TuringLang · yebai · Jun 4, 2026 · Apr 24, 2026 · May 19, 2026 · May 19, 2026
diff --git a/.buildkite/pipeline.yml b/.buildkite/pipeline.yml
@@ -16,3 +16,4 @@ steps:
       setup:
         julia:
            - "1.10"
+           - "1"
diff --git a/.github/workflows/Docs.yml b/.github/workflows/Docs.yml
@@ -25,7 +25,7 @@ jobs:
       - name: Build and deploy Documenter.jl docs
         uses: TuringLang/actions/DocsDocumenter@main
         with:
-          julia-version: 'lts'
+          julia-version: '1'
 
       - name: Run doctests
         shell: julia --project=docs --color=yes {0}

diff --git a/.github/workflows/Format.yml b/.github/workflows/Format.yml
@@ -1,8 +1,11 @@
-name: Format suggestions
+name: Format
 
 on:
+  push:
+    branches:
+      - main
   pull_request:
-  
+
 concurrency:
   # Skip intermediate builds: always.
   # Cancel intermediate builds: only if it is a pull request build.
@@ -12,15 +15,7 @@ concurrency:
 jobs:
   format:
     runs-on: ubuntu-latest
+
     steps:
-      - uses: actions/checkout@v4
-      - uses: julia-actions/setup-julia@v2
-        with:
-          version: 1
-      - run: |
-          julia  -e 'using Pkg; Pkg.add("JuliaFormatter")'
-          julia  -e 'using JuliaFormatter; format("."; verbose=true)'
-      - uses: reviewdog/action-suggester@v1
-        with:
-          tool_name: JuliaFormatter
-          fail_on_error: true
+      - name: Format code
+        uses: TuringLang/actions/Format@main
diff --git a/Project.toml b/Project.toml
@@ -4,10 +4,10 @@ version = "0.7.0"
 
 [deps]
 ADTypes = "47edcb42-4c32-4615-8424-f2b9edc5f35b"
+AbstractPPL = "7a57a42e-76ec-4ea3-a279-07e840d6d9cf"
 Accessors = "7d9f7c33-5ae7-4f3b-8dc6-eff91059b697"
 ChainRulesCore = "d360d2e6-b24c-11e9-a2a3-2a2ae2dbcce4"
 DiffResults = "163ba53b-c6d8-5494-b064-1a9d43ac40c5"
-DifferentiationInterface = "a0c0ee7d-e4b9-4e03-894e-1c5f64a51d63"
 Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
 DocStringExtensions = "ffbed154-4ef7-542d-bbb7-c09d3a79fcae"
 FillArrays = "1a297f60-69ca-5386-bcde-b61e274b549b"
@@ -29,23 +29,23 @@ DynamicPPL = "366bfd00-2699-11ea-058f-f148b4cae6d8"
 AdvancedVIEnzymeExt = ["Enzyme", "ChainRulesCore"]
 AdvancedVIMooncakeExt = ["Mooncake", "ChainRulesCore"]
 AdvancedVIReverseDiffExt = ["ReverseDiff", "ChainRulesCore"]
-AdvancedVIDynamicPPLExt = ["DynamicPPL", "Accessors", "Distributions", "DifferentiationInterface", "LogDensityProblems"]
+AdvancedVIDynamicPPLExt = ["DynamicPPL", "Accessors", "Distributions", "LogDensityProblems"]
 
 [compat]
 ADTypes = "1"
+AbstractPPL = "0.15"
 Accessors = "0.1"
 ChainRulesCore = "1"
 DiffResults = "1"
-DifferentiationInterface = "0.6, 0.7"
 Distributions = "0.25.111"
 DocStringExtensions = "0.8, 0.9"
-DynamicPPL = "0.40, 0.41"
+DynamicPPL = "0.42"
 Enzyme = "0.13"
 FillArrays = "1.3"
 Functors = "0.4, 0.5"
 LinearAlgebra = "1"
 LogDensityProblems = "2"
-Mooncake = "0.4, 0.5"
+Mooncake = "0.5.31"
 Optimisers = "0.2.16, 0.3, 0.4"
 ProgressMeter = "1.6"
 Random = "1"

diff --git a/bench/Project.toml b/bench/Project.toml
@@ -2,6 +2,7 @@
 ADTypes = "47edcb42-4c32-4615-8424-f2b9edc5f35b"
 AdvancedVI = "b5ca4192-6429-45e5-a2d9-87aec30a685c"
 BenchmarkTools = "6e4b80f9-dd63-53aa-95a3-0cdb28fa8baf"
+DifferentiationInterface = "a0c0ee7d-e4b9-4e03-894e-1c5f64a51d63"
 Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
 DistributionsAD = "ced4e74d-a319-5a8a-b0ac-84af2272839c"
 Enzyme = "7da242da-08ed-463a-9acd-ee780be4f1d9"
@@ -15,12 +16,12 @@ Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 ReverseDiff = "37e2e3b7-166d-5795-8a7a-e32c996b4267"
 SimpleUnPack = "ce78b400-467f-4804-87d8-8f486da07d0a"
 StableRNGs = "860ef19b-820b-49d6-a774-d7a799459cd3"
-Zygote = "e88e6eb3-aa80-5325-afca-941959d7151f"
 
 [compat]
 ADTypes = "1"
 AdvancedVI = "0.7, 0.6"
 BenchmarkTools = "1"
+DifferentiationInterface = "0.6, 0.7"
 Distributions = "0.25.111"
 DistributionsAD = "0.6"
 Enzyme = "0.13.7"
@@ -34,5 +35,4 @@ Random = "1"
 ReverseDiff = "1"
 SimpleUnPack = "1"
 StableRNGs = "1"
-Zygote = "0.6, 0.7"
 julia = "1.10, 1.11.2"
diff --git a/bench/benchmarks.jl b/bench/benchmarks.jl
@@ -3,7 +3,8 @@ using AdvancedVI
 using BenchmarkTools
 using Distributions
 using DistributionsAD
-using Enzyme, ForwardDiff, ReverseDiff, Zygote, Mooncake
+using DifferentiationInterface  # provides AbstractPPL.prepare for non-Mooncake, non-ForwardDiff backends
+using Enzyme, ForwardDiff, Mooncake, ReverseDiff
 using FillArrays
 using InteractiveUtils
 using LinearAlgebra
@@ -68,9 +69,8 @@ begin
                 ("RepGradELBO + STL", StickingTheLandingEntropy()),
             ],
             (adname, adtype) in [
-                ("Zygote", AutoZygote()),
+                ("Mooncake", AutoMooncake()),
                 ("ReverseDiff", AutoReverseDiff()),
-                ("Mooncake", AutoMooncake(; config=Mooncake.Config())),
                 # ("Enzyme", AutoEnzyme(; mode=Enzyme.set_runtime_activity(Enzyme.Reverse), function_annotation=Enzyme.Const)),
             ],
             (familyname, family) in [

diff --git a/docs/Project.toml b/docs/Project.toml
@@ -1,48 +1,46 @@
 [deps]
 ADTypes = "47edcb42-4c32-4615-8424-f2b9edc5f35b"
+AbstractPPL = "7a57a42e-76ec-4ea3-a279-07e840d6d9cf"
 Accessors = "7d9f7c33-5ae7-4f3b-8dc6-eff91059b697"
 AdvancedVI = "b5ca4192-6429-45e5-a2d9-87aec30a685c"
 Bijectors = "76274a88-744f-5084-9051-94815aaf08c4"
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
-DifferentiationInterface = "a0c0ee7d-e4b9-4e03-894e-1c5f64a51d63"
 Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
 Documenter = "e30172f5-a6a5-5a46-863b-614d45cd2de4"
 FillArrays = "1a297f60-69ca-5386-bcde-b61e274b549b"
 ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
 Functors = "d9f16b24-f501-4c13-a1f2-28368ffc5196"
 JSON = "682c06a0-de6a-54ab-a142-c8b1cf79cde6"
 LogDensityProblems = "6fdf6af0-433a-55f7-b3ed-c6c6e0b8df7c"
-LogDensityProblemsAD = "996a588d-648d-4e1f-a8f0-a84b347e47b1"
+Mooncake = "da2b9cff-9c12-43a0-ae48-6db2b0edb7d6"
 NormalizingFlows = "50e4474d-9f12-44b7-af7a-91ab30ff6256"
 OpenML = "8b6db2d4-7670-4922-a472-f9537c81ab66"
 Optimisers = "3bd65402-5787-11e9-1adc-39752487f4e2"
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
 QuasiMonteCarlo = "8a4e6c94-4038-4cdc-81c3-7e6ffdb2a71b"
-ReverseDiff = "37e2e3b7-166d-5795-8a7a-e32c996b4267"
 StanLogDensityProblems = "a545de4d-8dba-46db-9d34-4e41d3f07807"
 StatsFuns = "4c63d2b9-4356-54db-8cca-17b64c39e42c"
 
 [compat]
 ADTypes = "1"
+AbstractPPL = "0.15"
 Accessors = "0.1"
 AdvancedVI = "0.7, 0.6"
-Bijectors = "0.13.6, 0.14, 0.15"
+Bijectors = "0.15, 0.16"
 DataFrames = "1"
-DifferentiationInterface = "0.7"
 Distributions = "0.25"
 Documenter = "1"
 FillArrays = "1"
 ForwardDiff = "0.10, 1"
 Functors = "0.5"
 JSON = "0.21, 1"
 LogDensityProblems = "2.1.1"
-LogDensityProblemsAD = "1"
+Mooncake = "0.5.31"
 NormalizingFlows = "0.2.2"
 OpenML = "0.3"
 Optimisers = "0.3, 0.4"
 Plots = "1"
 QuasiMonteCarlo = "0.3"
-ReverseDiff = "1"
 StanLogDensityProblems = "0.1"
 StatsFuns = "1, 2"
 julia = "1.10, 1.11.2"
diff --git a/docs/make.jl b/docs/make.jl
@@ -1,4 +1,9 @@
 
+# Force GR (Plots.jl's default backend) into its no-display workstation so
+# `savefig` writes files but no interactive window is opened. Must be set
+# before GR loads.
+ENV["GKSwstype"] = "100"
+
 using AdvancedVI
 using Documenter
 

diff --git a/docs/src/families.md b/docs/src/families.md
@@ -138,7 +138,7 @@ using LinearAlgebra
 using LogDensityProblems
 using Optimisers
 using Plots
-using ForwardDiff, ReverseDiff
+using ForwardDiff, Mooncake
 
 struct Target{D}
     dist::D
@@ -184,7 +184,7 @@ D     = ones(n_dims)
 U     = zeros(n_dims, 3)
 q0_lr = LowRankGaussian(μ, D, U)
 
-alg = KLMinRepGradDescent(AutoReverseDiff(); optimizer=Adam(0.01), operator=ClipScale())
+alg = KLMinRepGradDescent(AutoMooncake(); optimizer=Adam(0.01), operator=ClipScale())
 
 max_iter = 10^4
 

diff --git a/docs/src/klminrepgraddescent.md b/docs/src/klminrepgraddescent.md
@@ -103,7 +103,7 @@ using Plots
 using Random
 
 using Optimisers
-using ADTypes, ForwardDiff, ReverseDiff
+using ADTypes, ForwardDiff, Mooncake
 using AdvancedVI
 
 struct Dist{D}
@@ -150,10 +150,7 @@ Recall that the original ADVI objective with a closed-form entropy (CFE) is give
 n_montecarlo = 16;
 
 cfe = KLMinRepGradDescent(
-    AutoReverseDiff();
-    entropy=ClosedFormEntropy(),
-    optimizer=Adam(1e-2),
-    operator=ClipScale(),
+    AutoMooncake(); entropy=ClosedFormEntropy(), optimizer=Adam(1e-2), operator=ClipScale()
 )
 nothing
 ```
@@ -162,7 +159,7 @@ The repgradelbo estimator can instead be created as follows:
 
 ```@example repgradelbo
 stl = KLMinRepGradDescent(
-    AutoReverseDiff();
+    AutoMooncake();
     entropy=StickingTheLandingEntropy(),
     optimizer=Adam(1e-2),
     operator=ClipScale(),
@@ -262,6 +259,10 @@ In this case, it suffices to override its `rand` specialization as follows:
 using QuasiMonteCarlo
 using StatsFuns
 
+# QMC samples are inputs to AD, not parameters; declare them non-differentiable
+# so Mooncake doesn't trace through Sobol/Owen bit-twiddling.
+Mooncake.@zero_derivative Mooncake.DefaultCtx Tuple{typeof(QuasiMonteCarlo.sample),Vararg}
+
 qmcrng = SobolSample(; R=OwenScramble(; base=2, pad=32))
 
 function Distributions.rand(
@@ -281,7 +282,7 @@ nothing
 
 ```@setup repgradelbo
 _, info_qmc, _ = AdvancedVI.optimize(
-    KLMinRepGradDescent(AutoReverseDiff(); n_samples=n_montecarlo, optimizer=Adam(1e-2), operator=ClipScale()),
+    KLMinRepGradDescent(AutoMooncake(); n_samples=n_montecarlo, optimizer=Adam(1e-2), operator=ClipScale()),
     max_iter,
     model,
     q0;

diff --git a/docs/src/tutorials/basic.md b/docs/src/tutorials/basic.md
@@ -143,10 +143,10 @@ nothing
 For the VI algorithm, we will use `KLMinRepGradDescent`:
 
 ```@example basic
-using ADTypes, ReverseDiff
+using ADTypes, Mooncake
 using AdvancedVI
 
-alg = KLMinRepGradDescent(ADTypes.AutoReverseDiff(); operator=ClipScale());
+alg = KLMinRepGradDescent(ADTypes.AutoMooncake(); operator=ClipScale());
 nothing
 ```
 
@@ -159,19 +159,7 @@ Location-scale family distributions require the scale matrix to have strictly po
 Here, the projection operator `ClipScale` ensures this.
 
 `KLMinRepGradDescent`, in particular, assumes that the target `LogDensityProblem` is differentiable.
-If the `LogDensityProblem` has a differentiation [capability](https://www.tamaspapp.eu/LogDensityProblems.jl/dev/#LogDensityProblems.capabilities) of at least first-order, we can take advantage of this.
-For this example, we will use `LogDensityProblemsAD` to equip our problem with a first-order capability:
-
-[^TL2014]: Titsias, M., & Lázaro-Gredilla, M. (2014, June). Doubly stochastic variational Bayes for non-conjugate inference. In *International Conference on Machine Learning*. PMLR.
-[^RMW2014]: Rezende, D. J., Mohamed, S., & Wierstra, D. (2014, June). Stochastic backpropagation and approximate inference in deep generative models. In *International Conference on Machine Learning*. PMLR.
-[^KW2014]: Kingma, D. P., & Welling, M. (2014). Auto-encoding variational bayes. In *International Conference on Learning Representations*.
-```@example basic
-using DifferentiationInterface: DifferentiationInterface
-using LogDensityProblemsAD: LogDensityProblemsAD
-
-prob_trans_ad = LogDensityProblemsAD.ADgradient(ADTypes.AutoForwardDiff(), prob_trans)
-nothing
-```
+If the problem provides only a zeroth-order [capability](https://www.tamaspapp.eu/LogDensityProblems.jl/dev/#LogDensityProblems.capabilities), `AdvancedVI` will differentiate through `LogDensityProblems.logdensity` directly using the AD backend supplied to the algorithm.
 
 For the variational family, we will consider a `FullRankGaussian` approximation:
 
@@ -190,7 +178,7 @@ We can now run VI:
 
 ```@example basic
 max_iter = 10^4
-q_out, info, _ = AdvancedVI.optimize(alg, max_iter, prob_trans_ad, q; show_progress=false)
+q_out, info, _ = AdvancedVI.optimize(alg, max_iter, prob_trans, q; show_progress=false)
 nothing
 ```
 
@@ -286,7 +274,7 @@ The `callback` can be supplied to `optimize`:
 ```@example basic
 max_iter = 10^4
 q_out, info, _ = AdvancedVI.optimize(
-    alg, max_iter, prob_trans_ad, q; show_progress=false, callback=callback
+    alg, max_iter, prob_trans, q; show_progress=false, callback=callback
 )
 nothing
 ```
@@ -331,3 +319,7 @@ nothing
 ![](basic_example_acc.svg)
 
 Clearly, the accuracy is improving over time.
+
+[^TL2014]: Titsias, M., & Lázaro-Gredilla, M. (2014, June). Doubly stochastic variational Bayes for non-conjugate inference. In *International Conference on Machine Learning*. PMLR.
+[^RMW2014]: Rezende, D. J., Mohamed, S., & Wierstra, D. (2014, June). Stochastic backpropagation and approximate inference in deep generative models. In *International Conference on Machine Learning*. PMLR.
+[^KW2014]: Kingma, D. P., & Welling, M. (2014). Auto-encoding variational bayes. In *International Conference on Learning Representations*.
-Original file line number
+Diff line change
@@ Expand Up / @@ -16,3 +16,4 @@ steps: @@
           setup:
             julia:
                - "1.10"
+               - "1"