diff --git a/litellm/__pycache__/utils.cpython-311.pyc b/litellm/__pycache__/utils.cpython-311.pyc
index bf7b112ca4..6bebd64510 100644
Binary files a/litellm/__pycache__/utils.cpython-311.pyc and b/litellm/__pycache__/utils.cpython-311.pyc differ
diff --git a/litellm/llms/ai21.py b/litellm/llms/ai21.py
index 6a22b99e89..882e217e6b 100644
--- a/litellm/llms/ai21.py
+++ b/litellm/llms/ai21.py
@@ -102,7 +102,7 @@ class AI21LLM:
                 try:
                     model_response["choices"][0]["message"]["content"] = completion_response["completions"][0]["data"]["text"]
                 except:
-                    raise ValueError(f"Unable to parse response. Original response: {response.text}")
+                    raise AI21Error(message=json.dumps(completion_response), status_code=response.status_code)
 
             ## CALCULATING USAGE - baseten charges on time, not tokens - have some mapping of cost here. 
             prompt_tokens = len(
diff --git a/litellm/tests/test_exceptions.py b/litellm/tests/test_exceptions.py
index 396cdc0790..263d4e0d2a 100644
--- a/litellm/tests/test_exceptions.py
+++ b/litellm/tests/test_exceptions.py
@@ -33,8 +33,8 @@ litellm.failure_callback = ["sentry"]
 # Approach: Run each model through the test -> assert if the correct error (always the same one) is triggered
 
 # models = ["gpt-3.5-turbo", "chatgpt-test",  "claude-instant-1", "command-nightly"]
-test_model = "togethercomputer/CodeLlama-34b-Python"
-models = ["togethercomputer/CodeLlama-34b-Python"]
+test_model = "j2-light"
+models = ["j2-light"]
 
 
 def logging_fn(model_call_dict):
@@ -98,6 +98,9 @@ def invalid_auth(model):  # set the model key to an invalid key, depending on th
         elif model == "command-nightly":
             temporary_key = os.environ["COHERE_API_KEY"]
             os.environ["COHERE_API_KEY"] = "bad-key"
+        elif "j2" in model:
+            temporary_key = os.environ["AI21_API_KEY"]
+            os.environ["AI21_API_KEY"] = "bad-key"
         elif "togethercomputer" in model:
             temporary_key = os.environ["TOGETHERAI_API_KEY"]
             os.environ["TOGETHERAI_API_KEY"] = "84060c79880fc49df126d3e87b53f8a463ff6e1c6d27fe64207cde25cdfcd1f24a"
@@ -138,6 +141,8 @@ def invalid_auth(model):  # set the model key to an invalid key, depending on th
             == "replicate/llama-2-70b-chat:2c1608e18606fad2812020dc541930f2d0495ce32eee50074220b87300bc16e1"
         ):
             os.environ["REPLICATE_API_KEY"] = temporary_key
+        elif "j2" in model:
+            os.environ["AI21_API_KEY"] = temporary_key
         elif ("togethercomputer" in model):
             os.environ["TOGETHERAI_API_KEY"] = temporary_key
     return
diff --git a/litellm/utils.py b/litellm/utils.py
index 994b4916c9..18c32d58b7 100644
--- a/litellm/utils.py
+++ b/litellm/utils.py
@@ -1446,6 +1446,37 @@ def exception_type(model, original_exception, custom_llm_provider):
                             message=f"HuggingfaceException - {original_exception.message}",
                             llm_provider="huggingface",
                         )
+            elif custom_llm_provider == "ai21":
+                print(f"e: {original_exception}")
+                if hasattr(original_exception, "message"):
+                    if "Prompt has too many tokens" in original_exception.message:
+                        exception_mapping_worked = True
+                        raise ContextWindowExceededError(
+                            message=f"AI21Exception - {original_exception.message}",
+                            model=model,
+                            llm_provider="ai21"
+                        )
+                if hasattr(original_exception, "status_code"):
+                    print(f"status code: {original_exception.status_code}")
+                    if original_exception.status_code == 401:
+                        exception_mapping_worked = True
+                        raise AuthenticationError(
+                            message=f"AI21Exception - {original_exception.message}",
+                            llm_provider="ai21",
+                        )
+                    if original_exception.status_code == 422 or "Prompt has too many tokens" in original_exception.message:
+                        exception_mapping_worked = True
+                        raise InvalidRequestError(
+                            message=f"AI21Exception - {original_exception.message}",
+                            model=model,
+                            llm_provider="ai21",
+                        )
+                    elif original_exception.status_code == 429:
+                        exception_mapping_worked = True
+                        raise RateLimitError(
+                            message=f"AI21Exception - {original_exception.message}",
+                            llm_provider="ai21",
+                        )
             elif custom_llm_provider == "together_ai":
                 error_response = json.loads(error_str)
                 if "error" in error_response and "`inputs` tokens + `max_new_tokens` must be <=" in error_response["error"]: