Commit Graph

  • d3f05eae8c Move some models to candle-transformers so that it's easier to re-use. (#794) Laurent Mazare 2023-09-10 09:40:27 +01:00
  • b2796ce6ef Reduce the number of threads. conv-transpose-groups laurent 2023-09-09 21:10:30 +01:00
  • c22f23e568 Bugfix. laurent 2023-09-09 19:11:59 +01:00
  • b283b3e181 Merge branch 'main' into conv-transpose-groups laurent 2023-09-09 18:50:13 +01:00
  • 258ac32c38 Fix cuda randn when generating an odd number of values. (#793) Laurent Mazare 2023-09-09 18:44:21 +01:00
  • b4fe316fa1 Group support in conv-transpose2d. laurent 2023-09-09 18:11:41 +01:00
  • 31936c08fe ViT tracing. (#790) Laurent Mazare 2023-09-09 17:26:39 +01:00
  • 74ad4deb42 Get the MobileSAM TinyViT based version to work. (#789) Laurent Mazare 2023-09-09 16:21:44 +01:00
  • b7cd58473b TinyViT backbone for segment-anything. (#787) Laurent Mazare 2023-09-09 15:10:06 +01:00
  • 3cd7e7b51d Fuse the rel-pos additions via a custom-op. (#786) Laurent Mazare 2023-09-09 10:46:09 +01:00
  • 722c50bb0c Use byteorder in mnist. (#785) Laurent Mazare 2023-09-09 09:03:59 +01:00
  • 976a1086ee feat: u32 from_be_bytes (#765) YangNianYi 2023-09-09 15:55:35 +08:00
  • c88d6fd4b9 Remove set_training. (#784) Laurent Mazare 2023-09-09 08:27:37 +01:00
  • 057f7909bc Accelerate support for gelu. (#782) Laurent Mazare 2023-09-08 21:58:56 +01:00
  • a910ec5993 CustomOp for einsum. einsum-custom-op laurent 2023-09-08 20:46:30 +01:00
  • acf8f10ae1 Get the comparison operation to work on scalar values. (#780) Laurent Mazare 2023-09-08 20:13:29 +01:00
  • 0906acab91 Automatic mask generation (#779) Laurent Mazare 2023-09-08 19:11:34 +01:00
  • 158ff3c609 Add tracing to segment-anything (#777) Laurent Mazare 2023-09-08 15:31:29 +01:00
  • e5703d2f56 Draw the mask on a merged image. (#775) Laurent Mazare 2023-09-08 14:04:34 +01:00
  • 98172d46fa Fix some errors about BlockQ8_1 (#776) zmlcc 2023-09-08 20:29:40 +08:00
  • 28c87f6a34 Automatic mask generator + point base mask (#773) Laurent Mazare 2023-09-08 12:26:56 +01:00
  • c1453f00b1 Improve the safetensor loading in the segment-anything example. (#772) Laurent Mazare 2023-09-08 09:39:10 +01:00
  • 989a4807b1 Use shape with holes. (#771) Laurent Mazare 2023-09-08 08:50:27 +01:00
  • 0e250aee4f Shape with holes (#770) Laurent Mazare 2023-09-08 08:38:13 +01:00
  • cfcbec9fc7 Add small customization to the build (#768) Zsombor 2023-09-08 09:15:14 +02:00
  • 3898e500de Generate a mask image + the scaled input image. (#769) Laurent Mazare 2023-09-08 05:53:08 +01:00
  • 79c27fc489 Segment-anything fixes: avoid normalizing twice. (#767) Laurent Mazare 2023-09-07 21:45:16 +01:00
  • 7396b8ed1a Segment Anything - process images (#766) Laurent Mazare 2023-09-07 19:22:45 +01:00
  • 7b50f3e106 More segment-anything again. (#764) Laurent Mazare 2023-09-07 13:06:55 +02:00
  • 8c991df394 More segment-anything. (#763) Laurent Mazare 2023-09-07 08:28:30 +02:00
  • 000fa00e31 Expose the conv2d-transpose layers. (#761) Laurent Mazare 2023-09-07 07:04:52 +02:00
  • a17a7c42c1 Add a nn layer for conv-transpose2d. (#760) Laurent Mazare 2023-09-07 06:47:28 +02:00
  • 6527ab81a3 Sketch the segment anything model. (#759) Laurent Mazare 2023-09-07 06:34:05 +02:00
  • 7b1f2da828 Cudnn fix. (#758) Laurent Mazare 2023-09-06 18:39:39 +02:00
  • bdc9d46fe3 Use an arc in the varbuilder rather than rc. (#757) Laurent Mazare 2023-09-06 16:29:09 +02:00
  • dcf708559d Fix for cudnn to work with img2img. (#753) Laurent Mazare 2023-09-06 08:49:28 +02:00
  • 7299a68353 img2img pipeline for stable diffusion. (#752) Laurent Mazare 2023-09-06 08:06:49 +02:00
  • 16bf44f6e9 force model cache (#751) Radamés Ajna 2023-09-05 20:53:31 -07:00
  • a4f40f3dc8 Use rayon directly rather than constraining the number of threads. (#749) Laurent Mazare 2023-09-05 21:26:15 +02:00
  • 6a40decc76 Minor WASM UI improvements (#748) Radamés Ajna 2023-09-05 11:24:43 -07:00
  • a0d65585db Softmax implementation for cuda. (#747) Laurent Mazare 2023-09-05 19:38:03 +02:00
  • 94c6a8d3d3 Add a dedicated cuda kernel for softmax. (#746) Laurent Mazare 2023-09-05 17:53:20 +02:00
  • 6615daf242 Tweaks to softmax. (#745) Laurent Mazare 2023-09-05 16:22:27 +02:00
  • 1c9e5394a5 Add a custom softmax implementation. (#744) Laurent Mazare 2023-09-05 15:20:23 +02:00
  • a8410bf35e Add some documentation. (#743) Laurent Mazare 2023-09-05 10:51:12 +02:00
  • cda45a7443 Let outside CustomOp2 implementations use binary_map/binary_map_vec (#741) Gonzalo 2023-09-05 05:27:32 -03:00
  • 4698eb5cb6 Fix typo in the nll function document (#742) Masato Mori 2023-09-05 17:25:11 +09:00
  • 000487c36f Add a python function to save as safetensors. (#740) Laurent Mazare 2023-09-04 21:32:14 +02:00
  • ab0d9fbdd1 Properly set the is_bf16 flag. (#738) Laurent Mazare 2023-09-04 17:45:26 +02:00
  • f80fd44201 BF16 support for flash-attn. (#737) Laurent Mazare 2023-09-04 17:35:43 +02:00
  • 0d00c06a83 Fix clippy lint. (#736) Laurent Mazare 2023-09-04 17:09:19 +02:00
  • 8395152d20 Llama2c WASM UI improvements (#732) Radamés Ajna 2023-09-04 07:59:22 -07:00
  • e2f9f60ac2 Avoid some redundant clone. (#731) Laurent Mazare 2023-09-04 09:18:32 +02:00
  • d0cdea95a5 Add back the bf16 flash-attn kernels. (#730) Laurent Mazare 2023-09-04 08:50:52 +02:00
  • 20512ba408 Return the metadata in the gguf pyo3 bindings. (#729) Laurent Mazare 2023-09-04 08:07:00 +02:00
  • 9c61b0fc9b Proper log buckets for t5. (#727) Laurent Mazare 2023-09-03 21:33:50 +02:00
  • 26cd266e65 Musicgen text embeddings. (#726) Laurent Mazare 2023-09-03 19:27:48 +02:00
  • bbec527bb9 Fix the musicgen example. (#724) Laurent Mazare 2023-09-03 15:50:39 +02:00
  • f7980e07e0 Add ggufv2 support (#725) Lukas Kreussel 2023-09-03 15:41:57 +02:00
  • 74a82c358a Add the mse loss. (#723) Laurent Mazare 2023-09-03 11:51:40 +02:00
  • 84d003ff53 Handle arbitrary shapes in Tensor::new. (#718) Laurent Mazare 2023-09-02 20:59:21 +02:00
  • 21109e1983 Recommend using maturin. (#717) Laurent Mazare 2023-09-02 17:19:35 +02:00
  • ad796eb4be More quantized llama in python. (#716) Laurent Mazare 2023-09-02 14:41:48 +02:00
  • e8e33752f4 Sketch a quantized llama using the pyo3 api. (#715) Laurent Mazare 2023-09-02 12:26:05 +02:00
  • dabaa479b9 Update README.md (#714) Laurent Mazare 2023-09-02 08:56:12 +02:00
  • 2c1df6bba1 Add a repeat penality to the llama2-c command line example. (#713) Laurent Mazare 2023-09-01 21:38:58 +02:00
  • 4d56cef583 Handle the empty sequence case properly. (#712) Laurent Mazare 2023-09-01 21:12:30 +02:00
  • 19042962d5 Whisper fix (#711) Laurent Mazare 2023-09-01 21:04:07 +02:00
  • 731e3ffb03 Remove unnecessary file. (#710) Laurent Mazare 2023-09-01 20:42:23 +02:00
  • 2fef14cb14 Add a repeat penalty to the llama2.c wasm example. (#709) Laurent Mazare 2023-09-01 20:32:28 +02:00
  • 1e5b2cc1d5 Add some quantized functions to pyo3. (#708) Laurent Mazare 2023-09-01 19:45:36 +02:00
  • 2ed78ab336 Support for quantized tensors in the python api. (#706) Laurent Mazare 2023-09-01 16:53:42 +02:00
  • 237323c2bc Cleanup the pyo3 setup. (#705) Laurent Mazare 2023-09-01 15:26:18 +02:00
  • af552a5274 Fix the rnn tests for accelerate. (#704) Laurent Mazare 2023-09-01 14:21:38 +02:00
  • 7529531056 Add the optimizer trait. (#702) Laurent Mazare 2023-09-01 13:55:39 +02:00
  • f2d476ca65 Replace the discord link. (#701) Laurent Mazare 2023-09-01 10:43:55 +02:00
  • f9f482d4e5 Add some doc to the varbuilder. (#700) Laurent Mazare 2023-09-01 09:28:35 +02:00
  • 9736236175 Allow retrieving and setting prefix of VarBuilder (#699) Lennard 2023-09-01 09:08:41 +02:00
  • 30a4b593d7 More ops again. (#697) Laurent Mazare 2023-08-31 23:28:48 +02:00
  • 949f1eae6f Implement a couple more binary ops. (#693) Laurent Mazare 2023-08-31 22:30:15 +02:00
  • 7cef35c84d Tweak some quantized args (#692) Laurent Mazare 2023-08-31 18:25:21 +02:00
  • 7509c98970 Interactive mode for the quantized model. (#690) Laurent Mazare 2023-08-31 11:52:42 +02:00
  • 94aa234dfd Add the kv-cache to the whisper wasm version. (#689) Laurent Mazare 2023-08-31 10:37:44 +02:00
  • db59816087 Add a GRU layer. (#688) Laurent Mazare 2023-08-31 09:43:10 +02:00
  • d210c71d77 Set the learning rate. (#687) Laurent Mazare 2023-08-31 09:03:40 +02:00
  • 8e84d8a59b Llama2.c wasm module. (#686) Laurent Mazare 2023-08-31 08:44:32 +02:00
  • 9bd486fb96 Add Yolo Pose to JS Example (#684) Radamés Ajna 2023-08-30 22:32:57 -07:00
  • eaf760a751 Add a python variant for the lstm test. (#682) Laurent Mazare 2023-08-30 23:32:08 +02:00
  • 1d0bb48fae Improve Whisper WASM UI example (#669) Radamés Ajna 2023-08-30 11:35:41 -07:00
  • 21e1c73892 Add a LSTM test. (#681) Laurent Mazare 2023-08-30 20:05:42 +02:00
  • 2047d34b7c More robust tests (so that they pass on accelerate). (#679) Laurent Mazare 2023-08-30 19:10:10 +02:00
  • 9874d843f1 Fix the accelerate build (#678) Laurent Mazare 2023-08-30 18:31:14 +02:00
  • 7d753d3acd Mnist training dropout (#677) Laurent Mazare 2023-08-30 16:41:01 +01:00
  • 3159982a89 Add a Dropout layer (#676) Laurent Mazare 2023-08-30 16:19:28 +01:00
  • ad8a62dbf5 Add tanh. (#675) Laurent Mazare 2023-08-30 13:54:50 +01:00
  • f35b9f6baa Add some recurrent neural networks (#674) Laurent Mazare 2023-08-30 13:27:09 +01:00
  • 618f4e4c78 Add some documentation. (#673) Laurent Mazare 2023-08-30 11:54:00 +01:00
  • 5ac0a98f01 Changelog update. (#672) Laurent Mazare 2023-08-30 09:27:56 +01:00
  • 393690387f Support dilation in conv-transpose2d. (#671) Laurent Mazare 2023-08-30 09:22:00 +01:00
  • 9b25113393 Small cleanups (avoid some possible mutations) (#670) Laurent Mazare 2023-08-30 08:54:00 +01:00