Implement Libtask.might_produce_if_sig_contains#218
Merged
penelopeysm merged 3 commits intomainfrom Mar 2, 2026
Merged
Conversation
Member
Author
|
I don't expect this to cause any issues with Libtask. The main question is whether adding this to Turing.jl Libtask.might_produce_if_sig_contains(::Type{<:DynamicPPL.Model}) = truecauses any slowdown in pMCMC, which I'll test out now. |
|
Libtask.jl documentation for PR #218 is available at: |
penelopeysm
added a commit
to TuringLang/Turing.jl
that referenced
this pull request
Mar 2, 2026
This needs a Libtask release and version bump, which I'll handle once JuliaRegistrator does its things. This essentially implements the plan described in TuringLang/Libtask.jl#217. A lot of the issues stemming from Libtask not picking up model evaluators either with keyword arguments, or in submodels, can be fixed by simply declaring that **every** method that dispatches on `DynamicPPL.Model` is produceable. The mechanism for this is implemented in TuringLang/Libtask.jl#218, and this PR makes use of that. **For the end-user, this means that we guarantee correctness where models either have submodels or where models have keyword arguments. The user no longer has to mark models with keyword arguments as `@might_produce`.** I tested performance, and there is no regression — in fact there is a small speedup (although that is probably benchmark noise): ## Submodel case This was the issue #2772 where non-inlined submodels were not correctly picked up. #2778 fixed this with a strategy that was similar to that in this PR, but was slightly more limited (this PR handles both submodels and keyword arguments together). ```julia using Turing, StableRNGs, Test @model function inner(y, x) @noinline y ~ Normal(x) end @model function nested(y) x ~ Normal() a ~ to_submodel(inner(y, x)) end m1 = nested(1.0) @time sample(StableRNG(468), m1, PG(10), 2000; chain_type=Any, progress=false); # 2.585299 seconds on #2778 # 2.523017 seconds on this PR ``` ## Keyword argument case This was the long-standing issue where models with keyword arguments were originally not picked up by Libtask, and since v0.42.5, could be, but relied on the user themselves manually declaring `Libtask.@might_produce`. ```julia @model function withkw(; y=0.0) x ~ Normal() y ~ Normal(x) end m1 = withkw(y=10.0); # Turing.@might_produce(withkw) @time sample(StableRNG(468), m1, PG(10), 2000; chain_type=Any, progress=false); # withkw case # 4.741234 seconds on main (requiring @might_produce) # 4.441797 seconds on this PR (and not requiring @might_produce) ```
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
1 participant
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.
Closes #217. Please see that issue for explanation.