Commit Graph

  • 70d06ab4b0 Add support for the phi-hermes finetuned model. (#1192) Laurent Mazare 2023-10-27 05:57:08 +01:00
  • 0ec5ebcec4 Use the hub model file when possible. (#1190) Laurent Mazare 2023-10-26 20:00:50 +01:00
  • c8e197f68c Fixes for jina-bert. (#1189) Laurent Mazare 2023-10-26 18:52:30 +01:00
  • 5f20697918 Add the jina-bert embeddings model. (#1187) Laurent Mazare 2023-10-26 16:54:36 +01:00
  • e37b487767 Add Blip to online demos README.md (#1184) Blanchon 2023-10-26 12:07:01 +02:00
  • e5dc8cb4f4 [Wasm] BLIP Example (#1183) Radamés Ajna 2023-10-25 23:24:02 -07:00
  • e7b886d56f Add a link to the optimisers crate. (#1180) Laurent Mazare 2023-10-25 21:51:45 +01:00
  • 6a446d9d73 convert pytorch's tensor in Python API (#1172) andrew 2023-10-26 01:39:14 +07:00
  • 0acd16751d Expose the fields from batch-norm. (#1176) Laurent Mazare 2023-10-25 15:35:32 +01:00
  • c698e17619 Enable the test for meshgrid + fix the implementation. (#1175) Laurent Mazare 2023-10-25 13:47:54 +01:00
  • 20da4f44ef Enable the test for meshgrid + fix the implementation. meshgrid-fix Laurent 2023-10-25 13:38:42 +01:00
  • e4c9adfdbe Implemented meshgrid (#1174) Wouter Doppenberg 2023-10-25 13:49:11 +02:00
  • b6053b938b [Wasm] Add puffin phi model to wasm (#1166) Radamés Ajna 2023-10-24 23:09:03 -07:00
  • 45dbe541bc fix ucopy for f64 tensors (#1170) Ibiyemi Abiodun 2023-10-24 12:06:03 -04:00
  • 7bd0faba75 Add support for accelerate in the pyo3 bindings. (#1167) Laurent Mazare 2023-10-24 06:34:37 +01:00
  • 807e3f9f52 derivative for GELU (#1160) KGrewal1 2023-10-23 20:23:45 +01:00
  • eae94a451b PyO3: Add mkl support (#1159) Lukas Kreussel 2023-10-23 21:10:59 +02:00
  • 86e1803191 Add Binary Cross Entropy With Logit Loss to nn crate (#1157) Ogundepo Odunayo 2023-10-23 12:12:44 -04:00
  • 25c3cc4149 Mention the flash-attention restriction in the readme. (#1158) Laurent Mazare 2023-10-23 10:26:56 +01:00
  • a11af79e23 Add a quantized blip model. (#1155) Laurent Mazare 2023-10-22 20:33:25 +01:00
  • 8a82d623e5 Handle LongStorage in pytorch checkpoints. (#1152) Laurent Mazare 2023-10-22 18:34:36 +01:00
  • df2f89b6cf Add some KV cache to blip. (#1150) Laurent Mazare 2023-10-22 09:44:48 +01:00
  • 62fc965617 Expose the track-op method. (#1148) Laurent Mazare 2023-10-22 06:57:03 +01:00
  • 5b32c2a41e Remove the unused pragma and properly apply the bias. (#1147) Laurent Mazare 2023-10-22 06:47:40 +01:00
  • 3115fe42e4 Blip attention mask + readme (#1146) Laurent Mazare 2023-10-21 22:44:13 +01:00
  • 2531b13bf8 Blip fixes (#1145) Laurent Mazare 2023-10-21 21:34:48 +01:00
  • 0d9bb4eb18 Add the blip example. (#1144) Laurent Mazare 2023-10-21 20:05:02 +01:00
  • e8f760ee44 Add get_on_dim. (#1142) Laurent Mazare 2023-10-21 15:01:38 +01:00
  • 94e3373883 Blip forward pass (#1141) Laurent Mazare 2023-10-21 10:19:23 +01:00
  • 34d9e91748 Add the blip image captioning model (#1140) Laurent Mazare 2023-10-20 22:09:11 +01:00
  • cfb423ab76 PyO3: Add CI (#1135) Lukas Kreussel 2023-10-20 20:05:14 +02:00
  • 7366aeac21 Make func cloneable. (#1137) Laurent Mazare 2023-10-20 16:28:50 +01:00
  • 99cf13e8e2 Add the sequential layer. (#1136) Laurent Mazare 2023-10-20 16:08:50 +01:00
  • b43ab6cd1d PyO3: Add None and Tensor indexing to candle.Tensor (#1098) Lukas Kreussel 2023-10-20 10:59:00 +02:00
  • 31ca4897bb Readme updates. (#1134) Laurent Mazare 2023-10-20 09:08:39 +01:00
  • 55351ef57d Add some vision transformers models (#1132) Laurent Mazare 2023-10-19 22:24:18 +01:00
  • 6684b7127a PyO3: Add pytorch like .to() operator to candle.Tensor (#1100) Lukas Kreussel 2023-10-19 22:46:21 +02:00
  • 93c25e8844 Expose the larger resnets (50/101/152) in the example. (#1131) Laurent Mazare 2023-10-19 13:48:28 +01:00
  • cd53c472df Support ResNet 50/101/152. (#1130) Laurent Mazare 2023-10-19 10:48:31 +01:00
  • 6f76383f38 Add a readme for the resnet example. (#1129) Laurent Mazare 2023-10-19 09:58:50 +01:00
  • 8e773cc0c6 Experiment with resnet (#1128) Laurent Mazare 2023-10-19 09:25:03 +01:00
  • 87eb1658e1 Add pad_with_same. (#1127) Laurent Mazare 2023-10-18 23:13:37 +01:00
  • 902d0b9166 More model cloning. (#1126) Laurent Mazare 2023-10-18 21:55:46 +01:00
  • 185b54a33b Make some model cloneable. (#1125) Laurent Mazare 2023-10-18 19:30:47 +01:00
  • 620c94d12e Add support for Zephyr-7b in the quantized model. (#1124) Laurent Mazare 2023-10-18 17:31:26 +01:00
  • 86e7d539d2 Add the quantized mpt model. (#1123) Laurent Mazare 2023-10-18 16:29:38 +01:00
  • cb034506cd Remove the unused pragma in mpt. (#1122) Laurent Mazare 2023-10-18 15:47:50 +01:00
  • 63c204c79e Add a mention to the replit-code model in the readme. (#1121) Laurent Mazare 2023-10-18 11:27:23 +01:00
  • 767a6578f1 MPT alibi fixes. (#1120) Laurent Mazare 2023-10-18 10:58:05 +01:00
  • 662c186fd5 Better error message when overflowing in narrow. (#1119) Laurent Mazare 2023-10-18 08:40:14 +01:00
  • 2cd745a97c MPT fixes. (#1117) Laurent Mazare 2023-10-17 21:53:31 +01:00
  • a72b50e2c0 Build alibi bias. (#1115) Laurent Mazare 2023-10-17 20:41:37 +01:00
  • 872c3f14b0 Add the MPT model. (#1114) Laurent Mazare 2023-10-17 16:06:48 +01:00
  • f9e93f5b69 Extend stub.py to accept external typehinting (#1102) Lukas Kreussel 2023-10-17 12:07:26 +02:00
  • b355ab4e2e Always broadcast magic methods (#1101) Lukas Kreussel 2023-10-17 11:57:12 +02:00
  • 2fe24ac5b1 Rework the cuda casting bits. (#1112) Laurent Mazare 2023-10-17 09:44:51 +01:00
  • 00948eb656 Formatting tweak. (#1111) Laurent Mazare 2023-10-16 21:02:53 +01:00
  • af67672207 Add support for Puffin-Phi-v2. (#1110) Laurent Mazare 2023-10-16 20:54:21 +01:00
  • 6c588c4792 Refactor the pth tensor exctraction. (#1109) Laurent Mazare 2023-10-16 18:16:34 +01:00
  • 122da87580 feat: add pth varbuilder (#1108) OlivierDehaene 2023-10-16 17:20:36 +02:00
  • 75629981bc feat: parse Cuda compute cap from env (#1066) OlivierDehaene 2023-10-16 16:37:38 +02:00
  • 0106b0b04c Read all the tensors in a PyTorch pth file. (#1106) Laurent Mazare 2023-10-16 13:50:07 +01:00
  • 588ad4835a Fix the verbose prompt for phi. (#1097) Laurent Mazare 2023-10-15 10:53:25 +01:00
  • b73c35cc57 Improve the reshape error messages. (#1096) Laurent Mazare 2023-10-15 10:43:10 +01:00
  • 8f310cc666 Avoid trying to backprop through non-differentiable layers. (#1094) Laurent Mazare 2023-10-14 22:03:41 +01:00
  • 8921d5027c Add support for phi-1.0 (#1093) Laurent Mazare 2023-10-14 20:15:43 +01:00
  • f524755634 DDPG checkout. ddpg laurent 2023-10-14 18:57:20 +01:00
  • 29c7f2565d Add some reinforcement learning example. (#1090) Laurent Mazare 2023-10-14 16:46:43 +01:00
  • 9309cfc47d Create a new curand instead of reseeding. (#1089) Laurent Mazare 2023-10-14 10:03:59 +01:00
  • a193bf5f60 Another gemm update. (#1088) Laurent Mazare 2023-10-14 09:36:52 +01:00
  • 2c110ac7d9 Add the pooling operators to the pyo3 layer. (#1086) Laurent Mazare 2023-10-13 21:18:10 +02:00
  • 75989fc3b7 Use an attention mask in the e5 padding case. (#1085) Laurent Mazare 2023-10-13 19:53:40 +02:00
  • 07af87a1d8 Typos. (#1084) Laurent Mazare 2023-10-13 17:21:20 +02:00
  • eefad2b95f Update to gemm 0.16.1 (#1083) Laurent Mazare 2023-10-13 07:40:20 +02:00
  • 5e6df4a3f7 Update to gemm-0.16. (#1082) Laurent Mazare 2023-10-12 22:56:59 +02:00
  • 7473c4ceca Fix the npy read function and add some testing. (#1080) Laurent Mazare 2023-10-12 15:25:05 +02:00
  • c096f02411 Add a matvec cpu benchmark. (#1076) Laurent Mazare 2023-10-12 10:29:18 +02:00
  • e7560443e4 Convmixer example (#1074) Laurent Mazare 2023-10-11 20:51:10 +02:00
  • 89b525b5e7 Convmixer (#1073) Laurent Mazare 2023-10-11 19:24:32 +02:00
  • 69c1fb1ee8 Add a benchmark for the matmul slowness. matmul-slowness Laurent 2023-10-11 15:49:42 +02:00
  • 37dbbff261 Use full tensors for zeros and ones (#1071) Laurent Mazare 2023-10-11 09:16:04 +02:00
  • c55ebaf477 Use full tensors for zeros and ones. Laurent 2023-10-11 08:50:43 +02:00
  • 9fea56d28e Only optimize float tensors. (#1069) Laurent Mazare 2023-10-10 10:05:41 +02:00
  • 4c91dd2ff4 Only optimize float tensors. Laurent 2023-10-10 09:45:49 +02:00
  • bc3351bce4 Tracing for StableLM and quantized StableLM. (#1068) Laurent Mazare 2023-10-10 08:09:25 +02:00
  • b34d7f0248 Remove some unusued bits. (#1067) Laurent Mazare 2023-10-09 20:49:57 +02:00
  • 4d04ac83c7 Override the repo for SDXL f16 vae weights. (#1064) Laurent Mazare 2023-10-09 06:52:28 +01:00
  • 392fe02fba Move the common quantized-nn code to a shared module. (#1063) Laurent Mazare 2023-10-09 06:22:22 +01:00
  • 59ab6d7832 Quantized version of StableLM. (#1058) Laurent Mazare 2023-10-08 15:42:38 +01:00
  • 783735cf22 Use softmax-last-dim where possible. (#1057) Laurent Mazare 2023-10-08 13:16:42 +01:00
  • 9abeddd750 Make the cuda rng seedable. (#1056) Laurent Mazare 2023-10-08 09:32:36 +01:00
  • 2e5fb0b251 Do not use the kv-cache on external key-value states. (#1054) Laurent Mazare 2023-10-07 22:37:19 +01:00
  • 823fe23f9b Add flash-attn support for stable-lm. (#1052) Laurent Mazare 2023-10-07 21:12:54 +01:00
  • d833527fda Use candle_nn::LSTM in encodec. (#1051) Laurent Mazare 2023-10-07 19:43:06 +01:00
  • a4967600d0 More general seq forward functions for RNNs. (#1050) Laurent Mazare 2023-10-07 15:08:01 +01:00
  • aa53368aeb Better control on the optional dequantization in QMatMul (#1049) Laurent Mazare 2023-10-07 10:16:18 +01:00
  • 955e00b2e8 Add to the readmes for stable-lm. (#1047) Laurent Mazare 2023-10-06 21:26:04 +01:00
  • d5f7267087 Add the stable-lm example. (#1046) Laurent Mazare 2023-10-06 19:20:35 +01:00
  • 904bbdae65 Make the Python Wrapper more Hackable and simplify Quantization (#1010) Lukas Kreussel 2023-10-06 20:01:07 +02:00
  • b0442eff8a Sketch the stable-lm model. (#1045) Laurent Mazare 2023-10-06 18:19:06 +01:00