Remove BreakEpochException #2759
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
eracah
approved these changes
Dec 5, 2023
cli99
added a commit
that referenced
this pull request
Jan 17, 2024
* Bump torch to 2.1.1 version (#2717) * Add more info when run doesnt complete (#2751) * Lower sequence generation length on code gen to be dependent on max canonical solution length (#2682) * sequentialize generations_per_sample * fix bug * lower generation length * lower generation length * lower generation length * fix gen len * restore * restore * restore * fix tests * fix test * Remove flatten params (#2761) * remove flatten params * simplify tests * simplify tests * clean * fix more tests * rerun tests * speed up icl * fix tests * fix cpu tests * add more fixtures * fix tests * token count * fix vocab size * remove logger * remove clears * fix mosaicml logger * change codeowners * clean up codeowners * rerun tests * shrink dataset * fix tests * fix test * rerun tests * fix tests * fix tests * fix seed * set to 0 * rerun tests * rerun tests * change threshold * rerun tests * rerun tests * logs * remove changes * logs * logs * remove logs * rerun tests * rerun tests * logs * rerun * logs * rerun * rerun * rerun tests * many more logs * rerun tests * strip logs * enable tests * remove opt * rerun tests * add test * lint * rerun tests * fix lint * lint * filter warnings * rerun tests * fixture * add fixture * change * logs * rerun tests * add logs * rerun tests * fixture * lint * lint * rerun tests * fix ignore warning * logs * regex * regex * regex * fix * logs * reformat * fix lint (#2767) * lint (#2768) * Use time.tokens for speedmonitor instead of dataset length (#2762) * change token math * tokens * add test * fix tests * remove exception (#2759) * time to clean up time parsing 😉 (#2770) * time to clean up time parsing * fix type error * updates * Upgrade RunConfig compute specification (#2772) * Upgrade RunConfig compute specification * extra cluster * Use async logging in MLflowLogger (#2693) * async mlflow logging Signed-off-by: chenmoneygithub <chen.qian@databricks.com> * small fix Signed-off-by: chenmoneygithub <chen.qian@databricks.com> * clean up * fix test * fix tests * deflake * pin mlflow --------- Signed-off-by: chenmoneygithub <chen.qian@databricks.com> * Fix FSDP _param_init_fn to not reinit parameters multiple times (#2765) * Gate FSDP param init test on torch 2.1 (#2774) * Parallelize OCI multipart download (#2750) * [UCVolumes] Add support for list API (#2769) * Add the memory timeline profiling support through the PyTorch profiler. (#2771) * v1 * fix issues * add logs * change names * comment * add device * uncomment original trace * add custome plot * fix pyright * Update composer/profiler/torch_profiler.py Co-authored-by: Charles Tang <j316chuck@users.noreply.github.com> * address comments * fix code check * fix formatting * address comments * add unit test * fix check * fix check * fix check * fix check * fix print * add test comment * add test comment --------- Co-authored-by: Mihir Patel <mihir.v.patel7@gmail.com> Co-authored-by: Charles Tang <j316chuck@users.noreply.github.com> * Improve torch memory profiling arguments processing (#2777) * improve torch profile args * improve torch profile args * change default torch_prof_memory_filename * add memory profiling arg test * fix check * fix check * fix check * fix check * fix check * fix check * Add platform AWS and bump aws ofi nccl version (#2776) * Extend checkpoint loading to accept a validation function (#2726) * Fix checkpoint validation tests for torch 1.13 (#2779) * fix checkpoint validation tests for torch 1.13 * Fix * Bump version to 0.17.2 (#2780) * bump version * 0.17.2 * update matrix * bump transformers version (#2781) * Bump sphinxext-opengraph from 0.9.0 to 0.9.1 (#2784) Bumps [sphinxext-opengraph](https://github.com/wpilibsuite/sphinxext-opengraph) from 0.9.0 to 0.9.1. - [Release notes](https://github.com/wpilibsuite/sphinxext-opengraph/releases) - [Commits](wpilibsuite/sphinxext-opengraph@v0.9.0...v0.9.1) --- updated-dependencies: - dependency-name: sphinxext-opengraph dependency-type: direct:development update-type: version-update:semver-patch ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * Bump coverage[toml] from 7.3.0 to 7.3.3 (#2783) Bumps [coverage[toml]](https://github.com/nedbat/coveragepy) from 7.3.0 to 7.3.3. - [Release notes](https://github.com/nedbat/coveragepy/releases) - [Changelog](https://github.com/nedbat/coveragepy/blob/master/CHANGES.rst) - [Commits](nedbat/coveragepy@7.3.0...7.3.3) --- updated-dependencies: - dependency-name: coverage[toml] dependency-type: direct:development update-type: version-update:semver-patch ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * Update torch requirement from <2.1.2,>=1.13.1 to >=1.13.1,<2.1.3 (#2785) Updates the requirements on [torch](https://github.com/pytorch/pytorch) to permit the latest version. - [Release notes](https://github.com/pytorch/pytorch/releases) - [Changelog](https://github.com/pytorch/pytorch/blob/main/RELEASE.md) - [Commits](pytorch/pytorch@v1.13.1...v2.1.2) --- updated-dependencies: - dependency-name: torch dependency-type: direct:production ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * [UCVolumes] Rely on databricks-sdk auth for the right requirements (#2789) * Enable system metrics in mosaic mlflow logger (#2775) * Enable system metrics in mosaic mlflow logger * remove fixture * Update composer/loggers/mlflow_logger.py Co-authored-by: Mihir Patel <mihir.v.patel7@gmail.com> * Update composer/loggers/mlflow_logger.py Co-authored-by: Mihir Patel <mihir.v.patel7@gmail.com> * Update composer/loggers/mlflow_logger.py Co-authored-by: Mihir Patel <mihir.v.patel7@gmail.com> --------- Co-authored-by: Mihir Patel <mihir.v.patel7@gmail.com> Co-authored-by: Daniel King <43149077+dakinggg@users.noreply.github.com> * Update parse_uri (#2787) * default-no-memory-timeline (#2790) * Add eot token to ICL generate kwargs (#2782) * add custome gen kwargs and stopping on eos token * modify test * modify test * finish * finish * finish * finish * Add nightly image for torch 2.2.0 12-20-23 (#2791) * Add torch nightly 12-13 (#2792) * Add process group as arg to FSDP (#2794) * add test * only cast if PG is specified * add to docstring * filter warning * filter warning * docs * support lists * remove warnings * lint * hsdp monkeypatch * logs * change log * fix patch * typo * clean up logs * Bump coverage[toml] from 7.3.3 to 7.3.4 (#2798) Bumps [coverage[toml]](https://github.com/nedbat/coveragepy) from 7.3.3 to 7.3.4. - [Release notes](https://github.com/nedbat/coveragepy/releases) - [Changelog](https://github.com/nedbat/coveragepy/blob/master/CHANGES.rst) - [Commits](nedbat/coveragepy@7.3.3...7.3.4) --- updated-dependencies: - dependency-name: coverage[toml] dependency-type: direct:development update-type: version-update:semver-patch ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * Fix load_ignore_keys with rng (#2803) * fix rng load * lint * Bump ipykernel from 6.26.0 to 6.28.0 (#2806) Bumps [ipykernel](https://github.com/ipython/ipykernel) from 6.26.0 to 6.28.0. - [Release notes](https://github.com/ipython/ipykernel/releases) - [Changelog](https://github.com/ipython/ipykernel/blob/main/CHANGELOG.md) - [Commits](ipython/ipykernel@v6.26.0...v6.28.0) --- updated-dependencies: - dependency-name: ipykernel dependency-type: direct:development update-type: version-update:semver-minor ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * Bump junitparser from 3.1.0 to 3.1.1 (#2805) Bumps [junitparser](https://github.com/weiwei/junitparser) from 3.1.0 to 3.1.1. - [Changelog](https://github.com/weiwei/junitparser/blob/master/CHANGELOG.md) - [Commits](weiwei/junitparser@3.1.0...3.1.1) --- updated-dependencies: - dependency-name: junitparser dependency-type: direct:development update-type: version-update:semver-patch ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * Bump pytest from 7.4.3 to 7.4.4 (#2807) Bumps [pytest](https://github.com/pytest-dev/pytest) from 7.4.3 to 7.4.4. - [Release notes](https://github.com/pytest-dev/pytest/releases) - [Changelog](https://github.com/pytest-dev/pytest/blob/main/CHANGELOG.rst) - [Commits](pytest-dev/pytest@7.4.3...7.4.4) --- updated-dependencies: - dependency-name: pytest dependency-type: direct:development update-type: version-update:semver-patch ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * Avoid futures on close for MosaicML logger (#2804) * avoid futures on close * typo * logs * logs * check (#2812) * Better communication computation overlap (#2811) * patched torch * fixed torch imports * fixed torch imports * fixed torch imports * patching through composer * patching through composer * patching typingr * comment added * don't patch torch 2.1.0 * patch torch 2.1.1 and 2.2.0 * linting fix * Improve error message for speed monitor (#2801) * fix flops * stacklevel * bump torch version (#2814) * bump vision (#2815) * fix rng load (#2816) * Correct multi-unshard stream patching for torch 2.2.0dev, and stream waiting correctness. (#2817) * patched torch * fixed torch imports * fixed torch imports * fixed torch imports * patching through composer * patching through composer * patching typingr * comment added * don't patch torch 2.1.0 * patch torch 2.1.1 and 2.2.0 * linting fix * waiting on computation stream from unshard stream * waiting on computation stream from unshard stream * less waiting * no waiting * all unshard streams wait on computation stream now * 2.2.0 dev change * fix profiler (#2818) * Bump traitlets from 5.13.0 to 5.14.1 (#2822) Bumps [traitlets](https://github.com/ipython/traitlets) from 5.13.0 to 5.14.1. - [Release notes](https://github.com/ipython/traitlets/releases) - [Changelog](https://github.com/ipython/traitlets/blob/main/CHANGELOG.md) - [Commits](ipython/traitlets@v5.13.0...v5.14.1) --- updated-dependencies: - dependency-name: traitlets dependency-type: direct:development update-type: version-update:semver-minor ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * All unshard streams wait on computation every step (#2823) * patched torch * fixed torch imports * fixed torch imports * fixed torch imports * patching through composer * patching through composer * patching typingr * comment added * don't patch torch 2.1.0 * patch torch 2.1.1 and 2.2.0 * linting fix * waiting on computation stream from unshard stream * waiting on computation stream from unshard stream * less waiting * no waiting * all unshard streams wait on computation stream now * 2.2.0 dev change * correct waiting on computation stream * fsdp state typiung * patching root pre forward * patching root pre forward * fsdp state typing * patch forward * correct waiting * linting * Add encoding=utf-8 (#2824) * Fix import for daily test (#2826) * patched torch * fixed torch imports * fixed torch imports * fixed torch imports * patching through composer * patching through composer * patching typingr * comment added * don't patch torch 2.1.0 * patch torch 2.1.1 and 2.2.0 * linting fix * waiting on computation stream from unshard stream * waiting on computation stream from unshard stream * less waiting * no waiting * all unshard streams wait on computation stream now * 2.2.0 dev change * correct waiting on computation stream * fsdp state typiung * patching root pre forward * patching root pre forward * fsdp state typing * patch forward * correct waiting * linting * daily test change * daily test fix * [MLFlowObjectStore] [1/2] Base implementation for MLFlowObjectStore (#2802) * Implementation of MLFlowObjectStore * Update object store test settings * Import mlflow dependencies inline * Fix tests and ignore some pyright * Bugfix * Enforce experiment and run in get_artifact_path * Update placeholders * Make logs debug instead of info * Minor PR comments * MLflow casing * tracking_uri fixes * Update comments * Update placeholders * Fix tests * Fix pyright * Use tempfile for temp dirs * Read tracking uri env var directly * Remove dist from MLFlowObjectStore --------- Co-authored-by: Daniel King <43149077+dakinggg@users.noreply.github.com> * Remove fused layernorm (already deprecated for 2 versions) (#2827) * remove fused layernorm * remove import * remove import * remove * fix * remove docs * all * fix * filter warnings * norm * lint * refactor --------- Co-authored-by: Your Name <you@example.com> * checkpoint saver tracks all checkpoints/intervals in state (#2819) * checkpoint tracking state * fix some tests * Update tests/callbacks/test_checkpoint_saver.py * Checkpoint itself should be included in state, dont pickle timestamp object * patch the key error (doesnt fix the bug though :sad:) * avoid slashes in state, adjust tests * fix gpu test, probably * formatting * feedback * add a comment * Apply suggestions from code review Co-authored-by: Mihir Patel <mihir.v.patel7@gmail.com> --------- Co-authored-by: Mihir Patel <mihir.v.patel7@gmail.com> * code-quality timeout update (#2830) Timed out after 10 minutes here https://github.com/mosaicml/composer/actions/runs/7465107219/job/20313553654?pr=2819 Bumps runtime up to 15min * [S] Fix how single value tensors are logged (#2831) Co-authored-by: Daniel King <43149077+dakinggg@users.noreply.github.com> * Adds DTensor Support (#2821) * fixes to get dtensor to work * more fixes * Change state dict materialization for new version of torch * get load working for new set_state_dict api * use device_mesh * Add fsdp init monkeypatch for DTensor * Add checkpoint profiling logs * attempt * working single node * fix optimizer * allow 3d device mesh * attempt to use different pg during 3d mesh save * undo 3d mesh changes * load_state_dict -> load * allow parent mesh in FSDP init * allow override of force_sync_module_states * remove unnecessary exit * ignore _validate_and_get_shard_state() * save/load hsdp-moe working * remove prints * v1 * v2 * lint * add more tests * switch to PRs * ignore warning * fix lint * version error * fix version * fix state dict * update versions * lint * lint * disable lint for mosaic fsdp utils * remove bad line * move around for legacy * device mesh * ignore warning * fix import * always init * fix error * fix load planner * remove * fix lint * lint * delay state dict * test checkpoint * checkpoint * fix cpu tests * fix rotate tests * fix precision * lint * fix alibi * cleanup * cleanup * remove force sync * fix type * merge * lint * fix gpt * comment * fix test * lint * minor optimizations * Update composer/core/state.py Co-authored-by: Evan Racah <evan@mosaicml.com> * revert tests --------- Co-authored-by: Evan Racah <ejracah@gmail.com> Co-authored-by: Abhinav Venigalla <abhi.venigalla@databricks.com> Co-authored-by: root <23239305+b-chu@users.noreply.github.com> Co-authored-by: Abhinav Venigalla <abhi@mosaicml.com> Co-authored-by: Your Name <you@example.com> Co-authored-by: Evan Racah <evan@mosaicml.com> * Remove duplicate checkpoint verifications (#2828) * Fix seed for FSDP wrap (#2833) * first try * add context * lint * more lint * remove comment --------- Co-authored-by: Daniel King <daniel@mosaicml.com> Co-authored-by: Your Name <you@example.com> * Remove fsdp patch for comm overlap (#2836) * allow hsdp (#2838) * Bump torch 2.1.2 (#2840) * bump torch * bump * bump * Upgrade pyright to 1.1.310 (#2841) * [MLFlowObjectStore] [2/2] Support checkpointing with MLFlow (#2810) * Support checkpoint uploads to MLFlow (untested) Use MLFlow run tag for autoresume Add MLFlowLogger test for existing composer run tag * Try formatting mlflow save folder after INIT Make MLFlow experiment and run ID available on all ranks Fix path issue Format mlflow placeholders in remote filenames * Unit tests for partial_format * Log mlflow info as hyperparams * partial_format doc update * Fix formatting * Pull distributed logic out of MLFlowObjectStore Add debug tracebacks Bugfix Add path to debug info Try fixing RUD object store init Pyright * Partial format in format_name helpers * Fix import * Add extra partial_format test * Fix mlflow RUD check * Fix test pyright No longer expect KeyError for format_with_dist using partial_format Refactor partial_format for readability * Max iters on partial_format * Fix partial_format * Clean up * fix test import * Fix test * update nightly to torch 2.3 (#2842) * update nightly to torch 2.3 * tighten --------- Co-authored-by: Mihir Patel <mihir.v.patel7@gmail.com> * Pin sphinxcontrib applehelp (#2854) * pin release * bump * break pypi * tighter pin * pin * pin * pin * Update setup.py (#2855) * Torch 2.3 patch (#2849) * add monkeypatch for verify_options * patch * fix * fix * partial precommit * bit of cleanup * doc * debug * fix version pinning * precommit * checkdown * lint --------- Co-authored-by: Evan Racah <ejracah@gmail.com> Co-authored-by: Mihir Patel <mihir.v.patel7@gmail.com> * Update mosaicml-cli requirement from <0.6,>=0.5.25 to >=0.5.25,<0.7 (#2866) Updates the requirements on [mosaicml-cli](https://github.com/mosaicml/mosaicml-cli) to permit the latest version. - [Commits](https://github.com/mosaicml/mosaicml-cli/commits) --- updated-dependencies: - dependency-name: mosaicml-cli dependency-type: direct:production ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * Rewrite to use individual state functions (#2860) * checkdown * checkdown * lint * fix * load ignore keys * fix * resolve comments * fix load ignore keys * offload * fix gate * merge * lint * use flag * force trye * Add custom stopping criteria to ICL generate tasks (#2800) * add custome gen kwargs and stopping on eos token * modify test * modify test * finish * finish * finish * finish * finish pr * implement early stop * add tesT * fix bug * bug fix * add keys * diff split * fix typo * fix precommit * fix precommit * fix precommit * fix precommit * fix precommit * fix precommit * fix conditional import * add nlp metrics * remove code gen changes * fix nits --------- Co-authored-by: Daniel King <43149077+dakinggg@users.noreply.github.com> * Add save_ignore_keys (#2868) * comment * add it * debug * add the keys * debug * debug * remove print statement * docs and tests * fix tests --------- Co-authored-by: Daniel King <daniel@mosaicml.com> --------- Signed-off-by: chenmoneygithub <chen.qian@databricks.com> Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: Charles Tang <j316chuck@users.noreply.github.com> Co-authored-by: Anna <anna@mosaicml.com> Co-authored-by: Jeremy D <115047575+bmosaicml@users.noreply.github.com> Co-authored-by: Mihir Patel <mihir.v.patel7@gmail.com> Co-authored-by: Chen Qian <chenmoney@google.com> Co-authored-by: Daniel King <43149077+dakinggg@users.noreply.github.com> Co-authored-by: coryMosaicML <83666378+coryMosaicML@users.noreply.github.com> Co-authored-by: Harsh Panchal <68880048+panchalhp-db@users.noreply.github.com> Co-authored-by: willgleich <22464726+willgleich@users.noreply.github.com> Co-authored-by: Irene Dea <deaairene@gmail.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: snarayan21 <saaketh@mosaicml.com> Co-authored-by: Jerry Chen <jerry.chen@databricks.com> Co-authored-by: Your Name <you@example.com> Co-authored-by: Evan Racah <ejracah@gmail.com> Co-authored-by: Abhinav Venigalla <abhi.venigalla@databricks.com> Co-authored-by: root <23239305+b-chu@users.noreply.github.com> Co-authored-by: Abhinav Venigalla <abhi@mosaicml.com> Co-authored-by: Evan Racah <evan@mosaicml.com> Co-authored-by: Daniel King <daniel@mosaicml.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
Remove BreakEpochException. It's never been used and is ugly design
Along for ride, cleans up ICL tests to minimize GPU time
Along for ride, disables mosaicml logger to remove flaky tests. Currently blocking CI/CD