Datasets:

---

renwei2024

/

nemo-github-issues

like 0

[ { "id": 1719393562, "node_id": "MDU6TGFiZWwxNzE5MzkzNTYy", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/feature", "name": "feature", "color": "68F3F0", "default": false, "description": "request/PR for a new feature" } ]

[ { "login": "okuchaiev", "id": 22035961, "node_id": "MDQ6VXNlcjIyMDM1OTYx", "avatar_url": "https://avatars.githubusercontent.com/u/22035961?v=4", "gravatar_id": "", "url": "https://api.github.com/users/okuchaiev", "html_url": "https://github.com/okuchaiev", "followers_url": "https://api.github.com/users/okuchaiev/followers", "following_url": "https://api.github.com/users/okuchaiev/following{/other_user}", "gists_url": "https://api.github.com/users/okuchaiev/gists{/gist_id}", "starred_url": "https://api.github.com/users/okuchaiev/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/okuchaiev/subscriptions", "organizations_url": "https://api.github.com/users/okuchaiev/orgs", "repos_url": "https://api.github.com/users/okuchaiev/repos", "events_url": "https://api.github.com/users/okuchaiev/events{/privacy}", "received_events_url": "https://api.github.com/users/okuchaiev/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "Hi @MelikaBahmanabadi Thanks for your question.\r\nWe have Voice Activity Detection support ([tutorial](https://github.com/NVIDIA/NeMo/blob/r1.0.0rc1/tutorials/asr/06_Voice_Activiy_Detection.ipynb)) in NeMo. For KWS I think you can start with that and train on data to have a classier to handle `the keyword vs. others`. \r\n\r\nif you are interested multi-word keyword spotting. You can use the combination of VAD and Keyword detection. If you are interested in this approach, have a look at the [03 tutorial](https://github.com/NVIDIA/NeMo/blob/r1.0.0rc1/tutorials/asr/03_Speech_Commands.ipynb) that talks about keyword detection training and finetuning on the new dataset, and [04 tutorial](https://github.com/NVIDIA/NeMo/blob/r1.0.0rc1/tutorials/asr/04_Online_Offline_Speech_Commands_Demo.ipynb) that talks about VAD+keyword detection inference. " ]

[]

[]

[ "This pull request **introduces 2 alerts** when merging da62342a19182f09fee4ec3d31e5b23bd0d7004c into e7977fd52de17f27befb4d36237584373adfe154 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-28e3ff6a1ceac31e742f1feca442469642e2da8b)\n\n**new alerts:**\n\n* 2 for Unused import" ]

[]

[]

[]

[]

[]

[ "This pull request **introduces 12 alerts** when merging af437cf4df8f2897529e1aa911b9288daa67d555 into 0b495ccbfc3bade395678fd3d0524febb799a9cb - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-bd47bbdee5cc95948826a8844a4ff81bc1d84c0b)\n\n**new alerts:**\n\n* 9 for Unused import\n* 2 for Unused local variable\n* 1 for Signature mismatch in overriding method", "This pull request **introduces 3 alerts** when merging 61e0b4e803e11cca3a0e39c86bea9317ed37d103 into 0b495ccbfc3bade395678fd3d0524febb799a9cb - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-c25dcc72d84a600981873466023c1a620d05414d)\n\n**new alerts:**\n\n* 2 for Unused import\n* 1 for Signature mismatch in overriding method", "This pull request **introduces 2 alerts** when merging 2cb00c22a46de35f96d6ca191107cf9483d569be into b4040fb37350ae86b64a5f53be911371d7a3879d - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-536a2ddac195c789efbf6f2bc3256e1cdbc13639)\n\n**new alerts:**\n\n* 2 for Signature mismatch in overriding method" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "@taufique74 first of all, thanks for catching the bug and PR! Since we would prefer to keep the line breaks which fits more the user's expectations i will replace this PR my this one https://github.com/NVIDIA/NeMo/pull/2101\r\nAnd close this one." ]

[]

[]

[]

[]

[]

[]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" }, { "id": 4910680898, "node_id": "LA_kwDOC_bI7s8AAAABJLMLQg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/TN/ITN", "name": "TN/ITN", "color": "0052cc", "default": false, "description": "" } ]

[]

[ "@yzhang123 ", "@taufique74 this is true, we do not handle upper cased words at this point, since we assume its the output of the asr model. please lower case your input" ]

[]

[]

[]

[]

[]

[]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" }, { "id": 4910680898, "node_id": "LA_kwDOC_bI7s8AAAABJLMLQg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/TN/ITN", "name": "TN/ITN", "color": "0052cc", "default": false, "description": "" } ]

[]

[ "Hey @yzhang123 \r\nLet me know if you'd be interested in a PR, it's a quick fix.", "@taufique74 sure go ahead! thanks!", "closing since https://github.com/NVIDIA/NeMo/pull/2101 is merged. Please re-open if the bug persists" ]

[]

[]

[ "This pull request **introduces 1 alert** when merging f5bf7fe4827595ac8b290ca8e002357ac4e29814 into 88af7a24a7ebb4418cb51c328e2c03aed8388e03 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-b5d3cd734613603214c2207eab3527795ce634af)\n\n**new alerts:**\n\n* 1 for Unused import" ]

[]

[]

[ "This pull request **introduces 2 alerts** when merging 6383d773b8cef1cb0206d9cdadec9b0abe610cf4 into 98bdf7a16e27cf4157f0ec0fcf78e6b0d6373c02 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-f4cf5d868dae323cc31e029daacb46a2d10f1e31)\n\n**new alerts:**\n\n* 2 for Unused import" ]

[]

[]

[]

[]

[]

[ "This pull request **fixes 1 alert** when merging e6ae53a4b3dde3e74a3b34778b6a0ffcf50a81f8 into cc1553b253e9d17369c9771ccd1af196bfac70a5 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-f9683a00beb9844ce1192d72bb3702e9183e045d)\n\n**fixed alerts:**\n\n* 1 for Unused import", "This pull request **fixes 1 alert** when merging 85839109d19b8c09e818f096ca0660b9c22bc09f into cc1553b253e9d17369c9771ccd1af196bfac70a5 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-7d08219436246b8357c295a4d6f5f684a6ce96d0)\n\n**fixed alerts:**\n\n* 1 for Unused import" ]

[]

[]

[]

[]

[]

[ "This pull request **introduces 24 alerts** when merging fcbfbdd0a2934da73a8898d6688a615c4431cde4 into fbc5c2b4033896b75f643cf169976c1d38505d65 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-719061033791abc13da3e5efcb9ec3fd7f2232e4)\n\n**new alerts:**\n\n* 20 for Unused import\n* 2 for Wrong name for an argument in a class instantiation\n* 1 for First parameter of a method is not named 'self'\n* 1 for Module is imported with 'import' and 'import from'", "This pull request **introduces 24 alerts** when merging e92091de87d1e4075119d8ddbf6bbef9d4637f85 into fbc5c2b4033896b75f643cf169976c1d38505d65 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-6a7597d154e1828be7a6bbe16fb002d9cc3841c5)\n\n**new alerts:**\n\n* 20 for Unused import\n* 2 for Wrong name for an argument in a class instantiation\n* 1 for First parameter of a method is not named 'self'\n* 1 for Module is imported with 'import' and 'import from'", "This pull request **introduces 2 alerts** when merging 6f651e929b25df79861c72ab4b8d96218a600d49 into 56649822d824eba946f243e96dbe215996e03794 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-f7a24d2d58168e700c327d02e02381645de149a9)\n\n**new alerts:**\n\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 2 alerts** when merging 98e85727baeb2d04870dd9db09138a50ead45247 into 56649822d824eba946f243e96dbe215996e03794 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-6dfebbace6d0dc646a131844e5d0e94e672fc2a9)\n\n**new alerts:**\n\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 2 alerts** when merging 5f30bdccaf633e1b84f852c78b71ad90a66e6279 into e59dd39f6f847155a82d40ebe2264a0b7f2b7a13 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-0c7417a07a9c793e952a899be06612c3c6b86ae9)\n\n**new alerts:**\n\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 2 alerts** when merging 7205ac6bbaeeca0ff33dcada6a75ac576efdea36 into 627956e12c349fa9fde8a71d20d997ff5060294a - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-faa1de9a1cb854f3c5d870c8d746db743ad2932a)\n\n**new alerts:**\n\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 2 alerts** when merging e21f4a98e065c6d7f153c6729c0f693695129630 into 627956e12c349fa9fde8a71d20d997ff5060294a - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-bbb9751b5f5c51509c5275e10d6a24f5e8650943)\n\n**new alerts:**\n\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 2 alerts** when merging 52c07569485b56ae5aff1ad8c89d11c048ff8bc3 into 12a7e66dc90c5a604477ecb1ce37fee7320ba70b - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-91c475c72c7f16693ffcb3cd64837c1ad9cfe5a1)\n\n**new alerts:**\n\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 2 alerts** when merging d9ed96a77cffab46235b549192189b9952c3956e into 855ce265b80c0dc40f4f06ece76d2c9d6ca1be8d - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-1dc4d2efff9b42387291d6f1d2b769f3ae6b211b)\n\n**new alerts:**\n\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 3 alerts** when merging 1efebb387963cad9a49c5d1a3c35166d7ac4bf78 into 3d92dcf5ec6972b63c30faaff0aca4ffae16671d - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-146bd724d5ec4ab34fb2e82fa17017c65b15a294)\n\n**new alerts:**\n\n* 2 for Wrong name for an argument in a class instantiation\n* 1 for Unused import", "This pull request **introduces 3 alerts** when merging 681b73229178e16a05fb472a0df1dc2e74bfd3cd into 6d53452eb1487102c0d5ae657815ec2aecc12076 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-14d6d7ec52d383cd0f58db0a5327536297db43bb)\n\n**new alerts:**\n\n* 2 for Wrong name for an argument in a class instantiation\n* 1 for Unused import", "This pull request **introduces 5 alerts** when merging b5c9cc405503f2ae1c3ae5082af5f41df76f50b6 into f0b3624dc1aee9670d85b82251ad4e488a622bc7 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-0729e653242fddcd97df8ff03a9e7ca01192206b)\n\n**new alerts:**\n\n* 2 for Unused import\n* 2 for Wrong name for an argument in a class instantiation\n* 1 for Unused local variable", "This pull request **introduces 5 alerts** when merging 26d8bbfaf04ba028f8f51a98e609c94397c119f0 into 08f3c65837adfcd092ad921c050ce1628cfcb8c4 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-84c4f8c0a8a7b64f84e05ea42872c57672675777)\n\n**new alerts:**\n\n* 2 for Unused import\n* 2 for Wrong name for an argument in a class instantiation\n* 1 for Unused local variable", "This pull request **introduces 5 alerts** when merging 78900c51c5876b8edf3af889c60d77b0d4e98c29 into c036cec3acc9c2cfb39fbbead20deb6bbdde7b94 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-7fca2a7efd53a775693cc81dc790fd11884b2a75)\n\n**new alerts:**\n\n* 3 for Unused import\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 5 alerts** when merging 1da5af7556b585226c2c9f707adda831eb5c5d56 into c036cec3acc9c2cfb39fbbead20deb6bbdde7b94 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-6e931d7898ca2d3bd25df3a6700063f62e57fb19)\n\n**new alerts:**\n\n* 3 for Unused import\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 3 alerts** when merging b89addb530138e43e63eff16ff32da04eace035c into cfe2548d4db9dcc4ca393ee63604139c9f681cd8 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-5097ff851790a2e1b3bed6b7257d4e01517d71af)\n\n**new alerts:**\n\n* 2 for Wrong name for an argument in a class instantiation\n* 1 for Unused import", "This pull request **introduces 4 alerts** when merging b4d60a12ef604bad8a742961832c32a07fc1e399 into cfe2548d4db9dcc4ca393ee63604139c9f681cd8 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-f906ed361f365dc5590d965d4fb3e3b05576b74f)\n\n**new alerts:**\n\n* 2 for Unused import\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 4 alerts** when merging 4b40a781139485255b49456d3c58594b5d89dbfd into e070e04ea219a53842bc6c3adb461a7b96c52519 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-04450883cae12e75669a7edc4f635fb5549c893a)\n\n**new alerts:**\n\n* 2 for Unused import\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 4 alerts** when merging 27742c0940775f6a0e07710f3a3fa4a40cf0d309 into 5fcfa9e7bf2da39d622071fc0cac592f993f89ea - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-b10899532ace4825ee1546889c2643817917aa4f)\n\n**new alerts:**\n\n* 2 for Unused import\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 4 alerts** when merging ce7f32161b2b7bfc157124b08902939f1b448e71 into 44a3d02b87692b45a1906972f63f127a6043c92b - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-7bf08e0a6bf7c379630a07f3bcb0f0c1aa612683)\n\n**new alerts:**\n\n* 2 for Unused import\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 4 alerts** when merging 6c56055038d556beb14213f7774f1adac4917665 into a8b6a1a4dd0d89ea76e5c3ee5e75a8d9104700dd - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-20973e3f34bba46b58b04f049d727d8d3eb2937d)\n\n**new alerts:**\n\n* 2 for Unused import\n* 2 for Wrong name for an argument in a class instantiation", "Hi @titu1994 ,\r\n\r\nThanks a lot for this awesome work.\r\nDo you have any idea when this pull request will be ready and merged to the main ?", "It will take time. There are experiments underway to see the effectiveness of this approach. ", "This pull request **introduces 4 alerts** when merging d14259af2a158beebd1f5fcd4c143f8005a0b956 into cd6a691607894f13cafcbe7994d1b2809ccb7aa4 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-316012550de5366a989699e74a0d486814898ba2)\n\n**new alerts:**\n\n* 2 for Unused import\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 4 alerts** when merging 8cb9374293a596b86650a75a1d0e917da4716e75 into aad6bbf143331c43b08816ea316533fc6c98dc2a - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-c037d3440a2ba8eed2860d4d689fcc5465201c71)\n\n**new alerts:**\n\n* 2 for Unused import\n* 2 for Wrong name for an argument in a class instantiation", "This pull request **introduces 1 alert** when merging 4fd7b9c09e9d41e48c855ee29838a5810d4090c3 into aad6bbf143331c43b08816ea316533fc6c98dc2a - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-cb1f4e1db9bdfc3840bbad87c5a6f56210b63b06)\n\n**new alerts:**\n\n* 1 for Unused import", "This pull request **fixes 1 alert** when merging 49fde6bcb30e119a852dfb46b7dbdf873e228f38 into aad6bbf143331c43b08816ea316533fc6c98dc2a - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-0bd56d32d0eab057e6edd21947aa52dbaae6efd0)\n\n**fixed alerts:**\n\n* 1 for Unused import", "This pull request **introduces 4 alerts** and **fixes 1** when merging 8225938eef1e1d01b0f8e8f65f18196fd8263127 into ec6591e76ab5d33e050e0dded25bed43d5d6c3e4 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-dff6fede60f9b2ce118407d88684cf8d43548f7f)\n\n**new alerts:**\n\n* 4 for Unused import\n\n**fixed alerts:**\n\n* 1 for Unused import", "Closing due to lack of necessity at the moment." ]

[]

[]

[ "This pull request **introduces 1 alert** and **fixes 2** when merging 0e2ecc61c2b1caa9eac17b42025c9df2b9bad6cf into 4ce74f134ca5c318f421a54e5fbd571879bbb418 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-3895299e38b5f30ef6af355511dca3a9d47a19f1)\n\n**new alerts:**\n\n* 1 for Unused import\n\n**fixed alerts:**\n\n* 2 for Unused import", "This pull request **introduces 1 alert** when merging 5d65457fd92da253990cc2ac2b4842d2924ba25d into fbfe24bc548cd88c1f87fa66b00b12ddfbdeab29 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-8a62b94a4caaf0cc2e95c2c739b8b7cbc516a4c4)\n\n**new alerts:**\n\n* 1 for Unused import" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "This pull request **introduces 2 alerts** when merging a4efaa132eae55f2663972c1d8f368b23419aff6 into c2d9559e0cf36dfeb88c1848e8360ae9a6b08776 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-2714799267083096cd0a3b6f4f534df19d9bcabb)\n\n**new alerts:**\n\n* 2 for Unused import", "is it draft or ready to review?\r\nAre there any feature models that uses these loaders, if isn't better to have it merge along with it ?!", "> is it draft or ready to review?\r\n> Are there any feature models that uses these loaders, if isn't better to have it merge along with it ?!\r\n \r\n@nithinraok ready to review. Well I thought about it. Tried to split huge PR into small things. And the feature loader should be less correlated with models. ", "General questions on overall feature loader:\r\ninstead of seq label inputted through manifest should we support reading from file?\r\nany helper function to get seq_label from rttm file?\r\ndoes each row in manifest correspond to each session or part of session? if its part of session can that be handled automatically for seq_label inputs (grouth truth) ", "> General questions on overall feature loader:\r\n> instead of seq label inputted through manifest should we support reading from file?\r\n\r\nNot for now. It will make data augmentation much complicated.\r\n> any helper function to get seq_label from rttm file?\r\n\r\nNo right now the emb and seq_label is generated offline. That is **not related to loader but data preparation imo** and shall include in the following PRs. \r\n\r\n\r\n> does each row in manifest correspond to each session or part of session? if its part of session can that be handled automatically for seq_label inputs (grouth truth)\r\n\r\n\r\nCould be each or part. Not quite understand this \"if its part of session can that be handled automatically for seq_label inputs (grouth truth)\". If I understand your question correctly, theoretically yes but as in above question, the augmentation will be difficult. This will be handled in neural clustering in clustering diarizer.\r\n", "This pull request **introduces 2 alerts** when merging 76265fa76f528d54098dbab4423701d950ad6770 into 6d53452eb1487102c0d5ae657815ec2aecc12076 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-329bc8e6c7700dfe43401c574fc0d3d7b214b4b2)\n\n**new alerts:**\n\n* 1 for Unused import\n* 1 for Illegal raise", "> Add the docstring, LGTM otherwise\r\n\r\n@titu1994 Thanks! Added. will merge once the checks finished." ]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" }, { "id": 4910680898, "node_id": "LA_kwDOC_bI7s8AAAABJLMLQg", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/TN/ITN", "name": "TN/ITN", "color": "0052cc", "default": false, "description": "" } ]

[ { "login": "yzhang123", "id": 4204271, "node_id": "MDQ6VXNlcjQyMDQyNzE=", "avatar_url": "https://avatars.githubusercontent.com/u/4204271?v=4", "gravatar_id": "", "url": "https://api.github.com/users/yzhang123", "html_url": "https://github.com/yzhang123", "followers_url": "https://api.github.com/users/yzhang123/followers", "following_url": "https://api.github.com/users/yzhang123/following{/other_user}", "gists_url": "https://api.github.com/users/yzhang123/gists{/gist_id}", "starred_url": "https://api.github.com/users/yzhang123/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/yzhang123/subscriptions", "organizations_url": "https://api.github.com/users/yzhang123/orgs", "repos_url": "https://api.github.com/users/yzhang123/repos", "events_url": "https://api.github.com/users/yzhang123/events{/privacy}", "received_events_url": "https://api.github.com/users/yzhang123/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "@mschonwe we changed a few things regarding normalization since 1.0.0rc.\r\n please refer to RC tutorial https://github.com/NVIDIA/NeMo/blob/r1.0.0rc1/tutorials/tools/Text_Normalization_Tutorial.ipynb which is tested to work in nvcr.io/nvidia/nemo:1.0.0rc1.", "Thank you. Got it working following the updated tutorial.", "@mschonwe we will be soon changing text normalization to WFST just like inverse text normalization, just FYI" ]

[]

[]

[ "It is normal to get 100% WER for first few epochs. How many epochs did train?", "Also, there might be a mismatch between the datasets. For example, is the sampling rate in your data the same as it is in the LibriSpeech dataset?", "around 50 epoch, after 50 epochs its crashing\r\n> It is normal to get 100% WER for first few epochs. How many epochs did train?\r\n\r\n", "> Also, there might be a mismatch between the datasets. For example, is the sampling rate in your data the same as it is in the LibriSpeech dataset?\r\n\r\nsample rate is same but when I make a test run on lesser data i.e. around 200 hours at that time it was decreasing\r\n\r\n\r\n\r\n\r\n\r\n\r\nBlue = ~200hours data\r\npink = ~400 hours data", "60s is unnecessarily long. We typically use max 16.7s. Btw, by default, it may ignore everything longer than that.\r\nSuch long fragments doesn't help with training too much but will require large amount of GPU ram.\r\n\r\nAlso, what is your character set in your transcriptions? For QuartzNet, it is lowercase En letters and apostrophe ('), no punctuation or numbers.\r\n\r\nTry lowering your learning rate and increasing batch size too.", "Several things - disable spec augment, reduce Weight Decay to 0 and see if the model can learn with no regularization first. \r\nThe fact that 200 hours works and 400 doesn't means that your learning rate is too high - with 200 hours you have roughly half the number of steps per epoch so warmup and all other parameters should be adjusted. " ]

[]

[ { "login": "fayejf", "id": 36722593, "node_id": "MDQ6VXNlcjM2NzIyNTkz", "avatar_url": "https://avatars.githubusercontent.com/u/36722593?v=4", "gravatar_id": "", "url": "https://api.github.com/users/fayejf", "html_url": "https://github.com/fayejf", "followers_url": "https://api.github.com/users/fayejf/followers", "following_url": "https://api.github.com/users/fayejf/following{/other_user}", "gists_url": "https://api.github.com/users/fayejf/gists{/gist_id}", "starred_url": "https://api.github.com/users/fayejf/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/fayejf/subscriptions", "organizations_url": "https://api.github.com/users/fayejf/orgs", "repos_url": "https://api.github.com/users/fayejf/repos", "events_url": "https://api.github.com/users/fayejf/events{/privacy}", "received_events_url": "https://api.github.com/users/fayejf/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "Thanks for your question.\r\nYes. Have a look at **_VAD diarization_** part in diarization [tutorial](https://github.com/NVIDIA/NeMo/blob/r1.0.0rc1/tutorials/speaker_recognition/Speaker_Diarization_Inference.ipynb ) which will generate VAD RTTM-like json file that could suit your request.\r\n", "Yes, thank you for the response." ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[ "This pull request **introduces 1 alert** when merging 760f3012e9f086af1d1893d0a1c378bd2b2aabab into bee43e840837d7e30e17a65b972e9fe89ecb9ca5 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-7b0cbe4483c3653568bbc31df8b151ec2086aefc)\n\n**new alerts:**\n\n* 1 for Unused local variable" ]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" } ]

[]

[ "@titu1994 any thoughts? ", "Just to text, can you add accelerator=\"ddp\" (even for 1 gpu, just a test). \r\n\r\nAs far as I can parse the error, it's due to a deep copy of a model which can't be pickled, so it is raising a bound function wrapper error. \r\n\r\nYou'll probably have to come up with a toy example (which is unpickleable) and try the same to see if that's indeed the reason. ", "That actually did something. I got this other exception instead:\r\n\r\n---------------------------------------------------------------------------\r\nModuleAttributeError Traceback (most recent call last)\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py in run_train(self)\r\n 636 # run train epoch\r\n--> 637 self.train_loop.run_training_epoch()\r\n 638 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/training_loop.py in run_training_epoch(self)\r\n 576 if should_check_val:\r\n--> 577 self.trainer.run_evaluation(on_epoch=True)\r\n 578 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py in run_evaluation(self, max_batches, on_epoch)\r\n 751 # hook\r\n--> 752 self.evaluation_loop.on_evaluation_end()\r\n 753 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/evaluation_loop.py in on_evaluation_end(self, *args, **kwargs)\r\n 99 else:\r\n--> 100 self.trainer.call_hook('on_validation_end', *args, **kwargs)\r\n 101 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py in call_hook(self, hook_name, *args, **kwargs)\r\n 1094 trainer_hook = getattr(self, hook_name)\r\n-> 1095 trainer_hook(*args, **kwargs)\r\n 1096 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/callback_hook.py in on_validation_end(self)\r\n 184 for callback in self.callbacks:\r\n--> 185 callback.on_validation_end(self, self.lightning_module)\r\n 186 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/callbacks/model_checkpoint.py in on_validation_end(self, trainer, pl_module)\r\n 211 \"\"\"\r\n--> 212 self.save_checkpoint(trainer, pl_module)\r\n 213 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/callbacks/model_checkpoint.py in save_checkpoint(self, trainer, pl_module)\r\n 258 if self.save_top_k:\r\n--> 259 self._save_top_k_checkpoints(trainer, pl_module, monitor_candidates)\r\n 260 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/callbacks/model_checkpoint.py in _save_top_k_checkpoints(self, trainer, pl_module, metrics)\r\n 562 if self.check_monitor_top_k(trainer, current):\r\n--> 563 self._update_best_and_save(current, epoch, step, trainer, pl_module, metrics)\r\n 564 elif self.verbose:\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/callbacks/model_checkpoint.py in _update_best_and_save(self, current, epoch, step, trainer, pl_module, ckpt_name_metrics)\r\n 604 )\r\n--> 605 self._save_model(filepath, trainer, pl_module)\r\n 606 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/callbacks/model_checkpoint.py in _save_model(self, filepath, trainer, pl_module)\r\n 334 if self.save_function is not None:\r\n--> 335 self.save_function(filepath, self.save_weights_only)\r\n 336 else:\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/properties.py in save_checkpoint(self, filepath, weights_only)\r\n 326 def save_checkpoint(self, filepath, weights_only: bool = False) -> None:\r\n--> 327 self.checkpoint_connector.save_checkpoint(filepath, weights_only)\r\n 328 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/connectors/checkpoint_connector.py in save_checkpoint(self, filepath, weights_only)\r\n 396 # dump states as a checkpoint dictionary object\r\n--> 397 checkpoint = self.dump_checkpoint(weights_only)\r\n 398 if self.trainer.is_global_zero:\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/connectors/checkpoint_connector.py in dump_checkpoint(self, weights_only)\r\n 283 # dump callbacks\r\n--> 284 checkpoint['callbacks'] = self.trainer.on_save_checkpoint(checkpoint)\r\n 285 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/callback_hook.py in on_save_checkpoint(self, checkpoint)\r\n 221 else:\r\n--> 222 state = callback.on_save_checkpoint(self, self.lightning_module, checkpoint)\r\n 223 if state:\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/callbacks/pruning.py in on_save_checkpoint(self, trainer, pl_module, checkpoint)\r\n 399 # prune a copy so training can continue with the same buffers\r\n--> 400 copy = deepcopy(pl_module.to(\"cpu\"))\r\n 401 self.make_pruning_permanent(copy)\r\n\r\n/opt/conda/lib/python3.8/copy.py in deepcopy(x, memo, _nil)\r\n 171 else:\r\n--> 172 y = _reconstruct(x, memo, *rv)\r\n 173 \r\n\r\n/opt/conda/lib/python3.8/copy.py in _reconstruct(x, memo, func, args, state, listiter, dictiter, deepcopy)\r\n 269 if deep:\r\n--> 270 state = deepcopy(state, memo)\r\n 271 if hasattr(y, '__setstate__'):\r\n\r\n/opt/conda/lib/python3.8/copy.py in deepcopy(x, memo, _nil)\r\n 145 if copier is not None:\r\n--> 146 y = copier(x, memo)\r\n 147 else:\r\n\r\n/opt/conda/lib/python3.8/copy.py in _deepcopy_dict(x, memo, deepcopy)\r\n 229 for key, value in x.items():\r\n--> 230 y[deepcopy(key, memo)] = deepcopy(value, memo)\r\n 231 return y\r\n\r\n/opt/conda/lib/python3.8/copy.py in deepcopy(x, memo, _nil)\r\n 171 else:\r\n--> 172 y = _reconstruct(x, memo, *rv)\r\n 173 \r\n\r\n/opt/conda/lib/python3.8/copy.py in _reconstruct(x, memo, func, args, state, listiter, dictiter, deepcopy)\r\n 269 if deep:\r\n--> 270 state = deepcopy(state, memo)\r\n 271 if hasattr(y, '__setstate__'):\r\n\r\n/opt/conda/lib/python3.8/copy.py in deepcopy(x, memo, _nil)\r\n 145 if copier is not None:\r\n--> 146 y = copier(x, memo)\r\n 147 else:\r\n\r\n/opt/conda/lib/python3.8/copy.py in _deepcopy_dict(x, memo, deepcopy)\r\n 229 for key, value in x.items():\r\n--> 230 y[deepcopy(key, memo)] = deepcopy(value, memo)\r\n 231 return y\r\n\r\n/opt/conda/lib/python3.8/copy.py in deepcopy(x, memo, _nil)\r\n 171 else:\r\n--> 172 y = _reconstruct(x, memo, *rv)\r\n 173 \r\n\r\n/opt/conda/lib/python3.8/copy.py in _reconstruct(x, memo, func, args, state, listiter, dictiter, deepcopy)\r\n 269 if deep:\r\n--> 270 state = deepcopy(state, memo)\r\n 271 if hasattr(y, '__setstate__'):\r\n\r\n/opt/conda/lib/python3.8/copy.py in deepcopy(x, memo, _nil)\r\n 145 if copier is not None:\r\n--> 146 y = copier(x, memo)\r\n 147 else:\r\n\r\n/opt/conda/lib/python3.8/copy.py in _deepcopy_dict(x, memo, deepcopy)\r\n 229 for key, value in x.items():\r\n--> 230 y[deepcopy(key, memo)] = deepcopy(value, memo)\r\n 231 return y\r\n\r\n/opt/conda/lib/python3.8/copy.py in deepcopy(x, memo, _nil)\r\n 171 else:\r\n--> 172 y = _reconstruct(x, memo, *rv)\r\n 173 \r\n\r\n/opt/conda/lib/python3.8/copy.py in _reconstruct(x, memo, func, args, state, listiter, dictiter, deepcopy)\r\n 269 if deep:\r\n--> 270 state = deepcopy(state, memo)\r\n 271 if hasattr(y, '__setstate__'):\r\n\r\n/opt/conda/lib/python3.8/copy.py in deepcopy(x, memo, _nil)\r\n 145 if copier is not None:\r\n--> 146 y = copier(x, memo)\r\n 147 else:\r\n\r\n/opt/conda/lib/python3.8/copy.py in _deepcopy_dict(x, memo, deepcopy)\r\n 229 for key, value in x.items():\r\n--> 230 y[deepcopy(key, memo)] = deepcopy(value, memo)\r\n 231 return y\r\n\r\n/opt/conda/lib/python3.8/copy.py in deepcopy(x, memo, _nil)\r\n 171 else:\r\n--> 172 y = _reconstruct(x, memo, *rv)\r\n 173 \r\n\r\n/opt/conda/lib/python3.8/copy.py in _reconstruct(x, memo, func, args, state, listiter, dictiter, deepcopy)\r\n 271 if hasattr(y, '__setstate__'):\r\n--> 272 y.__setstate__(state)\r\n 273 else:\r\n\r\n/opt/conda/lib/python3.8/site-packages/torch/nn/parallel/distributed.py in __setstate__(self, state)\r\n 549 self.__dict__.setdefault('require_backward_grad_sync', True)\r\n--> 550 self._ddp_init_helper()\r\n 551 \r\n\r\n/opt/conda/lib/python3.8/site-packages/torch/nn/parallel/distributed.py in _ddp_init_helper(self)\r\n 479 \r\n--> 480 self.modules_params = [list(parameters(m)) for m in self._module_copies]\r\n 481 self.modules_buffers = [list(m.buffers()) for m in self._module_copies]\r\n\r\n/opt/conda/lib/python3.8/site-packages/torch/nn/parallel/distributed.py in <listcomp>(.0)\r\n 479 \r\n--> 480 self.modules_params = [list(parameters(m)) for m in self._module_copies]\r\n 481 self.modules_buffers = [list(m.buffers()) for m in self._module_copies]\r\n\r\n/opt/conda/lib/python3.8/site-packages/torch/nn/parallel/distributed.py in parameters(m, recurse)\r\n 441 for m in m.modules() if recurse else [m]:\r\n--> 442 for p in model_parameters(m):\r\n 443 yield p\r\n\r\n/opt/conda/lib/python3.8/site-packages/torch/nn/parallel/distributed.py in model_parameters(m)\r\n 437 else m.parameters(recurse=False)\r\n--> 438 for p in ps:\r\n 439 yield p\r\n\r\n/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py in parameters(self, recurse)\r\n 1088 \"\"\"\r\n-> 1089 for name, param in self.named_parameters(recurse=recurse):\r\n 1090 yield param\r\n\r\n/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py in named_parameters(self, prefix, recurse)\r\n 1114 prefix=prefix, recurse=recurse)\r\n-> 1115 for elem in gen:\r\n 1116 yield elem\r\n\r\n/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py in _named_members(self, get_members_fn, prefix, recurse)\r\n 1059 for module_prefix, module in modules:\r\n-> 1060 members = get_members_fn(module)\r\n 1061 for k, v in members:\r\n\r\n/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py in <lambda>(module)\r\n 1112 gen = self._named_members(\r\n-> 1113 lambda module: module._parameters.items(),\r\n 1114 prefix=prefix, recurse=recurse)\r\n\r\n/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py in __getattr__(self, name)\r\n 777 return modules[name]\r\n--> 778 raise ModuleAttributeError(\"'{}' object has no attribute '{}'\".format(\r\n 779 type(self).__name__, name))\r\n\r\nModuleAttributeError: 'EncDecCTCModel' object has no attribute '_parameters'\r\n\r\nDuring handling of the above exception, another exception occurred:\r\n\r\nRuntimeError Traceback (most recent call last)\r\n<ipython-input-17-2116e952a23e> in <module>\r\n 1 # Start training!!!\r\n----> 2 trainer.fit(first_asr_model)\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py in fit(self, model, train_dataloader, val_dataloaders, datamodule)\r\n 497 \r\n 498 # dispath `start_training` or `start_testing` or `start_predicting`\r\n--> 499 self.dispatch()\r\n 500 \r\n 501 # plugin will finalized fitting (e.g. ddp_spawn will load trained model)\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py in dispatch(self)\r\n 544 \r\n 545 else:\r\n--> 546 self.accelerator.start_training(self)\r\n 547 \r\n 548 def train_or_test_or_predict(self):\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/accelerators/accelerator.py in start_training(self, trainer)\r\n 71 \r\n 72 def start_training(self, trainer):\r\n---> 73 self.training_type_plugin.start_training(trainer)\r\n 74 \r\n 75 def start_testing(self, trainer):\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/plugins/training_type/training_type_plugin.py in start_training(self, trainer)\r\n 112 def start_training(self, trainer: 'Trainer') -> None:\r\n 113 # double dispatch to initiate the training loop\r\n--> 114 self._results = trainer.run_train()\r\n 115 \r\n 116 def start_testing(self, trainer: 'Trainer') -> None:\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py in run_train(self)\r\n 668 finally:\r\n 669 # hook\r\n--> 670 self.train_loop.on_train_end()\r\n 671 \r\n 672 def run_evaluation(self, max_batches=None, on_epoch=False):\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/training_loop.py in on_train_end(self)\r\n 132 # when a checkpoint was saved at the last step\r\n 133 self.trainer.global_step -= 1\r\n--> 134 self.check_checkpoint_callback(should_update=True, is_last=True)\r\n 135 self.trainer.global_step += 1\r\n 136 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/training_loop.py in check_checkpoint_callback(self, should_update, is_last)\r\n 162 \r\n 163 for cb in callbacks:\r\n--> 164 cb.on_validation_end(self.trainer, model)\r\n 165 \r\n 166 def check_early_stopping_callback(self, should_update):\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/callbacks/model_checkpoint.py in on_validation_end(self, trainer, pl_module)\r\n 210 checkpoints can be saved at the end of the val loop\r\n 211 \"\"\"\r\n--> 212 self.save_checkpoint(trainer, pl_module)\r\n 213 \r\n 214 def on_save_checkpoint(self, trainer, pl_module, checkpoint: Dict[str, Any]) -> Dict[str, Any]:\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/callbacks/model_checkpoint.py in save_checkpoint(self, trainer, pl_module)\r\n 257 # Mode 1: save all checkpoints OR only the top k\r\n 258 if self.save_top_k:\r\n--> 259 self._save_top_k_checkpoints(trainer, pl_module, monitor_candidates)\r\n 260 \r\n 261 # Mode 2: save the last checkpoint\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/callbacks/model_checkpoint.py in _save_top_k_checkpoints(self, trainer, pl_module, metrics)\r\n 560 step = metrics.get(\"step\")\r\n 561 \r\n--> 562 if self.check_monitor_top_k(trainer, current):\r\n 563 self._update_best_and_save(current, epoch, step, trainer, pl_module, metrics)\r\n 564 elif self.verbose:\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/callbacks/model_checkpoint.py in check_monitor_top_k(self, trainer, current)\r\n 360 \r\n 361 # If using multiple devices, make sure all processes are unanimous on the decision.\r\n--> 362 should_update_best_and_save = trainer.training_type_plugin.reduce_boolean_decision(should_update_best_and_save)\r\n 363 \r\n 364 return should_update_best_and_save\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/plugins/training_type/parallel.py in reduce_boolean_decision(self, decision)\r\n 90 def reduce_boolean_decision(self, decision: bool) -> bool:\r\n 91 decision = torch.tensor(int(decision), device=self.lightning_module.device)\r\n---> 92 decision = self.reduce(decision, reduce_op=ReduceOp.SUM)\r\n 93 decision = bool(decision == self.world_size)\r\n 94 return decision\r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/plugins/training_type/ddp.py in reduce(self, output, group, reduce_op)\r\n 308 def reduce(self, output, group: Optional[Any] = None, reduce_op: Optional[Union[ReduceOp, str]] = None):\r\n 309 if isinstance(output, torch.Tensor):\r\n--> 310 output = sync_ddp_if_available(output, group, reduce_op)\r\n 311 return output\r\n 312 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/utilities/distributed.py in sync_ddp_if_available(result, group, reduce_op)\r\n 130 \"\"\"\r\n 131 if torch.distributed.is_available() and torch.distributed.is_initialized():\r\n--> 132 return sync_ddp(result, group=group, reduce_op=reduce_op)\r\n 133 return result\r\n 134 \r\n\r\n/opt/conda/lib/python3.8/site-packages/pytorch_lightning/utilities/distributed.py in sync_ddp(result, group, reduce_op)\r\n 163 # sync all processes before reduction\r\n 164 torch.distributed.barrier(group=group)\r\n--> 165 torch.distributed.all_reduce(result, op=op, group=group, async_op=False)\r\n 166 \r\n 167 if divide_by_world_size:\r\n\r\n/opt/conda/lib/python3.8/site-packages/torch/distributed/distributed_c10d.py in all_reduce(tensor, op, group, async_op)\r\n 952 if group == GroupMember.WORLD:\r\n 953 _check_default_pg()\r\n--> 954 work = _default_pg.allreduce([tensor], opts)\r\n 955 else:\r\n 956 work = group.allreduce([tensor], opts)\r\n\r\nRuntimeError: Tensors must be CUDA and dense", "BTW I changed docker settings to check if the the issue also happens in other python versions. I downloaded the latest pytorch docker, and added the following lines to the requirements so the pytorch versions will match:\r\n```bash\r\ntorch==1.7.1\r\ntorchtext==0.8.1\r\n```\r\nI'll try that again with the 20.11 docker later.", "What version of cuda and cudnn is currently installed on your system? It's trying to all reduce the integer flag which acts as a boolean, but it fails cause ints are not dense and the step value is not on cuda. \r\n\r\nI think you can't use ddp + this pruning. We haven't tried and don't support it since it's not tested. I would suggest creating a minimal unpickleable PT lightning module and then trying with and without ddp if it works. If even that doesn't work, it should be brought up to Pytorch lightning team. ", "I understand what's the problem, and I made a hacky solution.\r\nApparently, to make the pruning permanent in the saved model, during checkpoints the ModelPruning callback deep- copies the model, makes the pruning permanent in the copy and then replaces the checkpoint's state dict by the copy's dict:\r\n```python\r\n def on_save_checkpoint(self, trainer, pl_module: LightningModule, checkpoint: Dict[str, Any]):\r\n if self._make_pruning_permanent:\r\n rank_zero_debug(\"`ModelPruning.on_save_checkpoint`. Pruning is made permanent for this checkpoint.\")\r\n prev_device = pl_module.device\r\n # prune a copy so training can continue with the same buffers \r\n copy = deepcopy(pl_module.to(\"cpu\"))\r\n \r\n self.make_pruning_permanent(copy)\r\n checkpoint[\"state_dict\"] = copy.state_dict()\r\n \r\n pl_module.to(prev_device)\r\n```\r\n\r\nThe EncDecCTCModel module has several attributes that cannot be deep- copied for various reasons (I tried deep copying all of them by themselves and they caused exceptions, and removing all of them was required in my solution):\r\n- `trainer` (has an attribute dev_debugger which is a weakref, which has a weakref back to itself)\r\n- `_trainer` (same)\r\n- `_train_dl` (Lightning Data Loader, I'm not sure why it's uncopiable)\r\n- `_validation_dl` (same)\r\n- `training_step` (a function wrapper, and so uncopiable)\r\n\r\nMy hacky solution is replacing all the offending attributes by `None`, letting the callback do its thing and then place them back in the module:\r\n```python\r\n def on_save_checkpoint(self, trainer, pl_module: LightningModule, checkpoint: Dict[str, Any]):\r\n if self._make_pruning_permanent:\r\n rank_zero_debug(\"`ModelPruning.on_save_checkpoint`. Pruning is made permanent for this checkpoint.\")\r\n prev_device = pl_module.device\r\n # prune a copy so training can continue with the same buffers\r\n \r\n trainer = pl_module.trainer\r\n _trainer = pl_module._trainer\r\n _train_dl = pl_module._train_dl\r\n _validation_dl = pl_module._validation_dl\r\n training_step = pl_module.training_step\r\n \r\n pl_module.trainer = None\r\n pl_module._trainer = None\r\n pl_module._train_dl = None\r\n pl_module._validation_dl = None\r\n pl_module.training_step = None\r\n \r\n copy = deepcopy(pl_module.to(\"cpu\"))\r\n \r\n self.make_pruning_permanent(copy)\r\n checkpoint[\"state_dict\"] = copy.state_dict()\r\n \r\n pl_module.trainer = trainer\r\n pl_module._trainer = _trainer\r\n pl_module._train_dl = _train_dl\r\n pl_module._validation_dl = _validation_dl\r\n pl_module.training_step = training_step\r\n \r\n pl_module.to(prev_device)\r\n```\r\nA less hacky solution would be to replace the `__deepcolpy__` method of EncDecCTCModel with something that does not try to copy uncopyable attributes. An even less hacky solution would be to have a ModelPruning implementation which doesn't deepcopies the whole model just to operate on its weights, but I'm not sure how that would work.\r\nThanks for your help! " ]

[]

[]

[ "This question is a little bit vague. Would you mind expanding on what you're trying to do? For example, what NeMo model are you working with and what would you like to do with the hidden states?" ]

[]

[]

[ "We have completely changed NeMo since .11. We highly recommend to start using our latest stable version. Please see our User Guide: https://docs.nvidia.com/deeplearning/nemo/user-guide/docs/en/main/ and our Tutorials: https://docs.nvidia.com/deeplearning/nemo/user-guide/docs/en/main/starthere/tutorials.html for how to get started with NeMo. " ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[ "This pull request **introduces 2 alerts** when merging a5a058a6d13568253e1a2e49d996cdc6cda8c522 into 21a17b267fac68d4cdd20f3969a580a0a40dbdb4 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-d8d2c21d7f0c2b2b08d057ea93e8c9524bb8826e)\n\n**new alerts:**\n\n* 1 for Unused local variable\n* 1 for Variable defined multiple times", "This pull request **introduces 4 alerts** when merging f65dcb8fc43a37883b2343b497f3b18df9d9f543 into 21a17b267fac68d4cdd20f3969a580a0a40dbdb4 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-5835aa9bd664c563372cbc59bb7d9fb020b375b7)\n\n**new alerts:**\n\n* 2 for Unused local variable\n* 1 for Unused import\n* 1 for Variable defined multiple times", "This pull request **introduces 4 alerts** when merging 2430c462a3149d4f23282ee4656669d72583466b into 21a17b267fac68d4cdd20f3969a580a0a40dbdb4 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-75fdc635019f8a33e0eb0f85f8233f3dd0a07f64)\n\n**new alerts:**\n\n* 3 for Unused local variable\n* 1 for Variable defined multiple times", "This pull request **introduces 1 alert** when merging 5b4ec59ec2d5ba1d08cee6480a6ed1fe7b67cda4 into 21a17b267fac68d4cdd20f3969a580a0a40dbdb4 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-b971f3c77917d12253297b3fdb3113f788478f88)\n\n**new alerts:**\n\n* 1 for Unused local variable", "This pull request **introduces 1 alert** when merging 52939fe8ce2bd485df8ac801f452d96b4d43a4e0 into ba027d0e88a588ed4bd7e4142366db127c396b2d - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-38e635a5f00f4add2b108de3d5ba48046f96a9a9)\n\n**new alerts:**\n\n* 1 for Unused local variable", "This pull request **introduces 1 alert** when merging abb0cbed21a24e9684f154cb59386224f138f8b1 into 54e6f6ee688f09810d3e54661275fd5c8718db00 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-cf3527523c3b634b3e0cdec6d178dad54e2d265f)\n\n**new alerts:**\n\n* 1 for Testing equality to None" ]

[]

[]

[ "Duplicate of https://github.com/NVIDIA/NeMo/discussions/2072" ]

[]

[]

[ "I think @ericharper may have had this in place on purpose to make sure both metadata and tarfiles were provided and not just the metadata. I'm fine with providing just the metadata JSON file (even better would be to allow specifying preproc_out_dir to avoid specifying tokenizers as well). \r\n\r\nCan you also handle the case where `cfg.train_ds.get('tar_files')` is not None and `cfg.train_ds.get('metadata_file')` is None and throw an error in that case since we always require the metadata file.", "> I think @ericharper may have had this in place on purpose to make sure both metadata and tarfiles were provided and not just the metadata. I'm fine with providing just the metadata JSON file (even better would be to allow specifying preproc_out_dir to avoid specifying tokenizers as well).\r\n> \r\n> Can you also handle the case where `cfg.train_ds.get('tar_files')` is not None and `cfg.train_ds.get('metadata_file')` is None and throw an error in that case since we always require the metadata file.\r\n\r\nYes, I can handle that case as well. " ]

[]

[]

[]

[]

[]

[ "yes, we support other languages https://docs.nvidia.com/deeplearning/nemo/user-guide/docs/en/main/asr/results.html \r\nBut at the moment, we don't have Hindi ASR model." ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[ "@yzhang123 there is merge conflict\r\n", "> @yzhang123 there is merge conflict\r\n\r\n@okuchaiev fixed" ]

[]

[]

[ "Hi, Ulucsahin\r\n\r\nThe `speech_to_text_quant_infer.py` script is meant for evaluating numerical property and accuracy of the quantized network, not speed/throughput. It is expected to be slower because it runs emulated quantization.\r\n\r\nThe script can generate an ONNX graph with quantization nodes. To deploy the quantized model for optimal performance, it will need TensorRT 8.0 (Refert this [GTC talk](https://gtc21.event.nvidia.com/media/Quantization%20Aware%20Training%20in%20PyTorch%20with%20TensorRT%20%5BS31653%5D/1_qdvvff64) for more detail) to compile the model to int8 I/O and compute. `speech_to_text_quant_infer_trt.py` is the script to run the quantized model by TensorRT and get optimal performance. Note that quantization features are designed for GPU and TensorRT only. To run the quantized model on CPU, it will need a different inference engine that supports the model\r\n" ]

[]

[]

[ "This pull request **introduces 18 alerts** when merging ed2b68b7bc10a626a2ae82fb0cbcc0e3f83a456c into 52121810b428e011666e656d313b42c954def26b - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-8fbe4d26b21589a2bf3447334aae25b590af73e0)\n\n**new alerts:**\n\n* 16 for Unused import\n* 1 for Redundant assignment\n* 1 for Wrong number of arguments in a call", "This pull request **introduces 18 alerts** when merging d2e88876defc4ed63869d6039c56dda0608ce3de into 6a3753b3013dc92a3587853d60c5086e2e64d98f - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-c3536e429970831dfc99cb496ffa96fb4a00429a)\n\n**new alerts:**\n\n* 16 for Unused import\n* 1 for Redundant assignment\n* 1 for Wrong number of arguments in a call", "This pull request **introduces 18 alerts** when merging 317ea4f076e29e3b908e1a3dd607ae2cb7b1f0f1 into 6a3753b3013dc92a3587853d60c5086e2e64d98f - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-94e5609abbf38db7fee58e2c7d92bf7bc3f431f3)\n\n**new alerts:**\n\n* 16 for Unused import\n* 1 for Redundant assignment\n* 1 for Wrong number of arguments in a call", "This pull request **introduces 2 alerts** when merging 11416e6697e79ce9f2b56103e89d6ac46b4ec1b5 into 6a3753b3013dc92a3587853d60c5086e2e64d98f - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-150f56213661189bb9ce76917fae1a37eaba1abf)\n\n**new alerts:**\n\n* 2 for Unused import", "This pull request **introduces 2 alerts** when merging 84e4ef4b4942977db7663f57e3f4cb823d7c508e into 6a3753b3013dc92a3587853d60c5086e2e64d98f - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-475e98e26e811f2d7bdf5842b5c7b5eb9c0ada8b)\n\n**new alerts:**\n\n* 2 for Unused import", "This pull request **introduces 1 alert** when merging 7fac138aab27c318fe8d641cff936de277d217ab into 400632a229d35e42d6a29f398fd16bea11fb897d - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-09217f2d0ec0910ce5c7a80b1e0e68d7f9174ae8)\n\n**new alerts:**\n\n* 1 for Syntax error", "This pull request **introduces 1 alert** when merging cb2deb82ce521026af4153e2054f4f523c6a124e into 400632a229d35e42d6a29f398fd16bea11fb897d - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-e22828d3e8eb3af5882eab32e7fc78395443e272)\n\n**new alerts:**\n\n* 1 for Syntax error", "This pull request **introduces 2 alerts** when merging 7e8ea27f55293f169fe43b8967cd34cbfb0437ff into 6ce670b7ab3ab5c84ce388fa1968748cd859b8f8 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-80b46b15d87dfaed714c65d36a615d5bc2bd9283)\n\n**new alerts:**\n\n* 2 for Unused import", "This pull request **introduces 2 alerts** when merging a942bfc42f3dee060fb576fbba95de8b9a0d25de into 6ce670b7ab3ab5c84ce388fa1968748cd859b8f8 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-95a7925bf90f3ea3fb6201b277158573954c658d)\n\n**new alerts:**\n\n* 2 for Unused import", "This pull request **introduces 2 alerts** when merging 7a649b5a4ba8db70b6b801f42e473e0ecb44bba3 into a806e044c6a1c583ea14caec7b219241c6924fa5 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-6e4add3d9214ec7d6ad6374c4e2f31d38f50a274)\n\n**new alerts:**\n\n* 2 for Unused import", "This pull request **introduces 2 alerts** when merging 79db128ea469a01f57a6807679c47a8d5bba32e5 into bf9f9ec29e1add0548e40f8b62d100f5e49f1b57 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-20d989a5d893f8743fd18d9ef8b88fee072e5f45)\n\n**new alerts:**\n\n* 2 for Unused import", "This pull request **introduces 2 alerts** when merging 26d25571bdd3b51b6e6f93023e82511efb2195c8 into d017cfcd16e4eb714b9188ccf4df2a8fc69471b7 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-72194d9c524383e779cbfbddd9aca9a06ca3fef6)\n\n**new alerts:**\n\n* 2 for Unused import", "This pull request **introduces 2 alerts** when merging 39f300fe2c7dc13e43fd70bf8d5a04bee4fce01f into 26e5d3a9b3cc934db0004b1dd9263fb7bce9a964 - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-f88c4f49aed8af7aa9bd7fcfd1cc423a02616413)\n\n**new alerts:**\n\n* 2 for Unused import", "This pull request **introduces 2 alerts** when merging a8615668105bf7eee8359382883f0ed387f06028 into 5316f2bcd0e326d336cb35bd8222b00914a5b77f - [view on LGTM.com](https://lgtm.com/projects/g/NVIDIA/NeMo/rev/pr-acb4c73634f3755b2f7f76c4fecd143ce88cf349)\n\n**new alerts:**\n\n* 2 for Unused import" ]

[ { "id": 1485815557, "node_id": "MDU6TGFiZWwxNDg1ODE1NTU3", "url": "https://api.github.com/repos/NVIDIA/NeMo/labels/bug", "name": "bug", "color": "d73a4a", "default": true, "description": "Something isn't working" } ]

[]

[ "this is not a bug. The container version looks like 1.8.0\"something\" - where \"something\" is commit id pre 1.8.0 release but a little after 1.7.1" ]

[]

[ { "login": "fayejf", "id": 36722593, "node_id": "MDQ6VXNlcjM2NzIyNTkz", "avatar_url": "https://avatars.githubusercontent.com/u/36722593?v=4", "gravatar_id": "", "url": "https://api.github.com/users/fayejf", "html_url": "https://github.com/fayejf", "followers_url": "https://api.github.com/users/fayejf/followers", "following_url": "https://api.github.com/users/fayejf/following{/other_user}", "gists_url": "https://api.github.com/users/fayejf/gists{/gist_id}", "starred_url": "https://api.github.com/users/fayejf/starred{/owner}{/repo}", "subscriptions_url": "https://api.github.com/users/fayejf/subscriptions", "organizations_url": "https://api.github.com/users/fayejf/orgs", "repos_url": "https://api.github.com/users/fayejf/repos", "events_url": "https://api.github.com/users/fayejf/events{/privacy}", "received_events_url": "https://api.github.com/users/fayejf/received_events", "type": "User", "user_view_type": "public", "site_admin": false } ]

[ "Thanks for your question. \r\nOf course! You can change the `diarizer.vad.model_path` and `diarizer.speaker_embeddings.model_path` in [config file](https://github.com/NVIDIA/NeMo/blob/main/examples/speaker_recognition/conf/speaker_diarization.yaml) to use your own models." ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]