Support Loading to GPU#10710
Closed
xush6528 wants to merge 1 commit intopytorch:masterfrom
xush6528:export-D9372950
Closed
Support Loading to GPU#10710xush6528 wants to merge 1 commit intopytorch:masterfrom xush6528:export-D9372950
xush6528 wants to merge 1 commit intopytorch:masterfrom
xush6528:export-D9372950
Conversation
Summary: Pull Request resolved: #10710 Can't resume from checkpoint for workflows that use GPU. The problem is just we didn't leverage the already-provided GPU deserialization of Caffe2. `keep_device` arg of LoadOp. See https://fburl.com/y27ltaxw How a serialized BlobProto (contraining TensorProto) is loaded into GPU memory? - Load BlobProto from DB. https://fburl.com/pe1qaeyf - Deserialize the BlobProto into a Blob instance. https://fburl.com/5dirjuuh and https://fburl.com/stoho0x1 - Call Blob->Deserialized. https://fburl.com/bnureu32 - Deserializer Registration. https://fburl.com/wbu95ry7 https://fburl.com/ycetud8u - Create TensorCUDA Deserializer. https://fburl.com/2lirfuqj - Create Tensor on GPU and get TensorProto of BlobProto. https://fburl.com/7dre82zg - Copy TensorProto in CPU to Tensor on GPU. https://fburl.com/fr0qk2oe Cloned the GPU workflows for testing in D9125520. Differential Revision: D9372950 fbshipit-source-id: d8cb72608a8b7174e8763ee4c74428297a1cb99b
petrex
pushed a commit
to petrex/pytorch
that referenced
this pull request
Aug 21, 2018
* upstream/master: (147 commits) Support Loading to GPU (pytorch#10710) More changes for hidden visibility (pytorch#10692) Add arguments __repr__ in Distribution base class Add util function from core type to dtype (pytorch#10716) Make ONNX_ATEN_FALLBACK as internal default option Set the BUILD_ENVIRONMENT variable before installing sccache. (pytorch#10640) Avoid shadowing i, j vars in GeneralProposals test (pytorch#10721) Move THNN Reduction to ATen/core. (pytorch#10703) Completely remove build_aten and use_aten (pytorch#10469) Make empty list literals construct empty Tensor[] (pytorch#10705) Soumith's last few patches to v0.4.1 Fix issues link in Caffe2 readme (pytorch#10711) Update the onnx Gemm op to FC/FCTransposed logic in caffe2 onnx backend (pytorch#10108) Adding ATEN_NO_TEST option to root level cmake for propogation to aten Allow method-style casts on tensors (pytorch#10641) Fix pytorch#10698 build failure (pytorch#10704) Add support for Log() Add a bisect percentile operator (pytorch#10563) Fix EnsureCPUOutputOp (pytorch#10651) Nomnigraph - rename some APIs that invole Subtree to Subgraph (pytorch#10551) ...
PenghuiCheng
pushed a commit
to PenghuiCheng/pytorch
that referenced
this pull request
Sep 11, 2018
Summary: Pull Request resolved: pytorch#10710 Can't resume from checkpoint for workflows that use GPU. The problem is just we didn't leverage the already-provided GPU deserialization of Caffe2. `keep_device` arg of LoadOp. See https://fburl.com/y27ltaxw How a serialized BlobProto (contraining TensorProto) is loaded into GPU memory? - Load BlobProto from DB. https://fburl.com/pe1qaeyf - Deserialize the BlobProto into a Blob instance. https://fburl.com/5dirjuuh and https://fburl.com/stoho0x1 - Call Blob->Deserialized. https://fburl.com/bnureu32 - Deserializer Registration. https://fburl.com/wbu95ry7 https://fburl.com/ycetud8u - Create TensorCUDA Deserializer. https://fburl.com/2lirfuqj - Create Tensor on GPU and get TensorProto of BlobProto. https://fburl.com/7dre82zg - Copy TensorProto in CPU to Tensor on GPU. https://fburl.com/fr0qk2oe Cloned the GPU workflows for testing in D9125520. Reviewed By: mraway Differential Revision: D9372950 fbshipit-source-id: 2bf70747bd71e8da16239197f7d2761d63f09ff8
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary:
Can't resume from checkpoint for workflows that use GPU.
The problem is just we didn't leverage the already-provided GPU deserialization of Caffe2.
keep_devicearg of LoadOp. See https://fburl.com/y27ltaxwHow a serialized BlobProto (contraining TensorProto) is loaded into GPU memory?
Cloned the GPU workflows for testing in D9125520.
Differential Revision: D9372950