diff --git a/nvshmem4py/nvshmem/core/collective.py b/nvshmem4py/nvshmem/core/collective.py
index f678561..ad2630f 100644
--- a/nvshmem4py/nvshmem/core/collective.py
+++ b/nvshmem4py/nvshmem/core/collective.py
@@ -43,6 +43,13 @@
 # Collectives
 valid_collectives = ["reduce", "reducescatter", "alltoall", "fcollect", "broadcast"]
 
+# Mapping from user-facing dtype aliases to NVSHMEM binding dtype names
+# This allows users to use common shorthand like "fp16" and "bf16"
+dtype_aliases = {
+    "fp16": "half",
+    "bf16": "bfloat16",
+}
+
 # Mapping from Cupy/Torch dtypes to NVSHMEM dtype names
 external_to_nvshmem_dtypes = {
     # --------------------
@@ -178,9 +185,12 @@ def collective_on_buffer(coll: str, team: Teams, dest: Buffer, src: Buffer, dtyp
     else:
         size_elem = max(1, size // (n_pes() * dtype_nbytes(dtype)))
 
+
     func_name = ""
     if dtype:
-        func_name += f"{dtype}_"
+        # Normalize dtype alias (e.g., fp16 -> half, bf16 -> bfloat16)
+        normalized_dtype = dtype_aliases.get(dtype, dtype)
+        func_name += f"{normalized_dtype}_"
     if op:
         func_name += f"{op}_"
     func_name += f"{coll}_on_stream"