diff --git a/.github/workflows/CI.yml b/.github/workflows/CI.yml
index a54b850a1..a36e61b4f 100644
--- a/.github/workflows/CI.yml
+++ b/.github/workflows/CI.yml
@@ -4,9 +4,11 @@ on:
   push:
     branches:
       - master
+      - backport-*
   pull_request:
     branches:
       - master
+      - backport-*
   merge_group:
     types: [checks_requested]
 
diff --git a/Project.toml b/Project.toml
index 5830de123..8e2cb92e2 100644
--- a/Project.toml
+++ b/Project.toml
@@ -1,6 +1,6 @@
 name = "DynamicPPL"
 uuid = "366bfd00-2699-11ea-058f-f148b4cae6d8"
-version = "0.28.4"
+version = "0.28.5"
 
 [deps]
 ADTypes = "47edcb42-4c32-4615-8424-f2b9edc5f35b"
diff --git a/src/utils.jl b/src/utils.jl
index 4bf652363..4bf5d183b 100644
--- a/src/utils.jl
+++ b/src/utils.jl
@@ -811,9 +811,9 @@ end
 """
     float_type_with_fallback(x)
 
-Return type corresponding to `float(typeof(x))` if possible; otherwise return `Real`.
+Return type corresponding to `float(typeof(x))` if possible; otherwise return `float(Real)`.
 """
-float_type_with_fallback(::Type) = Real
+float_type_with_fallback(::Type) = float(Real)
 float_type_with_fallback(::Type{T}) where {T<:Real} = float(T)
 
 """
diff --git a/test/turing/Project.toml b/test/turing/Project.toml
index 5dab655cb..798421c8a 100644
--- a/test/turing/Project.toml
+++ b/test/turing/Project.toml
@@ -15,5 +15,5 @@ DynamicPPL = "0.24, 0.25, 0.26, 0.27, 0.28"
 HypothesisTests = "0.11"
 MCMCChains = "6"
 ReverseDiff = "1.15"
-Turing = "0.33"
+Turing = "0.34"
 julia = "1.7"
diff --git a/test/turing/varinfo.jl b/test/turing/varinfo.jl
index 30408e598..0c6879186 100644
--- a/test/turing/varinfo.jl
+++ b/test/turing/varinfo.jl
@@ -342,4 +342,19 @@
         model = state_space(y, length(t))
         @test size(sample(model, NUTS(; adtype=AutoReverseDiff(true)), n), 1) == n
     end
+
+    if Threads.nthreads() > 1
+        @testset "DynamicPPL#684: OrderedDict with multiple types when multithreaded" begin
+            @model function f(x)
+                ns ~ filldist(Normal(0, 2.0), 3)
+                m ~ Uniform(0, 1)
+                x ~ Normal(m, 1)
+            end
+            model = f(1)
+            chain = sample(model, NUTS(), MCMCThreads(), 10, 2);
+            loglikelihood(model, chain)
+            logprior(model, chain)
+            logjoint(model, chain)
+        end
+    end
 end