[Driver] Make compilation more compatible with multi-processing #350 #351
Conversation
This file contains 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
…del support (hidet-org#347) 1. Enhance support for `__setitem__` and` __getitem__` of Tensor; Add SetStridedSlice Op, Roll Op. 2. Add/Update torch mapping for adaptive_avg_pool3d, eq, pad, roll, matmul, new_zeros, batch_norm, MultiHeadAttention. 3. Update torch Linear mapping to optionally accept transposed weights. 4. Fix a bug where a empty graph will output a zero tensor instead of the input/weight.
…hidet-org#345) Encountered a few minor issues when compiling a transformer-based model using torch.compile with very large batch sizes, submitting the fix here.
This is a continuation of hidet-org#347. 1. Add LP normalization task (ToDo: schedule template) 2. Add torch mappings for normalize, clone, zero_, exp, chunk 3. Add ceil_mode=True support for pool2d 4. Fix dtype issue in resize 5. Fix other bugs in pad, conv2d_pattern
Add an ad-hoc implementation of einsum based on pattern matching. Only supports batched matmul.
…uild can work with spawned processes
|
Hi @soodoshll maybe a quick review ? |
|
@xinli-git LGTM! |
vadiklyutiy
pushed a commit
that referenced
this pull request
Jul 22, 2024
… for conv-bert-base model (#351) Added support for `torch.multiply` and `torch.nn.functional.unfold` These ops are needed in `conv-bert-base` models --------- Co-authored-by: Zhumakhan <nazirzhumakhan@gmail,.com>
vadiklyutiy
pushed a commit
that referenced
this pull request
Jul 23, 2024
… for conv-bert-base model (#351) Added support for `torch.multiply` and `torch.nn.functional.unfold` These ops are needed in `conv-bert-base` models --------- Co-authored-by: Zhumakhan <nazirzhumakhan@gmail,.com>
vadiklyutiy
pushed a commit
that referenced
this pull request
Dec 26, 2024
… for conv-bert-base model (#351) Added support for `torch.multiply` and `torch.nn.functional.unfold` These ops are needed in `conv-bert-base` models --------- Co-authored-by: Zhumakhan <nazirzhumakhan@gmail,.com>
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.
This change adds a filelock to task compilation so that workflows such as distributed inference only builds the task once and avoids any potential data (file) races.
Currently, only task building is included in the filelock because in general, compiled graphs will be different and compiled modules is already protected by task building.