New Release (IBM#60)

* Informative error messages in async generator and type change in GenAI exception (IBM#57)

* update greedy examples removing incompatible params (IBM#59)

* Fix/setup cfg (IBM#49)

* fix: including README in pypi deployment

* fix: adding build dependency and changing workflow to use it


Signed-off-by: James Sutton <[email protected]>
Co-authored-by: Mírian Silva <[email protected]>
Co-authored-by: James Sutton <[email protected]>

Author: 3 people

.github/workflows/python-publish.yml

@@ -24,7 +24,7 @@ jobs:
         python -m pip install -e ".[dev]"
     - name: Version and package
       run: |
-        pip wheel -w dist -e . --no-deps
+        python -m build
     - name: Publish package
       uses: pypa/gh-action-pypi-publish@f9ed8ba9ad06d20b1ebb6002ffb93050ed9a1951
       with:

examples/user/grid_search_params.py

@@ -21,7 +21,7 @@
 # use the dictionary to define the search space, keep the keys as
 # the same from GenerateParams and use a list for the values to search
 my_space_params = {
-    "decoding_method": ["sample", "greedy"],
+    "decoding_method": ["sample"],
     "max_new_tokens": [10, 20],
     "min_new_tokens": [1, 2],
     "temperature": [0.7, 0.8, 0.9, 1.5],

examples/user/prompt_csv_random_rows.py

@@ -28,10 +28,6 @@
     max_new_tokens=15,
     min_new_tokens=1,
     stream=False,
-    temperature=0.7,
-    top_k=50,
-    top_p=1,
-    random_seed=2,
 )
 
 creds = Credentials(api_key, api_endpoint)

examples/user/prompt_from_all_csv.py

@@ -28,10 +28,6 @@
     max_new_tokens=15,
     min_new_tokens=1,
     stream=False,
-    temperature=0.7,
-    top_k=50,
-    top_p=1,
-    random_seed=2,
 )
 
 creds = Credentials(api_key, api_endpoint)

examples/user/prompt_from_all_dataframe.py

@@ -35,10 +35,6 @@
     max_new_tokens=15,
     min_new_tokens=1,
     stream=False,
-    temperature=0.7,
-    top_k=50,
-    top_p=1,
-    random_seed=2,
 )
 
 creds = Credentials(api_key, api_endpoint)

examples/user/prompt_from_dataframe.py

@@ -34,10 +34,6 @@
     max_new_tokens=15,
     min_new_tokens=1,
     stream=False,
-    temperature=0.7,
-    top_k=50,
-    top_p=1,
-    random_seed=2,
 )
 
 creds = Credentials(api_key, api_endpoint)

examples/user/self-reflection.py

@@ -23,9 +23,6 @@
     max_new_tokens=20,
     min_new_tokens=1,
     stream=False,
-    temperature=0.7,
-    top_k=50,
-    top_p=1,
 )
 
 creds = Credentials(api_key, api_endpoint)

pyproject.toml

@@ -1,5 +1,5 @@
 [build-system]
-requires = ["setuptools>=67.6.0", "wheel>=0.40.0", "setuptools_scm[toml]>=7.1.0"]
+requires = ["setuptools>=68.0.0", "wheel>=0.40.0", "setuptools_scm[toml]>=7.1.0"]
 build-backend = "setuptools.build_meta"
 
 [project]
@@ -15,7 +15,7 @@ authors = [
   {name="James Sutton", email="[email protected]"},
 ]
 description = "IBM Generative AI is a Python library built on IBM's large language model REST interface."
-readme = "README.MD"
+readme = "README.md"
 dependencies = [
     "urllib3<2", # https://github.com/psf/requests/issues/6432
     "requests>=2.31.0",
@@ -41,7 +41,8 @@ dev = [
   "pytest-asyncio>=0.21.0",
   "pytest-mock>=3.10.0",
   "setuptools-scm>=7.1.0",
-  "pytest-httpx>=0.22.0"
+  "pytest-httpx>=0.22.0",
+  "build>=0.10.0"
 ]
 langchain = [
   "langchain>=0.0.200"

src/genai/exceptions/genai_exception.py

@@ -1,8 +1,8 @@
 import logging
 from typing import Union
 
+from httpx import Response
 from pydantic import ValidationError
-from requests import Response
 
 from genai.schemas.responses import ErrorResponse
 

src/genai/services/async_generator.py

@@ -97,6 +97,7 @@ async def _get_response_json(self, model, inputs, params):
         return response
 
     async def _task(self, inputs, batch_num):
+        response = None
         try:
             response = await self._get_response_json(self.model_id, inputs, self.params)
             logger.debug("Received response = {}".format(response))
@@ -105,7 +106,11 @@ async def _task(self, inputs, batch_num):
             response = self.message_type_(**response)
             logger.debug("Cast to Response = {}".format(response))
         except Exception as e:
-            logger.error("Exception raised async_generate and casting : {}, inputs = {}".format(str(e), inputs))
+            logger.error(
+                "Exception raised async_generate and casting : {}, response = {}, inputs = {}".format(
+                    str(e), response, inputs
+                )
+            )
             self.queue_.put_nowait((batch_num, len(inputs), None))
             return
         try:
@@ -114,7 +119,9 @@ async def _task(self, inputs, batch_num):
                 for result in response.results:
                     self.callback(result)
         except Exception as e:
-            logger.error("Exception raised in callback : {}, inputs = {}".format(str(e), inputs))
+            logger.error(
+                "Exception raised in callback : {}, response = {}, inputs = {}".format(str(e), response, inputs)
+            )
 
     async def _schedule_requests(self):
         tasks = []

tests/test_concurrent.py

@@ -309,7 +309,7 @@ async def test_concurrent_tokenize_not_inorder(self, mock_tokenize_json, tokeniz
     @pytest.mark.asyncio
     async def test_concurrent_tokenize_nones(self, mock_tokenize_json, tokenize_params, mocker):
         # test that if one request gets dropped, we get appropriate number of nones
-        num_prompts = 14
+        num_prompts = 18
         inputs = ["This is input number " + str(i) for i in range(num_prompts)]
         expected = SimpleResponse.tokenize_response_array_async(model=self.model, inputs=inputs)
         mock_tokenize_json.side_effect = expected

tests/test_generate_schema.py

@@ -15,7 +15,7 @@ def setup_method(self):
 
         # test all GenerateParams fields
         self.params = GenerateParams(
-            decoding_method="greedy",
+            decoding_method="sample",
             length_penalty=LengthPenalty(decay_factor=1.5, start_index=2),
             max_new_tokens=3,
             min_new_tokens=1,

tests/test_request_handler.py

@@ -14,7 +14,7 @@ def setup_method(self):
 
     @pytest.fixture
     def params(seld):
-        return GenerateParams(decoding_method="greedy", temperature=0.8).dict(by_alias=True, exclude_none=True)
+        return GenerateParams(decoding_method="greedy").dict(by_alias=True, exclude_none=True)
 
     def test_metadata_post(self, params):
         headers, json_data = RequestHandler._metadata(
@@ -30,7 +30,7 @@ def test_metadata_post(self, params):
         assert json_data == {
             "model_id": self.model,
             "inputs": self.inputs,
-            "parameters": {"decoding_method": "greedy", "temperature": 0.8},
+            "parameters": {"decoding_method": "greedy"},
         }
 
     def test_metadata_get(self):

tests/test_service_interface.py

@@ -17,7 +17,7 @@ def setup_method(self):
 
     @pytest.fixture
     def params(seld):
-        return GenerateParams(decoding_method="greedy", temperature=0.8, return_options=ReturnOptions(input_text=True))
+        return GenerateParams(decoding_method="greedy", return_options=ReturnOptions(input_text=True))
 
     @patch("genai.services.RequestHandler.patch")
     def test_tou(self, mocked_post_request):

tests/test_service_utils.py

@@ -14,29 +14,9 @@
 @pytest.mark.unit
 class TestServiceUtils(unittest.TestCase):
     def test_sanitize_params_with_params(self):
-        input_genParams = GenerateParams(
-            decoding_method="greedy", temperature=1.0, return_options=ReturnOptions(input_text=True)
-        )
+        input_genParams = GenerateParams(decoding_method="greedy", return_options=ReturnOptions(input_text=True))
 
-        expected_genParams = {"decoding_method": "greedy", "temperature": 1.0, "return_options": {"input_text": True}}
+        expected_genParams = {"decoding_method": "greedy", "return_options": {"input_text": True}}
 
         sanitized_dict = ServiceInterface._sanitize_params(input_genParams)
         self.assertEqual(expected_genParams, sanitized_dict)
-
-    def test_sanitize_params_with_dict(self):
-        # TODO : Confirm what approach we want to have for user passing dict
-
-        # Test with dictionary
-        # input_dict = {
-        #     "fieldA": None,
-        #     "fieldB": None,
-        #     "returns_options": "something",
-        # }
-
-        # expected_dict = {
-        #     "return": "something",
-        # }
-
-        # sanitized_dict = sanitize_params(input_dict)
-        # self.assertEqual(expected_dict, sanitized_dict)
-        pass