{"payload":{"feedbackUrl":"https://github.com/orgs/community/discussions/53140","repo":{"id":659838034,"defaultBranch":"main","name":"text-generation-inference","ownerLogin":"Vinno97","currentUserCanPush":false,"isFork":true,"isEmpty":false,"createdAt":"2023-06-28T17:07:33.000Z","ownerAvatar":"https://avatars.githubusercontent.com/u/9059378?v=4","public":true,"private":false,"isOrgOwned":false},"refInfo":{"name":"","listCacheKey":"v0:1694682133.0","currentOid":""},"activityList":{"items":[{"before":null,"after":"2a16b4101f5c265ba7c759d7d04d8881c07acf48","ref":"refs/heads/fix/top_n_token_logprobs","pushedAt":"2023-09-14T09:02:13.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"Vinno97","name":"Vincent Brouwers","path":"/Vinno97","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/9059378?s=80&v=4"},"commit":{"message":"Fix top_n_tokens returning non-log probs for some models","shortMessageHtmlLink":"Fix top_n_tokens returning non-log probs for some models"}},{"before":null,"after":"9b4545f2797d94fa59f20343ee6c71e360fb70a8","ref":"refs/heads/fix/galactica-missing-top-n-tokens","pushedAt":"2023-09-14T08:43:28.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"Vinno97","name":"Vincent Brouwers","path":"/Vinno97","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/9059378?s=80&v=4"},"commit":{"message":"Fix missing arguments in Galactica's from_pb\n\nFixes #1004","shortMessageHtmlLink":"Fix missing arguments in Galactica's from_pb"}},{"before":null,"after":"e864b956564f886098bff17a8878b125ae64ee54","ref":"refs/heads/feature/falcon-fix-weight-alias","pushedAt":"2023-08-30T10:18:34.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"Vinno97","name":"Vincent Brouwers","path":"/Vinno97","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/9059378?s=80&v=4"},"commit":{"message":"Fix Falcon weight mapping for H2O.ai checkpoints\n\nDuring the safetensor conversion, duplicate weights are removed.\nHowever, which of the duplicates gets removed, differs per checkpoint.\nIn some, like `h2oai/h2ogpt-oig-oasst1-falcon-40b`, the weight\n`transformer.word_embeddings.weightSafetensor` gets removed. In others,\n`lm_head.weight` gets removed. Long story long, we need to support both.\n\nOriginally, f018143 mapped `lm_head` to `word_embeddings`. Then ac736fd\nswitched this around. This commit merges them and allows for both.","shortMessageHtmlLink":"Fix Falcon weight mapping for H2O.ai checkpoints"}},{"before":"e30f4f61e7958ba7a29924705cb6ffcc9766ec3d","after":"ea78c5c26a590a56b42dcfdafa68b64105c992db","ref":"refs/heads/feat/return-top-tokens","pushedAt":"2023-08-09T08:48:49.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"Vinno97","name":"Vincent Brouwers","path":"/Vinno97","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/9059378?s=80&v=4"},"commit":{"message":"Fix typo in batch concatination","shortMessageHtmlLink":"Fix typo in batch concatination"}},{"before":"8471e1862db9f9eb4add36365291120e0a659fac","after":"e30f4f61e7958ba7a29924705cb6ffcc9766ec3d","ref":"refs/heads/feat/return-top-tokens","pushedAt":"2023-08-02T13:06:44.000Z","pushType":"push","commitsCount":2,"pusher":{"login":"Vinno97","name":"Vincent Brouwers","path":"/Vinno97","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/9059378?s=80&v=4"},"commit":{"message":"Only return top_tokens field when requested\n\nMimics the behaviour of `best_of`. Also allows client compatibility with\nolder versions","shortMessageHtmlLink":"Only return top_tokens field when requested"}},{"before":"d16298b8d4b00eafb2342748a540fa49d107a89b","after":"8471e1862db9f9eb4add36365291120e0a659fac","ref":"refs/heads/feat/return-top-tokens","pushedAt":"2023-08-01T15:05:39.000Z","pushType":"push","commitsCount":2,"pusher":{"login":"Vinno97","name":"Vincent Brouwers","path":"/Vinno97","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/9059378?s=80&v=4"},"commit":{"message":"Defer building top-token objects to Rust","shortMessageHtmlLink":"Defer building top-token objects to Rust"}},{"before":"95d0fba7de062840dde979c043581085d4a7ac9d","after":"d16298b8d4b00eafb2342748a540fa49d107a89b","ref":"refs/heads/feat/return-top-tokens","pushedAt":"2023-07-31T13:12:49.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"Vinno97","name":"Vincent Brouwers","path":"/Vinno97","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/9059378?s=80&v=4"},"commit":{"message":"Allocate top_n_token tensor in Batch","shortMessageHtmlLink":"Allocate top_n_token tensor in Batch"}},{"before":"6b29a3298a8dca1f3cbe7e4408c88639661b7a9e","after":"95d0fba7de062840dde979c043581085d4a7ac9d","ref":"refs/heads/feat/return-top-tokens","pushedAt":"2023-07-28T14:28:15.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"Vinno97","name":"Vincent Brouwers","path":"/Vinno97","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/9059378?s=80&v=4"},"commit":{"message":"Return more top-n-tokens when probabilities are equal","shortMessageHtmlLink":"Return more top-n-tokens when probabilities are equal"}},{"before":"d45982114af1646e2171d4ff4e89ff7e070f68a8","after":"6b29a3298a8dca1f3cbe7e4408c88639661b7a9e","ref":"refs/heads/feat/return-top-tokens","pushedAt":"2023-07-26T15:14:45.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"Vinno97","name":"Vincent Brouwers","path":"/Vinno97","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/9059378?s=80&v=4"},"commit":{"message":"Implement top-n-tokens for all models","shortMessageHtmlLink":"Implement top-n-tokens for all models"}},{"before":"d013577c3ff1e32936393e62c050253dbe510778","after":"d45982114af1646e2171d4ff4e89ff7e070f68a8","ref":"refs/heads/feat/return-top-tokens","pushedAt":"2023-07-25T14:57:31.000Z","pushType":"push","commitsCount":2,"pusher":{"login":"Vinno97","name":"Vincent Brouwers","path":"/Vinno97","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/9059378?s=80&v=4"},"commit":{"message":"Share computation for top-n-token decoding","shortMessageHtmlLink":"Share computation for top-n-token decoding"}},{"before":"789d809e515e506b64b6d7d48e1dfa21b57a2f8a","after":"d013577c3ff1e32936393e62c050253dbe510778","ref":"refs/heads/feat/return-top-tokens","pushedAt":"2023-07-24T14:04:34.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"Vinno97","name":"Vincent Brouwers","path":"/Vinno97","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/9059378?s=80&v=4"},"commit":{"message":"Add top-n-tokens support to benchmark","shortMessageHtmlLink":"Add top-n-tokens support to benchmark"}},{"before":null,"after":"789d809e515e506b64b6d7d48e1dfa21b57a2f8a","ref":"refs/heads/feat/return-top-tokens","pushedAt":"2023-07-14T19:53:05.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"Vinno97","name":"Vincent Brouwers","path":"/Vinno97","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/9059378?s=80&v=4"},"commit":{"message":"Add WIP support for returning top tokens\n\nInitial support returning the most probable tokens.\nNote that it is currently only implemented for seq-to-seq models. It is\nalso always enabled, regardless of whether it is used or not.","shortMessageHtmlLink":"Add WIP support for returning top tokens"}},{"before":"d6bb10f2025913dd3f4af8de93bd8efbcf4bd2c2","after":"742199aa0d9b4d4dc410f90572f78e9a5d776265","ref":"refs/heads/feature/automatic-duplicated-weight-mapping","pushedAt":"2023-07-04T09:31:03.000Z","pushType":"push","commitsCount":2,"pusher":{"login":"Narsil","name":"Nicolas Patry","path":"/Narsil","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/204321?s=80&v=4"},"commit":{"message":"Modified fix.","shortMessageHtmlLink":"Modified fix."}},{"before":null,"after":"d6bb10f2025913dd3f4af8de93bd8efbcf4bd2c2","ref":"refs/heads/feature/automatic-duplicated-weight-mapping","pushedAt":"2023-06-28T17:28:19.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"Vinno97","name":"Vincent Brouwers","path":"/Vinno97","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/9059378?s=80&v=4"},"commit":{"message":"Map deduplicated tensors via metadata\n\nThis PR automatically points tensors that were removed due to\ndeduplication to their still existing twin.\n\nIn `server.text_generation_server.utils.convert.py#convert_file`,\nduplicated tensors are removed and logged to the \"metadata\" dictionary.\nHowever, this dictionary was not yet used during loading. This requires\nexplicit remapping when loading the models (as mentioned in the\ndocstring).\n\nWhat does this fix?\nWe currently cannot load `h2oai/h2ogpt-oig-oasst1-falcon-40b` with the\nunmodified server, since the `transformer.word_embeddings.weight` weight\nis equal to `lm_head.weight` and is automatically removed.","shortMessageHtmlLink":"Map deduplicated tensors via metadata"}}],"hasNextPage":false,"hasPreviousPage":false,"activityType":"all","actor":null,"timePeriod":"all","sort":"DESC","perPage":30,"cursor":"djE6ks8AAAADgGe_JAA","startCursor":null,"endCursor":null}},"title":"Activity · Vinno97/text-generation-inference"}