aiverify-moonshot 0.4.5__py3-none-any.whl → 0.4.6__py3-none-any.whl

{aiverify_moonshot-0.4.5.dist-info → aiverify_moonshot-0.4.6.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: aiverify-moonshot
-Version: 0.4.5
+Version: 0.4.6
 Summary: AI Verify advances Gen AI testing with Project Moonshot.
 Project-URL: Repository, https://github.com/aiverify-foundation/moonshot
 Project-URL: Documentation, https://aiverify-foundation.github.io/moonshot/
@@ -18,6 +18,7 @@ Requires-Python: >=3.11
 Requires-Dist: datasets==2.20.0
 Requires-Dist: ijson==3.3.0
 Requires-Dist: jinja2==3.1.4
+Requires-Dist: numpy==1.26.4
 Requires-Dist: pandas==2.2.2
 Requires-Dist: pydantic==2.8.2
 Requires-Dist: pyparsing==3.1.2
@@ -45,7 +46,7 @@ Description-Content-Type: text/markdown
 
 ![Moonshot Logo](https://github.com/aiverify-foundation/moonshot/raw/main/misc/aiverify-moonshot-logo.png)
 
-**Version 0.4.5**
+**Version 0.4.6**
 
 A simple and modular tool to evaluate any LLM application.
 

{aiverify_moonshot-0.4.5.dist-info → aiverify_moonshot-0.4.6.dist-info}/RECORD

@@ -6,21 +6,22 @@ moonshot/integrations/cli/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJW
 moonshot/integrations/cli/__main__.py,sha256=0VnYSj2AayvDCZ3uXpldPcjMHt2Yd7BWojWzFOGSSl4,679
 moonshot/integrations/cli/active_session_cfg.py,sha256=n8hOFxFjvz26qbEFY4q7iPUZYrGLoeCmXJxmOb_xWUE,20
 moonshot/integrations/cli/cli.py,sha256=9tnzcxcSOjblxCUpyh3pK0ke0bLs3s-63OxXtYoZI2g,2769
+moonshot/integrations/cli/cli_errors.py,sha256=BhA7k7leuPe3ms2W25iip_yqjrhWKmTaZ2iz1OSjBvE,22981
 moonshot/integrations/cli/benchmark/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 moonshot/integrations/cli/benchmark/benchmark.py,sha256=QUxr6DU11-XeH6Y3j1uPsZsotshgy64G_cWNf0Rn2_U,6303
-moonshot/integrations/cli/benchmark/cookbook.py,sha256=9mNuqsaVFCfXnLgVxJQ-fzwSgH8j082NR2gE-GPQbG8,22183
-moonshot/integrations/cli/benchmark/datasets.py,sha256=E-kreuXkrNU39sFuVWGhutFHfoEhsVjFK4tmq7YQwxs,7193
-moonshot/integrations/cli/benchmark/metrics.py,sha256=6wLn3WyqgUJKCRvblxiyx8X4nQ7gA1sKGcC_DNBIWBo,6351
-moonshot/integrations/cli/benchmark/recipe.py,sha256=EYv94Qjq2hqn1zz9NzWaINEY9DNdY7TkSS57xHv6ymQ,22698
-moonshot/integrations/cli/benchmark/result.py,sha256=U0SsjJYNQJYP_EZ3pAil4VaOQLMZHchfzq982nXU-jE,9486
-moonshot/integrations/cli/benchmark/run.py,sha256=Rqdb_FspgQQhkLORjgdiyaijk7HTxM39giOzXpBXA1k,6161
-moonshot/integrations/cli/benchmark/runner.py,sha256=nAnNKihQD0AScl6vPFiFH_9MZU4ppeMG34QaLRidqvU,6602
+moonshot/integrations/cli/benchmark/cookbook.py,sha256=h9exPhLKLRa9a1qWB2WptZ35gcVov332tjHDO9ECs0o,29639
+moonshot/integrations/cli/benchmark/datasets.py,sha256=Uq5XMNWUp775sz9jCZUZHHmkumPFI7cHVRueHgWm70Q,8965
+moonshot/integrations/cli/benchmark/metrics.py,sha256=ATZtVOj3kjGuyxOgfAO1QzTDeAOEjbSeQYr3NOCmTCQ,8070
+moonshot/integrations/cli/benchmark/recipe.py,sha256=yVFX3pwNzDL0a95rjlQd4cek06M9blv5jIeYB26jOKk,32481
+moonshot/integrations/cli/benchmark/result.py,sha256=TGlIuDkPB1Cfu1u2uyDsWtgQxTOoNdGDzc0vCnGC1zk,11152
+moonshot/integrations/cli/benchmark/run.py,sha256=HBztvG_Zkg1ZAWsFv0QDE43FaEmx92vTWc4h1U3VesU,7438
+moonshot/integrations/cli/benchmark/runner.py,sha256=Y4Vt6Qqn9QzsM6eLUM9m2_XKkW3ctu-2jMTSei_TDPU,7098
 moonshot/integrations/cli/common/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 moonshot/integrations/cli/common/common.py,sha256=hSvxj-T91nhKfJtQqxrIPgYdS8P_g7wbOwpFpqBg90I,3090
-moonshot/integrations/cli/common/connectors.py,sha256=8_U7KJzfsS9L13dMbouk5JKOy_ow4UhrILICCw27Q5c,13909
-moonshot/integrations/cli/common/dataset.py,sha256=KZuwSV4Di7C378Aa2bRxIKCd03oo6HdMTEZpBZSQb5w,3113
+moonshot/integrations/cli/common/connectors.py,sha256=0KoCji-a6AJ0GLqpDdgm4mcfj-FzNW2DOsuQ0Fi_gwc,19445
+moonshot/integrations/cli/common/dataset.py,sha256=6v5qXgWBGhT4Rl5uuqPMwLeFF0c5m7ECzyGe1IFXNW0,4743
 moonshot/integrations/cli/common/display_helper.py,sha256=8rVowW33XK0j0C_X_H1jUbFlFk1Y2WpzxmIUE3Ca5Co,1459
-moonshot/integrations/cli/common/prompt_template.py,sha256=KzawF2VaiWUKzRIrXX_TaHboiPVllSiqz5I_o2TqGiA,4812
+moonshot/integrations/cli/common/prompt_template.py,sha256=2cXWeHh16EUvzYFre7juMxT2R0x3kqDJVucH0bKLmCU,6351
 moonshot/integrations/cli/initialisation/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 moonshot/integrations/cli/initialisation/initialisation.py,sha256=zMjklhoBlYLPshe7Q16vBBbOfhbPjZc4_ooywAFp2XY,387
 moonshot/integrations/cli/redteam/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -28,12 +29,12 @@ moonshot/integrations/cli/redteam/attack_module.py,sha256=sjjQdBno7rXNfrXqns55ER
 moonshot/integrations/cli/redteam/context_strategy.py,sha256=gLNgnd1oHWD1X93BQoKlR7G8b5g9P6AcKFLlE3WvETk,7065
 moonshot/integrations/cli/redteam/prompt_template.py,sha256=rWINvMFLax8ynIFJoX93xsB1cWJ8-dQ1nRyqgdri2yQ,2079
 moonshot/integrations/cli/redteam/redteam.py,sha256=144tVuiSWBLseC0EvvIuevHYJhqn39xvg36-1cLgGWQ,4257
-moonshot/integrations/cli/redteam/session.py,sha256=auZ73LfMAGmT1jLJUjqY6oAsMWoSzHstiRGA8DKF6A4,33652
+moonshot/integrations/cli/redteam/session.py,sha256=lucUS8Ejcg9PQKVhVj1VUZECe2pvFwbwoSbsVl0BazI,37672
 moonshot/integrations/cli/utils/process_data.py,sha256=QVL5vp2_8ZgGicmCAdeYEHkeb6f-NC775-JCzWziNiU,1901
 moonshot/integrations/web_api/.env.dev,sha256=0z5_Ut8rF-UqFZtgjkH2qoqORhD5_nSs2w_OeX2SteI,182
 moonshot/integrations/web_api/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 moonshot/integrations/web_api/__main__.py,sha256=MdnLi_ZF-olAAEJwTPU1iGYFYwo-fNWNT2qfchkH3y4,2050
-moonshot/integrations/web_api/app.py,sha256=4EzGEwvFh6any62ZJNGH2GCXUOPjqNYURL0Toe2rTNI,3651
+moonshot/integrations/web_api/app.py,sha256=x9QTqkHy4zfnQTA0UJVFHht0vSuDs1BZphjLfhFjsVY,3651
 moonshot/integrations/web_api/container.py,sha256=DVkJG_qm7ItcG6tgMYOqIj07wpKhPWOOfy6-bEv72y4,5915
 moonshot/integrations/web_api/logging_conf.py,sha256=t3EGRV6tZhV732KXe8_Tiy0fiwVAWxZX5Tt8VTgrrfg,3388
 moonshot/integrations/web_api/log/.gitkeep,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -95,8 +96,9 @@ moonshot/integrations/web_api/status_updater/interface/redteam_progress_callback
 moonshot/integrations/web_api/temp/.gitkeep,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 moonshot/integrations/web_api/types/types.py,sha256=AN0Xf61lx2c5AFAYoXA8mVL5iufVBpwYlIPdo8gv-ls,2395
 moonshot/src/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+moonshot/src/messages_constants.py,sha256=pFhm7c2UkdX-XPr62bj6mArf5xrRmktuh_ZiUICJHd0,2352
 moonshot/src/api/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-moonshot/src/api/api_bookmark.py,sha256=Nr0B3S6UUnNbuvXyL-dz8IMyFxCywU1u4UL2D8TjHik,3004
+moonshot/src/api/api_bookmark.py,sha256=FhtSojdw8wfzJXQnH9VnkwsuqRf_c_y6LH0oOQA48Dc,2926
 moonshot/src/api/api_connector.py,sha256=JpwLFd7Vh1LRz6oB8fhO2iufMbSTt5RmrMe7nt56bPM,2257
 moonshot/src/api/api_connector_endpoint.py,sha256=UUuHzrgoj0bWcbOXPUJh7FEg-pzBOSo6DLgwqCZnLmw,5503
 moonshot/src/api/api_context_strategy.py,sha256=WqzUEw3f7gROJlc_G7bsGkuP3nO1HycFzn3dfcxzaoc,2153
@@ -111,8 +113,8 @@ moonshot/src/api/api_result.py,sha256=M5zKF7ytKp237UZusLSYJ7QVfui85Ys0WEaYySGcAK
 moonshot/src/api/api_run.py,sha256=3PrETAVcFnJ09R0-xhWiFkEfqL6eYj4B2voEGJDPznU,2936
 moonshot/src/api/api_runner.py,sha256=cH0rxWREjc2qKmt4Tuwr-fEMrYDBE_TKRw0jOohNEgU,4179
 moonshot/src/api/api_session.py,sha256=OGH05ZxAwo_hKI-RNaJ-jCp_v-zcTm-9bHUclpq2z4Q,10978
-moonshot/src/bookmark/bookmark.py,sha256=KZoKOyJseW02IS5KqfCIMusFiNwESBxOAvOZTwO7YUw,8867
-moonshot/src/bookmark/bookmark_arguments.py,sha256=rwgUPMXmDLNaHfJvRTR4GaZ8vwRilchzOg9bZJQoe_8,1130
+moonshot/src/bookmark/bookmark.py,sha256=5TJ-q7mGpULqaH5osdB2kThP4bKMIO6nWqrXhvjAy4U,11270
+moonshot/src/bookmark/bookmark_arguments.py,sha256=cB5m2zB8255WVdacmC2-ZYNyaoK4-gOM_Qwb_JDR-34,1449
 moonshot/src/configs/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 moonshot/src/configs/env_variables.py,sha256=eF__UJN37LCzIB4pv_T7G-kQHlOa657QA7IpL1d_0MM,7150
 moonshot/src/connectors/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -151,7 +153,7 @@ moonshot/src/results/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hS
 moonshot/src/results/result.py,sha256=o56SdhYH-XVfpeeKhN495dJPkU035MmTjRUx48q53lo,4527
 moonshot/src/results/result_arguments.py,sha256=mTR7yajY72PFglfAaa1ajJfvYNV4IBGLXS4VaD53-8c,1334
 moonshot/src/runners/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-moonshot/src/runners/runner.py,sha256=Wsvdzcw4KOa-ppgGVRGSK2YxLfXK-daHDw3k5jETv-4,21197
+moonshot/src/runners/runner.py,sha256=nnYiMHZUSu35jzHoeiEqADhc7iHLKnViyUbd_Qg8WZs,21203
 moonshot/src/runners/runner_arguments.py,sha256=Bg4OPSmgr9jZKNAwPH0T3epEHw-6qGrflszFc6oMyEU,1640
 moonshot/src/runners/runner_type.py,sha256=jOfnAnaCYp-rPTRJXhM8hin_dinlR0sMwmimQXvLcJ0,100
 moonshot/src/runs/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -169,9 +171,9 @@ moonshot/src/utils/import_modules.py,sha256=T9zTN59PFnvY2rjyWhSV9KSIAHxWV1pyBemF
 moonshot/src/utils/log.py,sha256=YNgD7Eh2OT36XlmVBKCGUTAh9TRp4Akfe4kDdvHASgs,2502
 moonshot/src/utils/pagination.py,sha256=5seymyRoqyENIhKllAatr1T91kMCGFslcvRnJHyMSvc,814
 moonshot/src/utils/timeit.py,sha256=TvuF0w8KWhp0oZFY0cUU3UY0xlGKjchb0OkfYfgVTlc,866
-aiverify_moonshot-0.4.5.dist-info/METADATA,sha256=d8sIDGy6Ywuov6QnMj62fa6Qu5Uvbekv4_rBGeLObWM,12368
-aiverify_moonshot-0.4.5.dist-info/WHEEL,sha256=1yFddiXMmvYK7QYTqtRNtX66WJ0Mz8PYEiEUoOUUxRY,87
-aiverify_moonshot-0.4.5.dist-info/licenses/AUTHORS.md,sha256=mmAbe3i3sT8JZHJMBhxp3i1xRehV0g7WB4T_eyIBuBs,59
-aiverify_moonshot-0.4.5.dist-info/licenses/LICENSE.md,sha256=mDOKOkWFbJmUORaAchXByEVGC1jw37QRn-zS14wY_wM,11347
-aiverify_moonshot-0.4.5.dist-info/licenses/NOTICES.md,sha256=0Ikx6IBGGQEOJeNb2MkRoXxTXwrtlMz6EDgLBFIz6v0,179593
-aiverify_moonshot-0.4.5.dist-info/RECORD,,
+aiverify_moonshot-0.4.6.dist-info/METADATA,sha256=PM15lwqxY3nBr-fbxQp44E_wMZyxKBOCq0r_hYiSFTI,12397
+aiverify_moonshot-0.4.6.dist-info/WHEEL,sha256=1yFddiXMmvYK7QYTqtRNtX66WJ0Mz8PYEiEUoOUUxRY,87
+aiverify_moonshot-0.4.6.dist-info/licenses/AUTHORS.md,sha256=mmAbe3i3sT8JZHJMBhxp3i1xRehV0g7WB4T_eyIBuBs,59
+aiverify_moonshot-0.4.6.dist-info/licenses/LICENSE.md,sha256=mDOKOkWFbJmUORaAchXByEVGC1jw37QRn-zS14wY_wM,11347
+aiverify_moonshot-0.4.6.dist-info/licenses/NOTICES.md,sha256=0Ikx6IBGGQEOJeNb2MkRoXxTXwrtlMz6EDgLBFIz6v0,179593
+aiverify_moonshot-0.4.6.dist-info/RECORD,,

moonshot/integrations/cli/benchmark/cookbook.py

@@ -19,8 +19,33 @@ from moonshot.api import (
     api_update_cookbook,
 )
 from moonshot.integrations.cli.benchmark.recipe import (
-    display_view_grading_scale_format,
-    display_view_statistics_format,
+    _display_view_grading_scale_format,
+    _display_view_statistics_format,
+)
+from moonshot.integrations.cli.cli_errors import (
+    ERROR_BENCHMARK_ADD_COOKBOOK_DESC_VALIDATION,
+    ERROR_BENCHMARK_ADD_COOKBOOK_NAME_VALIDATION,
+    ERROR_BENCHMARK_ADD_COOKBOOK_RECIPES_LIST_STR_VALIDATION,
+    ERROR_BENCHMARK_ADD_COOKBOOK_RECIPES_VALIDATION,
+    ERROR_BENCHMARK_DELETE_COOKBOOK_COOKBOOK_VALIDATION,
+    ERROR_BENCHMARK_LIST_COOKBOOK_FIND_VALIDATION,
+    ERROR_BENCHMARK_LIST_COOKBOOK_PAGINATION_VALIDATION,
+    ERROR_BENCHMARK_LIST_COOKBOOK_PAGINATION_VALIDATION_1,
+    ERROR_BENCHMARK_RUN_COOKBOOK_COOKBOOKS_VALIDATION,
+    ERROR_BENCHMARK_RUN_COOKBOOK_COOKBOOKS_VALIDATION_1,
+    ERROR_BENCHMARK_RUN_COOKBOOK_ENDPOINTS_VALIDATION,
+    ERROR_BENCHMARK_RUN_COOKBOOK_ENDPOINTS_VALIDATION_1,
+    ERROR_BENCHMARK_RUN_COOKBOOK_NAME_VALIDATION,
+    ERROR_BENCHMARK_RUN_COOKBOOK_NO_RESULT,
+    ERROR_BENCHMARK_RUN_COOKBOOK_NUM_OF_PROMPTS_VALIDATION,
+    ERROR_BENCHMARK_RUN_COOKBOOK_RANDOM_SEED_VALIDATION,
+    ERROR_BENCHMARK_RUN_COOKBOOK_RESULT_PROC_MOD_VALIDATION,
+    ERROR_BENCHMARK_RUN_COOKBOOK_RUNNER_PROC_MOD_VALIDATION,
+    ERROR_BENCHMARK_RUN_COOKBOOK_SYS_PROMPT_VALIDATION,
+    ERROR_BENCHMARK_UPDATE_COOKBOOK_COOKBOOK_VALIDATION,
+    ERROR_BENCHMARK_UPDATE_COOKBOOK_UPDATE_VALUES_VALIDATION,
+    ERROR_BENCHMARK_UPDATE_COOKBOOK_UPDATE_VALUES_VALIDATION_1,
+    ERROR_BENCHMARK_VIEW_COOKBOOK_COOKBOOK_VALIDATION,
 )
 from moonshot.integrations.cli.common.display_helper import display_view_list_format
 from moonshot.integrations.cli.utils.process_data import filter_data
@@ -45,11 +70,38 @@ def add_cookbook(args) -> None:
             description (str): The description of the cookbook.
             recipes (str): A string representation of a list of recipes. Each recipe is represented by its ID.
 
+    Raises:
+        TypeError: If the 'name', 'description', or 'recipes' arguments are not strings or are None.
+        ValueError: If the 'recipes' argument is not a list after evaluation.
+
     Returns:
         None
     """
     try:
+        if not isinstance(args.name, str) or not args.name or args.name is None:
+            raise TypeError(ERROR_BENCHMARK_ADD_COOKBOOK_NAME_VALIDATION)
+
+        if (
+            not isinstance(args.description, str)
+            or not args.description
+            or args.description is None
+        ):
+            raise TypeError(ERROR_BENCHMARK_ADD_COOKBOOK_DESC_VALIDATION)
+
+        if (
+            not isinstance(args.recipes, str)
+            or not args.recipes
+            or args.recipes is None
+        ):
+            raise TypeError(ERROR_BENCHMARK_ADD_COOKBOOK_RECIPES_VALIDATION)
+
         recipes = literal_eval(args.recipes)
+        if not (
+            isinstance(recipes, list)
+            and all(isinstance(recipe, str) for recipe in recipes)
+        ):
+            raise ValueError(ERROR_BENCHMARK_ADD_COOKBOOK_RECIPES_LIST_STR_VALIDATION)
+
         new_cookbook_id = api_create_cookbook(args.name, args.description, recipes)
         print(f"[add_cookbook]: Cookbook ({new_cookbook_id}) created.")
     except Exception as e:
@@ -61,22 +113,48 @@ def list_cookbooks(args) -> list | None:
     List all available cookbooks.
 
     This function retrieves all available cookbooks by calling the api_get_all_cookbook function from the
-    moonshot.api module.
-    It then displays the retrieved cookbooks using the _display_cookbooks function.
+    moonshot.api module. It then filters the retrieved cookbooks based on the provided 'find' keyword and
+    'pagination' parameters, and displays the filtered cookbooks using the _display_cookbooks function.
 
     Args:
-        args: A namespace object from argparse. It should have an optional attribute:
-        find (str): Optional field to find cookbook(s) with a keyword.
-        pagination (str): Optional field to paginate cookbooks.
+        args: A namespace object from argparse. It should have the following optional attributes:
+            find (str): Optional field to find cookbook(s) with a keyword.
+            pagination (str): Optional field to paginate cookbooks. It should be a string representation of a tuple
+                              containing two integers (page number and page size).
+
+    Raises:
+        TypeError: If the 'find' or 'pagination' arguments are not strings or are None.
+        ValueError: If the 'pagination' argument is not a tuple of two integers after evaluation.
 
     Returns:
-        list | None: A list of Cookbook or None if there is no result.
+        list | None: A list of filtered cookbooks or None if there is no result.
     """
 
     try:
+        if args.find is not None:
+            if not isinstance(args.find, str) or not args.find:
+                raise TypeError(ERROR_BENCHMARK_LIST_COOKBOOK_FIND_VALIDATION)
+
+        if args.pagination is not None:
+            if not isinstance(args.pagination, str) or not args.pagination:
+                raise TypeError(ERROR_BENCHMARK_LIST_COOKBOOK_PAGINATION_VALIDATION)
+            try:
+                pagination = literal_eval(args.pagination)
+                if not (
+                    isinstance(pagination, tuple)
+                    and len(pagination) == 2
+                    and all(isinstance(i, int) for i in pagination)
+                ):
+                    raise ValueError(
+                        ERROR_BENCHMARK_LIST_COOKBOOK_PAGINATION_VALIDATION_1
+                    )
+            except (ValueError, SyntaxError):
+                raise ValueError(ERROR_BENCHMARK_LIST_COOKBOOK_PAGINATION_VALIDATION_1)
+        else:
+            pagination = ()
+
         cookbooks_list = api_get_all_cookbook()
         keyword = args.find.lower() if args.find else ""
-        pagination = literal_eval(args.pagination) if args.pagination else ()
 
         if cookbooks_list:
             filtered_cookbooks_list = filter_data(cookbooks_list, keyword, pagination)
@@ -89,6 +167,7 @@ def list_cookbooks(args) -> list | None:
 
     except Exception as e:
         print(f"[list_cookbooks]: {str(e)}")
+        return None
 
 
 def view_cookbook(args) -> None:
@@ -96,19 +175,30 @@ def view_cookbook(args) -> None:
     View a specific cookbook.
 
     This function retrieves a specific cookbook by calling the api_read_cookbook function from the
-    moonshot.api module using the cookbook name provided in the args.
+    moonshot.api module using the cookbook ID provided in the args.
     It then displays the retrieved cookbook using the display_view_cookbook function.
 
     Args:
         args: A namespace object from argparse. It should have the following attribute:
-            cookbook (str): The id of the cookbook to view.
+            cookbook (str): The ID of the cookbook to view.
+
+    Raises:
+        TypeError: If the 'cookbook' argument is not a string or is None.
 
     Returns:
         None
     """
     try:
+        if (
+            not isinstance(args.cookbook, str)
+            or not args.cookbook
+            or args.cookbook is None
+        ):
+            raise TypeError(ERROR_BENCHMARK_VIEW_COOKBOOK_COOKBOOK_VALIDATION)
+
         cookbook_info = api_read_cookbook(args.cookbook)
-        display_view_cookbook(cookbook_info)
+        _display_view_cookbook(cookbook_info)
+
     except Exception as e:
         print(f"[view_cookbook]: {str(e)}")
 
@@ -132,48 +222,105 @@ def run_cookbook(args) -> None:
             runner_proc_module (str): The runner processing module to use.
             result_proc_module (str): The result processing module to use.
 
+    Raises:
+        TypeError: If any of the required arguments are not of the expected type or are None.
+        ValueError: If the 'cookbooks' or 'endpoints' arguments are not lists of strings after evaluation.
+        RuntimeError: If no results are found after running the cookbooks.
+
     Returns:
         None
     """
     try:
-        name = args.name
+        if not isinstance(args.name, str) or not args.name or args.name is None:
+            raise TypeError(ERROR_BENCHMARK_RUN_COOKBOOK_NAME_VALIDATION)
+
+        if (
+            not isinstance(args.cookbooks, str)
+            or not args.cookbooks
+            or args.cookbooks is None
+        ):
+            raise TypeError(ERROR_BENCHMARK_RUN_COOKBOOK_COOKBOOKS_VALIDATION)
+
+        if (
+            not isinstance(args.endpoints, str)
+            or not args.endpoints
+            or args.endpoints is None
+        ):
+            raise TypeError(ERROR_BENCHMARK_RUN_COOKBOOK_ENDPOINTS_VALIDATION)
+
+        if isinstance(args.num_of_prompts, bool) or not isinstance(
+            args.num_of_prompts, int
+        ):
+            raise TypeError(ERROR_BENCHMARK_RUN_COOKBOOK_NUM_OF_PROMPTS_VALIDATION)
+
+        if isinstance(args.random_seed, bool) or not isinstance(args.random_seed, int):
+            raise TypeError(ERROR_BENCHMARK_RUN_COOKBOOK_RANDOM_SEED_VALIDATION)
+
+        if (
+            not isinstance(args.system_prompt, str)
+            or not args.system_prompt
+            or args.system_prompt is None
+        ):
+            raise TypeError(ERROR_BENCHMARK_RUN_COOKBOOK_SYS_PROMPT_VALIDATION)
+
+        if (
+            not isinstance(args.runner_proc_module, str)
+            or not args.runner_proc_module
+            or args.runner_proc_module is None
+        ):
+            raise TypeError(ERROR_BENCHMARK_RUN_COOKBOOK_RUNNER_PROC_MOD_VALIDATION)
+
+        if (
+            not isinstance(args.result_proc_module, str)
+            or not args.result_proc_module
+            or args.result_proc_module is None
+        ):
+            raise TypeError(ERROR_BENCHMARK_RUN_COOKBOOK_RESULT_PROC_MOD_VALIDATION)
+
         cookbooks = literal_eval(args.cookbooks)
+        if not (
+            isinstance(cookbooks, list)
+            and all(isinstance(item, str) for item in cookbooks)
+        ):
+            raise TypeError(ERROR_BENCHMARK_RUN_COOKBOOK_COOKBOOKS_VALIDATION_1)
+
         endpoints = literal_eval(args.endpoints)
-        num_of_prompts = args.num_of_prompts
-        random_seed = args.random_seed
-        system_prompt = args.system_prompt
-        runner_proc_module = args.runner_proc_module
-        result_proc_module = args.result_proc_module
+        if not (
+            isinstance(endpoints, list)
+            and all(isinstance(item, str) for item in endpoints)
+        ):
+            raise TypeError(ERROR_BENCHMARK_RUN_COOKBOOK_ENDPOINTS_VALIDATION_1)
 
         # Run the cookbooks with the defined endpoints
-        slugify_id = slugify(name, lowercase=True)
+        slugify_id = slugify(args.name, lowercase=True)
         if slugify_id in api_get_all_runner_name():
             cb_runner = api_load_runner(slugify_id)
         else:
-            cb_runner = api_create_runner(name, endpoints)
+            cb_runner = api_create_runner(args.name, endpoints)
 
-        loop = asyncio.get_event_loop()
-        loop.run_until_complete(
-            cb_runner.run_cookbooks(
+        async def run():
+            await cb_runner.run_cookbooks(
                 cookbooks,
-                num_of_prompts,
-                random_seed,
-                system_prompt,
-                runner_proc_module,
-                result_proc_module,
+                args.num_of_prompts,
+                args.random_seed,
+                args.system_prompt,
+                args.runner_proc_module,
+                args.result_proc_module,
             )
-        )
-        cb_runner.close()
+            await cb_runner.close()
+
+        loop = asyncio.get_event_loop()
+        loop.run_until_complete(run())
 
         # Display results
         runner_runs = api_get_all_run(cb_runner.id)
         result_info = runner_runs[-1].get("results")
         if result_info:
-            show_cookbook_results(
+            _show_cookbook_results(
                 cookbooks, endpoints, result_info, result_info["metadata"]["duration"]
             )
         else:
-            raise RuntimeError("no run result generated")
+            raise RuntimeError(ERROR_BENCHMARK_RUN_COOKBOOK_NO_RESULT)
 
     except Exception as e:
         print(f"[run_cookbook]: {str(e)}")
@@ -183,8 +330,8 @@ def update_cookbook(args) -> None:
     """
     Update a specific cookbook.
 
-    This function updates a specific cookbook by calling the api_update_cookbook function from the
-    moonshot.api module using the cookbook name and update values provided in the args.
+    This function updates a specific cookbook by calling the api_update_cookbook function using the
+    cookbook name and update values provided in the args.
 
     Args:
         args: A namespace object from argparse. It should have the following attributes:
@@ -192,13 +339,36 @@ def update_cookbook(args) -> None:
             update_values (str): A string representation of a list of tuples. Each tuple contains a key
             and a value to update in the cookbook.
 
+    Raises:
+        ValueError: If the 'cookbook' or 'update_values' arguments are not of the expected type or are None.
+
     Returns:
         None
     """
     try:
+        if (
+            args.cookbook is None
+            or not isinstance(args.cookbook, str)
+            or not args.cookbook
+        ):
+            raise ValueError(ERROR_BENCHMARK_UPDATE_COOKBOOK_COOKBOOK_VALIDATION)
+
+        if (
+            args.update_values is None
+            or not isinstance(args.update_values, str)
+            or not args.update_values
+        ):
+            raise ValueError(ERROR_BENCHMARK_UPDATE_COOKBOOK_UPDATE_VALUES_VALIDATION)
+
         cookbook = args.cookbook
-        update_values = dict(literal_eval(args.update_values))
+        if literal_eval(args.update_values) and all(
+            isinstance(i, tuple) for i in literal_eval(args.update_values)
+        ):
+            update_values = dict(literal_eval(args.update_values))
+        else:
+            raise ValueError(ERROR_BENCHMARK_UPDATE_COOKBOOK_UPDATE_VALUES_VALIDATION_1)
         api_update_cookbook(cookbook, **update_values)
+
         print("[update_cookbook]: Cookbook updated.")
     except Exception as e:
         print(f"[update_cookbook]: {str(e)}")
@@ -218,6 +388,9 @@ def delete_cookbook(args) -> None:
         args: A namespace object from argparse. It should have the following attribute:
             cookbook (str): The identifier of the cookbook to delete.
 
+    Raises:
+        ValueError: If the 'cookbook' argument is not a string or is None.
+
     Returns:
         None
     """
@@ -228,7 +401,15 @@ def delete_cookbook(args) -> None:
     if confirmation.lower() != "y":
         console.print("[bold yellow]Cookbook deletion cancelled.[/]")
         return
+
     try:
+        if (
+            args.cookbook is None
+            or not isinstance(args.cookbook, str)
+            or not args.cookbook
+        ):
+            raise ValueError(ERROR_BENCHMARK_DELETE_COOKBOOK_COOKBOOK_VALIDATION)
+
         api_delete_cookbook(args.cookbook)
         print("[delete_cookbook]: Cookbook deleted.")
     except Exception as e:
@@ -248,6 +429,9 @@ def _display_cookbooks(cookbooks_list):
 
     Args:
         cookbooks_list (list): A list of dictionaries, where each dictionary contains the details of a cookbook.
+
+    Returns:
+        None
     """
     table = Table(
         title="List of Cookbooks", show_lines=True, expand=True, header_style="bold"
@@ -265,7 +449,7 @@ def _display_cookbooks(cookbooks_list):
     console.print(table)
 
 
-def display_view_cookbook(cookbook_info):
+def _display_view_cookbook(cookbook_info):
     """
     Display the cookbook information in a formatted table.
 
@@ -313,10 +497,10 @@ def display_view_cookbook(cookbook_info):
             attack_strategies_info = display_view_list_format(
                 "Attack Strategies", attack_strategies
             )
-            grading_scale_info = display_view_grading_scale_format(
+            grading_scale_info = _display_view_grading_scale_format(
                 "Grading Scale", grading_scale
             )
-            stats_info = display_view_statistics_format("Statistics", stats)
+            stats_info = _display_view_statistics_format("Statistics", stats)
 
             recipe_info = (
                 f"[red]id: {id}[/red]\n\n[blue]{name}[/blue]\n{description}\n\n"
@@ -331,11 +515,11 @@ def display_view_cookbook(cookbook_info):
         console.print("[red]There are no recipes found for the cookbook.[/red]")
 
 
-def show_cookbook_results(cookbooks, endpoints, cookbook_results, duration):
+def _show_cookbook_results(cookbooks, endpoints, cookbook_results, duration):
     """
     Show the results of the cookbook benchmarking.
 
-    This function takes the cookbooks, endpoints, cookbook results, results file, and duration as arguments.
+    This function takes the cookbooks, endpoints, cookbook results, and duration as arguments.
     If there are results, it generates a table with the cookbook results and prints a message indicating
     where the results are saved. If there are no results, it prints a message indicating that no results were found.
     Finally, it prints the duration of the run.
@@ -351,7 +535,7 @@ def show_cookbook_results(cookbooks, endpoints, cookbook_results, duration):
     """
     if cookbook_results:
         # Display recipe results
-        generate_cookbook_table(cookbooks, endpoints, cookbook_results)
+        _generate_cookbook_table(cookbooks, endpoints, cookbook_results)
     else:
         console.print("[red]There are no results.[/red]")
 
@@ -361,7 +545,7 @@ def show_cookbook_results(cookbooks, endpoints, cookbook_results, duration):
     console.print(run_stats)
 
 
-def generate_cookbook_table(cookbooks: list, endpoints: list, results: dict) -> None:
+def _generate_cookbook_table(cookbooks: list, endpoints: list, results: dict) -> None:
     """
     Generate and display a table with the cookbook benchmarking results.
 

moonshot/integrations/cli/benchmark/datasets.py

@@ -9,6 +9,13 @@ from moonshot.api import (
     api_get_all_datasets,
     api_get_all_datasets_name,
 )
+from moonshot.integrations.cli.cli_errors import (
+    ERROR_BENCHMARK_DELETE_DATASET_DATASET_VALIDATION,
+    ERROR_BENCHMARK_LIST_DATASETS_FIND_VALIDATION,
+    ERROR_BENCHMARK_LIST_DATASETS_PAGINATION_VALIDATION,
+    ERROR_BENCHMARK_LIST_DATASETS_PAGINATION_VALIDATION_1,
+    ERROR_BENCHMARK_VIEW_DATASET_DATASET_FILENAME_VALIDATION,
+)
 from moonshot.integrations.cli.common.display_helper import display_view_str_format
 from moonshot.integrations.cli.utils.process_data import filter_data
 
@@ -23,22 +30,43 @@ def list_datasets(args) -> list | None:
     List all available datasets.
 
     This function retrieves all available datasets by calling the api_get_all_datasets function from the
-    moonshot.api module. It then displays the datasets using the _display_datasets function. If an exception occurs,
-    it prints an error message.
+    moonshot.api module. It then filters the datasets based on the provided keyword and pagination arguments.
+    If there are no datasets, it prints a message indicating that no datasets were found.
 
     Args:
-        args: A namespace object from argparse. It should have an optional attribute:
-        find (str): Optional field to find dataset(s) with a keyword.
-        pagination (str): Optional field to paginate datasets.
+        args: A namespace object from argparse. It should have optional attributes:
+            find (str): Optional keyword to filter datasets.
+            pagination (str): Optional tuple to paginate datasets.
 
     Returns:
-        list | None: A list of Dataset or None if there is no result.
+        list | None: A list of datasets or None if there are no datasets.
     """
     try:
         print("Listing datasets may take a while...")
+        if args.find is not None:
+            if not isinstance(args.find, str) or not args.find:
+                raise TypeError(ERROR_BENCHMARK_LIST_DATASETS_FIND_VALIDATION)
+
+        if args.pagination is not None:
+            if not isinstance(args.pagination, str) or not args.pagination:
+                raise TypeError(ERROR_BENCHMARK_LIST_DATASETS_PAGINATION_VALIDATION)
+            try:
+                pagination = literal_eval(args.pagination)
+                if not (
+                    isinstance(pagination, tuple)
+                    and len(pagination) == 2
+                    and all(isinstance(i, int) for i in pagination)
+                ):
+                    raise ValueError(
+                        ERROR_BENCHMARK_LIST_DATASETS_PAGINATION_VALIDATION_1
+                    )
+            except (ValueError, SyntaxError):
+                raise ValueError(ERROR_BENCHMARK_LIST_DATASETS_PAGINATION_VALIDATION_1)
+        else:
+            pagination = ()
+
         datasets_list = api_get_all_datasets()
         keyword = args.find.lower() if args.find else ""
-        pagination = literal_eval(args.pagination) if args.pagination else ()
 
         if datasets_list:
             filtered_datasets_list = filter_data(datasets_list, keyword, pagination)
@@ -48,8 +76,10 @@ def list_datasets(args) -> list | None:
 
         console.print("[red]There are no datasets found.[/red]")
         return None
+
     except Exception as e:
         print(f"[list_datasets]: {str(e)}")
+        return None
 
 
 def view_dataset(args) -> None:
@@ -69,6 +99,13 @@ def view_dataset(args) -> None:
     """
     try:
         print("Viewing datasets may take a while...")
+        if (
+            not isinstance(args.dataset_filename, str)
+            or not args.dataset_filename
+            or args.dataset_filename is None
+        ):
+            raise TypeError(ERROR_BENCHMARK_VIEW_DATASET_DATASET_FILENAME_VALIDATION)
+
         datasets_list = api_get_all_datasets()
         datasets_name_list = api_get_all_datasets_name()
 
@@ -92,7 +129,7 @@ def delete_dataset(args) -> None:
 
     Args:
         args: A namespace object from argparse. It should have the following attribute:
-            dataset_name (str): The name of the dataset to delete.
+            dataset (str): The name of the dataset to delete.
 
     Returns:
         None
@@ -104,7 +141,15 @@ def delete_dataset(args) -> None:
     if confirmation.lower() != "y":
         console.print("[bold yellow]Dataset deletion cancelled.[/]")
         return
+
     try:
+        if (
+            args.dataset is None
+            or not isinstance(args.dataset, str)
+            or not args.dataset
+        ):
+            raise ValueError(ERROR_BENCHMARK_DELETE_DATASET_DATASET_VALIDATION)
+
         api_delete_dataset(args.dataset)
         print("[delete_dataset]: Dataset deleted.")
     except Exception as e: