https://github.com/facebookresearch/pythia
- HEAD
- refs/heads/0.1
- refs/heads/additional_validate
- refs/heads/airstore
- refs/heads/audio_video_encoders
- refs/heads/config_mmft_doc
- refs/heads/debug_ln_masked_lm
- refs/heads/dependabot/npm_and_yarn/website/ansi-regex-4.1.1
- refs/heads/dependabot/npm_and_yarn/website/http-cache-semantics-4.1.1
- refs/heads/dependabot/npm_and_yarn/website/terser-5.16.2
- refs/heads/dependabot/npm_and_yarn/website/webpack-5.76.1
- refs/heads/es
- refs/heads/fix
- refs/heads/fix_base_model
- refs/heads/fix_ci
- refs/heads/fix_circleci_tests
- refs/heads/fix_lint
- refs/heads/fixblack
- refs/heads/fsdp
- refs/heads/fsdp_asg
- refs/heads/fsdp_fairseq
- refs/heads/fsdp_support
- refs/heads/gh-pages
- refs/heads/gh/ebsmothers/1/base
- refs/heads/gh/ryan-qiyu-jiang/30/base
- refs/heads/gh/ryan-qiyu-jiang/30/head
- refs/heads/gh/ryan-qiyu-jiang/30/orig
- refs/heads/gh/ryan-qiyu-jiang/32/base
- refs/heads/gh/ryan-qiyu-jiang/32/head
- refs/heads/gh/ryan-qiyu-jiang/32/orig
- refs/heads/gh/ryan-qiyu-jiang/39/base
- refs/heads/gh/ryan-qiyu-jiang/39/head
- refs/heads/gh/ryan-qiyu-jiang/39/orig
- refs/heads/gh/ryan-qiyu-jiang/40/base
- refs/heads/gh/ryan-qiyu-jiang/40/head
- refs/heads/gh/ryan-qiyu-jiang/40/orig
- refs/heads/gh/ryan-qiyu-jiang/42/base
- refs/heads/gh/ryan-qiyu-jiang/42/head
- refs/heads/gh/ryan-qiyu-jiang/42/orig
- refs/heads/gh/ytsheng/2/base
- refs/heads/gh/ytsheng/2/orig
- refs/heads/gh/ytsheng/21/base
- refs/heads/gh/ytsheng/21/orig
- refs/heads/gh/ytsheng/3/base
- refs/heads/gh/ytsheng/3/orig
- refs/heads/gh/ytsheng/4/base
- refs/heads/gh/ytsheng/5/base
- refs/heads/gh/ytsheng/6/base
- refs/heads/gh/ytsheng/7/base
- refs/heads/gh/ytsheng/8/base
- refs/heads/hydra
- refs/heads/itm
- refs/heads/main
- refs/heads/mask_key_mmft
- refs/heads/mm_alignamet
- refs/heads/mmf_interactive
- refs/heads/mmft_output_key
- refs/heads/multitask_training
- refs/heads/notebooks
- refs/heads/pl_upgrade
- refs/heads/project/cycle-consistency
- refs/heads/project/m4c
- refs/heads/pt_19
- refs/heads/skip_optimizer_update
- refs/heads/stable_mmft
- refs/heads/v0.4
- refs/heads/vilbert_multimodal
- refs/heads/vilbert_multitask
- refs/heads/vilbert_multitask_flickr30k
- refs/heads/vilbert_multitask_visual7w
- refs/heads/xla-checkpoint-fix
- refs/tags/v0.3
- refs/tags/v0.3.1
Take a new snapshot of a software origin
If the archived software origin currently browsed is not synchronized with its upstream version (for instance when new commits have been issued), you can explicitly request Software Heritage to take a new snapshot of it.
Use the form below to proceed. Once a request has been submitted and accepted, it will be processed as soon as possible. You can then check its processing state by visiting this dedicated page.Processing "take a new snapshot" request ...
Permalinks
To reference or cite the objects present in the Software Heritage archive, permalinks based on SoftWare Hash IDentifiers (SWHIDs) must be used.
Select below a type of object currently browsed in order to display its associated SWHID and permalink.
Revision | Author | Date | Message | Commit Date |
---|---|---|---|---|
c42f208 | omkar | 10 June 2020, 00:38:56 UTC | Initial Commit | 10 June 2020, 00:39:53 UTC |
9a96e5e | Vedanuj Goswami | 09 June 2020, 19:34:22 UTC | [docs] Update readme for Hateful Memes with more instructions (#296) Summary: Pull Request resolved: https://github.com/facebookresearch/mmf/pull/296 Reviewed By: apsdehal Differential Revision: D21910249 Pulled By: vedanuj fbshipit-source-id: 555794b47e20e22aae98cb11535795e7a354abbd | 09 June 2020, 19:36:30 UTC |
9bbcc40 | Omkar Parkhi | 08 June 2020, 23:22:40 UTC | [fix] Tensorboard logging folder fix and Database visualization fix (#307) Summary: - Fixes log_folder attribute in TensorBoard logging - Fixes visualization when resize sizes are specified Pull Request resolved: https://github.com/facebookresearch/mmf/pull/307 Reviewed By: apsdehal Differential Revision: D21939755 Pulled By: omkar-fb fbshipit-source-id: 7ca44e1a680582ae0f0b210cbae8e1d734b94921 | 08 June 2020, 23:24:38 UTC |
1f7be78 | Vedanuj Goswami | 06 June 2020, 07:34:51 UTC | [fix] mmimdb dataset fixes (#297) Summary: - Update paths and proper annotation files Test Plan : Test with mmimdb training starting with no cache Pull Request resolved: https://github.com/facebookresearch/mmf/pull/297 Reviewed By: apsdehal Differential Revision: D21918955 Pulled By: vedanuj fbshipit-source-id: 161d9966ae11a2376daf855a8c563d06ae86ee5f | 06 June 2020, 07:36:40 UTC |
9e778dc | Vatsal Goel | 05 June 2020, 18:15:39 UTC | [tests] Fix device error in checkpoint test (#257) Summary: There was an error in running pytest ./tests that it expected cuda object but received cpu object. This PR assigns right device to the object tested according to the weight's device. Fixes #293 Pull Request resolved: https://github.com/facebookresearch/mmf/pull/257 Reviewed By: vedanuj Differential Revision: D21900928 Pulled By: apsdehal fbshipit-source-id: 3a06416f3bc3e4233e01339a0536b30b35d08b6c | 05 June 2020, 18:18:11 UTC |
f11adf0 | Vedanuj Goswami | 03 June 2020, 19:26:08 UTC | [feature, fix] Capture stdout in logger, fix logger initialization (#284) Summary: This PR addresses several logger related issues - Capture `stdout` by the logger. Adds a StreamToLogger class to write `stdout` to logger. Behavior can be overridden using config `training.stdout_capture`. - Remove sending the writer/warning logs to stdout when we are capturing stdout logs in the logger - Refactor out `build_config` as a separate method and decouple from `build_trainer`. - Adds basic tests to check log files are created in proper place - Fixes slurm script Pull Request resolved: https://github.com/facebookresearch/mmf/pull/284 Reviewed By: apsdehal Differential Revision: D21856097 Pulled By: vedanuj fbshipit-source-id: e6b312f410e084bf1bb51280712e2de3666e173e | 03 June 2020, 19:30:31 UTC |
168d78f | Vedanuj Goswami | 03 June 2020, 17:57:19 UTC | [fix] Fix Hateful Memes output format (#289) Summary: Fixes https://github.com/facebookresearch/mmf/issues/288 Pull Request resolved: https://github.com/facebookresearch/mmf/pull/289 Reviewed By: apsdehal Differential Revision: D21862440 Pulled By: vedanuj fbshipit-source-id: b11bbd0c4d5c3ee5362924f58a7f4cb434288a8a | 03 June 2020, 17:58:37 UTC |
23898ef | Amanpreet Singh | 03 June 2020, 04:22:53 UTC | [feature] Add MMF website, unify docs; move docs to pytorch_sphinx_theme (#287) Summary: - Adds website based on docusaurus 2 - Uses Pytorch Sphinx Theme with a custom version of it - Unify docs into the website - Add first version of the landing page - Adapt all of the links to mmf - Add redirect to api as well TODOs in next PR: - Add versioning support - Add continuous integration checks for website - Add continuous integration deploy for website - Setup rest of the stuff regarding website deploy Pull Request resolved: https://github.com/facebookresearch/mmf/pull/287 Reviewed By: vedanuj Differential Revision: D21849352 Pulled By: apsdehal fbshipit-source-id: 154f8c1455d7e5054cafc70eda1eb25ec2fe3b21 | 03 June 2020, 04:27:17 UTC |
ad32366 | Vedanuj Goswami | 03 June 2020, 00:53:40 UTC | [chores] Update project readme files (#286) Summary: - Add project Readme for MMBT - Update BAN, BUTD, Lorra, Pythia, Vilbert, Visual BERT readme files Pull Request resolved: https://github.com/facebookresearch/mmf/pull/286 Reviewed By: apsdehal Differential Revision: D21835976 Pulled By: vedanuj fbshipit-source-id: 10125790ffe4d7f70fa6a2a9648e3fa605a312ab | 03 June 2020, 00:54:55 UTC |
7f3692f | Amanpreet Singh | 02 June 2020, 16:43:12 UTC | [fix] phoc build during setup.py (#285) Summary: - Also specifies the exact variation for detectron weights - Builds the cphoc extension properly in setup.py Pull Request resolved: https://github.com/facebookresearch/mmf/pull/285 Reviewed By: vedanuj Differential Revision: D21835564 Pulled By: apsdehal fbshipit-source-id: 6ecc71fb008c4ca82064d02556b546486df08752 | 02 June 2020, 16:45:05 UTC |
3734880 | Amanpreet Singh | 01 June 2020, 19:08:16 UTC | [feature] MMBT Interface for pretrained models (#278) Summary: - Add interface for MMBT which has classify function - Create build_processors function for easy use with interfaces - Allow no_tqdm in download - Add tests for these interfaces Pull Request resolved: https://github.com/facebookresearch/mmf/pull/278 Test Plan: Added tests for testing the pretrained model Reviewed By: vedanuj Differential Revision: D21773357 Pulled By: apsdehal fbshipit-source-id: 58fc98fbe5580ecb14d95bf90f6880a5272a9315 | 01 June 2020, 19:10:28 UTC |
7084ae5 | Amanpreet Singh | 01 June 2020, 16:38:24 UTC | [feat] Automatic downloads for VQA2,COCO,CC,SNLI,VizWiz (#282) Summary: - Updates configs for pretraining project - Updates default configs to use automatic download paths - Updates README as well Pull Request resolved: https://github.com/facebookresearch/mmf/pull/282 Test Plan: Tested with zoo test Reviewed By: vedanuj Differential Revision: D21808251 Pulled By: apsdehal fbshipit-source-id: 51a7b5c9c4de83a10a2af42862f107388a1ffe8c | 01 June 2020, 16:41:26 UTC |
d176851 | Vedanuj Goswami | 30 May 2020, 01:00:12 UTC | [mmf] Fblearner changes for Hateful Memes, HM sweep script, some fixes (#281) Summary: Pull Request resolved: https://github.com/facebookresearch/mmf/pull/281 Pull Request resolved: https://github.com/fairinternal/mmf-internal/pull/121 - Added new squashfs hateful_memes.img to `/mnt/fair/hateful_memes.img` and `/mnt/fair-flash3-east/hateful_memes.img` . This contains all data files necessary for hateful memes v5 - Added Hateful Memes sweep script for mmbt. This can be modified to run with other models - set env variable for `TORCH_HOME` to read cached files for models, tokenizer configs etc. - add `projects` files to target so that fblearner can find relative configs - fblearner fixes after file name was changed Reviewed By: mamhamed Differential Revision: D21759302 fbshipit-source-id: eec83b519a46b7e0c05c465836e1a9309e71031a | 30 May 2020, 01:02:18 UTC |
08e5416 | Vedanuj Goswami | 28 May 2020, 01:07:44 UTC | [docs] Update Quickstart documentation (#277) Summary: Fixes https://github.com/facebookresearch/mmf/issues/272 #276 Pull Request resolved: https://github.com/facebookresearch/mmf/pull/277 Reviewed By: apsdehal Differential Revision: D21734443 Pulled By: vedanuj fbshipit-source-id: 3d3488a3eb4cbd11e042cc6fd9c3487e1d42c9dc | 28 May 2020, 01:10:27 UTC |
6495f7d | Amanpreet Singh | 26 May 2020, 17:14:48 UTC | [fix] Regression due to OmegaConf upgrade in pop (#275) Summary: - Fixes https://github.com/facebookresearch/mmf/issues/274 - Side: Fix Load pretrained command in readme - Side: Fix precommit warning Pull Request resolved: https://github.com/facebookresearch/mmf/pull/275 Reviewed By: vedanuj Differential Revision: D21710272 Pulled By: apsdehal fbshipit-source-id: b7a8dcd68ffd48e9bc6d18be9d69bb46aef8420a | 26 May 2020, 17:16:37 UTC |
6d89e1d | Amanpreet Singh | 23 May 2020, 03:51:43 UTC | [chore] Add licenses to files missing it along with precommit hook (#270) Summary: - Adds pre-commit hook to automatically detect and add license if missing - Updates all of the current files Pull Request resolved: https://github.com/facebookresearch/mmf/pull/270 Reviewed By: vedanuj Differential Revision: D21706344 Pulled By: apsdehal fbshipit-source-id: 8cfb706015868817228401157884c01c7b2dd4a5 | 23 May 2020, 03:53:30 UTC |
e5d7964 | Amanpreet Singh | 23 May 2020, 00:32:41 UTC | [feat,refactor] build,visualize dataset; refactor MultiDataset, fix HM (#267) Summary: - Adds builder utils for dataset and dataset loader - setup_imports now have a check whether they were already setup and always happen in the case of registry import - Add typings file which contains multiple typings used build - Adds visualize function to hateful memes dataset and to base dataset class as well which can be overriden in the child datasets - MultiDataset has been refactored to be more clean and use build utils added in this PR. - MultiDataset has been renamed to MultiDatasetLoader as it is actually a loader and not a dataset - Change accordingly in trainer - Move setup imports function to env as that is the correct location - Adds function load_yaml_with_defaults which will merge the yaml passed with default MMF configuration file - Adds option to image database's from_path function to not apply transform - Adds a new utils file visualize which contains visualize_images function as of now - HM dataset fix: Should not return target in case where label is not present Pull Request resolved: https://github.com/facebookresearch/mmf/pull/267 Test Plan: Test with regular visual_bert on hateful_memes command, pip install and then test in jupyter notebook Reviewed By: vedanuj Differential Revision: D21680440 Pulled By: apsdehal fbshipit-source-id: d8458495c4af84b0cf53067fdfb7c7f0ea591d42 | 23 May 2020, 00:34:11 UTC |
3e192ea | Amanpreet Singh | 22 May 2020, 21:16:34 UTC | [feat] SampleList pin_memory and to_dict (#271) Summary: - Adds pin_memory to properly pin_memory in case of custom batch - Adds to_dict which will be useful for torchscript Pull Request resolved: https://github.com/facebookresearch/mmf/pull/271 Test Plan: Adds tests to test these two on SampleList Reviewed By: vedanuj Differential Revision: D21701640 Pulled By: apsdehal fbshipit-source-id: 61752032364cf63231dc9246953ed6066593653c | 22 May 2020, 21:18:13 UTC |
396cd0e | Amanpreet Singh | 22 May 2020, 20:55:38 UTC | [fix,refactor] Upgrade to Python3.6+ syntax and pyupgrade hook (#269) Summary: - Upgrade to Python3.6+ syntax - Use pyupgrade pre-commit hook to make sure everything stays on same version Pull Request resolved: https://github.com/facebookresearch/mmf/pull/269 Reviewed By: vedanuj Differential Revision: D21699916 Pulled By: apsdehal fbshipit-source-id: 2e70682e508da05166e09a9439974ac6344f0d85 | 22 May 2020, 20:57:19 UTC |
9cc6cbc | Vedanuj Goswami | 22 May 2020, 04:26:23 UTC | [chores, fix] Add pretrain VL right models to zoo, fixes for masked datasets (#263) Summary: - Adds all pretrained models for Pretrain VL project - Fixes masked coco dataset - Add few missing configs symlinks - Fix few configs for train+val Pull Request resolved: https://github.com/facebookresearch/mmf/pull/263 Reviewed By: apsdehal Differential Revision: D21658125 Pulled By: vedanuj fbshipit-source-id: cab72c0e9e7cda48d8bb0476585e4719220a5579 | 22 May 2020, 04:28:16 UTC |
7cb8334 | Vedanuj Goswami | 22 May 2020, 04:10:38 UTC | [chores] Upgrade to Pytorch 1.5, torchvision 0.6 (#265) Summary: - Upgrade to Pytorch 1.5, torchvision 0.6 - Update setup.py and hence rc version - Update CircleCI for windows build test. Since windows instance we are using does not have GPU enabled we install cpu version of torch only. Tests: - Tests passing - Tested with VisualBERT model training Pull Request resolved: https://github.com/facebookresearch/mmf/pull/265 Reviewed By: apsdehal Differential Revision: D21682111 Pulled By: vedanuj fbshipit-source-id: 8fafeeecec2aae118d4aa31cdb4e08a2e34e1500 | 22 May 2020, 04:12:05 UTC |
17495ad | Amanpreet Singh | 20 May 2020, 22:02:22 UTC | [fix,refactor] Add mmf_cli cmds, fix,speed up hm_convert, pip package (#262) Summary: - Python zipfile module uses Python interpretation to unzip files which are encrypted; so improve performance we first use unzip command if present otherwise fall back to zipfile implementation - Bump the version number - Add a new folder mmf_cli which will contain all cli commands in one place but usually should symlink to tools folder - Add proper __init__.py files for folder which were not getting imported properly - get_absolute_path now actually returns absolute path if the path is not uri instead of directly returning the path. This fixes cases where USER_DIR is set to "." - If the folder is not git repository such as package folder git repo creation in checkpoint used to throw error. Now, we catch that error and just pass on it in case it is not a git repo - Manifest file has been removed as package_data handles its uses - Symlinks to packages that need to be included in mmf wheel have been added in package subfolder so that they are easily includable as per setuptools recommendation - Automatically discovers the package data now - Updates checksum for visual_bert coco - Fix predict command for single GPU Pull Request resolved: https://github.com/facebookresearch/mmf/pull/262 Test Plan: Tested first by publishing on test pypi Reviewed By: vedanuj Differential Revision: D21653287 Pulled By: apsdehal fbshipit-source-id: 06b3b9d818b992b6f015a10d1b0215c9e07f3405 | 20 May 2020, 22:07:13 UTC |
93cd2a7 | Amanpreet Singh | 16 May 2020, 17:21:46 UTC | [enhancement] Speed up build and docs by removing tensorboard dep (#254) Summary: - Tensorboard has been slowing down our building time by huge factor due to grpcio build times - Expose version in the mmf's __init__.py - Now, the warning will only be thrown in case if user tries to use Tensorboard Pull Request resolved: https://github.com/facebookresearch/mmf/pull/254 Test Plan: Tested on CircleCI as well as RTD Reviewed By: vedanuj Differential Revision: D21599427 Pulled By: apsdehal fbshipit-source-id: f7cd2e745f44b9123035ae67ef07e26d19f2c438 | 16 May 2020, 17:24:17 UTC |
d9d1d9a | Amanpreet Singh | 14 May 2020, 20:30:13 UTC | [docs] Add docs for configuration system, fix rtd build (#253) Summary: - Details hierarchy, overrides and other things - Removes fasttext from rtd build to fix build and improve building time Pull Request resolved: https://github.com/facebookresearch/mmf/pull/253 Test Plan: Built the documentation locally and tested the specific branch on RTD as well. Reviewed By: vedanuj Differential Revision: D21567163 Pulled By: apsdehal fbshipit-source-id: 2ad46a8a4cd493c6cb794c079af9f878fe7ac8f9 | 14 May 2020, 20:31:42 UTC |
0035271 | Amanpreet Singh | 14 May 2020, 03:17:26 UTC | [fix] Windows build and add its tests for CircleCI (#252) Summary: - CircleCI now uses `python setup.py install` for installation - Fixes https://github.com/facebookresearch/mmf/issues/250 Pull Request resolved: https://github.com/facebookresearch/mmf/pull/252 Test Plan: Tested on CircleCI Reviewed By: vedanuj Differential Revision: D21564135 Pulled By: apsdehal fbshipit-source-id: 00f276dbe4f1cac3886dcdcda8c2b8682793eccc | 14 May 2020, 03:19:00 UTC |
ae1689c | Vedanuj Goswami | 14 May 2020, 00:52:36 UTC | [fix] Fix slurm scripts after config changes (#119) Summary: Pull Request resolved: https://github.com/fairinternal/mmf-internal/pull/119 Pull Request resolved: https://github.com/facebookresearch/mmf/pull/251 Fixing slurm scripts after configuration changes. Reviewed By: apsdehal Differential Revision: D21558124 fbshipit-source-id: dad32574bd5ee19d52e17ef6a21b70e392809891 | 14 May 2020, 00:55:58 UTC |
fc9d0f7 | Amanpreet Singh | 13 May 2020, 17:50:29 UTC | [fix] User directory related changes; see summary (#249) Summary: - `get_absolute_path` also checks for user_dir - Fix bug in import_user_module - Adds MMF citation and fixes code block in HM project - Also, use `get_absolute_path` for model_config, dataset_config and user_config Pull Request resolved: https://github.com/facebookresearch/mmf/pull/249 Reviewed By: vedanuj Differential Revision: D21550924 Pulled By: apsdehal fbshipit-source-id: b05c51361edc518fc8006a2b4b93a70649a6c565 | 13 May 2020, 17:52:07 UTC |
b09a1bb | Vedanuj Goswami | 13 May 2020, 03:49:21 UTC | [docs] Hateful Memes challenge tutorial (#247) Summary: Adds a tutorial and submission instructions for Hateful Memes Challenge. Pull Request resolved: https://github.com/facebookresearch/mmf/pull/247 Reviewed By: apsdehal Differential Revision: D21536037 Pulled By: vedanuj fbshipit-source-id: 25869cd3e253b88a853001f74e1e4bee54099c6d | 13 May 2020, 03:51:21 UTC |
1abcc45 | Amanpreet Singh | 12 May 2020, 22:41:15 UTC | [docs] Add installation docs for MMF (#248) Summary: Pull Request resolved: https://github.com/facebookresearch/mmf/pull/248 Reviewed By: vedanuj Differential Revision: D21526237 Pulled By: apsdehal fbshipit-source-id: 2218bec750114cd0dcd26bd7087c9e49c2f1dc1d | 12 May 2020, 22:45:26 UTC |
972e7ee | Amanpreet Singh | 12 May 2020, 22:40:38 UTC | [chores,feat] HM Readme along with conversion tools for manual downloads (#246) Summary: - Adds conversion tool for manual downloads needed in HM - Adds README for the project - Add some missing configs under HM - Add command directly to setup for easy accessibility Pull Request resolved: https://github.com/facebookresearch/mmf/pull/246 Reviewed By: vedanuj Differential Revision: D21526040 Pulled By: apsdehal fbshipit-source-id: 696c683ed0c01d1cf6b6af8e650aeead508fa515 | 12 May 2020, 22:42:55 UTC |
9a74f07 | Vedanuj Goswami | 12 May 2020, 16:06:42 UTC | [feature] HM prediction support, new predict cli cmd, support for csv predict file etc (#245) Summary: PR adds the following capabilities : - New `mmf_predict` cli command for pypi. Can also be run as `python tools/predict.py .. ` - Support for saving predictions to a `csv` file in addition to default `json` - New config keys `evaluation.predict` and `evalaution.predict_file_format`. Removes `evalai_inference` - Predictions can be modified from models as well - Add HM predict support Test Plan : ``` mmf_predict config=projects/visual_bert/configs/hateful_memes/defaults.yaml dataset=hateful_memes model=visual_bert evaluation.predict=true evaluation.predict_file_format=csv ``` Pull Request resolved: https://github.com/facebookresearch/mmf/pull/245 Reviewed By: apsdehal Differential Revision: D21520274 Pulled By: vedanuj fbshipit-source-id: da8293d64289d6efe74d70e4857caba629accf1b | 12 May 2020, 16:08:42 UTC |
5773fe8 | Amanpreet Singh | 12 May 2020, 05:18:44 UTC | [feat,refactor] Resume from zoo, config, class refactor, more robust (#243) Summary: - Directly load from a zoo model with `resume_zoo` option - Optionally, reset optimizers or counts - Use `load_pretrained` to load using `zoo_override` otherwise correct config must be known - `resume_pretrained` added to clarify confusion on loading pretrained mapping - `pretrained_mapping` -> `pretrained_state_mapping` - Complete some pending TODOs in checkpoint class Pull Request resolved: https://github.com/facebookresearch/mmf/pull/243 Test Plan: Yet to be added Also need to update some configs Reviewed By: vedanuj Differential Revision: D21517816 Pulled By: apsdehal fbshipit-source-id: 35eda3a057b12dffcafd30247344ad29afed700b | 12 May 2020, 05:20:35 UTC |
087a9e5 | Vedanuj Goswami | 11 May 2020, 08:16:23 UTC | [feature] Add hateful memes model zoo (#244) Summary: - Adds pretrained model files for HM to zoo - Adds CC pretrained model from original ViLBERT paper to zoo - Adds a `load_requirements` method to base_model when `zoo_requirements` for the model contains other models for example pretrained models that can be finetuned. See example for `from_coco.yaml` and `from_cc.yaml` in VisualBERT and ViLBERT Test Plan : - Tested all uploaded models by loading them with `.from_pretrained` from `mmf` lib - Tested training HM with configs that require other pretrained models for pretrained weights Pull Request resolved: https://github.com/facebookresearch/mmf/pull/244 Reviewed By: apsdehal Differential Revision: D21499197 Pulled By: vedanuj fbshipit-source-id: 7cdf25d21af6ecdf6f7559e41e809dfd5f666bb9 | 11 May 2020, 14:42:43 UTC |
da5d884 | Amanpreet Singh | 09 May 2020, 07:35:20 UTC | [chores] Remove Pythia logo, update version number, mmf_bert (#242) Summary: - Removes Pythia logo and update documentation as well - Update version number - Also, rename pythia_bert to mmf_bert Pull Request resolved: https://github.com/facebookresearch/mmf/pull/242 Reviewed By: vedanuj Differential Revision: D21490411 Pulled By: apsdehal fbshipit-source-id: c54d7385fafc99ba43928969b72f4fbadeb1e5f5 | 09 May 2020, 08:24:15 UTC |
87c4b83 | dependabot[bot] | 09 May 2020, 06:13:34 UTC | Bump nltk from 3.4.1 to 3.4.5 (#240) Summary: Bumps [nltk](https://github.com/nltk/nltk) from 3.4.1 to 3.4.5. <details> <summary>Changelog</summary> <p><em>Sourced from <a href="https://github.com/nltk/nltk/blob/develop/ChangeLog">nltk's changelog</a>.</em></p> <blockquote> <p>Version 3.5 2020-04-13</p> <ul> <li>add support for Python 3.8</li> <li>drop support for Python 2</li> <li>create NLTK's own Tokenizer class distinct from the Treebank reference tokeniser</li> <li>update Vader sentiment analyser</li> <li>fix JSON serialization of some PoS taggers</li> <li>minor improvements in grammar.CFG, Vader, pl196x corpus reader, StringTokenizer</li> <li>change implementation <= and >= for FreqDist so they are partial orders</li> <li>make FreqDist iterable</li> <li>correctly handle Penn Treebank trees with a unlabeled branching top node.</li> </ul> <p>Thanks to the following contributors to 3.5: Nicolas Darr, Gerhard Kremer, Liling Tan, Christopher Hench, Alexandre Dias, Hervé Nicol, Pierpaolo Pantone, Bonifacio de Oliveira, Maciej Gawinecki, BLKSerene, hoefling, alvations, pyfisch, srhrshr</p> <p>Version 3.4.5 2019-08-20</p> <ul> <li>Fixed security bug in downloader: Zip slip vulnerability - for the unlikely situation where a user configures their downloader to use a compromised server <a href="https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2019-14751">https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2019-14751</a>)</li> </ul> <p>Thanks to the following contributors to 3.4.5: Mike Salvatore</p> <p>Version 3.4.4 2019-07-04</p> <ul> <li>fix bug in plot function (probability.py)</li> <li>add improved PanLex Swadesh corpus reader</li> </ul> <p>Thanks to the following contributors to 3.4.4: Devashish Lal, Liling Tan</p> <p>Version 3.4.3 2019-06-07</p> <ul> <li>add Text.generate()</li> <li>add QuadgramAssocMeasures</li> <li>add SSP to tokenizers</li> <li>return confidence of best tag from AveragedPerceptron</li> <li>make plot methods return Axes objects</li> <li>don't require list arguments to PositiveNaiveBayesClassifier.train</li> <li>fix Tree classes to work with native Python copy library</li> <li>fix inconsistency for NomBank</li> <li>fix random seeding in LanguageModel.generate</li> <li>fix ConditionalFreqDist mutation on tabulate/plot call</li> <li>fix broken links in documentation</li> <li>fix misc Wordnet issues</li> <li>update installation instructions</li> </ul> <p>Thanks to the following contributors to 3.4.3: alvations, Bharat123rox, cifkao, drewmiller, free-variation, henchc irisxzhou, nick-ulle, ppartarr, simonepri, yigitsever, zhaoyanpeng</p> </tr></table> ... (truncated) </blockquote> </details> <details> <summary>Commits</summary> <ul> <li><a href="https://github.com/nltk/nltk/commit/6b0312a41e9403111c9cfb5673947df7416b1f43"><code>6b0312a</code></a> Merge pull request <a href="https://github-redirect.dependabot.com/nltk/nltk/issues/2446">#2446</a> from sildar/3.4.5</li> <li><a href="https://github.com/nltk/nltk/commit/3d5e59707ac0a6928cd54da7db0e935ec0d65b88"><code>3d5e597</code></a> Fix devnull error on python2.7</li> <li><a href="https://github.com/nltk/nltk/commit/acca8d573878e173379fe190d7f14f298b12dac9"><code>acca8d5</code></a> updates for 3.4.5</li> <li><a href="https://github.com/nltk/nltk/commit/083bbf00fcc3b5e944fa5e2708a9a06769af0d5b"><code>083bbf0</code></a> updates for 3.4.5</li> <li><a href="https://github.com/nltk/nltk/commit/f59d7ed8df2e0e957f7f247fe218032abdbe9a10"><code>f59d7ed</code></a> CVE-2019-14751:</li> <li><a href="https://github.com/nltk/nltk/commit/2554ff48feed878ba7e830ada9825196f3eaa86a"><code>2554ff4</code></a> updates for 3.4.4</li> <li><a href="https://github.com/nltk/nltk/commit/fbda91914fee1968cd18a6c3d3ef21871476daee"><code>fbda919</code></a> drop comment about implementation which is no longer accurate, and which did ...</li> <li><a href="https://github.com/nltk/nltk/commit/8bcc98a28d3447ca8eaf3f75ae6fb03165319244"><code>8bcc98a</code></a> Merge pull request <a href="https://github-redirect.dependabot.com/nltk/nltk/issues/2319">#2319</a> from BLaZeKiLL/BLaZeKiLL-polt-bug-fix</li> <li><a href="https://github.com/nltk/nltk/commit/f6a4f38351b1a195fa220776b89c982ff8f55ba3"><code>f6a4f38</code></a> Merge pull request <a href="https://github-redirect.dependabot.com/nltk/nltk/issues/2291">#2291</a> from alvations/better-panlex</li> <li><a href="https://github.com/nltk/nltk/commit/8c75c567d8bacc2df2b7c375130153a8a428455f"><code>8c75c56</code></a> Merge pull request <a href="https://github-redirect.dependabot.com/nltk/nltk/issues/2324">#2324</a> from minho42/Fix-typo</li> <li>Additional commits viewable in <a href="https://github.com/nltk/nltk/compare/3.4.1...3.4.5">compare view</a></li> </ul> </details> <br /> [![Dependabot compatibility score](https://dependabot-badges.githubapp.com/badges/compatibility_score?dependency-name=nltk&package-manager=pip&previous-version=3.4.1&new-version=3.4.5)](https://help.github.com/articles/configuring-automated-security-fixes) Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting `dependabot rebase`. [//]: # (dependabot-automerge-start) [//]: # (dependabot-automerge-end) --- <details> <summary>Dependabot commands and options</summary> <br /> You can trigger Dependabot actions by commenting on this PR: - `dependabot rebase` will rebase this PR - `dependabot recreate` will recreate this PR, overwriting any edits that have been made to it - `dependabot merge` will merge this PR after your CI passes on it - `dependabot squash and merge` will squash and merge this PR after your CI passes on it - `dependabot cancel merge` will cancel a previously requested merge and block automerging - `dependabot reopen` will reopen this PR if it is closed - `dependabot close` will close this PR and stop Dependabot recreating it. You can achieve the same result by closing it manually - `dependabot ignore this major version` will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself) - `dependabot ignore this minor version` will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself) - `dependabot ignore this dependency` will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself) - `dependabot use these labels` will set the current labels as the default for future PRs for this repo and language - `dependabot use these reviewers` will set the current reviewers as the default for future PRs for this repo and language - `dependabot use these assignees` will set the current assignees as the default for future PRs for this repo and language - `dependabot use this milestone` will set the current milestone as the default for future PRs for this repo and language You can disable automated security fix PRs for this repo from the [Security Alerts page](https://github.com/facebookresearch/mmf/network/alerts). </details> Pull Request resolved: https://github.com/facebookresearch/mmf/pull/240 Reviewed By: apsdehal Differential Revision: D21489985 Pulled By: vedanuj fbshipit-source-id: ceda20960fb6e723baa5295a47f23b4d56308569 | 09 May 2020, 06:25:36 UTC |
355f61f | vedanuj | 09 May 2020, 04:45:15 UTC | [fix] FB Shipit Sync fbshipit-source-id: ebe8266069f5c2d1e28b588697b5f8f3587ed2f1 | 09 May 2020, 04:45:15 UTC |
4517e13 | Amanpreet Singh | 08 May 2020, 21:37:12 UTC | [feature] CLI commands for PyPI package along with user_dir option (#104) Summary: - Allows users to specify their own directory as a one step forward towards MMF as a lib - mmf_run commmand to run mmf training from virtual anywhere - Add functions for easy imports on user end - Fixes test_results upload issue on circleci - Address some comments from previous PR - User dir can also be specified via MMF_USER_DIR Pull Request resolved: https://github.com/fairinternal/mmf-internal/pull/104 Test Plan: Install with `python setup.py develop` and use mmf_train instead of `python -u tools/run.py` to run all of your commands now. Reviewed By: vedanuj Differential Revision: D21173056 Pulled By: apsdehal fbshipit-source-id: de24b990e5c18e478f413a7c3f6b23b6abce6949 | 08 May 2020, 22:07:19 UTC |
95340d1 | Vedanuj Goswami | 08 May 2020, 07:33:04 UTC | [feature] Config updates for HM, Unimodal image with feature changes (#118) Summary: - Config updates for Hateful Memes - Change unimodal image to mean output features in spatial dimension before feeding to classifier. Pull Request resolved: https://github.com/fairinternal/mmf-internal/pull/118 Reviewed By: apsdehal Differential Revision: D21473057 Pulled By: vedanuj fbshipit-source-id: 47f1d47816d40b12847afa370ce71db4d8c3a13d | 08 May 2020, 22:07:19 UTC |
c98800a | Amanpreet Singh | 08 May 2020, 06:34:45 UTC | [fix] Support for torch.distributed.launch (#110) Summary: - Some people still will use torch.distributed.launch for distributed setup when they won't have slurm or some other launcher. - This PR aims to fix the support for it Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/110 Test Plan: - Tested on devfair with following command ``` OMP_NUM_THREADS=1 python -m torch.distributed.launch --nproc_per_node=2 tools/run.py config="projects/pythia/configs/vqa2/defaults.yaml" model=pythia dataset=vqa2 training.num_workers=0 ``` Reviewed By: vedanuj Differential Revision: D21447059 Pulled By: apsdehal fbshipit-source-id: 21df45152f7acbc4ed52c9116529ab480208c722 | 08 May 2020, 22:07:19 UTC |
d981094 | Amanpreet Singh | 08 May 2020, 01:40:34 UTC | [chores] Updates to docs/readme regarding MMF (#117) Summary: - Adding banners of top of documentation to clarify about the outdated stuff. - Explicitly mentions to download v0.3 for the meanwhile - Adds redirect for learnpythia to mmf documentation Pull Request resolved: https://github.com/fairinternal/mmf-internal/pull/117 Reviewed By: vedanuj Differential Revision: D21450908 Pulled By: apsdehal fbshipit-source-id: 3370bbc53bb5012383f193447307fd3637258305 | 08 May 2020, 22:07:19 UTC |
ef04cc2 | Amanpreet Singh | 08 May 2020, 01:38:52 UTC | [feature] Hateful Memes Dataset (#115) Summary: - Dataset zoo for HM - All of model configurations - Image processors have been added - Properly use MMFDataset for the HM dataset - Some bug fixes in the models Pull Request resolved: https://github.com/fairinternal/mmf-internal/pull/115 Test Plan: All of the models present in the paper are tested to be working Same as the previous commit you can set the data dir to my data dir and test with it. Reviewed By: vedanuj Differential Revision: D21447066 Pulled By: apsdehal fbshipit-source-id: 8890503e95075ebe33eac02a3be540ff980c6b6b | 08 May 2020, 22:07:19 UTC |
b28b6e5 | Vedanuj Goswami | 07 May 2020, 23:47:52 UTC | [chore] Add updated citations for pretrain VL paper (#111) Summary: Pull Request resolved: https://github.com/fairinternal/mmf-internal/pull/111 Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/111 Update citations for Pretrain VL paper Reviewed By: apsdehal Differential Revision: D21368998 fbshipit-source-id: 2736d673fafc54837ef833151ca29e51f7f88d1e | 08 May 2020, 22:07:19 UTC |
a9d63ab | Vedanuj Goswami | 07 May 2020, 17:59:21 UTC | [refactor] Decouple metrics from models and refactor (#101) Summary: - New config key `evaluation` added for evaluation related configurations. `metrics` added to `evaluation` - Remove metrics from `model_config` entirely. Model and metrics decoupled. - Remove `training` from dataset configs and moved to project specific configs. Dataset and metrics decoupled. - Load metrics in trainer class only - Remove metrics from training logs. We calculate metrics only when in evaluation mode. Calculating metrics in batches during training isn't correct. Training should only be related to losses. - Metrics updated to `meter` only when eval_mode=True and will show in logs when full evaluation is run - Metrics calculated on whole evaluation set. Training can monitor a metric in evaluation to pick the best model. Test Plan : - Tested with mmimdb dataset training and evaluation Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/101 Reviewed By: apsdehal Differential Revision: D21440720 Pulled By: vedanuj fbshipit-source-id: db585a3b70d32300064e00d7b0806fade605926e | 08 May 2020, 22:07:19 UTC |
b19a526 | Vedanuj Goswami | 07 May 2020, 14:05:45 UTC | [fix] Fix LMDBFeatureReader (#116) Summary: Fix LMDBFeatureReader Test Plan : Test with mmimdb dataset. Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/116 Reviewed By: apsdehal Differential Revision: D21445705 Pulled By: vedanuj fbshipit-source-id: af96f3cc32f50d3c97ebc2df8ca815f97bea0c0d | 08 May 2020, 22:07:19 UTC |
f6abb44 | Amanpreet Singh | 07 May 2020, 06:31:16 UTC | [enhancement,refactor,fix] M4C Captioner, TextCaps related changes (#113) Summary: - Fix test reporter according to new API, distributed sampler - MultiDataset distributed sampler will be loaded now - Refactor M4C TextCaps to TextCaps - Remove the confusion between COCO and TextCaps by introducing annotation_style config parameter that will allow to load other datasets than default dataset. In long term, we want datasets to be able to specify the dataset class that will be used to build the dataset - Allow `zoo_requirements` in the model zoo as well - Upload detectron weights as well so that they can be included as requirement - Update all models to use zoo based detectron weights - Add dataset zoo for textcaps, coco-m4c version and coco-ocr_en version Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/113 Test Plan: Same as the previous PR, you can use my data dir to test it out. I have tested all of the variations that come with m4c_textcaps so we should be good to go Reviewed By: vedanuj Differential Revision: D21421244 Pulled By: apsdehal fbshipit-source-id: 5a89ee516f7e4f4cac68f7a977b893ace9af4a92 | 08 May 2020, 22:07:19 UTC |
7bb07b9 | Amanpreet Singh | 01 May 2020, 16:11:45 UTC | [feature,refactor] M4C Zoo, M4CTextVQA/TextVQA merge, see message (#109) Summary: This PR adds and refactors a lot of things around M4C paper and MMF utilties to make sure we don't miss anything around the running challenges. It is part of 3 PR series. The other two would involving renaming m4c_textcaps to m4c_textcaps and proper prediction followups for evalai and in general. This PR introduces and changes following stuff: - LMDB features for TextVQA, STVQA, OCRVQA - Merges M4C_TextVQA and TextVQA datasets. M4C and LoRRA both should work with these - Automatic downloads for all three datasets. - Renames object to byte tensors utilities and moves them inside distributed - Performance optimizations for TextVQA dataset - Model Zoo for M4C along with all variations - Custom load_pretrained method for M4C - Allow registry hacks in load_pretrained - Fixes EvalAI predictions setup, along with TestReporter - Adds functionality of zoo_requirements, which can allow one datasets to be dependent on resources from other datasets - Remove VERSION and RESOURCES property from MMFDatasetBuilder for simplicity - Fixes performance and cuda issues in M4C metrics - Merges M4CBERTTokenizer and MMF's BERT Tokenizer - Decouples TextVQA dataset from VQA dataset - Updates all of the configuration - Adds tests to check if zoo configurations are valid and do contains actual working URLs - Adds tests for get_zoo_config and handles more cases in it - Updates LMDB conversion tool to handle more generic cases, and default to file path as the key for LMDB - Handle these different cases in the feature reader as well - Handle ResNet152 features also in the feature reader - Throw an error if lmdb file doesn't exist - Change TextVQA defaults to use detectron features only and allow resnet features via separate config - Also adds sensible defaults for OCRVQA and STVQA without BERT based assumptions Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/109 Test Plan: As usual, install the repo with `python setup.py develop`. If you don't want to test automatic downloads and want to test working, you can test with my features on devfair by setting MMF_DATA_DIR="/private/home/asg/pythia/data" and MMF_CACHE_DIR="/private/home/asg/pythia/mmf/.mmf_cache" An example run on M4C with TextVQA: ``` export MMF_DATA_DIR="/private/home/asg/pythia/data"; export MMF_CACHE_DIR="/private/home/asg/pythia/mmf/.mmf_cache"; export MMF_MODEL_DATA_DIR="/private/home/asg/pythia/data"; python -u tools/run.py config="projects/m4c/configs/textvqa/defaults.yaml" model=m4c dataset=textvqa training.num_workers=5 training.run_type=val ``` For pretrained models test the following example: ```py from mmf.models import M4C model = M4C.from_pretrained("m4c.textvqa") ``` Things I have tested: - M4C on TextVQA - M4C on TextVQA with STVQA - M4C on TextVQA with OCR ML - Pythia on OCRVQA - Pythia on TextVQA - LoRRA on TextVQA - M4C on STVQA - M4C on OCRVQA Reviewed By: vedanuj Differential Revision: D21323346 Pulled By: apsdehal fbshipit-source-id: d8045e06729802d20e84b18cd2029c9926ae1b9a | 08 May 2020, 22:07:19 UTC |
e245bec | Vedanuj Goswami | 28 April 2020, 06:04:39 UTC | [refactor] Refactor Vilbert and Visualbert (#67) Summary: Refactor of our BERT based models. This removes the use of sample list from our `*Pretraining` or `*Classification` heads and these modules can be called directly with tensors or strings inputs. This will help to make these modules scriptable. Enabling scripting for these models will come in next set of PRs. This PR also consolidates all classification heads for different datasets. `training_head_type` is set to `classification` for all along with a `num_labels` configuration. For nlvr2 we are keeping the `training_head_type` as `nlvr2` as we need to specially modify the `hidden_size` for that dataset. Tested with current visual bert/vilbert models as well as loading old models. Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/67 Reviewed By: apsdehal Differential Revision: D21272728 Pulled By: vedanuj fbshipit-source-id: 715af8be62caa1e4e10f84a63eb45499f30a6362 | 08 May 2020, 22:07:19 UTC |
0ef144f | Amanpreet Singh | 26 April 2020, 07:04:50 UTC | [feature] Model Zoo and load_pretrained (#108) Summary: - This change ability to load pretrained models from load_pretrained method - Moves configuration based general functions now to configuration for streamlined flow - Allows configuration object to be initialized without any args with only default config - hack_imports moves to a separate function now - Header will only checked while downloading if there is actually a need to download Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/108 Test Plan: Install MMF with python setup.py develop In a Python interpretor do the following: ``` from mmf.models import VisualBERT visual_bert = VisualBERT.from_pretrained("visual_bert.pretrained.coco") ``` Reviewed By: vedanuj Differential Revision: D21210637 Pulled By: apsdehal fbshipit-source-id: 3ee7776355a401a8b69a2b33049d468e18e87c0d | 08 May 2020, 22:07:19 UTC |
1e54ce0 | Vedanuj Goswami | 24 April 2020, 06:29:20 UTC | [mmf][feature] Add Fblearner workflow (#107) Summary: Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/107 This diff enables fblearner training for mmf as well as an example sweep script for running vizwiz training. Reviewed By: apsdehal Differential Revision: D21193072 fbshipit-source-id: 6151406ed6b4a07865258320162fbb1cc9c6ca69 | 08 May 2020, 22:07:19 UTC |
93e93a6 | Vedanuj Goswami | 23 April 2020, 06:06:14 UTC | [feature] Adding fvcore PathManager support for handling internal backends (#106) Summary: - Adding fvcore PathManager for file operations on URIs (like manifold, everstore etc) as well as regular paths while keeping python builtin path support for OSS. For regular paths it will fall back to python builtin operations. - Currently adding PathManager support only to checkpoints, logging and cache. It can be later extended to datasets, vocabs, processors etc if we want to read from manifold. - import git only when the git module is installed. i.e for OSS. Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/106 Reviewed By: apsdehal Differential Revision: D21187474 Pulled By: vedanuj fbshipit-source-id: dc06d7f6a148aeeafe3069e31013a86805ec24b4 | 08 May 2020, 22:07:19 UTC |
c79727b | Amanpreet Singh | 23 April 2020, 02:56:49 UTC | [feature] Move folder based configs to env; cache_dir in torch home (#103) Summary: - Adds mmf cache directory inside torch cache directory - Adds env vars and defaults in cache directory for data directory - Set model_data_dir to be same as data_dia - Move save_dir, log_dir, report_dir, tensorboard_logdir to env as well - All envs now can be overridden through Environmental Variables - Update config files to use these - Renames data_root_dir to data_dir for consistency Replicating older behavior can be done via: Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/103 Test Plan: ``` MMF_CACHE_DIR="./mmf/.mmf_cache/" MMF_DATA_DIR="../data" python -u tools/run.py config="projects/pythia/configs/vizwiz/defaults.yaml" model=pythia dataset=vizwiz ``` Reviewed By: vedanuj Differential Revision: D21173063 Pulled By: apsdehal fbshipit-source-id: abe4b8bf37dc161be4d33604ef4d5aec62e3e9e7 | 08 May 2020, 22:07:19 UTC |
ad8ecff | Amanpreet Singh | 22 April 2020, 19:23:41 UTC | [feature] Add dataset zoo for easy download configuration (#102) Summary: - This is also extensible to model zoo - Needs latest OmegaConf master to work Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/102 Reviewed By: vedanuj Differential Revision: D21150233 Pulled By: apsdehal fbshipit-source-id: 4b7433e59e32296ba2d7a831d7552e3fd33697d4 | 08 May 2020, 22:07:19 UTC |
53c67b0 | Amanpreet Singh | 22 April 2020, 03:40:41 UTC | [chores] Changes for proper PyPI packaging (#97) Summary: - Manifest.in - Fix setup.py options - Remove YAML dependency as we are using OmegaConf now - Remove test options from requirements and setup as they are deprecated - Don't build extensions in READTHEDOCS enviroment - Python versions specified - Bumps the version - Fixes reading of README - Separates out version in a seprate file - Adds proper imports in init files as well Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/97 Test Plan: Try installing via `pip install mmf==0.9a2`. Reviewed By: vedanuj Differential Revision: D21103876 Pulled By: apsdehal fbshipit-source-id: 0b335bb2599b3eee577fc3bc64c456fd6f3fdcb5 | 08 May 2020, 22:07:19 UTC |
ed0f84c | Amanpreet Singh | 21 April 2020, 22:58:45 UTC | [docs] Fix docs build and some refactoring (#98) Summary: Docs builds has been broken since the update. This fixes it and also does some refactoring based on our current needs for tutorial, quickstart and other things. Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/98 Reviewed By: vedanuj Differential Revision: D21106729 Pulled By: apsdehal fbshipit-source-id: 475812cf640255a3f969a28593f7669585cc9501 | 08 May 2020, 22:07:19 UTC |
4e5b5c1 | Amanpreet Singh | 21 April 2020, 21:35:50 UTC | [feature] Add MMFDatasetBuilder (#95) Summary: - This builder builds datasets and downloads the resources specified in them - The builder will AnnotationDatabase - This will also build ImageDatabase and FeaturesDatabase conditionally on `use_features` and `use_images` if they are present in the configuration - PR adapts VQA builder to use this - PR also adapts MMIMDb dataset to use download resource and MMFBuilder Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/95 Test Plan: Tested on MMIMDb dataset Reviewed By: vedanuj Differential Revision: D21150224 Pulled By: apsdehal fbshipit-source-id: 63fd11af74b7dd4fdc5d907b19993c6251d2ee53 | 08 May 2020, 22:07:19 UTC |
563a62d | Amanpreet Singh | 21 April 2020, 08:10:42 UTC | [feature] MMF Dataset for external datasets (#92) Summary: PR (https://github.com/fairinternal/pythia-internal/issues/6) in the series of PR for better dataset API and E2E tests. MMF dataset will be used for external facing datasets, while we can create a new dataset called FBDataset for internal use cases. MMFDataset, builds annotation db, image db and features db for the user but __getitem__ still needs to be written. This PR in turn cleans up VQA2 dataset a lot so that it doesn't handle the weight of all responsibility. Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/92 Reviewed By: vedanuj Differential Revision: D21150002 Pulled By: apsdehal fbshipit-source-id: 45223684e2e6d6d43bd227739b2012d3d9478fc5 | 08 May 2020, 22:07:19 UTC |
b7dbbbe | Amanpreet Singh | 20 April 2020, 18:40:55 UTC | [chores] Add flake8 to pre-commit hooks (#100) Summary: - This will save time with internal and CI flake8 checks when they fail - I didn't fix all flake8 errors currently present as they need to be manually fixed. With time as we commit on those files they should be fixed. Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/100 Reviewed By: vedanuj Differential Revision: D21110113 Pulled By: apsdehal fbshipit-source-id: 0e4a65202905b6a434a3336b75c5be93b25be1d9 | 08 May 2020, 22:07:19 UTC |
50cc5c7 | Amanpreet Singh | 19 April 2020, 02:23:32 UTC | [enhancement] Repurpose FeatureDatabase to better conform to Database API (#90) Summary: PR (5) in the sequence of PR needed for better dataset API required for E2E testing. Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/90 Reviewed By: vedanuj Differential Revision: D21106643 Pulled By: apsdehal fbshipit-source-id: be6097163c84f6a88fe6fa4ff0fc0de1e563ab1f | 08 May 2020, 22:07:19 UTC |
cfa31e2 | Amanpreet Singh | 19 April 2020, 01:46:45 UTC | [fix,test] Loss inconsistency with metrics and PythiaLoss tests (#75) Summary: Now the losses can also be used directly with a string instead of always specifying a "type" dict. Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/75 Reviewed By: vedanuj Differential Revision: D21105430 Pulled By: apsdehal fbshipit-source-id: f15b2cf9521af80312674e7be9838a4dfdb1351b | 08 May 2020, 22:07:19 UTC |
fcdcde8 | Vedanuj Goswami | 19 April 2020, 01:25:02 UTC | [fix] Change mmf_config to env (#99) Summary: Addressing some remaining TODOs after https://github.com/fairinternal/pythia-internal/issues/91 Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/99 Reviewed By: apsdehal Differential Revision: D21109825 Pulled By: vedanuj fbshipit-source-id: 47c326dc79935d7b9b1b59f63f89d1fea6c6984c | 08 May 2020, 22:07:19 UTC |
55e9b71 | Vedanuj Goswami | 19 April 2020, 00:51:09 UTC | [feature] Add Late fusion model (#77) Summary: Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/77 Reviewed By: apsdehal Differential Revision: D21107711 Pulled By: vedanuj fbshipit-source-id: bc1e7bda05ab6bdee8aaaafe95d878116d978d20 | 08 May 2020, 22:07:19 UTC |
47c62b6 | Amanpreet Singh | 18 April 2020, 07:37:02 UTC | [feature] Add ImageDatabase for loading images (#89) Summary: PR (4) in the series of dataset simplification PRs towards E2E tests. Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/89 Reviewed By: vedanuj Differential Revision: D21103067 Pulled By: apsdehal fbshipit-source-id: 771c1a23d4d30db723a83b8eb8676b2201db5c1e | 08 May 2020, 22:07:19 UTC |
82fd8cd | Amanpreet Singh | 18 April 2020, 07:31:47 UTC | [fix] Seeding fixes, seed random, generate strong seed (#76) Summary: Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/76 Reviewed By: vedanuj Differential Revision: D21105265 Pulled By: apsdehal fbshipit-source-id: 4f991a81a9614bf4ff2536fe7aff6dac7e2b341a | 08 May 2020, 22:07:19 UTC |
d3ee615 | Amanpreet Singh | 18 April 2020, 03:51:27 UTC | [fix] Older checkpoint loading fixed for once and all (#93) Summary: This will ensure older checkpoints from Pythia can be loaded. This also enables possibility of backwards' compatibility to the name "pythia" Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/93 Reviewed By: vedanuj Differential Revision: D21086207 Pulled By: apsdehal fbshipit-source-id: 7165279e108e90f5f1c965b7bca4e0fa521ca80a | 08 May 2020, 22:07:19 UTC |
75372d2 | Amanpreet Singh | 18 April 2020, 00:16:56 UTC | [feature] Change ImageDatabase to AnnotationDatabase (#88) Summary: PR (3) in the series of PRs towards dataset simplification. - Change to annotation database as image database is coming in. More to come. Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/88 Reviewed By: vedanuj Differential Revision: D21086055 Pulled By: apsdehal fbshipit-source-id: ebf461b9acc85426ac35c38df018cd40befe22ab | 08 May 2020, 22:07:19 UTC |
06292e7 | Amanpreet Singh | 17 April 2020, 23:17:43 UTC | [fix] Only finalize in main process for checkpoint (#94) Summary: This should fix the problem of unusable final pth files. Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/94 Reviewed By: vedanuj Differential Revision: D21086250 Pulled By: apsdehal fbshipit-source-id: ae835bea90171c18c7632f1e7a47d9642ba52695 | 08 May 2020, 22:07:19 UTC |
608bc70 | Amanpreet Singh | 17 April 2020, 07:19:50 UTC | [feature] DownloadFile class for simplifying dataset downloads (#78) Summary: PR https://github.com/fairinternal/pythia-internal/issues/1 in the series of dataset simplification PRs towards E2E tests. - This will be also useful for pretrained model download Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/78 Test Plan: Tests have been added some of which depend on network. Reviewed By: vedanuj Differential Revision: D21031225 Pulled By: apsdehal fbshipit-source-id: 5f191eb8dd7463297685b6a3c3fb41c707c8fbf3 | 08 May 2020, 22:07:19 UTC |
5f3a876 | Amanpreet Singh | 17 April 2020, 07:03:32 UTC | [refactor] API simplification for Dataset (#86) Summary: PR (2) in the series of dataset simplification PRs towards E2E tests. Following changes has been made to the dataset api for simplification - Move from _load and _build to _load and _build for the DatasetBuilder class - get_item is changed to PyTorch default __getitem__ now - BatchCollator now handles adding dataset_name and dataset_type - DatasetLoader is always assigned by MultiDataset now - Updates to documentation Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/86 Test Plan: No new tests have been added, but this is a change in a series of changes towards E2E tests Reviewed By: vedanuj Differential Revision: D21035124 Pulled By: apsdehal fbshipit-source-id: 0cb4f3b30a18cbcb25babad5950e48d548df3523 | 08 May 2020, 22:07:19 UTC |
1e980c0 | Amanpreet Singh | 17 April 2020, 06:54:53 UTC | [feature] Add UniModal baselines (#72) Summary: Depends https://github.com/fairinternal/pythia-internal/issues/71 #70 https://github.com/fairinternal/pythia-internal/issues/64 Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/72 Reviewed By: vedanuj Differential Revision: D21082074 Pulled By: apsdehal fbshipit-source-id: 291b7a9de31f8e377a708809409b03867b3f9a7b | 08 May 2020, 22:07:19 UTC |
0f6decb | Vedanuj Goswami | 17 April 2020, 01:27:26 UTC | [refactor] Universal cache directory for mmf (#91) Summary: Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/91 Single `cache_dir` for everything Reviewed By: apsdehal Differential Revision: D21038738 fbshipit-source-id: 746078e2b23582f694b360b3fc756c3b0c43ec22 | 08 May 2020, 22:07:19 UTC |
39edad6 | Amanpreet Singh | 15 April 2020, 06:23:46 UTC | [fix] Vocab download in case of distributed settings (#87) Summary: - Only download if the node is master and then synchronize On a fresh install, the vocab won't work in distributed settings as all of the processes tried to install the vocab. Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/87 Reviewed By: vedanuj Differential Revision: D21035112 Pulled By: apsdehal fbshipit-source-id: 4e48b625d5014e1191997711b6229e70831217f1 | 08 May 2020, 22:07:19 UTC |
6aaa98a | Vedanuj Goswami | 15 April 2020, 06:10:40 UTC | [feature] Add Pretraining V&L paper project (#47) Summary: Adding project readme. Adding LMDBs and IMDBs for the project. Update all absolute paths to relative paths. Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/47 Reviewed By: apsdehal Differential Revision: D21027320 Pulled By: vedanuj fbshipit-source-id: 2bf7f84abd38317d4e568730f51f7611beeea442 | 08 May 2020, 22:07:19 UTC |
de89da8 | Vedanuj Goswami | 15 April 2020, 04:56:51 UTC | [fix] Fix spellings with Spell linter (#85) Summary: Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/85 Run Spell checker for all files Reviewed By: apsdehal Differential Revision: D21014867 fbshipit-source-id: 1a90d6435f1b9c8e6e99f89043b5e96f76564274 | 08 May 2020, 22:07:19 UTC |
ddafe4e | Amanpreet Singh | 14 April 2020, 23:17:13 UTC | Add script for generating test data (#84) Summary: - This script can be used for generating smaller size test data in all formats for testing purposes. This will be generate data in a specific format which will be used to override configuration in the tests Pull Request resolved: https://github.com/fairinternal/pythia-internal/pull/84 Reviewed By: vedanuj Differential Revision: D21027676 Pulled By: apsdehal fbshipit-source-id: fcf4c28af5b825c541f1d172abefcb8b0b6c9380 | 08 May 2020, 22:07:19 UTC |
7dbcfb1 | Vedanuj Goswami | 14 April 2020, 07:17:37 UTC | Enable MMF tests in fbcode Summary: Add test target for mmf Reviewed By: mamhamed Differential Revision: D21006909 fbshipit-source-id: cf2396fc79660b378f46afe575c2b95568d0faeb | 08 May 2020, 22:07:19 UTC |
d5f2094 | Vedanuj Goswami | 14 April 2020, 00:24:03 UTC | [lint] Fix a unused variable lint error (#83) | 08 May 2020, 22:07:19 UTC |
9b2f791 | Vedanuj Goswami | 13 April 2020, 22:36:03 UTC | [fix] Downgrade to transformers v2.3 (#82) | 08 May 2020, 22:07:19 UTC |
e8f4e9f | Amanpreet Singh | 13 April 2020, 21:43:50 UTC | [feature] Add slurm sweep script (#73) | 08 May 2020, 22:07:19 UTC |
f5daf27 | Amanpreet Singh | 13 April 2020, 21:43:19 UTC | [feature] Add ConcatBOW baseline (#71) | 08 May 2020, 22:07:19 UTC |
b98616e | Vedanuj Goswami | 13 April 2020, 21:37:31 UTC | [chores] Enable tests on GPU in CircleCI (#79) | 08 May 2020, 22:07:19 UTC |
17d16c9 | Vedanuj Goswami | 13 April 2020, 21:14:28 UTC | [fix] Use dummy config for bert classifier head test (#81) | 08 May 2020, 22:07:19 UTC |
5f1516d | Amanpreet Singh | 13 April 2020, 20:25:50 UTC | [fix] Fix config print and config save in checkpoint (#74) - pretty needs to be resolved before printing - save should save primitive type to avoid any dependencies | 08 May 2020, 22:07:19 UTC |
fac2917 | Amanpreet Singh | 13 April 2020, 19:35:45 UTC | [feature] Add Concat BERT (#70) * [feature] Add ConcatBERT * [enhancement] Add optimizer parameter and freeze, finetune options | 08 May 2020, 22:07:19 UTC |
72db8cf | Amanpreet Singh | 13 April 2020, 18:54:53 UTC | [refactor] Rename to MMF (#80) | 08 May 2020, 22:07:19 UTC |
47918c8 | Amanpreet Singh | 09 April 2020, 19:21:35 UTC | [feature] Add MMBT model (#64) This PR implements MMBT model described in https://arxiv.org/pdf/1909.02950.pdf. Salient features: - Two training modes: pretraining and classification - Works with both images as well as features - Can use any underlying encoder - Starter configurations provided for Hateful Memes and MaskedCOCO We provide two modes of training mmbt, pretraining and classification which are configurable through `training_head_type` configuration parameters. We also provide starter configuration that can be used to compose training configs for your use case. Sample configurations are provided for Hateful Memes and Masked COCO dataset. The model can work with either direct images or features provided from faster rcnn as is used in other Pythia models. A base is provided for easy building of further baselines. Follow the steps in #62 to build the setup for hateful memes dataset and then test following commands: - For training on hateful memes with images: ```sh python -u tools/run.py training.batch_size=16 config=projects/mmbt/configs/hateful_memes/defaults.yaml dataset=hateful_memes model=mmbt training.log_interval=10 training.find_unused_parameters=True training.num_workers=2 ``` - For training on hateful memes but with preextracted features: ```sh python -u tools/run.py training.batch_size=16 config=projects/mmbt/configs/hateful_memes/with_features.yaml dataset=hateful_memes model=mmbt training.log_interval=10 training.find_unused_parameters=True training.num_workers=2 ``` Both of the above commands invoke the classification MMBT model and #62 to be landed before they are run - For running pretraining on masked_coco with preextracted features run: ```sh python -u tools/run.py training.batch_size=16 config=projects/mmbt/configs/masked_coco/defaults.yaml dataset=masked_coco model=mmbt training.log_interval=10 training.find_unused_parameters=True training.num_workers=2 ``` | 08 May 2020, 22:07:19 UTC |
a06cd32 | Amanpreet Singh | 08 April 2020, 16:51:36 UTC | [feature] ROC_AUC, AP metric, separate out multilabel/multiclass in F1 (#66) * [feature] ROC_AUC metric, add support for class indices based targets in others * [fix] Address Vedanuj's comments, add AP metrics and many others * [fix] Update configs | 08 May 2020, 22:07:19 UTC |
740fbdd | Amanpreet Singh | 08 April 2020, 16:30:30 UTC | [feature] Add BERTPredictionHead and MLP classifier layers (#69) * [feature] Add BERTPredictionHead and MLP classifier layers - Also add build_classifier_layer method * [tests] Add tests for new layers | 08 May 2020, 22:07:19 UTC |
2fab3a3 | Amanpreet Singh | 07 April 2020, 22:53:43 UTC | [fix] Add simple ConfigNode back to allow old models loading (#68) | 08 May 2020, 22:07:19 UTC |
9b83196 | Vedanuj Goswami | 07 April 2020, 20:07:18 UTC | [refactor] Remove _utils from utils filenames (#65) | 08 May 2020, 22:07:19 UTC |
52a638f | Amanpreet Singh | 06 April 2020, 05:11:52 UTC | [refactor] ImageEmbedding,Encoder -> ImageFeatureEmbedding,Encoder (#63) | 08 May 2020, 22:07:19 UTC |
0728eee | Amanpreet Singh | 06 April 2020, 05:11:30 UTC | [feature,fix] Exclude list, png, jpeg, cls_prob for feature extraction (#58) - Fixes support for exclude list and allows it to be passed as args - Add support for feature extraction from png and jpeg files - Add "cls_prob", class probabilities field back to output_dict | 08 May 2020, 22:07:19 UTC |
cc45b03 | Vedanuj Goswami | 04 April 2020, 20:14:16 UTC | [fix] Change some configs, fix broken M4C configs, and project Readme files (#61) * [fix] Change some configs and project Readme files * [fix] add missing dataset configs in M4C series A few configs were broken in #55. This commits adds them back. * [fix] Fix pythia BERT configs Co-authored-by: Ronghang <ronghang.hu@gmail.com> | 08 May 2020, 22:07:19 UTC |
41a3bcc | Vedanuj Goswami | 04 April 2020, 20:10:46 UTC | [enhancement] Cache CircleCI tests, add isort, black tests (#60) | 08 May 2020, 22:07:19 UTC |
ec1c81d | Amanpreet Singh | 04 April 2020, 17:52:42 UTC | [refactor] Change datasets folder structure to better fit our needs (#59) | 08 May 2020, 22:07:19 UTC |
073eaf3 | Vedanuj Goswami | 04 April 2020, 01:02:23 UTC | [feature] Add Conceptual Captions dataset (#57) * [feature] Add Conceptual Captions Dataset * Address comments | 08 May 2020, 22:07:19 UTC |
e684373 | Amanpreet Singh | 03 April 2020, 23:56:48 UTC | [refactor] Move configs into projects folder with proper structure (#55) | 08 May 2020, 22:07:19 UTC |
c43c0be | Vedanuj Goswami | 03 April 2020, 19:02:38 UTC | [fix] Fix configs for model data dir attr (#56) | 08 May 2020, 22:07:19 UTC |
3ea1497 | Amanpreet Singh | 03 April 2020, 17:31:08 UTC | [feature] Allow relative paths in config includes (#53) | 08 May 2020, 22:07:19 UTC |
7652d0c | Vedanuj Goswami | 03 April 2020, 07:08:48 UTC | [chores] Add BAN project folder (#54) | 08 May 2020, 22:07:19 UTC |
aa7e53f | Amanpreet Singh | 03 April 2020, 00:46:56 UTC | [tests] Add 2 new tests to test default configs for models and datasets (#52) | 08 May 2020, 22:07:19 UTC |