{"payload":{"feedbackUrl":"https://github.com/orgs/community/discussions/53140","repo":{"id":758230362,"defaultBranch":"main","name":"max","ownerLogin":"modularml","currentUserCanPush":false,"isFork":false,"isEmpty":false,"createdAt":"2024-02-15T21:58:11.000Z","ownerAvatar":"https://avatars.githubusercontent.com/u/39327063?v=4","public":true,"private":false,"isOrgOwned":true},"refInfo":{"name":"","listCacheKey":"v0:1715898084.0","currentOid":""},"activityList":{"items":[{"before":"782d65994f61afd2422c08dad0a45b98ddfee67c","after":"922935856bb6b2feee7ae35241b6625295be91c7","ref":"refs/heads/nightly","pushedAt":"2024-06-03T07:14:00.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[External] [stdlib] use `ref[_]` in Optional.value() return type (#40932)\n\n[External] [stdlib] use `ref[_]` in Optional.value() return type\n\nSince auto-deref is possible with the new `ref[_]` return annotation,\nthis PR wires it up for `Optional.value()`.\n\nCo-authored-by: Lukas Hermann <lukashermann28@gmail.com>\nCloses modularml/mojo#2875\nMODULAR_ORIG_COMMIT_REV_ID: 8bacdcb4205ab78ef54616e037e87b67167e551a","shortMessageHtmlLink":"[External] [stdlib] use <code>ref[_]</code> in Optional.value() return type (#40…"}},{"before":"4455071e202da5bc1d12ef6f18e5b7cc95c13ea8","after":"782d65994f61afd2422c08dad0a45b98ddfee67c","ref":"refs/heads/nightly","pushedAt":"2024-06-03T07:13:58.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[mojo-stdlib] Move global UnsafePointer functions to be methods.\n\nThis extends the compiler to allow the type of `self` to be `Self` with\nany parameters, allowing more flexible conditional conformance.\n\nThis allows us to move the global functions in UnsafePointer to being\nmethods on it instead, which is a much nicer API.  Specifically, this\nmoves:\n 1) `destroy_pointee(p)` => `p.destroy_pointee()`\n 2) `move_from_pointee(p)` => `p.take_pointee()`\n 3) `initialize_pointee_move(p, value)` => `p.init_pointee_move(value)`\n 4) `initialize_pointee_copy(p, value)` => `p.init_pointee_copy(value)`\n 5) `move_pointee(src=p1, dst=p2)` => `p.move_pointee_into(p2)`\n\nThis uses the word \"init\" instead of \"initialize\" since that is the term\nof\nart used in Python.\n\nMODULAR_ORIG_COMMIT_REV_ID: ba930ea1ef25c4f010bd27742e3afee047377764","shortMessageHtmlLink":"[mojo-stdlib] Move global UnsafePointer functions to be methods."}},{"before":"907796d959560661e40f45a8d9b81022de9047a0","after":"4455071e202da5bc1d12ef6f18e5b7cc95c13ea8","ref":"refs/heads/nightly","pushedAt":"2024-06-03T07:13:55.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[mojo-stdlib] Remove most remaining Reference self's\n\nThis is on the quest to push Reference down in visibility, and eliminate\nsupport for Reference self.\n\nMODULAR_ORIG_COMMIT_REV_ID: 3f9fc713bfa37bbb725c1ab1b1bda302a1ed0f76","shortMessageHtmlLink":"[mojo-stdlib] Remove most remaining Reference self's"}},{"before":"c7cbdf193d30e3241dc27f58af6d22691bfbfe2a","after":"907796d959560661e40f45a8d9b81022de9047a0","ref":"refs/heads/nightly","pushedAt":"2024-06-03T07:13:53.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[mojo-stdlib] Remove unneeded `borrowed` keywords\n\nThis keyword is the default and a noop in both `fn` and `def`s now,\nso it should not be explicitly provided.\n\nMODULAR_ORIG_COMMIT_REV_ID: d43d39d9779b9b6012451dd2a6402f537da45627","shortMessageHtmlLink":"[mojo-stdlib] Remove unneeded <code>borrowed</code> keywords"}},{"before":"e5c0811c5af2a1a9ff39c4c631664ec82fcb6260","after":"c7cbdf193d30e3241dc27f58af6d22691bfbfe2a","ref":"refs/heads/nightly","pushedAt":"2024-06-03T07:13:50.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[mojo-stdlib] Modernize a bunch of reference taking stuff\n\nThis moves a bunch of slice types to take their \"isMutable\" bit as\ninferred,\nusing `ref [_]` to tidy up a bunch of APIs, and fixes some mutability\nbugs\nin dictionary iteration.\n\nMODULAR_ORIG_COMMIT_REV_ID: 77003579b8773361d39259b10ce996c2e198361c","shortMessageHtmlLink":"[mojo-stdlib] Modernize a bunch of reference taking stuff"}},{"before":"cf89d5da52538191701c7bed873590490861a79b","after":"e5c0811c5af2a1a9ff39c4c631664ec82fcb6260","ref":"refs/heads/nightly","pushedAt":"2024-06-03T07:13:47.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[mojo-stdlib] Move the 'Reference. is_mutable' parameter to being\ninferred\n\nThis massively simplifies all users of the `Reference` type by allowing\nthe\n`is_mutable` parameter to be inferred, using Mojo's fancy new \"inferred\nonly\"\nparameter kind.\n\nMODULAR_ORIG_COMMIT_REV_ID: 1a1fce14e0584b5ae8dd29d2d2cb87b916ad50d8","shortMessageHtmlLink":"[mojo-stdlib] Move the 'Reference. is_mutable' parameter to being"}},{"before":"844f548af3c183059664e210ddf99a21cc60715e","after":"cf89d5da52538191701c7bed873590490861a79b","ref":"refs/heads/nightly","pushedAt":"2024-06-02T07:14:05.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[max-pipelines] Support Q6_K in Llama 3\n\nSupport `Q6_K` in the Llama 3 pipeline.\n\nChanges:\n- Add `Q6_K` dequantize op.\n- Fix incorrect BlockQ6K definition.\n- Change quantized `Q4_K` model to popular Bartowski quantized variant.\n- Also added `Q6_K` Bartowski quantized option, basically lossless.\nMODULAR_ORIG_COMMIT_REV_ID: 974d682e84ca133423d1e347e6b940aeb7fb2472","shortMessageHtmlLink":"[max-pipelines] Support Q6_K in Llama 3"}},{"before":"e257b4b65fb1b8e01b0c6847af851f3ce9744ec8","after":"844f548af3c183059664e210ddf99a21cc60715e","ref":"refs/heads/nightly","pushedAt":"2024-06-02T07:14:03.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[max-pipelines] Dispatch Linear on encoding id\n\nDispatch the `Linear` layer on the quantization encoding id.\nPass the id around as a mojo string metadata along with the staged\nconstant op result `Symbol` for now.\nFollow up work can tackle making that encoding id part of the underlying\ntensor type.\nThe encoding id originates from the tensor info stored in the GGUF file.\n\nThis is needed to cleanly support mixed precision models.\n\nMODULAR_ORIG_COMMIT_REV_ID: dadfc85e328d6f04c33f7a36ad800d0c634995fa","shortMessageHtmlLink":"[max-pipelines] Dispatch Linear on encoding id"}},{"before":"79fc0a0bbac546ba95a29ccb3c079cd48097afee","after":"e257b4b65fb1b8e01b0c6847af851f3ce9744ec8","ref":"refs/heads/nightly","pushedAt":"2024-06-02T07:14:01.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[graph-api] [max-pipelines] Support Q4_K in Llamas\n\nAdd support for Q4_K to MAX graph and the Llama 2 and 3 pipelines.\n\nChanges:\n- Add Q4_K dequantize implementation, needed for Q4_K embeddings.\n- Dispatch to Q4_K matmul and dequantize ops based on quantization\n  encoding.\n- Download pure Q4_K Llama 3 from Llama 3 pipeline.\n  This and the Q4_0 downloads are temporary until Q6_K is online, at\n  which point we can switch to a standard Llama 3 GGUF.\nMODULAR_ORIG_COMMIT_REV_ID: 2e689158f35893a839246f4954bb5e2b42e9689e","shortMessageHtmlLink":"[graph-api] [max-pipelines] Support Q4_K in Llamas"}},{"before":"326ed8adfe4720dcf5658ab04ff3d6b7e4ec4ec0","after":"79fc0a0bbac546ba95a29ccb3c079cd48097afee","ref":"refs/heads/nightly","pushedAt":"2024-06-01T07:19:16.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[max-pipelines] Add the missing setup and running requirements in the\npipeline READMEs\n\nMODULAR_ORIG_COMMIT_REV_ID: 97beaa05c8f1fb88f06ec3b15b0b6dafe3778648","shortMessageHtmlLink":"[max-pipelines] Add the missing setup and running requirements in the"}},{"before":"ddf2404f562a8c8075f5e1ff9283168554b90a49","after":"326ed8adfe4720dcf5658ab04ff3d6b7e4ec4ec0","ref":"refs/heads/nightly","pushedAt":"2024-06-01T07:19:14.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[stdlib] feature: Change `String.unsafe_ptr()` to return\n`UnsafePointer[UInt8]`\n\n* Add `alias C_char = Int8` in sys.ffi\n\n* Change `String.unsafe_ptr()` to return `UnsafePointer[UInt8]` (was\n`Int8`)\n* Remove `String.unsafe_uint8_ptr()`\n* Add `String.unsafe_cstr_ptr() -> UnsafePointer[C_char]` for\nconvenience\n\n* Change `String._steal_ptr()` to return `UnsafePointer[UInt8]` (was\n`DTypePointer[DType.int8]`)\n* Change `FileHandle._write` from taking `DTypePointer[int8]` to\n`UnsafePointer[UInt8]`\n* Change `DLHandle._get_function` to take `UnsafePointer[C_char]` (was\n`DTypePointer[DType.int8]`)\n\nMODULAR_ORIG_COMMIT_REV_ID: 6d49b224eeb2699aae52f659e197aab7f6b6c2eb","shortMessageHtmlLink":"[stdlib] feature: Change <code>String.unsafe_ptr()</code> to return"}},{"before":"66c65c21f147d657b6a58fc91c31faf48fa56ab5","after":"ddf2404f562a8c8075f5e1ff9283168554b90a49","ref":"refs/heads/nightly","pushedAt":"2024-06-01T07:19:12.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[AI Pipelines] Adds a new script that creates the Tiny Llama model from\nthe saved checkpoint.\n\n* A copy of the `TeenyTinyLlama` script is created to keep the\n`quantize_and_save` pipeline clean.\n* `run` now calls the save pipeline if the MAX checkpoint doesn't exist,\nand the load pipeline if it does.\n* The saved checkpoint now stores the hyperparams as well.\n* Minor changes to error reporting in TensorDict\n\nMODULAR_ORIG_COMMIT_REV_ID: 1ea3361d2ac0c1b628a9b32dc4398391df75c934","shortMessageHtmlLink":"[AI Pipelines] Adds a new script that creates the Tiny Llama model from"}},{"before":"eb5c509b9b0591cc61a1e53c5fa531b309f48b3b","after":"66c65c21f147d657b6a58fc91c31faf48fa56ab5","ref":"refs/heads/nightly","pushedAt":"2024-06-01T07:19:10.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"- Handle incorrect input when selecting a GGUF file\n- Exit gracefully if no GGUF files exist in folder\n\nMODULAR_ORIG_COMMIT_REV_ID: 58fc58e8231afecd3e21ba8179eb55aa464b987c","shortMessageHtmlLink":"- Handle incorrect input when selecting a GGUF file"}},{"before":"e563f602788d0e8c3f2d08951b663345c0bfdc84","after":"eb5c509b9b0591cc61a1e53c5fa531b309f48b3b","ref":"refs/heads/nightly","pushedAt":"2024-06-01T07:19:08.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"When running MAX pipelines add llama3 option to prompt, and exit\ngracefully if a model name isn't provided. It was segfaulting\npreviously.\n\nMODULAR_ORIG_COMMIT_REV_ID: b1d6a0e1a1d8e7e19741375ac784765d5350d7e4","shortMessageHtmlLink":"When running MAX pipelines add llama3 option to prompt, and exit"}},{"before":"2f731f42f7b8f3043ee5285ce0f3779a20c2c7f2","after":"e563f602788d0e8c3f2d08951b663345c0bfdc84","ref":"refs/heads/nightly","pushedAt":"2024-06-01T07:19:05.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"Graceful exit to fix the segfault in run_pipeline.mojo when no pipeline\nis given.\n\nMODULAR_ORIG_COMMIT_REV_ID: cd50176bf1e56e05957ea31de588839362ff09c8","shortMessageHtmlLink":"Graceful exit to fix the segfault in run_pipeline.mojo when no pipeline"}},{"before":"d83df261eab7c56ff47908dd4470c92ce69770b5","after":"2f731f42f7b8f3043ee5285ce0f3779a20c2c7f2","ref":"refs/heads/nightly","pushedAt":"2024-06-01T07:19:03.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"Mirrors the existing llama2/serve.:fire:.\n\nMODULAR_ORIG_COMMIT_REV_ID: 2a109464377ce28948e9a91d6d63e1de216ff274","shortMessageHtmlLink":"Mirrors the existing llama2/serve.:fire:."}},{"before":"8a123a29d6c222b0deeb78b9ec1e0d58242d4ac0","after":"d83df261eab7c56ff47908dd4470c92ce69770b5","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:32.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[mojo-stdlib] Add an \"empty()\" method to StringRef for convenience.\n\nMODULAR_ORIG_COMMIT_REV_ID: 0148ca35a8303a2037aaf158697388b8f7809c23","shortMessageHtmlLink":"[mojo-stdlib] Add an \"empty()\" method to StringRef for convenience."}},{"before":"7bec42396427f6c5d6a174d10b840218a48e8373","after":"8a123a29d6c222b0deeb78b9ec1e0d58242d4ac0","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:30.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"See the streaming output\n[here](ttps://platform.openai.com/docs/api-reference/chat/create) for\nreference.\n\nMODULAR_ORIG_COMMIT_REV_ID: 38ed9ce20f66005f549b1ad014094b3af2e9e3f9","shortMessageHtmlLink":"See the streaming output"}},{"before":"08e6620a60b58e083594ce8fce78329b05396c0c","after":"7bec42396427f6c5d6a174d10b840218a48e8373","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:27.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[max-pipelines] Enable Replit to output chunks of text, handle newlines\nin prompts\n\nRather than only printing the last character of a decoded token, print\nthe full chunked string that the tokenizer has produced. To properly\nhandle newlines in command-line prompt strings, replace them with actual\nnewlines in the prompt string.\n\nWhile fixing those, remove some of the no-longer-needed `var`s in the\nReplit model.\n\nMODULAR_ORIG_COMMIT_REV_ID: 4d104696bb6c702f3ec025e55f208acfaa3dd24f","shortMessageHtmlLink":"[max-pipelines] Enable Replit to output chunks of text, handle newlines"}},{"before":"3f1b787c1c3636fbd72a5c8e5e135e5a7a7f079d","after":"08e6620a60b58e083594ce8fce78329b05396c0c","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:25.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[max-pipelines] Test and fix llama3 BPE implementation\n\n- Adds an explicit test for BPE with a large test file\n- Test data was tokenized with tiktoken directly using\n  _encode_single_piece and the llama3 tokenizer vocab\n- Adds support for loading tiktoken-encoded vocabs to BPE\n- Fixes BPE to correctly use a MinHeap rather than a MaxHeap for merge\n  priority queue\n\nMODULAR_ORIG_COMMIT_REV_ID: bd8cc7cc43fc612e30e0ed23a2408772ac5bf892","shortMessageHtmlLink":"[max-pipelines] Test and fix llama3 BPE implementation"}},{"before":"3af038cf36cadedc4a6401eacdb988133435be44","after":"3f1b787c1c3636fbd72a5c8e5e135e5a7a7f079d","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:23.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"Introduces serving equivalents of run.:fire: and run_pipeline.:fire that use the new max.serve APIs.\n\nMODULAR_ORIG_COMMIT_REV_ID: 7ef6acfc685a79bcd01f687562cafe9b3b0c1edd","shortMessageHtmlLink":"Introduces serving equivalents of run.:fire: and run_pipeline.:fire t…"}},{"before":"c6e84795e9a709d0caf204032fd9ae1e10184ee9","after":"3af038cf36cadedc4a6401eacdb988133435be44","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:21.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[Serving] [SDK] [examples] Fix notebook and mojo examples for serving\nsdk\n\nThis PR fixes some minor issues with the serving example notebooks.\n\nMODULAR_ORIG_COMMIT_REV_ID: 98701e6b7cbbaecb22b719af68add8b95aaf7434","shortMessageHtmlLink":"[Serving] [SDK] [examples] Fix notebook and mojo examples for serving"}},{"before":"683cdbdac106e6e64d1a0b1a3b41bb1655cd9528","after":"c6e84795e9a709d0caf204032fd9ae1e10184ee9","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:19.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[max-pipelines] MaxHeap improvements\n\n- Add a basic test\n- Fix a correctness bug in `_sift_down` :this-is-fine:\n- Add a `MinHeap` variant\n\nMODULAR_ORIG_COMMIT_REV_ID: c0df3100d3613776f76e7437946c655f2c311edd","shortMessageHtmlLink":"[max-pipelines] MaxHeap improvements"}},{"before":"d60e13bb26e95e391c76e58dee468b8d85da2d22","after":"683cdbdac106e6e64d1a0b1a3b41bb1655cd9528","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:16.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"Parametrize models and layers in MAX graph pipelines on their\nquantization encoding rather than DType as done previously.\nThis allows supporting multiple quantization encodings.\nIn particular, follow up work will add `Q4_K` and `Q6_K` support.\n\nChanges:\n- Parametrize all Llama-related pipelines on `QuantizationEncoding`\n  instead of `DType`.\n- Add an identity float32 encoding as a base case.\n- As a drive by, fix an import bug in the Llama 2 pipeline.\nMODULAR_ORIG_COMMIT_REV_ID: b8cea8c5e632c13ec597dbf570ab2c3000928d67","shortMessageHtmlLink":"Parametrize models and layers in MAX graph pipelines on their"}},{"before":"37059d4ccd1c2accdf53751178a7fb95fc077164","after":"d60e13bb26e95e391c76e58dee468b8d85da2d22","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:14.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[AI Pipelines] Fix to TinyStories model following the Llama token\nsampler addition.\n\nThe model now returns logits instead of the token argmax.\n\nMODULAR_ORIG_COMMIT_REV_ID: 5aff8ae3538c9fd3aaed2448305891a678df3c13","shortMessageHtmlLink":"[AI Pipelines] Fix to TinyStories model following the Llama token"}},{"before":"6281dfff36b089a991cb08e12326a86e9f842dff","after":"37059d4ccd1c2accdf53751178a7fb95fc077164","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:12.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[max-pipelines] Fix UB in GGUF with initialize_pointee_copy. Such\nfunctions safely make the assignments work in case of uninitialized\nunsafe pointers.\n\nMODULAR_ORIG_COMMIT_REV_ID: f58f0651b87916223b7447ddcf7d9e14c7b40fa3","shortMessageHtmlLink":"[max-pipelines] Fix UB in GGUF with initialize_pointee_copy. Such"}},{"before":"14d7d8104f75baf0557adc0f080952821b5e6ba0","after":"6281dfff36b089a991cb08e12326a86e9f842dff","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:10.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[max-pipelines] Move Llama layers to pipelines.nn\n\nMove commonly used Llama layers to a shared `pipelines.nn` module.\nThese layers were mostly identical.\nRefactor only the `Attention` layer, to support both GQA and the custom\nRoPE ******.\nOtherwise, this PR is purely code motion and compression.\n\nMODULAR_ORIG_COMMIT_REV_ID: 71337f072842eefdd86916b75f476d978aecffb4","shortMessageHtmlLink":"[max-pipelines] Move Llama layers to pipelines.nn"}},{"before":"bad81005de5901104e4b04cbb45c8269415b1286","after":"14d7d8104f75baf0557adc0f080952821b5e6ba0","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:08.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[max-pipelines] Get hyperparams from GGUF metadata\n\nGet the Llama 2 and 3 ************** directly from the GGUF metadata in\ngguf.🔥.\nThis removes the need for hardcoding those **************.\n\nMODULAR_ORIG_COMMIT_REV_ID: 9e0054cdfed414b50dd885fe0e7807a2d1ba85e5","shortMessageHtmlLink":"[max-pipelines] Get hyperparams from GGUF metadata"}},{"before":"c79f75eac675be9eb2d664f0bf31e572a43c168c","after":"bad81005de5901104e4b04cbb45c8269415b1286","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:05.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"Style cleanup to Mojo example and add README\n\nRename the inference file for consistency with other examples.\nSwitch the main function to `def` and remove use of `var`.\nRename some variables for consistency with other examples and more\nlegibility.\n\nMODULAR_ORIG_COMMIT_REV_ID: 6705ceeba58e76d1697d56769ea00d6ecd5a2d0e","shortMessageHtmlLink":"Style cleanup to Mojo example and add README"}},{"before":"ca456d950731e7845f9fdd6711edb8e2d5b83f2e","after":"c79f75eac675be9eb2d664f0bf31e572a43c168c","ref":"refs/heads/nightly","pushedAt":"2024-05-31T21:12:03.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"modularbot","name":"modularbot","path":"/modularbot","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/116839051?s=80&v=4"},"commit":{"message":"[AI Pipelines] Llama2 token sampler\n\nAdd a basic weighted token sampler to the llama2 example. Supports min_p\nand temperature.\n\nMODULAR_ORIG_COMMIT_REV_ID: f14bf465993dc721ec1c8727bc6758969f5a83c1","shortMessageHtmlLink":"[AI Pipelines] Llama2 token sampler"}}],"hasNextPage":true,"hasPreviousPage":false,"activityType":"all","actor":null,"timePeriod":"all","sort":"DESC","perPage":30,"cursor":"djE6ks8AAAAEWp_PBAA","startCursor":null,"endCursor":null}},"title":"Activity · modularml/max"}