Stream: git-wasmtime

Topic: wasmtime / PR #9234 [WASI-NN] Add support for a PyTorch b...

Wasmtime GitHub notifications bot (Sep 12 2024 at 18:18):

rahulchaphalkar opened PR #9234 from rahulchaphalkar:pytorch-dyn-link-pr to bytecodealliance:main:

This change adds a PyTorch backend for wasi-nn.
tch crate is used for Libtorch bindings. I have added an image classification example to demonstrate its usage, which uses a torchscript model.
This backend is currently gated behind a wasi-nn feature flag --features pytorch as due to dynamic linking, a Libtorch v2.4.0 installation on the system (specified by LIBTORCH=/path/to/libtorch) is needed for building.

Wasmtime GitHub notifications bot (Sep 12 2024 at 18:18):

rahulchaphalkar requested alexcrichton for a review on PR #9234.

Wasmtime GitHub notifications bot (Sep 12 2024 at 18:18):

rahulchaphalkar requested wasmtime-core-reviewers for a review on PR #9234.

Wasmtime GitHub notifications bot (Sep 12 2024 at 18:18):

rahulchaphalkar requested wasmtime-default-reviewers for a review on PR #9234.

Wasmtime GitHub notifications bot (Sep 12 2024 at 18:41):

abrown assigned abrown to PR #9234.

Wasmtime GitHub notifications bot (Sep 12 2024 at 20:09):

alexcrichton requested abrown for a review on PR #9234.

Wasmtime GitHub notifications bot (Sep 20 2024 at 00:29):

abrown submitted PR review:

This is a good start. The main thing to fix is the handling of the input and output tensors.

Wasmtime GitHub notifications bot (Sep 20 2024 at 00:29):

abrown created PR review comment:

nit: merge these

Wasmtime GitHub notifications bot (Sep 20 2024 at 00:29):

abrown created PR review comment:

This function needs to handle the passed index: a model can have multiple inputs and the job this function needs to do is map the incoming tensor to the right one.

Wasmtime GitHub notifications bot (Sep 20 2024 at 00:29):

abrown created PR review comment:

This seems incorrect: load_from_dir is going to pass the path to a directory and this code will try to use it as a file.

Wasmtime GitHub notifications bot (Sep 20 2024 at 00:29):

abrown created PR review comment:

Storing the output tensor in the same location as the input tensor means that set_input followed immediately by get_output would return the input tensor... probably not what you want here. It looks like forward_ts only returns a single tensor so perhaps just create an output field for that and another input: Vec<Tensor> for the inputs.

Wasmtime GitHub notifications bot (Sep 20 2024 at 00:29):

abrown created PR review comment:

            unimplemented!("the pytorch backend does not yet support GPU execution targets")

Wasmtime GitHub notifications bot (Sep 20 2024 at 00:29):

abrown created PR review comment:

            unimplemented!("the pytorch backend does not yet support TPU execution targets")

Wasmtime GitHub notifications bot (Sep 20 2024 at 00:29):

abrown created PR review comment:

...since PyTorch backend does indeed support GPU execution.

Wasmtime GitHub notifications bot (Sep 20 2024 at 00:29):

abrown created PR review comment:

This is incorrect: we need to retrieve the data regardless of the type, so we need to first figure out how many bytes each Kind is before constructing the receiving buffer, like:

        let data = vec![0u8; size_of(ty) * numel];
        self.1.copy_data_u8(&mut data, numel);

Wasmtime GitHub notifications bot (Sep 20 2024 at 00:30):

abrown commented on PR #9234:

The cargo vet situation is a bit much:

    cargo vet diff zstd-safe 5.0.1+zstd.1.5.2 5.0.2+zstd.1.5.2
                                                          gyscos         zstd                        2 files changed, 4 insertions(+), 4 deletions(-)
    cargo vet diff zstd 0.11.1+zstd.1.5.2 0.11.2+zstd.1.5.2
                                                          gyscos         zip                         3 files changed, 5 insertions(+), 5 deletions(-)
    cargo vet diff num-complex 0.4.2 0.4.6                cuviper        ndarray                     6 files changed, 188 insertions(+), 48 deletions(-)
      NOTE: this project trusts Josh Stone (cuviper) - consider cargo vet trust num-complex or cargo vet trust --all cuviper
    cargo vet inspect constant_time_eq 0.1.5              cesarb         zip                         311 lines
    cargo vet diff sha1 0.10.5 0.10.6                     newpavlov      zip                         7 files changed, 302 insertions(+), 20 deletions(-)
    cargo vet inspect rawpointer 0.2.1                    bluss          ndarray and matrixmultiply  559 lines
    cargo vet diff zip 0.6.4 0.6.6                        Plecra         tch and torch-sys           14 files changed, 604 insertions(+), 109 deletions(-)
    cargo vet inspect inout 0.1.3                         newpavlov      cipher                      1112 lines
      NOTE: cargo vet import zcash would eliminate this
    cargo vet inspect pbkdf2 0.9.0                        tarcieri       zip                         1120 lines
    cargo vet inspect bzip2 0.4.4                         alexcrichton   zip                         2094 lines
      NOTE: this project trusts Alex Crichton (alexcrichton) - consider cargo vet trust bzip2 or cargo vet trust --all alexcrichton
    cargo vet inspect safetensors 0.3.3                   Narsil         tch                         2200 lines
    cargo vet inspect cipher 0.4.4                        newpavlov      aes                         2635 lines
      NOTE: cargo vet import zcash would reduce this to a [130](https://github.com/bytecodealliance/wasmtime/actions/runs/10836457564/job/30070281197?pr=9234#step:6:131)0-line diff
    cargo vet inspect password-hash 0.3.2                 tarcieri       pbkdf2                      3139 lines
    cargo vet inspect base64ct 1.6.0                      tarcieri       password-hash               3381 lines
    cargo vet diff half 1.8.2 2.4.1                       starkat99      tch                         19 files changed, 2546 insertions(+), 958 deletions(-)
    cargo vet inspect time 0.1.44                         jhpratt        zip                         3915 lines
    cargo vet inspect aes 0.7.5                           tarcieri       zip                         6822 lines
    cargo vet inspect matrixmultiply 0.3.8                bluss          ndarray                     7934 lines
    cargo vet inspect ndarray 0.15.6                      jturner314     tch                         41996 lines
    cargo vet inspect torch-sys 0.17.0                    LaurentMazare  tch                         52119 lines
    cargo vet inspect bzip2-sys 0.1.11+1.0.8              alexcrichton   bzip2                       264[133](https://github.com/bytecodealliance/wasmtime/actions/runs/10836457564/job/30070281197?pr=9234#step:6:134) lines
      NOTE: this project trusts Alex Crichton (alexcrichton) - consider cargo vet trust bzip2-sys or cargo vet trust --all alexcrichton
    cargo vet inspect tch 0.17.0                          LaurentMazare  wasmtime-wasi-nn            2287297 lines

Wasmtime GitHub notifications bot (Sep 20 2024 at 20:58):

rahulchaphalkar updated PR #9234 (assigned to abrown).

Wasmtime GitHub notifications bot (Sep 20 2024 at 21:01):

rahulchaphalkar submitted PR review.

Wasmtime GitHub notifications bot (Sep 20 2024 at 21:01):

rahulchaphalkar created PR review comment:

Changed to path.join("model.pt")

Wasmtime GitHub notifications bot (Sep 20 2024 at 21:02):

rahulchaphalkar submitted PR review.

Wasmtime GitHub notifications bot (Sep 20 2024 at 21:02):

rahulchaphalkar created PR review comment:

Done

Wasmtime GitHub notifications bot (Sep 20 2024 at 21:03):

rahulchaphalkar submitted PR review.

Wasmtime GitHub notifications bot (Sep 20 2024 at 21:03):

rahulchaphalkar created PR review comment:

Good point. I added inputs as vector of tensors. Also changed structs in general to use named fields to make code more readable.

Wasmtime GitHub notifications bot (Sep 20 2024 at 21:04):

rahulchaphalkar submitted PR review.

Wasmtime GitHub notifications bot (Sep 20 2024 at 21:04):

rahulchaphalkar created PR review comment:

Done, added kind_to_size() for conversions.

Wasmtime GitHub notifications bot (Sep 20 2024 at 21:05):

rahulchaphalkar submitted PR review.

Wasmtime GitHub notifications bot (Sep 20 2024 at 21:05):

rahulchaphalkar created PR review comment:

Done

Wasmtime GitHub notifications bot (Sep 20 2024 at 21:06):

rahulchaphalkar submitted PR review.

Wasmtime GitHub notifications bot (Sep 20 2024 at 21:06):

rahulchaphalkar created PR review comment:

This is one of the differences for this backend. The module's forward method should handle multiple inputs appropriately if it does support multiple inputs. The vector of input tensors being passed to forward should be sufficient, no index or name is needed.

Wasmtime GitHub notifications bot (Sep 20 2024 at 21:08):

rahulchaphalkar commented on PR #9234:

This is a good start. The main thing to fix is the handling of the input and output tensors.

Thanks for the review, Andrew. I've marked smaller Nits as resolved, and I've addressed other comments as well, but kept them 'unresolved' as of now until you take a look.

Wasmtime GitHub notifications bot (Sep 26 2024 at 20:25):

abrown submitted PR review.

Wasmtime GitHub notifications bot (Sep 26 2024 at 20:25):

abrown created PR review comment:

This is almost there: we're still ignoring the index, though. What might need to happen is that we set up inputs as a Vec<Option<TchTensor>> filled with None and then set the right item based on the index. We are able to retrieve the number of inputs, right?

Wasmtime GitHub notifications bot (Sep 27 2024 at 17:59):

rahulchaphalkar submitted PR review.

Wasmtime GitHub notifications bot (Sep 27 2024 at 17:59):

rahulchaphalkar created PR review comment:

The index is being ignored because jit-compiled pytorch models expect tensors in order, similar to what is achieved by vector of tensors in set_input(). Do we need to use index here to keep things consistent with wasi-nn? I don't think I saw a direct way to retrieve the number of inputs to a model, I can look into that further. However, someone using pytorch would probably not intend on using index.
This was my previous comment on ignoring the index:

This is one of the differences for this backend. The module's forward method should handle multiple inputs appropriately if it does support multiple inputs. The vector of input tensors being passed to forward should be sufficient, no index or name is needed.

Wasmtime GitHub notifications bot (Sep 27 2024 at 18:47):

rahulchaphalkar submitted PR review.

Wasmtime GitHub notifications bot (Sep 27 2024 at 18:47):

rahulchaphalkar created PR review comment:

Did you mean use the index as the position/index for the vector of tensors? That sounds good. I'll see if there's a way to determine the max number of inputs to the given model.

Wasmtime GitHub notifications bot (Sep 30 2024 at 05:58):

rahulchaphalkar updated PR #9234 (assigned to abrown).

Wasmtime GitHub notifications bot (Sep 30 2024 at 06:16):

rahulchaphalkar submitted PR review.

Wasmtime GitHub notifications bot (Sep 30 2024 at 06:16):

rahulchaphalkar created PR review comment:

Alright, so here's what I've changed -

• set_input() now uses the id passed to it. If it is u32, it assigns the tensor to the vector inputs[id], assigning None if there are any empty spots below the given id. This will give the user flexibility to set input tensors non sequentially.
• compute() will check the vector for any None values, and give an error if present, before calling forward_ts().
• There is no reliable way at this time to get max inputs for a model. However, assigning more inputs than available returns an error message detailing the max inputs the model expects at run time, so this is helpful for the user. The expectation, similar to other backends, is for the user to be aware of input size/shape etc.
• If id is a string, I've currently permitted only single input to be set, effectively ignoring the index. If more inputs are assigned, or if u32 and String indexes are used together, we give out an error.

Let me know if this looks fine. Thanks.

Wasmtime GitHub notifications bot (Sep 30 2024 at 15:18):

rahulchaphalkar updated PR #9234 (assigned to abrown).

Wasmtime GitHub notifications bot (Oct 02 2024 at 00:22):

abrown submitted PR review.

Wasmtime GitHub notifications bot (Oct 02 2024 at 00:22):

abrown created PR review comment:

Yeah, looking at 727a4aa, I think that makes more sense. I was kind of hoping that CModule::named_parameters would return the list of inputs by name, but, if that's not the case, then let's just return an error for Id::Name side of things. In any case, we probably don't need id_type: if we get an Id::Index we know were to put it in the vector and if we get an Id::Name then we either (a) look up its index in named_parameters or (b) if that's not possible, return an error.

Wasmtime GitHub notifications bot (Oct 02 2024 at 00:23):

abrown edited PR review comment.

Wasmtime GitHub notifications bot (Oct 03 2024 at 15:37):

rahulchaphalkar submitted PR review.

Wasmtime GitHub notifications bot (Oct 03 2024 at 15:37):

rahulchaphalkar created PR review comment:

I had added the Id:Name case because the wit test cases (nn_wit_image_classification_pytorch in this case) need a Name instead of Index to pass - nn.rs and wasi-nn.wit. Although it looks like set-input might be going away anyway.

Wasmtime GitHub notifications bot (Oct 08 2024 at 16:08):

rahulchaphalkar updated PR #9234 (assigned to abrown).

Wasmtime GitHub notifications bot (Oct 08 2024 at 23:11):

abrown submitted PR review.

Wasmtime GitHub notifications bot (Oct 08 2024 at 23:11):

abrown created PR review comment:

Here's the identical 44.7MB file checked in again.

Wasmtime GitHub notifications bot (Oct 08 2024 at 23:11):

abrown created PR review comment:

Can we find a smaller model or download this instead? Not all Wasmtime users probably want to download this file... twice.

Wasmtime GitHub notifications bot (Oct 08 2024 at 23:11):

abrown created PR review comment:

Can you re-add the new line?

Wasmtime GitHub notifications bot (Oct 08 2024 at 23:11):

abrown edited PR review comment.

Wasmtime GitHub notifications bot (Oct 08 2024 at 23:26):

abrown updated PR #9234 (assigned to abrown).

Wasmtime GitHub notifications bot (Oct 09 2024 at 00:17):

abrown updated PR #9234 (assigned to abrown).

Wasmtime GitHub notifications bot (Oct 09 2024 at 00:41):

abrown updated PR #9234 (assigned to abrown).

Wasmtime GitHub notifications bot (Oct 14 2024 at 15:06):

rahulchaphalkar updated PR #9234 (assigned to abrown).

Wasmtime GitHub notifications bot (Oct 14 2024 at 15:06):

rahulchaphalkar submitted PR review.

Wasmtime GitHub notifications bot (Oct 14 2024 at 15:06):

rahulchaphalkar created PR review comment:

Done

Wasmtime GitHub notifications bot (Oct 14 2024 at 15:12):

rahulchaphalkar submitted PR review.

Wasmtime GitHub notifications bot (Oct 14 2024 at 15:12):

rahulchaphalkar created PR review comment:

I've made the following changes -

• Created a repository to generate jit-compiled libtorch models, and download models from Release for tests and examples.
  https://github.com/rahulchaphalkar/libtorch-models
  https://github.com/rahulchaphalkar/libtorch-models/releases/tag/v0.1

• Switched from Resnet18(45MB) to Squeezenet1.1(4.5MB) as the file to download from above repo.

• This libtorch-models repo can issue new releases if we update the pytorch/libtorch version in this backend from current 2.4.0 to ensure model compatibility.

Wasmtime GitHub notifications bot (Oct 14 2024 at 15:12):

rahulchaphalkar created PR review comment:

Downloading from external repo, same as tests

Wasmtime GitHub notifications bot (Oct 14 2024 at 15:12):

rahulchaphalkar submitted PR review.

Wasmtime GitHub notifications bot (Oct 14 2024 at 15:50):

rahulchaphalkar updated PR #9234 (assigned to abrown).

Wasmtime GitHub notifications bot (Oct 14 2024 at 17:37):

abrown commented on PR #9234:

I think 51930fa and da1f467 have unintentionally updated some crates. Can you remove those commits?

Wasmtime GitHub notifications bot (Oct 14 2024 at 20:11):

rahulchaphalkar updated PR #9234 (assigned to abrown).

Wasmtime GitHub notifications bot (Oct 14 2024 at 20:31):

rahulchaphalkar commented on PR #9234:

I rolled back 2 commits, but there's an issue with the lock file. I had previously attempted to fix this by deleting my lockfile, rebasing off of latest main, and then doing a cargo build to generate any of my lock file changes on top of that.

Wasmtime GitHub notifications bot (Oct 14 2024 at 20:37):

rahulchaphalkar updated PR #9234 (assigned to abrown).

Wasmtime GitHub notifications bot (Oct 15 2024 at 16:26):

rahulchaphalkar commented on PR #9234:

The failing tests fail due to

error: failed retrieving file 'mingw-w64-x86_64-headers-git-12.0.0.r329.g8f7b5ce36-1-any.pkg.tar.zst.sig' from mirror.clarkson.edu : Operation too slow. Less than 1 bytes/sec transferred the last 10 seconds
 mingw-w64-x86_64-gcc-14.2.0-1-any downloading...
error: failed retrieving file 'mingw-w64-x86_64-mpfr-4.2.1-2-any.pkg.tar.zst' from mirror.msys2.org : Operation too slow. Less than 1 bytes/sec transferred the last 10 seconds

I'm hoping a rerun of CI would help.

Wasmtime GitHub notifications bot (Oct 17 2024 at 18:16):

rahulchaphalkar commented on PR #9234:

@abrown can you take a look

Wasmtime GitHub notifications bot (Oct 21 2024 at 21:49):

abrown submitted PR review.

Wasmtime GitHub notifications bot (Oct 21 2024 at 22:04):

abrown merged PR #9234.

Last updated: Dec 23 2024 at 12:05 UTC