diff --git a/Cargo.toml b/Cargo.toml
index 4f5bf2d9..e068382c 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -19,7 +19,7 @@ byteorder = "1.4.3"
 clap = { version = "4.2.4", features = ["derive"] }
 # Re-enable this once 0.9.13 as been released as it would include the cublas-f16 changes
 # cudarc = { version = "0.9.13", optional = true, features = ["f16"] }
-cudarc = { git = "https://github.com/coreylowman/cudarc.git", features = ["f16", "nccl"] }
+cudarc = { git = "https://github.com/coreylowman/cudarc.git", features = ["f16"] }
 # TODO: Switch back to the official gemm implementation once the following are available.
 # https://github.com/sarah-ek/gemm/pull/8.
 # https://github.com/sarah-ek/gemm/pull/9.
diff --git a/candle-examples/Cargo.toml b/candle-examples/Cargo.toml
index 2ecc8500..a595e74d 100644
--- a/candle-examples/Cargo.toml
+++ b/candle-examples/Cargo.toml
@@ -42,7 +42,7 @@ default = []
 cuda = ["candle/cuda", "candle-nn/cuda", "candle-transformers/cuda"]
 flash-attn = ["cuda", "dep:candle-flash-attn"]
 mkl = ["dep:intel-mkl-src", "candle/mkl", "candle-nn/mkl", "candle-transformers/mkl"]
-nccl = ["dep:cudarc", "dep:half"]
+nccl = ["cuda", "cudarc/nccl", "dep:half"]
 
 [[example]]
 name = "llama_multiprocess"