Make threadsafe evaluation opt-in #1151

penelopeysm · 2025-11-20T14:39:53Z

This adds a new type parameter to the model struct, which indicates whether threadsafe evaluation is required or not.

The default is false, so users must specifically opt-in. It's easy to make this opt-out instead, by changing the default to true. But that would probably tank performance on single-threaded sessions.

If a user declares a model with Threads.@threads or just @threads inside, it will issue a warning. This is not a foolproof detection method. But it will probably catch 95% of cases.

I believe this is the best technical solution. #1128 provides a global on/off switch, but that is quite slow. Furthermore, it's not technically correct either; threadsafe evaluation is a property of a model, not of a Julia session.

Here is a demo (launch Julia with > 1 thread, or else you won't observe this):

julia> using DynamicPPL, Distributions

julia> @model function f(x)
           a ~ Normal()
           Threads.@threads for i in eachindex(x)
               x[i] ~ Normal(a)
           end
       end
┌ Warning: It looks like you are using `Threads.@threads` in your model definition.
│
│ Note that since version 0.39 of DynamicPPL, threadsafe evaluation of models is disabled by default. If you need it, you will need to explicitly enable it by creating the model, and then running `model = setthreadsafe(model, true)`.
│
│ Threadsafe model evaluation is only needed when parallelising tilde-statements (not arbitrary Julia code), and avoiding it can often lead to significant performance improvements.
│
│ Please see https://turinglang.org/docs/usage/threadsafe-evaluation/ for more details of when threadsafe evaluation is actually required.
└ @ DynamicPPL ~/ppl/dppl/src/compiler.jl:383
f (generic function with 2 methods)

julia> x = randn(1000); unsafe_model = f(x);

julia> correct_logjoint(a) = logpdf(Normal(), a) + sum(logpdf.(Normal(a), x))
correct_logjoint (generic function with 1 method)

julia> correct_logjoint(0.5)
-1531.2868766878232

julia> logjoint(unsafe_model, (; a = 0.5))
-747.7508803635657

julia> logjoint(unsafe_model, (; a = 0.5))
-805.0710717034472

julia> logjoint(unsafe_model, (; a = 0.5))
-576.5499112280947

The above gives wrong results because threadsafe evaluation wasn't requested. This PR lets you do that:

julia> safe_model = setthreadsafe(unsafe_model, true)
Model{typeof(f), (:x,), (), (), Tuple{Vector{Float64}}, Tuple{}, DefaultContext, true}(f, (x = [1.7022725865773345, 0.24976102119304136, 0.3819156217493525, -2.4421990126257653, 1.0112387777431466, 0.0016047673674860264, -0.2252466150401273, 1.0744166063416623, -0.629808413260951, 0.041422940392546084  …  -0.06903422786623391, 0.7349873184231311, 1.1795733883972002, -0.8759768619489093, 1.1966743339419046, 0.850590281127217, 0.45075481294292064, -0.9911445478594456, 0.3609019495825119, 1.2313172493323474],), NamedTuple(), DefaultContext())

julia> logjoint(safe_model, (; a = 0.5))
-1531.286876687823

julia> logjoint(safe_model, (; a = 0.5))
-1531.286876687823

julia> logjoint(safe_model, (; a = 0.5))
-1531.286876687823

src/compiler.jl

github-actions · 2025-11-20T14:57:46Z

Benchmark Report

this PR's head: e839de88ab838d3db0e595793729a4abb1f97bde
base branch: a6d56a2b9074d9da27eea4a6e4a2ab9a3013913f

Computer Information

Julia Version 1.11.7
Commit f2b3dbda30a (2025-09-08 12:10 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 4 × Intel(R) Xeon(R) Platinum 8370C CPU @ 2.80GHz
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, icelake-server)
Threads: 1 default, 0 interactive, 1 GC (on 4 virtual cores)

Benchmark Results

┌───────────────────────┬───────┬─────────────┬───────────────────┬────────┬─────────────────────────────────┬────────────────────────────┬─────────────────────────────────┐
│                       │       │             │                   │        │        t(eval) / t(ref)         │     t(grad) / t(eval)      │        t(grad) / t(ref)         │
│                       │       │             │                   │        │ ──────────┬───────────┬──────── │ ───────┬─────────┬──────── │ ──────────┬───────────┬──────── │
│                 Model │   Dim │  AD Backend │           VarInfo │ Linked │      base │   this PR │ speedup │   base │ this PR │ speedup │      base │   this PR │ speedup │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│               Dynamic │    10 │    mooncake │             typed │   true │    400.15 │    372.71 │    1.07 │   9.71 │    8.98 │    1.08 │   3883.99 │   3345.54 │    1.16 │
│                   LDA │    12 │ reversediff │             typed │   true │   2647.33 │   2608.86 │    1.01 │   5.39 │    5.18 │    1.04 │  14259.47 │  13518.70 │    1.05 │
│   Loop univariate 10k │ 10000 │    mooncake │             typed │   true │ 108431.07 │ 111527.91 │    0.97 │   4.19 │    3.96 │    1.06 │ 453896.18 │ 441270.87 │    1.03 │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│    Loop univariate 1k │  1000 │    mooncake │             typed │   true │   9169.16 │   7945.89 │    1.15 │   4.28 │    4.67 │    0.92 │  39235.95 │  37141.43 │    1.06 │
│      Multivariate 10k │ 10000 │    mooncake │             typed │   true │  33886.61 │  31865.54 │    1.06 │   9.91 │   10.25 │    0.97 │ 335657.94 │ 326490.16 │    1.03 │
│       Multivariate 1k │  1000 │    mooncake │             typed │   true │   3681.23 │   3472.68 │    1.06 │   9.29 │    9.55 │    0.97 │  34187.52 │  33148.73 │    1.03 │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│ Simple assume observe │     1 │ forwarddiff │             typed │  false │      3.80 │      2.70 │    1.41 │   2.96 │    3.90 │    0.76 │     11.23 │     10.50 │    1.07 │
│           Smorgasbord │   201 │ forwarddiff │             typed │  false │   1221.81 │   1206.79 │    1.01 │  63.55 │  122.64 │    0.52 │  77649.38 │ 148000.70 │    0.52 │
│           Smorgasbord │   201 │ forwarddiff │       simple_dict │   true │       err │       err │     err │    err │     err │     err │       err │       err │     err │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│           Smorgasbord │   201 │ forwarddiff │ simple_namedtuple │   true │       err │       err │     err │    err │     err │     err │       err │       err │     err │
│           Smorgasbord │   201 │      enzyme │             typed │   true │   1681.55 │   1694.01 │    0.99 │   5.84 │    5.70 │    1.03 │   9826.93 │   9651.74 │    1.02 │
│           Smorgasbord │   201 │    mooncake │             typed │   true │   1673.56 │   1695.58 │    0.99 │   5.40 │    5.21 │    1.04 │   9034.67 │   8825.54 │    1.02 │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│           Smorgasbord │   201 │ reversediff │             typed │   true │   2010.82 │   1723.80 │    1.17 │  73.22 │   86.59 │    0.85 │ 147231.62 │ 149272.05 │    0.99 │
│           Smorgasbord │   201 │ forwarddiff │      typed_vector │   true │   1675.14 │   1679.82 │    1.00 │  57.47 │   55.00 │    1.05 │  96271.87 │  92382.62 │    1.04 │
│           Smorgasbord │   201 │ forwarddiff │           untyped │   true │   1663.29 │   1677.47 │    0.99 │ 120.71 │   55.83 │    2.16 │ 200780.30 │  93650.92 │    2.14 │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│           Smorgasbord │   201 │ forwarddiff │    untyped_vector │   true │   1672.06 │   1661.01 │    1.01 │  58.75 │   54.68 │    1.07 │  98225.72 │  90829.94 │    1.08 │
│              Submodel │     1 │    mooncake │             typed │   true │      8.59 │      7.07 │    1.22 │   4.46 │    5.17 │    0.86 │     38.30 │     36.51 │    1.05 │
└───────────────────────┴───────┴─────────────┴───────────────────┴────────┴───────────┴───────────┴─────────┴────────┴─────────┴─────────┴───────────┴───────────┴─────────┘

github-actions · 2025-11-20T15:04:11Z

DynamicPPL.jl documentation for PR #1151 is available at:
https://TuringLang.github.io/DynamicPPL.jl/previews/PR1151/

codecov · 2025-11-20T16:07:09Z

Codecov Report

❌ Patch coverage is 97.56098% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 79.99%. Comparing base (a6d56a2) to head (e839de8).
⚠️ Report is 1 commits behind head on breaking.

Files with missing lines	Patch %	Lines
src/model.jl	96.00%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##           breaking    #1151      +/-   ##
============================================
- Coverage     80.66%   79.99%   -0.67%     
============================================
  Files            41       41              
  Lines          3878     3874       -4     
============================================
- Hits           3128     3099      -29     
- Misses          750      775      +25

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

mhauru

Very happy with this. Only some minor style points to raise, and to wait for that docs page to be up.

docs/src/api.md

src/compiler.jl

mhauru · 2025-11-21T10:33:28Z

src/fasteval.jl

+    accs = map(
+        acc -> DynamicPPL.convert_eltype(float_type_with_fallback(eltype(params)), acc),
+        accs,
+    )
+    vi_wrapped = ThreadSafeVarInfo(OnlyAccsVarInfo(accs))
+    _, vi_wrapped = DynamicPPL._evaluate!!(model, vi_wrapped)
+    vi = OnlyAccsVarInfo(DynamicPPL.getaccs(vi_wrapped))


Would it be simpler to have a single method and wrap this part in a if is_threaded(f.model)? It should get constant propagated away at compile time. Very optional to change, the current version isn't bad either.

This probably purely a style question, but there could be a difference in that listing all the type parameters of Model in the function signature I think forces specialisation on all of them.

Yes, good idea, I was a bit concerned about the specialisation too. I thought about making threaded the first type parameter, which would also avoid this (and we ~~don't~~ rarely dispatch on any other type parameters in Model)... then decided against it because it might be too breaking

mhauru · 2025-11-21T10:36:48Z

src/model.jl

    args::NamedTuple{argnames,Targs},
    defaults::NamedTuple{kwargnames,Tkwargs},
    context::AbstractContext=DefaultContext(),
+    threadsafe::Bool=false,


Is there a particular meaning to sometimes calling this threadsafe and sometimes Threaded?

Threaded is a type parameter and threadsafe is a function argument 😄

A bit like Ctx and context.

Open to different names though!

src/model.jl

mhauru · 2025-11-21T10:45:50Z

src/compiler.jl

    # If it's a macro, we expand it
    if Meta.isexpr(expr, :macrocall)
-        return generate_mainbody!(mod, found, macroexpand(mod, expr; recursive=true), warn)
+        if expr.args[1] == Expr(:., :Threads, QuoteNode(Symbol("@threads"))) &&


I wonder how often people do

using Threads: @threads @threads

vs how often people call some other macro called @threads. I.e. false positives vs false negatives.

Another case where I wasn't too sure where to draw the line. I think Threads.@threads probably accounts for most usage, but I'm not averse to also handling @threads since after all the warning message is quite noncommittal.

I would err on the safe side in warning about @threads, but happy if you prefer otherwise.

Agreed, changed now

penelopeysm · 2025-11-27T12:36:03Z

Current benchmarks locally, all run without threadsafe eval:

unlinked @model f() = x ~ Normal()
                This PR                               breaking
eval      ----  8.754 ns                              10.417 ns
grad (FD) ----  51.638 ns (3 allocs: 96 bytes)        55.319 ns (3 allocs: 96 bytes)
grad (RD) ----  3.069 μs (44 allocs: 1.500 KiB)       3.051 μs (44 allocs: 1.500 KiB)
grad (MC) ----  127.976 ns (2 allocs: 64 bytes)       136.688 ns (2 allocs: 64 bytes)
grad (EN) ----  59.902 ns (2 allocs: 64 bytes)        101.934 ns (2 allocs: 64 bytes)

unlinked DynamicPPL.TestUtils.DEMO_MODELS[3]
                This PR                               breaking
eval      ----  234.252 ns (8 allocs: 352 bytes)      235.276 ns (8 allocs: 352 bytes)
grad (FD) ----  425.373 ns (13 allocs: 752 bytes)     483.333 ns (13 allocs: 752 bytes)
grad (RD) ----  14.792 μs (228 allocs: 8.438 KiB)     15.375 μs (230 allocs: 8.578 KiB)
grad (MC) ----  1.321 μs (18 allocs: 800 bytes)       1.363 μs (18 allocs: 800 bytes)
grad (EN) ----  899.719 ns (22 allocs: 1008 bytes)    1.231 μs (25 allocs: 1.141 KiB)

linked @model f() = x ~ Normal()
                This PR                               breaking
eval      ----  11.820 ns (1 allocs: 32 bytes)        14.505 ns (1 allocs: 32 bytes)
grad (FD) ----  64.845 ns (4 allocs: 144 bytes)       66.176 ns (4 allocs: 144 bytes)
grad (RD) ----  3.162 μs (51 allocs: 1.750 KiB)       3.097 μs (52 allocs: 1.781 KiB)
grad (MC) ----  199.830 ns (4 allocs: 128 bytes)      198.053 ns (4 allocs: 128 bytes)
grad (EN) ----  96.390 ns (5 allocs: 160 bytes)       174.799 ns (6 allocs: 208 bytes)

linked DynamicPPL.TestUtils.DEMO_MODELS[3]
                This PR                               breaking
eval      ----  295.918 ns (12 allocs: 528 bytes)     299.479 ns (12 allocs: 528 bytes)
grad (FD) ----  558.962 ns (17 allocs: 1.094 KiB)     575.521 ns (17 allocs: 1.094 KiB)
grad (RD) ----  16.834 μs (251 allocs: 9.188 KiB)     16.917 μs (253 allocs: 9.297 KiB)
grad (MC) ----  1.691 μs (26 allocs: 1.125 KiB)       1.654 μs (26 allocs: 1.125 KiB)
grad (EN) ----  1.052 μs (28 allocs: 1.203 KiB)       1.402 μs (34 allocs: 1.469 KiB)

~~I don't really understand why the last one is slower (and it's reproducible). Will try to profile.~~

I was wrong; it's not reproducible. I benchmarked again and the last one was 295 ns, which seems about right (cutting out the Threads.nthreads() > 1 check seems to shave a few nanoseconds off). The table above has been updated.

penelopeysm · 2025-11-27T13:48:15Z

src/model.jl

-@inline function init!!(
+function init!!(


re. #1139 (review)

Will the fragility of performance get better once TSVI is opt-in, and the num_threads > 1 check is replaced with something that can be evaluated at compile time? Also, even if it doesn't, I presume performance won't be worse than it was with the old LDF, just maybe roughly the same?

Turns out that yes, that helped :) So we can get rid of the @inline, AND also shift the convert_eltype thing inside the threadsafe bit. I checked with @code_typed and that stuff definitely gets compiled away if it's not needed.

penelopeysm · 2025-11-27T13:50:27Z

src/threadsafe.jl

-    # seems to provide an upper bound to maxthreadid(), so we use that here.
-    # See https://github.com/TuringLang/DynamicPPL.jl/pull/936
-    accs_by_thread = [map(split, getaccs(vi)) for _ in 1:(Threads.nthreads() * 2)]
+    accs_by_thread = [map(split, getaccs(vi)) for _ in 1:Threads.maxthreadid()]


Also very pleased to clean this up.

penelopeysm · 2025-11-27T14:19:04Z

src/debug_utils.jl

-    # Force single-threaded execution.
-    _, varinfo = DynamicPPL.evaluate_threadunsafe!!(model, varinfo)
+    # TODO(penelopeysm): Implement merge, etc. for DebugAccumulator, and then perform a
+    # check on the merged accumulator, rather than checking it in the accumulate_assume
+    # calls. That way we can also correctly support multi-threaded evaluation.
+    _, varinfo = DynamicPPL.evaluate!!(model, varinfo)


This is the last thing that I'm a bit displeased about, but it's still better than on main (I wrote up an issue here #1157), so I thinkkkkk we can leave the proper fix to another PR.

mhauru

Just a couple of trivialities.

Interesting how much Enzyme benchmarks care about this PR.

mhauru · 2025-11-28T16:33:51Z

src/compiler.jl

+            expr.args[1] == Expr(:., :Threads, QuoteNode(Symbol("@threads"))) &&
+            !warn_threads
+        )
+            warn_threads = true


Based on the name, I would have guessed this would work the other way around with true/false (I read "warn" as imperative).

Ah, yes, let's flip it.

src/model.jl

Fix outdated docstrings Co-authored-by: Markus Hauru <[email protected]>

penelopeysm · 2025-12-01T11:05:58Z

HISTORY.md

+When threadsafe evaluation is enabled for a model, an internal flag is set on the model.
+The value of this flag can be queried using `DynamicPPL.requires_threadsafe(model)`, which returns a boolean.
+This function is newly exported in this version of DynamicPPL.


I realised that Turing needs this function (basically PG/SMC should error any time they encounter a model that needs threadsafe eval -- TuringLang/Turing.jl#2658), so we need to export it. Other changes just follow on from review!

mhauru

Thanks!

* v0.39 * Update DPPL compats for benchmarks and docs * remove merge conflict markers * Remove `NodeTrait` (#1133) * Remove NodeTrait * Changelog * Fix exports * docs * fix a bug * Fix doctests * Fix test * tweak changelog * FastLDF / InitContext unified (#1132) * Fast Log Density Function * Make it work with AD * Optimise performance for identity VarNames * Mark `get_range_and_linked` as having zero derivative * Update comment * make AD testing / benchmarking use FastLDF * Fix tests * Optimise away `make_evaluate_args_and_kwargs` * const func annotation * Disable benchmarks on non-typed-Metadata-VarInfo * Fix `_evaluate!!` correctly to handle submodels * Actually fix submodel evaluate * Document thoroughly and organise code * Support more VarInfos, make it thread-safe (?) * fix bug in parsing ranges from metadata/VNV * Fix get_param_eltype for TSVI * Disable Enzyme benchmark * Don't override _evaluate!!, that breaks ForwardDiff (sometimes) * Move FastLDF to experimental for now * Fix imports, add tests, etc * More test fixes * Fix imports / tests * Remove AbstractFastEvalContext * Changelog and patch bump * Add correctness tests, fix imports * Concretise parameter vector in tests * Add zero-allocation tests * Add Chairmarks as test dep * Disable allocations tests on multi-threaded * Fast InitContext (#1125) * Make InitContext work with OnlyAccsVarInfo * Do not convert NamedTuple to Dict * remove logging * Enable InitFromPrior and InitFromUniform too * Fix `infer_nested_eltype` invocation * Refactor FastLDF to use InitContext * note init breaking change * fix logjac sign * workaround Mooncake segfault * fix changelog too * Fix get_param_eltype for context stacks * Add a test for threaded observe * Export init * Remove dead code * fix transforms for pathological distributions * Tidy up loads of things * fix typed_identity spelling * fix definition order * Improve docstrings * Remove stray comment * export get_param_eltype (unfortunatley) * Add more comment * Update comment * Remove inlines, fix OAVI docstring * Improve docstrings * Simplify InitFromParams constructor * Replace map(identity, x[:]) with [i for i in x[:]] * Simplify implementation for InitContext/OAVI * Add another model to allocation tests Co-authored-by: Markus Hauru <[email protected]> * Revert removal of dist argument (oops) * Format * Update some outdated bits of FastLDF docstring * remove underscores --------- Co-authored-by: Markus Hauru <[email protected]> * implement `LogDensityProblems.dimension` * forgot about capabilities... * use interpolation in run_ad * Improvements to benchmark outputs (#1146) * print output * fix * reenable * add more lines to guide the eye * reorder table * print tgrad / trel as well * forgot this type * Allow generation of `ParamsWithStats` from `FastLDF` plus parameters, and also `bundle_samples` (#1129) * Implement `ParamsWithStats` for `FastLDF` * Add comments * Implement `bundle_samples` for ParamsWithStats -> MCMCChains * Remove redundant comment * don't need Statistics? * Make FastLDF the default (#1139) * Make FastLDF the default * Add miscellaneous LogDensityProblems tests * Use `init!!` instead of `fast_evaluate!!` * Rename files, rebalance tests * Implement `predict`, `returned`, `logjoint`, ... with `OnlyAccsVarInfo` (#1130) * Use OnlyAccsVarInfo for many re-evaluation functions * drop `fast_` prefix * Add a changelog * Improve FastLDF type stability when all parameters are linked or unlinked (#1141) * Improve type stability when all parameters are linked or unlinked * fix a merge conflict * fix enzyme gc crash (locally at least) * Fixes from review * Make threadsafe evaluation opt-in (#1151) * Make threadsafe evaluation opt-in * Reduce number of type parameters in methods * Make `warned_warn_about_threads_threads_threads_threads` shorter * Improve `setthreadsafe` docstring * warn on bare `@threads` as well * fix merge * Fix performance issues * Use maxthreadid() in TSVI * Move convert_eltype code to threadsafe eval function * Point to new Turing docs page * Add a test for setthreadsafe * Tidy up check_model * Apply suggestions from code review Fix outdated docstrings Co-authored-by: Markus Hauru <[email protected]> * Improve warning message * Export `requires_threadsafe` * Add an actual docstring for `requires_threadsafe` --------- Co-authored-by: Markus Hauru <[email protected]> * Standardise `:lp` -> `:logjoint` (#1161) * Standardise `:lp` -> `:logjoint` * changelog * fix a test --------- Co-authored-by: Markus Hauru <[email protected]> Co-authored-by: Markus Hauru <[email protected]>

penelopeysm changed the base branch from main to breaking November 20, 2025 14:40

github-actions bot assigned penelopeysm Nov 20, 2025

penelopeysm force-pushed the py/opt-in-tsvi branch from 54f1b7b to b432bfc Compare November 20, 2025 14:40

penelopeysm commented Nov 20, 2025

View reviewed changes

src/compiler.jl Outdated Show resolved Hide resolved

penelopeysm force-pushed the py/opt-in-tsvi branch from b432bfc to 2369a5d Compare November 20, 2025 14:58

penelopeysm force-pushed the py/opt-in-tsvi branch from 2369a5d to f862acb Compare November 20, 2025 15:10

penelopeysm requested a review from mhauru November 20, 2025 18:46

mhauru reviewed Nov 21, 2025

View reviewed changes

penelopeysm force-pushed the py/opt-in-tsvi branch 4 times, most recently from 0333445 to 27b3c23 Compare November 26, 2025 12:49

penelopeysm added 9 commits November 27, 2025 12:09

Make threadsafe evaluation opt-in

33c3db1

Reduce number of type parameters in methods

cba5d29

Make warned_warn_about_threads_threads_threads_threads shorter

8d1fcaa

Improve setthreadsafe docstring

cb56280

warn on bare @threads as well

eeeb12f

fix merge

0688f11

Fix performance issues

1f1bb01

Use maxthreadid() in TSVI

f5e2e5a

Move convert_eltype code to threadsafe eval function

284dc88

penelopeysm force-pushed the py/opt-in-tsvi branch from 27b3c23 to 284dc88 Compare November 27, 2025 12:09

Point to new Turing docs page

74ea4ac

penelopeysm force-pushed the py/opt-in-tsvi branch from d3bce6e to 74ea4ac Compare November 27, 2025 12:22

penelopeysm added 2 commits November 27, 2025 13:45

Add a test for setthreadsafe

f8c45c2

Tidy up check_model

9d0edc4

penelopeysm mentioned this pull request Nov 27, 2025

Thread safety in check_model is handled poorly #1157

Open

penelopeysm commented Nov 27, 2025

View reviewed changes

penelopeysm requested a review from mhauru November 27, 2025 14:19

mhauru requested changes Nov 28, 2025

View reviewed changes

penelopeysm and others added 3 commits December 1, 2025 10:59

Apply suggestions from code review

8ef3355

Fix outdated docstrings Co-authored-by: Markus Hauru <[email protected]>

Improve warning message

180ced8

Export requires_threadsafe

79b5ca0

penelopeysm commented Dec 1, 2025

View reviewed changes

Add an actual docstring for requires_threadsafe

e839de8

mhauru approved these changes Dec 1, 2025

View reviewed changes

penelopeysm merged commit c27f5e0 into breaking Dec 1, 2025
20 of 21 checks passed

penelopeysm deleted the py/opt-in-tsvi branch December 1, 2025 13:22

penelopeysm mentioned this pull request Dec 1, 2025

PG silently gives incorrect results with Threads.@threads TuringLang/Turing.jl#2658

Closed

penelopeysm mentioned this pull request Dec 3, 2025

Remove TSVI or at least make TSVI opt-in (or or at least let TSVI be opt-out) #1086

Closed

Make threadsafe evaluation opt-in #1151

Make threadsafe evaluation opt-in #1151

Uh oh!

Conversation

penelopeysm commented Nov 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Nov 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Benchmark Report

Computer Information

Benchmark Results

Uh oh!

github-actions bot commented Nov 20, 2025

Uh oh!

codecov bot commented Nov 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

mhauru left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

penelopeysm Nov 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

penelopeysm commented Nov 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mhauru left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mhauru left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

penelopeysm commented Nov 20, 2025 •

edited

Loading

github-actions bot commented Nov 20, 2025 •

edited

Loading

codecov bot commented Nov 20, 2025 •

edited

Loading

penelopeysm Nov 21, 2025 •

edited

Loading

penelopeysm commented Nov 27, 2025 •

edited

Loading