https://huggingface.co/facebook/mms-1b is licensed by CC-By-NC which means the model is non commercial. I noticed people are using and sharing models finetuned on this including this
\nhttps://huggingface.co/Umong/wav2vec2-large-mms-1b-bengali by @umongsain
\nwhich achieves a really high score of 0.451
Can the competition organizers and kaggle please confirm if usage of such models is allowed ?
\nAlso if any of the top 5 finishes, use such a model, will they be removed from the leaderboard ?
\nOr only ineligible for prizes.
https://huggingface.co/facebook/mms-1b is licensed by CC-By-NC which means the model is non commercial. I noticed people are using and sharing models finetuned on this including this
\nhttps://huggingface.co/Umong/wav2vec2-large-mms-1b-bengali by @umongsain
\nwhich achieves a really high score of 0.451
Can the competition organizers and kaggle please confirm if usage of such models is allowed ?
\nAlso if any of the top 5 finishes, use such a model, will they be removed from the leaderboard ?
\nOr only ineligible for prizes.
Hi @benbla @tugstugi and @nikhilmishradev!
\nSorry for all the delays, after discussing with the teams we have decided to stick with the current rules, i.e., one would be eligible for the prize money only if their codes/models are commercially usable. Using NC licensed materials will disqualify from prize pool but they wont be removed from the leaderboard. Therefore, one would be able to win medals with NC stuff. I agree with tugstugi, it doesn't seem that doing good is impossible without NC methods.
\nThanks for the questions and asking for clarifications on this!! The main reason we want commercially usable submissions is because we want the competition proceedings to be widely usable. While Bengali.AI doesn't have any plans to commercially use this, but allowing commercial use will really accelerate Bengali ASR.
\nOfficial announcement here:
\nhttps://www.kaggle.com/competitions/bengaliai-speech/discussion/441423allowing
Thank you very much for this statement. I think everything is clear now.
", "rawMarkdown": "Thank you very much for this statement. I think everything is clear now.", "votes": 1 } ] }, { "id": 2438889, "postDate": "2023-09-14T14:39:24.960Z", "content": "I think, it should be allowed. From the competition rules:
\n\n\nOpen Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code. To the extent your Submission makes use of generally commercially available software not owned by you that you used to generate your submission, but that can be procured by the Competition Sponsor without undue expense, you do not grant the license in the preceding sentence to that software.
\n
The model is accessible by the competition hosts and they can also use the training scripts to train another wav2vec2 model with another license type.
\n@imtiazprio @reasat
\nPlease make a clear statement. The license topic led to big frustration in another competition. In the end competition hosts opened the field because most of the participants also used cc-by-nc models and datasets to achieve state of the art performance. Please have a look at this topic:
\nhttps://www.kaggle.com/competitions/benetech-making-graphs-accessible/discussion/417668
From my point of view, models and datasets with a cc-by-nc license should be allowed, if there are no disproportionate acquisition costs because techniques to achieve high performance results stay the same and can be followed by the competition hosts.
", "rawMarkdown": "I think, it should be allowed. From the competition rules:\n\n>Open Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code. **To the extent your Submission makes use of generally commercially available software not owned by you that you used to generate your submission, but that can be procured by the Competition Sponsor without undue expense, you do not grant the license in the preceding sentence to that software.**\n\nThe model is accessible by the competition hosts and they can also use the training scripts to train another wav2vec2 model with another license type.\n\n@imtiazprio @reasat \nPlease make a clear statement. The license topic led to big frustration in another competition. In the end competition hosts opened the field because most of the participants also used cc-by-nc models and datasets to achieve state of the art performance. Please have a look at this topic:\nhttps://www.kaggle.com/competitions/benetech-making-graphs-accessible/discussion/417668\n\nFrom my point of view, models and datasets with a cc-by-nc license should be allowed, if there are no disproportionate acquisition costs because techniques to achieve high performance results stay the same and can be followed by the competition hosts.", "votes": 2, "replies": [ { "id": 2438902, "postDate": "2023-09-14T14:50:19.743Z", "content": "thanks for this @benbla . Waiting for the competition hosts to clarify this soon, with only one month left for the competition to go
", "rawMarkdown": "thanks for this @benbla . Waiting for the competition hosts to clarify this soon, with only one month left for the competition to go", "votes": 2 }, { "id": 2439284, "postDate": "2023-09-14T18:33:06.640Z", "content": "Hi @benbla
\nThanks for tagging me here! Similar question was already answered here:
\nhttps://www.kaggle.com/competitions/bengaliai-speech/discussion/434179
Can you clarify why you think CC-BY-NC models like Facebook MMS-1B would fall under \"generally commercially available software not owned by you\"? Is it generally commercially available, i.e., can people obtain a commercial license for it without undue expense?
", "rawMarkdown": "Hi @benbla \n\nThanks for tagging me here! Similar question was already answered here: \nhttps://www.kaggle.com/competitions/bengaliai-speech/discussion/434179\n\nCan you clarify why you think CC-BY-NC models like Facebook MMS-1B would fall under \"generally commercially available software not owned by you\"? Is it generally commercially available, i.e., can people obtain a commercial license for it without undue expense?", "replies": [ { "id": 2439613, "postDate": "2023-09-15T02:52:07.127Z", "content": "So it is forbidden to use MMS-1B? The host should really clarify that ASAP.
", "rawMarkdown": "So it is forbidden to use MMS-1B? The host should really clarify that ASAP.", "votes": 4 }, { "id": 2440039, "postDate": "2023-09-15T08:30:33.040Z", "content": "Yeah, sorry, you are right in this case. However, please have a look at my comment below.
", "rawMarkdown": "Yeah, sorry, you are right in this case. However, please have a look at my comment below.", "votes": 1 }, { "id": 2443123, "postDate": "2023-09-17T14:35:42.300Z", "content": "@imtiazprio can you clarify that MMS-1B model can be used in the competition?
", "rawMarkdown": "@imtiazprio can you clarify that MMS-1B model can be used in the competition? ", "votes": 1 } ] }, { "id": 2439631, "postDate": "2023-09-15T03:17:40.737Z", "content": "@benbla @imtiazprio I don't use any model under CC-BY-NC and it is possible to achieve good score without MMS-1B. But the host really should clarify whether we can use CC-BY-NC ASAP because only a month is left.
", "rawMarkdown": "@benbla @imtiazprio I don't use any model under CC-BY-NC and it is possible to achieve good score without MMS-1B. But the host really should clarify whether we can use CC-BY-NC ASAP because only a month is left.", "votes": 6, "replies": [ { "id": 2440038, "postDate": "2023-09-15T08:27:28.587Z", "content": "I don't use this model either. However, it will be very difficult for the hosts to check if someone is using this model or not (Price Winner excluded because you can manually check their solutions). Therefore I plead for a less restrictive approach (see also my link to the Benetech Competition above), because otherwise it is really unfair in the bronze and silver area.
\nYou also can't control if datasets with a different license are used. I think it's also not very problematic for the host if solutions contain datasets or models which don´t allow a commercial usage: even if such models/datasets are used in the top solutions, the solutions can simply be reproduced by the host without these components.
\nThe problem with such license rules is that kagglers want top performance. There was once a competition where some of the prize winners were in the silver medal range, because medals are more important to most people than money.
\n", "rawMarkdown": "I don't use this model either. However, it will be very difficult for the hosts to check if someone is using this model or not (Price Winner excluded because you can manually check their solutions). Therefore I plead for a less restrictive approach (see also my link to the Benetech Competition above), because otherwise it is really unfair in the bronze and silver area.\n\nYou also can't control if datasets with a different license are used. I think it's also not very problematic for the host if solutions contain datasets or models which don´t allow a commercial usage: even if such models/datasets are used in the top solutions, the solutions can simply be reproduced by the host without these components.\n\nThe problem with such license rules is that kagglers want top performance. There was once a competition where some of the prize winners were in the silver medal range, because medals are more important to most people than money.\n\n@imtiazprio @reasat ", "votes": 4 } ] } ] }, { "id": 2436651, "postDate": "2023-09-13T17:24:33.130Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 2445830, "author_name": "Ahmed Imtiaz Humayun", "author_url": "", "post_date": "2023-09-19T05:26:08.760000", "content": "Hi @benbla @tugstugi and @nikhilmishradev!
\nSorry for all the delays, after discussing with the teams we have decided to stick with the current rules, i.e., one would be eligible for the prize money only if their codes/models are commercially usable. Using NC licensed materials will disqualify from prize pool but they wont be removed from the leaderboard. Therefore, one would be able to win medals with NC stuff. I agree with tugstugi, it doesn't seem that doing good is impossible without NC methods.
\nThanks for the questions and asking for clarifications on this!! The main reason we want commercially usable submissions is because we want the competition proceedings to be widely usable. While Bengali.AI doesn't have any plans to commercially use this, but allowing commercial use will really accelerate Bengali ASR.
\nOfficial announcement here:
\nhttps://www.kaggle.com/competitions/bengaliai-speech/discussion/441423allowing
Thank you very much for this statement. I think everything is clear now.
", "votes": 1, "replies": [] } ] }, { "id": 2438889, "author_name": "Benedikt Droste", "author_url": "", "post_date": "2023-09-14T14:39:24.960000", "content": "I think, it should be allowed. From the competition rules:
\n\n\nOpen Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code. To the extent your Submission makes use of generally commercially available software not owned by you that you used to generate your submission, but that can be procured by the Competition Sponsor without undue expense, you do not grant the license in the preceding sentence to that software.
\n
The model is accessible by the competition hosts and they can also use the training scripts to train another wav2vec2 model with another license type.
\n@imtiazprio @reasat
\nPlease make a clear statement. The license topic led to big frustration in another competition. In the end competition hosts opened the field because most of the participants also used cc-by-nc models and datasets to achieve state of the art performance. Please have a look at this topic:
\nhttps://www.kaggle.com/competitions/benetech-making-graphs-accessible/discussion/417668
From my point of view, models and datasets with a cc-by-nc license should be allowed, if there are no disproportionate acquisition costs because techniques to achieve high performance results stay the same and can be followed by the competition hosts.
", "votes": 2, "replies": [ { "id": 2438902, "author_name": "NikhilMishra", "author_url": "", "post_date": "2023-09-14T14:50:19.743000", "content": "thanks for this @benbla . Waiting for the competition hosts to clarify this soon, with only one month left for the competition to go
", "votes": 2, "replies": [] }, { "id": 2439284, "author_name": "Ahmed Imtiaz Humayun", "author_url": "", "post_date": "2023-09-14T18:33:06.640000", "content": "Hi @benbla
\nThanks for tagging me here! Similar question was already answered here:
\nhttps://www.kaggle.com/competitions/bengaliai-speech/discussion/434179
Can you clarify why you think CC-BY-NC models like Facebook MMS-1B would fall under \"generally commercially available software not owned by you\"? Is it generally commercially available, i.e., can people obtain a commercial license for it without undue expense?
", "votes": 0, "replies": [ { "id": 2439613, "author_name": "tugstugi", "author_url": "", "post_date": "2023-09-15T02:52:07.127000", "content": "So it is forbidden to use MMS-1B? The host should really clarify that ASAP.
", "votes": 4, "replies": [] }, { "id": 2440039, "author_name": "Benedikt Droste", "author_url": "", "post_date": "2023-09-15T08:30:33.040000", "content": "Yeah, sorry, you are right in this case. However, please have a look at my comment below.
", "votes": 1, "replies": [] }, { "id": 2443123, "author_name": "Raj Gothi", "author_url": "", "post_date": "2023-09-17T14:35:42.300000", "content": "@imtiazprio can you clarify that MMS-1B model can be used in the competition?
", "votes": 1, "replies": [] } ] }, { "id": 2439631, "author_name": "tugstugi", "author_url": "", "post_date": "2023-09-15T03:17:40.737000", "content": "@benbla @imtiazprio I don't use any model under CC-BY-NC and it is possible to achieve good score without MMS-1B. But the host really should clarify whether we can use CC-BY-NC ASAP because only a month is left.
", "votes": 6, "replies": [ { "id": 2440038, "author_name": "Benedikt Droste", "author_url": "", "post_date": "2023-09-15T08:27:28.587000", "content": "I don't use this model either. However, it will be very difficult for the hosts to check if someone is using this model or not (Price Winner excluded because you can manually check their solutions). Therefore I plead for a less restrictive approach (see also my link to the Benetech Competition above), because otherwise it is really unfair in the bronze and silver area.
\nYou also can't control if datasets with a different license are used. I think it's also not very problematic for the host if solutions contain datasets or models which don´t allow a commercial usage: even if such models/datasets are used in the top solutions, the solutions can simply be reproduced by the host without these components.
\nThe problem with such license rules is that kagglers want top performance. There was once a competition where some of the prize winners were in the silver medal range, because medals are more important to most people than money.
\n", "votes": 4, "replies": [] } ] } ] }, { "id": 2436651, "author_name": "", "author_url": "", "post_date": "2023-09-13T17:24:33.130000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2435855": "https://huggingface.co/facebook/mms-1b is licensed by CC-By-NC which means the model is non commercial. I noticed people are using and sharing models finetuned on this including this\nhttps://huggingface.co/Umong/wav2vec2-large-mms-1b-bengali by @umongsain \nwhich achieves a really high score of 0.451\n\nCan the competition organizers and kaggle please confirm if usage of such models is allowed ?\nAlso if any of the top 5 finishes, use such a model, will they be removed from the leaderboard ?\nOr only ineligible for prizes.\n\n@ashleychow @imtiazprio ", "2445830": "Hi @benbla @tugstugi and @nikhilmishradev!\n\nSorry for all the delays, after discussing with the teams we have decided to stick with the current rules, i.e., one would be eligible for the prize money only if their codes/models are commercially usable. Using NC licensed materials will disqualify from prize pool but they wont be removed from the leaderboard. Therefore, one would be able to win medals with NC stuff. I agree with tugstugi, it doesn't seem that doing good is impossible without NC methods.\n\nThanks for the questions and asking for clarifications on this!! The main reason we want commercially usable submissions is because we want the competition proceedings to be widely usable. While Bengali.AI doesn't have any plans to commercially use this, but allowing commercial use will really accelerate Bengali ASR.\n\nOfficial announcement here:\nhttps://www.kaggle.com/competitions/bengaliai-speech/discussion/441423allowing", "2438889": "I think, it should be allowed. From the competition rules:\n\n>Open Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code. **To the extent your Submission makes use of generally commercially available software not owned by you that you used to generate your submission, but that can be procured by the Competition Sponsor without undue expense, you do not grant the license in the preceding sentence to that software.**\n\nThe model is accessible by the competition hosts and they can also use the training scripts to train another wav2vec2 model with another license type.\n\n@imtiazprio @reasat \nPlease make a clear statement. The license topic led to big frustration in another competition. In the end competition hosts opened the field because most of the participants also used cc-by-nc models and datasets to achieve state of the art performance. Please have a look at this topic:\nhttps://www.kaggle.com/competitions/benetech-making-graphs-accessible/discussion/417668\n\nFrom my point of view, models and datasets with a cc-by-nc license should be allowed, if there are no disproportionate acquisition costs because techniques to achieve high performance results stay the same and can be followed by the competition hosts.", "2436651": "" } }