Skip to content

Conversation

@dcslin
Copy link
Member

@dcslin dcslin commented Aug 14, 2019

Hi this is not completely ready for merge, pending to fix a testing error.

@dcslin dcslin changed the title tc comprehension integration Ref. SINGA-482 SINGA-482 [WIP] tc comprehension integration Aug 14, 2019
@dcslin dcslin changed the title SINGA-482 [WIP] tc comprehension integration SINGA-482 tc comprehension integration Aug 15, 2019
@dcslin dcslin marked this pull request as ready for review August 15, 2019 09:09
@dcslin
Copy link
Member Author

dcslin commented Aug 15, 2019

Hi, this is the working version of tc integration for now, with following notes:

  1. tested in the environment specified in the dockerfile cuda9
  2. it implemented MatMul, Relu in TC for demonstration and usage is shown in following tests:
    ./bin/test_singa --gtest_filter=*TC*, and tests are passing.
  3. for now only working in Cuda environment, and only working with type float

P.S. softmax is implemented but still checking why the output is not correct.

@dcslin
Copy link
Member Author

dcslin commented Aug 22, 2019

TODO:

  1. move tc execution code into device DoExec, and support CPU
  2. package dependencies as conda package
  3. integer support

@dcslin
Copy link
Member Author

dcslin commented Sep 23, 2019

Integration is done except support CPU.
CPU support development is done, however the behavior was not expected even in clean TC env.
CPU backed TC return zero tensor as shown in https://gist.githubusercontent.com/dcslin/f3ed411012c144163f3bedbe621257f6/raw/9f6af3f5e62b5bb52a58c087967c813b3acd8872/tensor-comprehension-cpu-errors

@dcslin
Copy link
Member Author

dcslin commented Nov 1, 2019

Ready for review.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant