Commit Graph

  • fcf5c1af88 Merge pull request #114 from ufownl:experimental Copybara-Service 2024-03-22 05:36:07 -0700
  • b9829ccc3c revert hello_world gemma git tag weedge 2024-03-22 10:40:12 +0800
  • 2bdcf6f426 remove const argv weedge 2024-03-22 10:36:56 +0800
  • 61e031fe98 Towards building tests without GUnit Refs #29 Jan Wassenberg 2024-03-21 19:26:27 -0700
  • 24add61dd9 Fix SFP/NUQ for bf16 rounding in Highway SFP: Avoid rounding twice, and more robust TestDot. NUQ: also more robust SNR, minor touchups to header. Jan Wassenberg 2024-03-21 19:05:44 -0700
  • 9c66eca963 change main use char** argv weedge 2024-03-21 21:26:20 +0800
  • 9b7804c75b change main use char** argv weedge 2024-03-21 21:22:25 +0800
  • 7165806216 change example helloworld gemma.cpp git tag weedge 2024-03-21 19:56:53 +0800
  • 73db33580b fix: change hello_world cmake compiler use -std=c++17 && argv add const weedge 2024-03-21 19:50:50 +0800
  • 90b0e9fd7a Refactor the implementation of `Attention` RangerUFO 2024-03-21 14:40:56 +0800
  • a135bc1e47 Fix build for RPi, missing hn::. Refs #112, thanks long568 Jan Wassenberg 2024-03-21 04:07:12 +0100
  • ba86c8d590 Remove obsolete copybara tags, faster bazel builds (debug) Jan Wassenberg 2024-03-20 19:31:23 +0100
  • f8baac80f9 Fix msan error, uninitialized model_training Jan Wassenberg 2024-03-20 05:12:06 +0100
  • 52940d435f Connect "--weights" parameter to Gemma Eric Ye 2024-03-20 00:07:47 +0100
  • 89be4c3de8 No public description Eric Ye 2024-03-19 23:35:58 +0100
  • 30b8a3c1ac Fix build for RPi, missing hn::. Refs #112, thanks long568 Jan Wassenberg 2024-03-20 20:07:12 -0700
  • 546519c855
    Added a missing space in app.h Ville Sundell 2024-03-21 00:39:45 +0200
  • 06cea2bcdb Remove obsolete copybara tags, faster bazel builds (debug) Jan Wassenberg 2024-03-20 19:31:23 +0100
  • edaafe335f Fix msan error, uninitialized model_training Jan Wassenberg 2024-03-20 05:12:06 +0100
  • e2a04b79ed Connect "--weights" parameter to Gemma Eric Ye 2024-03-20 00:07:47 +0100
  • ffd02c59ad No public description Eric Ye 2024-03-19 23:35:58 +0100
  • 7d5364bb80 Remove obsolete copybara tags, faster bazel builds (debug) Jan Wassenberg 2024-03-20 11:31:23 -0700
  • b20e490dc3 Remove obsolete copybara tags, faster bazel builds (debug) Jan Wassenberg 2024-03-19 21:30:10 -0700
  • 8fc6959950 Move conditional branch out of `pos2` loop RangerUFO 2024-03-20 23:50:14 +0800
  • c75d2eb635 Add the missing `HWY_ATTR` of `ProjKV` RangerUFO 2024-03-20 23:21:43 +0800
  • ce32f4db81 Streamline the implementation RangerUFO 2024-03-20 22:39:31 +0800
  • 11d9c51473 Fix msan error, uninitialized model_training Jan Wassenberg 2024-03-20 05:12:06 +0100
  • 6865819bb7 Connect "--weights" parameter to Gemma Eric Ye 2024-03-20 00:07:47 +0100
  • fdc3812446 No public description Eric Ye 2024-03-19 23:35:58 +0100
  • 6923aec853 Add MQA support RangerUFO 2024-03-20 18:14:09 +0800
  • 130e1f678f Adjust vocab size to be the same as gemma_pytorch RangerUFO 2024-03-19 22:00:52 +0800
  • 5e0cafbdc2 Fix msan error, uninitialized model_training Jan Wassenberg 2024-03-19 21:12:06 -0700
  • fdb1091b9c Connect "--weights" parameter to Gemma Eric Ye 2024-03-19 16:07:47 -0700
  • 4400842337 Minor refactor in Softmax enum-class 2024-03-20 00:20:14 +0800
  • 858d5b08c2 Use highway in AddFrom, MulBy, MulByConst, MulByConstAndAdd, create_distribution enum-class 2024-03-19 08:38:09 +0800
  • 720f609d84 Merge pull request #102 from google:experimental Copybara-Service 2024-03-18 10:56:52 -0700
  • 810b5a0cc2 Update README with more details on contributing code, add experimental/ directory, add READMEs for subdirectories, clean up DEVELOPER notes austinvhuang 2024-03-15 14:10:24 -0400
  • 8fb44ed6dd Merge pull request #97 from ufownl:feature/topk_flag v0.1.1 Copybara-Service 2024-03-14 07:34:55 -0700
  • a2ef389897 Merge pull request #98 from zeerd:patch-2 Copybara-Service 2024-03-14 07:21:23 -0700
  • 46c1aca304
    Add missing log that point to a failed Generation Charles Chan 2024-03-14 10:03:25 +0800
  • 1408784d7b Remove unused ascii banner string Eric Ye 2024-03-13 12:21:06 -0700
  • a0f316d853 Merge pull request #95 from google:conversion Copybara-Service 2024-03-13 09:37:36 -0700
  • f520e5c25c
    Remove WIP messages. pculliton 2024-03-13 11:36:19 -0400
  • 83ec42954f Allow changing k parameter of `SampleTopK` as a compiler flag RangerUFO 2024-03-13 13:55:37 +0800
  • e0b912fc46
    Merge branch 'google:main' into main Sascha Ronnie Daoudia 2024-03-13 05:12:16 +0100
  • 5fa2eb1a86 Use bf16-rounded sqrt for scaling embeddings to match Gemma Thanks Daniel & Michael Han for pointing this out. https://unsloth.ai/blog/gemma-bugs Jan Wassenberg 2024-03-12 19:14:31 -0700
  • 0221956b2e Merge pull request #87 from google:refactor-tidy Copybara-Service 2024-03-12 16:10:47 -0700
  • b6831a2256 Fixed 7B conversion. Phil Culliton 2024-03-12 21:12:28 +0000
  • 72247614bb fix prefill feedback off-by-1, update fetch commit hash austinvhuang 2024-03-12 15:10:44 -0400
  • 4aa8d0584e Merge branch 'dev' into refactor-tidy austinvhuang 2024-03-12 15:01:46 -0400
  • a9aa63fd2e Fix bazel build attempt #2 (remove //base through automation change). https://github.com/google/gemma.cpp/actions/runs/8252325981/job/22571549312 Austin Huang 2024-03-12 10:42:50 -0700
  • 9345b0aed5 Fix bazel build failure: https://github.com/google/gemma.cpp/actions/runs/8252325981/job/22571549312 Austin Huang 2024-03-12 09:40:34 -0700
  • ccd055e06b Merge pull request #82 from google:examples Copybara-Service 2024-03-12 09:24:24 -0700
  • 0d406061c0 Detect and print build type. Refs #88 Jan Wassenberg 2024-03-11 21:57:35 -0700
  • a9be065c28 Merge pull request #77 from enum-class:rmsnorm Copybara-Service 2024-03-11 06:25:57 -0700
  • 6712f07ee7 Merge branch 'dev' into rmsnorm enum-class 2024-03-11 20:13:18 +0800
  • 683c1b5c32
    Merge branch 'google:main' into rmsnorm Saman 2024-03-11 20:01:16 +0800
  • 60d054e041 move arg definitions out of gemma.h to app.h austinvhuang 2024-03-10 23:49:25 -0400
  • e577198fc0 Merge pull request #78 from enum-class:rmsnorm2 Copybara-Service 2024-03-10 16:14:44 -0700
  • 415464b047 fix CMakeLists typo austinvhuang 2024-03-10 15:41:17 -0400
  • 5d323c00fe fix tokenizer scope austinvhuang 2024-03-10 13:23:16 -0400
  • 0fc80fad05 libgemma refactor - review changes austinvhuang 2024-03-10 12:55:08 -0400
  • 0e1aefdac1
    Merge branch 'google:main' into main Sascha Ronnie Daoudia 2024-03-10 11:59:40 +0100
  • cc5c24c4f8 remove app.h dependency + fix bazel build austinvhuang 2024-03-08 18:06:43 -0500
  • 8c7b2cf61b add README, license to hello_world austinvhuang 2024-03-08 17:59:54 -0500
  • 571a5449c4 update commit hash for gemma lib austinvhuang 2024-03-08 17:33:33 -0500
  • 03147effbd update loader arg names: cache -> compressed_weights, model -> weights austinvhuang 2024-03-08 17:32:36 -0500
  • dfd2fdc1dd Decouple gemma constructor from loader args, update hello_world example, add convenience version of constructor (no uncompressed weights) austinvhuang 2024-03-08 17:26:03 -0500
  • 42e53e2da8 [WIP] simplify hello world example, add convenience function. TODO: update git hash in CMakeLists.txt of hello world after push austinvhuang 2024-03-08 14:55:35 -0500
  • fce5c8c967 Avoid fadvise on older Android. Fixes #84 Jan Wassenberg 2024-03-07 22:18:46 -0800
  • b67e28d1a0 [WIP] remove args from GetWeights, GetCompressedWeights austinvhuang 2024-03-08 00:00:11 -0500
  • 2161908f50 Added 7B support and args parsing. Still todo: more testing of 7B conversion. Phil Culliton 2024-03-07 22:34:14 +0000
  • 3df06f64c2
    Merge pull request #81 from ufownl/feature/separated_kvcache Austin Huang 2024-03-07 10:10:11 -0500
  • 0f92464f14
    Auto-labels 'Gemma' on 'gemma' issues/PRs. Shivam Mishra 2024-03-07 14:45:56 +0530
  • 170a9b4690 Make `CreateKVCache` a free function rather than a method RangerUFO 2024-03-07 14:08:48 +0800
  • b841612e8c Separate KV cache from GemmaImpl RangerUFO 2024-03-05 17:50:24 +0800
  • 6c0388e049 [WIP] refine Runtime struct definition austinvhuang 2024-03-07 01:14:07 -0500
  • 49e654258d [WIP] clean up hello_world #includes and CMakeLists.txt austinvhuang 2024-03-07 01:04:25 -0500
  • e781007836 [WIP] Remove InferenceArgs from hello_world example, fix ordering of LoaderArgs validation, revert ReplGemma EOT token behavior austinvhuang 2024-03-06 23:21:13 -0500
  • 7042316013 [WIP] update GemmaInterface, Gemma, and Generate input parameter specs to remove InferenceArgs. TODO: update hello_world example after git commit hash is available for fetching austinvhuang 2024-03-06 22:22:59 -0500
  • 0f6a4b49d5 [WIP] quality tweaks - for constants, defer float cast and use double for intermediate computations, add `model` to EOT token austinvhuang 2024-03-06 15:34:11 -0500
  • 5b9d8a9936 [WIP] dev/examples branch merge austinvhuang 2024-03-06 15:10:48 -0500
  • 10f7a086aa [WIP] decouple GemmaImpl from CLI args austinvhuang 2024-03-06 15:06:41 -0500
  • 3cdd5e524a Fix loop iteration in GeluMulToBF16 Also attempt to speed up builders (parallel) Jan Wassenberg 2024-03-05 23:00:09 -0800
  • 843d9b0e1f fix for-loop bounderies enum-class 2024-03-06 08:25:50 +0800
  • c8b9675898 Merge pull request #74 from osanseviero:patch-1 Copybara-Service 2024-03-05 12:49:09 -0800
  • c93e1a1e4d Resolved layer ordering, reshaping, MQA->MHA, and quantization. Works only for 2B. Phil Culliton 2024-03-05 17:54:55 +0000
  • bc845515b7 fix style, add kCamelCase style for constexpr in clang-tidy enum-class 2024-03-05 20:45:30 +0800
  • 5f016fb433 use hwy/simd for RMSNorm(f, bf, f) calculation enum-class 2024-03-05 17:53:52 +0800
  • 507d64e3e6 use hwy/simd for SquaredL2 calculation enum-class 2024-03-05 17:37:09 +0800
  • bb9b023502 Support Bazel builds. Fixes #16 Also fix nuq/sfp-inl: warning, cast, and disable SCALAR Jan Wassenberg 2024-03-04 22:06:51 -0800
  • 90dd6c1f07 decoupling from args.h Rliop913 2024-03-05 01:42:12 +0900
  • 8c857b957e
    Update README.md Omar Sanseviero 2024-03-04 12:58:49 +0100
  • cd7468199c Merge pull request #65 from enum-class:narrowing-issues Copybara-Service 2024-03-03 18:51:59 -0800
  • c378ac2c56 [WIP] hello world example working. TODO: refactor interfaces to decouple arguments austinvhuang 2024-03-03 11:36:48 -0500
  • 39cd59caec [WIP] create skeleton for example frontend application austinvhuang 2024-03-03 10:33:29 -0500
  • 90c99c330d use hwy::SignedFromSize traits enum-class 2024-03-03 10:56:46 +0800
  • 3c69695c1e transformations and validations (wip) austinvhuang 2024-03-02 14:46:51 -0500
  • 7d7d43e661 converter transformations (wip) austinvhuang 2024-03-02 08:11:55 -0500
  • 86761dc113
    Update README.md Omar Sanseviero 2024-03-01 23:44:38 +0100