diff --git a/Project.toml b/Project.toml
index fc3452d074..5a4dbc8b97 100644
--- a/Project.toml
+++ b/Project.toml
@@ -1,7 +1,7 @@
 name = "Lux"
 uuid = "b2108857-7c20-44ae-9111-449ecde12c47"
 authors = ["Avik Pal <avikpal@mit.edu> and contributors"]
-version = "1.21.1"
+version = "1.21.2"
 
 [deps]
 ADTypes = "47edcb42-4c32-4615-8424-f2b9edc5f35b"
@@ -91,7 +91,7 @@ Enzyme = "0.13.74"
 EnzymeCore = "0.8.13"
 FastClosures = "0.3.2"
 Flux = "0.16.3"
-ForwardDiff = "0.10.36, 1"
+ForwardDiff = "0.10.36, =1"
 FunctionWrappers = "1.1.3"
 Functors = "0.5"
 GPUArraysCore = "0.2"
diff --git a/test/distributed/Project.toml b/test/distributed/Project.toml
index db487a7148..8cc79f4b0c 100644
--- a/test/distributed/Project.toml
+++ b/test/distributed/Project.toml
@@ -14,6 +14,7 @@ Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 Lux = {path = "../.."}
 LuxCUDA = {path = "../../lib/LuxCUDA"}
 MLDataDevices = {path = "../../lib/MLDataDevices"}
+NCCL = { url = "https://github.com/LuxDL/NCCL.jl", rev = "cd85e9f7830be425b64d2eee407df7d61a6a9635" }
 
 [compat]
 ComponentArrays = "0.15.29"
diff --git a/test/distributed/common_distributedtest.jl b/test/distributed/common_distributedtest.jl
index 696405501b..d96c0b59a2 100644
--- a/test/distributed/common_distributedtest.jl
+++ b/test/distributed/common_distributedtest.jl
@@ -11,7 +11,7 @@ end
 
 const backend_type = input_args[2] == "nccl" ? NCCLBackend : MPIBackend
 
-if input_args[1] == "nccl"
+if input_args[2] == "nccl"
     using NCCL
 end
 
diff --git a/test/distributed/data_distributedtest.jl b/test/distributed/data_distributedtest.jl
index ab6a31e71c..0eb36c627e 100644
--- a/test/distributed/data_distributedtest.jl
+++ b/test/distributed/data_distributedtest.jl
@@ -17,7 +17,7 @@ end
 
 const backend_type = input_args[2] == "nccl" ? NCCLBackend : MPIBackend
 
-if input_args[1] == "nccl"
+if input_args[2] == "nccl"
     using NCCL
 end
 
diff --git a/test/distributed/optimizer_distributedtest.jl b/test/distributed/optimizer_distributedtest.jl
index 4b728ec5c9..ef0ab5ee44 100644
--- a/test/distributed/optimizer_distributedtest.jl
+++ b/test/distributed/optimizer_distributedtest.jl
@@ -11,7 +11,7 @@ end
 
 const backend_type = input_args[2] == "nccl" ? NCCLBackend : MPIBackend
 
-if input_args[1] == "nccl"
+if input_args[2] == "nccl"
     using NCCL
 end
 
diff --git a/test/distributed/synchronize_distributedtest.jl b/test/distributed/synchronize_distributedtest.jl
index 5218dd927c..cda7ff3737 100644
--- a/test/distributed/synchronize_distributedtest.jl
+++ b/test/distributed/synchronize_distributedtest.jl
@@ -11,7 +11,7 @@ end
 
 const backend_type = input_args[2] == "nccl" ? NCCLBackend : MPIBackend
 
-if input_args[1] == "nccl"
+if input_args[2] == "nccl"
     using NCCL
 end