- 12 7月, 2021 1 次提交
-
-
由 Stas Bekman 提交于
-
- 10 7月, 2021 2 次提交
-
-
由 Stas Bekman 提交于
* post_init to be run only by a child module * better solution * add test * safer attr name * wants half() * improve doc Co-authored-by: NOlatunji Ruwase <olruwase@microsoft.com>
-
由 Stas Bekman 提交于
* [zero3] params_to_reduce isn't always there Trying to port HF's Electra model's to Deepspeed I'm getting this on the very first backward step (with some extra debug): ``` Incrementing with parameter id 42 ------ Before allocating allgather param name=generator_lm_head.weight id=41 shape=torch.Size([1]) status=ZeroParamStatus.NOT_AVAILABLE partition size=327680 ------allgather param with name=generator_lm_head.weight id=41 shape=torch.Size([1]) status=ZeroParamStatus.NOT_AVAILABLE partition size=327680 ------ Before allocating allgather param name=generator_lm_head.bias id=42 shape=torch.Size([1]) status=ZeroParamStatus.NOT_AVAILABLE partition size=5120 ------allgather param with name=generator_lm_head.bias id=42 shape=torch.Size([1]) status=ZeroParamStatus.NOT_AVAILABLE partition size=5120 Backward name=generator_lm_head.weight id=41 shape=torch.Size([5120, 64]) Inside reduce ipg buckets. name=generator_lm_head.weight id=41 shape=torch.Size([5120, 64]), ipg elements 0, reduce bucket size 4096 Params in ipg bucket [] Reducing [] GOT 1 torch.Size([4096]) Traceback (most recent call last): File "examples/pytorch/language-modeling/run_mlm.py", line 533, in <module> main() File "examples/pytorch/language-modeling/run_mlm.py", line 484, in main train_result = trainer.train(resume_from_checkpoint=checkpoint) File "/mnt/nvme1/code/huggingface/transformers-ds-zero_to_fp32-tests/src/transformers/trainer.py", line 1269, in train tr_loss += self.training_step(model, inputs) File "/mnt/nvme1/code/huggingface/transformers-ds-zero_to_fp32-tests/src/transformers/trainer.py", line 1778, in training_step loss = self.deepspeed.backward(loss) File "/mnt/nvme1/code/github/00optimize/DeepSpeed-zero-init-child-only-post_init/deepspeed/runtime/engine.py", line 1188, in backward self.optimizer.backward(loss) File "/mnt/nvme1/code/github/00optimize/DeepSpeed-zero-init-child-only-post_init/deepspeed/runtime/zero/stage3.py", line 2964, in backward self.loss_scaler.backward(loss.float(), retain_graph=retain_graph) File "/mnt/nvme1/code/github/00optimize/DeepSpeed-zero-init-child-only-post_init/deepspeed/runtime/fp16/loss_scaler.py", line 53, in backward scaled_loss.backward(retain_graph=retain_graph) File "/home/stas/anaconda3/envs/py38-pt19/lib/python3.8/site-packages/torch/_tensor.py", line 255, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/stas/anaconda3/envs/py38-pt19/lib/python3.8/site-packages/torch/autograd/__init__.py", line 147, in backward Variable._execution_engine.run_backward( File "/mnt/nvme1/code/github/00optimize/DeepSpeed-zero-init-child-only-post_init/deepspeed/runtime/zero/stage3.py", line 1867, in reduce_partition_and_remove_grads self.reduce_ready_partitions_and_remove_grads(param, i) File "/mnt/nvme1/code/github/00optimize/DeepSpeed-zero-init-child-only-post_init/deepspeed/runtime/zero/stage3.py", line 2212, in reduce_ready_partitions_and_remove_grads self.reduce_independent_p_g_buckets_and_remove_grads(param, i) File "/mnt/nvme1/code/github/00optimize/DeepSpeed-zero-init-child-only-post_init/deepspeed/runtime/zero/stage3.py", line 1897, in reduce_independent_p_g_buckets_and_remove_grads self.reduce_ipg_grads() File "/mnt/nvme1/code/github/00optimize/DeepSpeed-zero-init-child-only-post_init/deepspeed/runtime/zero/stage3.py", line 2193, in reduce_ipg_grads self.average_tensor(reduction_list, params_to_reduce) File "/mnt/nvme1/code/github/00optimize/DeepSpeed-zero-init-child-only-post_init/deepspeed/runtime/zero/stage3.py", line 1972, in average_tensor params_to_reduce[0].reduce_gradients_at_owner( ``` Is it always that `params_to_reduce` is populated? If I add this check the problem goes away it seems. * real fix
-
- 08 7月, 2021 1 次提交
-
-
由 Jeff Rasley 提交于
-
- 02 7月, 2021 2 次提交
-
-
由 Samyam Rajbhandari 提交于
* contiguous gradients should be set to True by default * Set contiguous gradients to True by default Features such as reduce_scatter depends on contiguous gradients being True. This is also the preferred default configuration.
-
由 Jeff Rasley 提交于
-
- 29 6月, 2021 1 次提交
-
-
由 Stas Bekman 提交于
Co-authored-by: NJeff Rasley <jerasley@microsoft.com>
-
- 26 6月, 2021 1 次提交
-
-
由 Stas Bekman 提交于
* undo noise * another
-
- 24 6月, 2021 5 次提交
-
-
由 Hyunwoong Ko 提交于
* Fix bugs about non-contiguous tensor broadcasting * Fix typo Co-authored-by: NJeff Rasley <jerasley@microsoft.com> Co-authored-by: NReza Yazdani <44502768+RezaYazdaniAminabadi@users.noreply.github.com>
-
由 Stas Bekman 提交于
Co-authored-by: NOlatunji Ruwase <olruwase@microsoft.com> Co-authored-by: NJeff Rasley <jerasley@microsoft.com>
-
由 Stas Bekman 提交于
Co-authored-by: NSamyam Rajbhandari <samyamr@microsoft.com> Co-authored-by: NJeff Rasley <jerasley@microsoft.com>
-
由 Stas Bekman 提交于
Co-authored-by: NOlatunji Ruwase <olruwase@microsoft.com> Co-authored-by: NJeff Rasley <jerasley@microsoft.com>
-
由 Jeff Rasley 提交于
-
- 22 6月, 2021 1 次提交
-
-
由 Jeff Rasley 提交于
-
- 18 6月, 2021 2 次提交
-
-
由 Olatunji Ruwase 提交于
-
由 Jeff Rasley 提交于
-
- 17 6月, 2021 4 次提交
-
-
由 Jeff Rasley 提交于
-
由 Jeff Rasley 提交于
-
由 Olatunji Ruwase 提交于
* Fix docstring * Make screenshots clickable for easier viewing * Navigation menu in alphabetical order; More clicable screenshots * Rename 1Cycle doc * Tweak naming * Remove no longer used flag * ZeRO3 Offload release * Single GPU results * Rearrange figures * Single GPU text * tweak intro * zero3-offload section * Add asynchronous i/o docs * Fix print_per_steps doc
-
由 Samyam Rajbhandari 提交于
* largest_partitioned_params calculation fix largest partitioned params was getting calculated incorrectly * Update stage3.py * Update stage3.py * formatting fix * changing sub-group size default to 1e9 Co-authored-by: NOlatunji Ruwase <olruwase@microsoft.com>
-
- 15 6月, 2021 1 次提交
-
-
由 Hyunwoong Ko 提交于
* Add `import os` to inference tutorials * assign deepspeed-initialized model to hf model
-
- 10 6月, 2021 2 次提交
-
-
由 eltonzheng 提交于
* Add Windows support in README, use c++17 on Windows to support latest vc build tool * Add detailed cpp build tools version in README Co-authored-by: NJeff Rasley <jerasley@microsoft.com>
-
由 Shaden Smith 提交于
* unit test for bugfix #1135 * formatter * fix test in presence of mpi4py Co-authored-by: NJeff Rasley <jerasley@microsoft.com>
-
- 09 6月, 2021 7 次提交
-
-
由 Jeff Rasley 提交于
-
由 Jeff Rasley 提交于
-
由 Jeff Rasley 提交于
-
由 Reza Yazdani 提交于
* fix links for inference tutorial * Fix automatic injection. Add the local-attention for GPT-Neo * fix the inference for generation of large sequences (>1K & <32K) * fix format Co-authored-by: NJeff Rasley <jerasley@microsoft.com>
-
由 Jeff Rasley 提交于
* remove old unit tests, should have been removed in rebase * formatting
-
由 Cody 提交于
Co-authored-by: NJeff Rasley <jerasley@microsoft.com>
-
由 Stas Bekman 提交于
Co-authored-by: NOlatunji Ruwase <olruwase@microsoft.com> Co-authored-by: NJeff Rasley <jerasley@microsoft.com>
-
- 08 6月, 2021 3 次提交
-
-
由 Stas Bekman 提交于
Co-authored-by: NOlatunji Ruwase <olruwase@microsoft.com>
-
由 Stas Bekman 提交于
* fix missed subclassed partitioning bug * fix on exit Co-authored-by: NOlatunji Ruwase <olruwase@microsoft.com>
-
由 Stas Bekman 提交于
* allow minor cuda version differences * cleanup * typo
-
- 04 6月, 2021 1 次提交
-
-
由 Stas Bekman 提交于
fixes: s/micro_batch_per_gpu/train_micro_batch_size_per_gpu/ Co-authored-by: NOlatunji Ruwase <olruwase@microsoft.com>
-
- 03 6月, 2021 3 次提交
-
-
由 Stas Bekman 提交于
-
由 Reza Yazdani 提交于
-
由 Reza Yazdani 提交于
* Change the sparse attention API to be compatible with latest changes on the triton side * remove compatibility checks for CUDA 11 * Update requirements-sparse_attn.txt Co-authored-by: NArash Ashari <arashari@microsoft.com>
-
- 28 5月, 2021 1 次提交
-
-
由 Reza Yazdani 提交于
-
- 26 5月, 2021 1 次提交
-
-
由 Reza Yazdani 提交于
-
- 25 5月, 2021 1 次提交
-
-
由 eltonzheng 提交于
Co-authored-by: NJeff Rasley <jerasley@microsoft.com>
-