[Testing][Models] Add gpt2 module in testing models #252
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
yaoyaoding
force-pushed
the
gpt2
branch
2 times, most recently
from
536592b to
96bac3a
Compare
vadiklyutiy
pushed a commit
that referenced
this pull request
Jul 22, 2024
…pass (#252) During the graph rewrite, we still keep constant tensors which could be deleted. For example, in [TwoMatmulFusion](https://github.com/CentML/hidet/blob/main/python/hidet/graph/transforms/graph_patterns/matmul_patterns.py#L36): Initially we have: ``` out1 = Matmul(x, c1) out2 = Matmul(x, c2) ``` After graph rewrite optimizations: ``` c = concat([c1, c2]) m = Matmul(x, c) out1, out2 = split(m) ``` We can safely remove `c1` and `c2` after computing `c` and set its trace to None (as if it is a terminal node). However `m` cannot me removed, thus compilation process with ***hidet*** inevitably consumes some additional memory. UPD: `m` is a symbolic tensor (because `x` is symbolic), it does not occupy any memory. Currently testing this approach, but for some reason, after removing those constant tensors `resolve_variant_pass` optimization causes all outputs to be `Nan`. If I exclude `resolve_variant_pass` optimization, it works --------- Co-authored-by: Zhumakhan <nazirzhumakhan@gmail,.com>
vadiklyutiy
pushed a commit
that referenced
this pull request
Jul 23, 2024
…pass (#252) During the graph rewrite, we still keep constant tensors which could be deleted. For example, in [TwoMatmulFusion](https://github.com/CentML/hidet/blob/main/python/hidet/graph/transforms/graph_patterns/matmul_patterns.py#L36): Initially we have: ``` out1 = Matmul(x, c1) out2 = Matmul(x, c2) ``` After graph rewrite optimizations: ``` c = concat([c1, c2]) m = Matmul(x, c) out1, out2 = split(m) ``` We can safely remove `c1` and `c2` after computing `c` and set its trace to None (as if it is a terminal node). However `m` cannot me removed, thus compilation process with ***hidet*** inevitably consumes some additional memory. UPD: `m` is a symbolic tensor (because `x` is symbolic), it does not occupy any memory. Currently testing this approach, but for some reason, after removing those constant tensors `resolve_variant_pass` optimization causes all outputs to be `Nan`. If I exclude `resolve_variant_pass` optimization, it works --------- Co-authored-by: Zhumakhan <nazirzhumakhan@gmail,.com>
vadiklyutiy
pushed a commit
that referenced
this pull request
Dec 26, 2024
…pass (#252) During the graph rewrite, we still keep constant tensors which could be deleted. For example, in [TwoMatmulFusion](https://github.com/CentML/hidet/blob/main/python/hidet/graph/transforms/graph_patterns/matmul_patterns.py#L36): Initially we have: ``` out1 = Matmul(x, c1) out2 = Matmul(x, c2) ``` After graph rewrite optimizations: ``` c = concat([c1, c2]) m = Matmul(x, c) out1, out2 = split(m) ``` We can safely remove `c1` and `c2` after computing `c` and set its trace to None (as if it is a terminal node). However `m` cannot me removed, thus compilation process with ***hidet*** inevitably consumes some additional memory. UPD: `m` is a symbolic tensor (because `x` is symbolic), it does not occupy any memory. Currently testing this approach, but for some reason, after removing those constant tensors `resolve_variant_pass` optimization causes all outputs to be `Nan`. If I exclude `resolve_variant_pass` optimization, it works --------- 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.
Added gpt2 to
hidet.testing.models.gpt2, and implemented the version that supports both initial generation and key-value cache.Enhancement: