sloth_stickers.py, fail-soft to the logo if the Studio asset folder ever moves).Notebook organization and Colab compatibility (base image)
unsloth_nb_view.py+unsloth_sync_notebooks.sh): a sibling tree of relative symlinks mirroring the README sections (01 Main Notebooks,02 GRPO and Reinforcement Learning Notebooks, ...) is rebuilt on every boot. The real.ipynbfiles are never moved or renamed, and the symlink tree is invisible to the sync state machine (which walksfind -type f), so edit-preservation and delete-restore are unaffected. JupyterLab opens straight into this view (studio_launch.shdefault_url/preferred_dir).AMD-*notebooks are shown only on an AMD/HIP host (autodetected; override withUNSLOTH_NB_GPU=amd|cuda).unsloth_nb_strip_colab.py): the Docker image removes the Colab-only "To run this, press Runtime, Run all..." sentence from the first cell of notebooks the user has not edited, updating their recorded hashes in place so refresh detection is unaffected. Opt out withUNSLOTH_KEEP_COLAB_INTRO=1. The upstream notebooks are unchanged; this is a Docker-only sync-time transform.unsloth_colab_compat.py): an IPython input transformer hoists%%capture(and other cell magics) above a leading#@titleform so the cell runs instead of raisingUsageError. Registered fromunsloth_ipython_startup.pyin its own try/except so it can never disable the transformers-sidecar hook.[unsloth-nb] activated transformers sidecar ...line is now silent unlessUNSLOTH_ENABLE_LOGGING=1(unsloth_nb_compat.py).Dependency pinning and image naming
jupyterlab==4.6.0,notebook==7.6.0,ipywidgets,matplotlib,soundfile,evaluate,jiwer,tensorboard,langid,easydict,protobuf,omegaconf,einx,librosa,ftfy) for reproducible rebuilds, matching the cu128 core convention.decordis split into its own fail-soft install since it ships no aarch64 wheel (amd64 gets it; on arm64 the ERNIE-VL video path is skipped rather than breaking the whole image build).:baseto:coreend to end: the user-facing helpers (run.sh,docker_confirm.sh,docker_confirm.ps1), theDockerfile.studioheader, anddocker-publish.ymlitself (which now publishes:corepluscore-<tag>/core-nightly/core-sha-*). The full Studio image keeps:latestand gains a stable:studioalias.Testing
tests/validate_studio_features.pyis a static self-test asserting the labextension plugins (theme, cell-nav, logo,#@titleforms, output-select-all, ui-chrome), theoverrides.jsonkeys, the branding/sticker wiring, and the login stickers are all present in the source tree.#@titleheading, Ctrl+A output-only select, hidden right bar, random sloth login, categorized landing view) was verified in a live container.Follow-up (separate PR)
The vLLM
_decompose_size_nodesgraph-erase crash that hitsfast_inference/ GRPO notebooks (vLLM PR #42543, unmerged upstream) is handled in a separate change: a monkeypatch inunsloth_zooplus a base-image bake. It is intentionally not in this PR to keep this focused on the JupyterLab and Studio UX.