diff --git a/bigcode_eval/tasks/mbppplus.py b/bigcode_eval/tasks/mbppplus.py
index 22fec8291..c9a1dfacf 100644
--- a/bigcode_eval/tasks/mbppplus.py
+++ b/bigcode_eval/tasks/mbppplus.py
@@ -48,7 +48,7 @@ def get_prompt(self, doc):
     #                is different from HumanEval(+) which further requires a `check` func
     def get_reference(self, doc):
         """Builds the reference solution for the doc (sample from the test dataset)."""
-        use_mbpp_tests = os.getenv("MBBPPLUS_USE_MBPP_TESTS", "0")
+        use_mbpp_tests = os.getenv("MBPPPLUS_USE_MBPP_TESTS", "0")
         if use_mbpp_tests == "1":
             return "\n".join(doc["test_list"])
         return "\n" + doc["test"]
diff --git a/docs/README.md b/docs/README.md
index 903c6a122..f3413c354 100644
--- a/docs/README.md
+++ b/docs/README.md
@@ -227,10 +227,10 @@ accelerate launch  main.py \
   --allow_code_execution
 ```
 
-By setting `MBBPPLUS_USE_MBPP_TESTS=1` when running MBPP+, one can run the 399 MBPP+ tasks (a subset of the 500 MBPP evaluation tasks) with the original MBPP base tests:
+By setting `MBPPPLUS_USE_MBPP_TESTS=1` when running MBPP+, one can run the 399 MBPP+ tasks (a subset of the 500 MBPP evaluation tasks) with the original MBPP base tests:
 
 ```bash
-MBBPPLUS_USE_MBPP_TESTS=1 accelerate launch main.py \
+MBPPPLUS_USE_MBPP_TESTS=1 accelerate launch main.py \
   --tasks mbppplus \
   --allow_code_execution \
   --load_generations_path generations_mbppplus.json \