langfun 0.0.2.dev20240408__tar.gz → 0.0.2.dev20240413__tar.gz

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langfun
-Version: 0.0.2.dev20240408
+Version: 0.0.2.dev20240413
 Summary: Langfun: Language as Functions.
 Home-page: https://github.com/google/langfun
 Author: Langfun Authors

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/eval/base.py

@@ -1551,8 +1551,33 @@ class Summary(pg.Object):
   def _repr_html_(self) -> str:
     return self.html()
 
+  def json(
+      self,
+  ) -> dict[
+      str,  # Task name
+      list[pg.Dict],  # List of pg.Dict with `experiment` and `metrics`.
+  ]:
+    """Returns the JSON representation of the summary."""
+    task_results = {}
+    for task in sorted(self.tasks(), key=lambda cls: cls.__name__):
+      results = []
+      for entry in self.select(task=task).evaluations:
+        results.append(
+            pg.Dict(
+                experiment=entry,
+                metrics=entry.result.metrics if entry.result else None,
+            )
+        )
+      task_results[task.__name__] = results
+    return task_results
+
   def save(self, file: str, pivot_field: str | None = None) -> None:
     pg.save(self.html(pivot_field), file, file_format='txt')
+    if file.endswith('.html'):
+      json_file = file.replace('.html', '.json')
+    else:
+      json_file = os.path.join(file, '.json')
+    pg.save(self.json(), json_file)
 
   @classmethod
   def from_dirs(

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/eval/base_test.py

@@ -101,7 +101,7 @@ class EvaluationTest(unittest.TestCase):
     self.assertEqual(s.dir, os.path.join(s.root_dir, s.id))
     self.assertEqual(s.hash, s.clone().hash)
     # Test persistent hash.
-    self.assertEqual(s.hash, 'abc7c29a')
+    self.assertEqual(s.hash, '436dc80c')
     self.assertEqual(
         s.hash, s.clone(override={'max_workers': 2, 'lm.timeout': 20}).hash
     )
@@ -209,7 +209,7 @@ class EvaluationTest(unittest.TestCase):
         s.result,
         dict(
             experiment_setup=dict(
-                id='Evaluation@17915dc6',
+                id='Evaluation@f1aa5126',
                 dir=s.dir,
                 model='StaticSequence',
                 prompt_template='{{example.question}}',
@@ -228,13 +228,23 @@ class EvaluationTest(unittest.TestCase):
         os.path.exists(os.path.join(s.dir, base.Evaluation.RESULT_JSON)))
     self.assertTrue(
         os.path.exists(os.path.join(s.dir, base.Evaluation.CACHE_JSON)))
-    self.assertTrue(
-        os.path.exists(os.path.join(s.root_dir, base.Evaluation.SUMMARY_HTML))
-    )
     self.assertTrue(
         os.path.exists(os.path.join(s.dir, base.Evaluation.INDEX_HTML)))
     self.assertTrue(
         os.path.exists(os.path.join(s.dir, base.Evaluation.FAILURES_HTML)))
+    self.assertTrue(
+        os.path.exists(os.path.join(s.root_dir, base.Evaluation.SUMMARY_HTML))
+    )
+    # Check summary JSON.
+    summary_json = os.path.join(
+        s.root_dir, base.Evaluation.SUMMARY_HTML.replace('.html', '.json')
+    )
+    self.assertTrue(os.path.exists(summary_json))
+    summary = pg.load(summary_json, force_dict=True)
+    self.assertIn('Evaluation', summary)
+    self.assertEqual(len(summary['Evaluation']), 1)
+    self.assertIsNotNone(summary['Evaluation'][0].experiment)
+    self.assertIsNotNone(summary['Evaluation'][0].metrics)
 
   def test_run_wihtout_save(self):
     lm = fake.StaticSequence([
@@ -321,7 +331,7 @@ class EvaluationTest(unittest.TestCase):
         s.children[0].dir, os.path.join(s.root_dir, s.children[0].id)
     )
     # Test persistent hash.
-    self.assertEqual(s.hash, 'ca7f722b')
+    self.assertEqual(s.hash, 'b66a4e88')
 
     summary = s.run(verbose=True)
     self.assertEqual(len(summary.evaluations), 2)
@@ -448,7 +458,7 @@ class SuiteTest(unittest.TestCase):
         lm=lm
     )
     # Test for persistent hash.
-    self.assertEqual(s.hash, '7285e52b')
+    self.assertEqual(s.hash, 'bbfdc7a8')
     s.run()
     expected = {
         s.children[0].id: dict(

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/eval/matching.py

@@ -155,19 +155,16 @@ class Matching(base.Evaluation):
     super().save(definition, result, report)
 
     if result:
-
-      def force_dict(v):
-        return pg.object_utils.json_conversion.strip_types(pg.to_json(v))
-
       # Save matches.
       pg.save(
           [
-              # We force the output to be dict as its type may be defined
-              # within functors which could be deserialized.
-              pg.Dict(input=input, output=force_dict(output))
+              pg.Dict(input=input, output=output)
               for input, output, _ in self.matches
           ],
           os.path.join(self.dir, Matching.MATCHES_JSON),
+          # We force the input and output to be dict so it does not depend on
+          # the downstream to serialize.
+          force_dict=True,
       )
 
       # Save mismatches.
@@ -175,10 +172,13 @@ class Matching(base.Evaluation):
           [
               # We force the output to be dict as its type may be defined
               # within functors which could be deserialized.
-              pg.Dict(input=input, output=force_dict(output))
+              pg.Dict(input=input, output=output)
               for input, output, _ in self.mismatches
           ],
           os.path.join(self.dir, Matching.MISMATCHES_JSON),
+          # We force the input and output to be dict so it does not depend on
+          # the downstream to serialize.
+          force_dict=True,
       )
 
     if report:

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/eval/matching_test.py

@@ -103,7 +103,7 @@ class MatchingTest(unittest.TestCase):
         s.result,
         dict(
             experiment_setup=dict(
-                id='MyTask@3d87f97f',
+                id='MyTask@acd56a61',
                 dir=s.dir,
                 model='StaticSequence',
                 prompt_template='{{example.question}}',

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/eval/scoring.py

@@ -118,19 +118,18 @@ class Scoring(base.Evaluation):
     super().save(definition, result, report)
 
     if result:
-
-      def force_dict(v):
-        return pg.object_utils.json_conversion.strip_types(pg.to_json(v))
-
       # Save scored.
       pg.save(
           [
               # We force the output to be dict as its type may be defined
               # within functors which could be deserialized.
-              pg.Dict(input=input, output=force_dict(output), score=score)
+              pg.Dict(input=input, output=output, score=score)
               for input, output, score, _ in self.scored
           ],
           os.path.join(self.dir, Scoring.SCORED_JSON),
+          # We force the input and output to be dict so it does not depend on
+          # the downstream to serialize.
+          force_dict=True,
       )
 
     if report:

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/eval/scoring_test.py

@@ -81,7 +81,7 @@ class ScoringTest(unittest.TestCase):
         s.result,
         dict(
             experiment_setup=dict(
-                id='ConstraintFollowing@9e51bb9e',
+                id='ConstraintFollowing@a44d8b89',
                 dir=s.dir,
                 model='StaticSequence',
                 prompt_template='{{example}}',

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/langfunc_test.py

@@ -92,8 +92,8 @@ class LangFuncCallTest(unittest.TestCase):
     self.assertEqual(
         repr(l),
         "LangFunc(template_str='Hello', clean=True,"
-        ' lm=ExcitedEchoer(sampling_options=LMSamplingOptions(temperature=0.0,'
-        ' max_tokens=1024, n=1, top_k=40, top_p=None, stop=None,'
+        ' lm=ExcitedEchoer(sampling_options=LMSamplingOptions(temperature=None,'
+        ' max_tokens=None, n=1, top_k=40, top_p=None, stop=None,'
         ' random_seed=None, logprobs=False, top_logprobs=None), cache=None,'
         ' max_concurrency=None, timeout=120.0, max_attempts=5,'
         ' retry_interval=(5, 60), exponential_backoff=True, debug=False))',

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/language_model.py

@@ -63,14 +63,24 @@ class LMSamplingOptions(component.Component):
   """Language model sampling options."""
 
   temperature: Annotated[
-      float,
+      float | None,
       (
           'Model temperature, which is usually between 0 and 1.0. '
-          'OpenAI models have temperature range from 0.0 to 2.0.'
+          'OpenAI models have temperature range from 0.0 to 2.0. '
+          'If None (default), honor the model\'s default behavior. '
       )
-  ] = 0.0
-  max_tokens: Annotated[int, 'Per example max tokens to generate.'] = 1024
+  ] = None
+
+  max_tokens: Annotated[
+      int | None,
+      (
+          'Per example max tokens to generate. '
+          'If None, use the model default.'
+      )
+  ] = None
+
   n: Annotated[int | None, 'Max number of samples to return.'] = 1
+
   top_k: Annotated[
       int | None,
       (
@@ -78,6 +88,7 @@ class LMSamplingOptions(component.Component):
           'Not applicable to OpenAI models.'
       )
   ] = 40
+
   top_p: Annotated[
       float | None,
       (
@@ -86,6 +97,7 @@ class LMSamplingOptions(component.Component):
           '`top_p` but not both.'
       ),
   ] = None
+
   stop: Annotated[
       list[str] | None,
       (
@@ -95,9 +107,11 @@ class LMSamplingOptions(component.Component):
           '`Model:` is reached.'
       ),
   ] = None
+
   random_seed: Annotated[
       int | None, 'A fixed random seed used during model inference.'
   ] = None
+
   logprobs: Annotated[
       bool,
       (
@@ -106,6 +120,7 @@ class LMSamplingOptions(component.Component):
           'in the content of message.'
       ),
   ] = False
+
   top_logprobs: Annotated[
       int | None,
       (

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/language_model_test.py

@@ -40,7 +40,7 @@ class MockModel(lm_lib.LanguageModel):
         return [
             lm_lib.LMSamplingResult([lm_lib.LMSample(  # pylint: disable=g-complex-comprehension
                 response=prompt.text * self.sampling_options.top_k,
-                score=self.sampling_options.temperature)])
+                score=self.sampling_options.temperature or -1.0)])
             for prompt in prompts
         ]
       context.attempt += 1
@@ -73,13 +73,13 @@ class LMSamplingOptionsTest(unittest.TestCase):
   def test_cache_key(self):
     options = lm_lib.LMSamplingOptions()
     key1 = options.cache_key()
-    self.assertEqual(key1, (0.0, 1024, 1, 40, None, None))
+    self.assertEqual(key1, (None, None, 1, 40, None, None))
     with options.override(temperature=1.0, max_tokens=256):
       key2 = options.cache_key()
       self.assertEqual(key2, (1.0, 256, 1, 40, None, None))
 
       # Make sure key1 does not change upon override.
-      self.assertEqual(key1, (0.0, 1024, 1, 40, None, None))
+      self.assertEqual(key1, (None, None, 1, 40, None, None))
 
 
 class LanguageModelTest(unittest.TestCase):
@@ -100,8 +100,8 @@ class LanguageModelTest(unittest.TestCase):
     self.assertEqual(
         lm.sample(prompts=['foo', 'bar']),
         [
-            lm_lib.LMSamplingResult([lm_lib.LMSample('foo', score=0.0)]),
-            lm_lib.LMSamplingResult([lm_lib.LMSample('bar', score=0.0)]),
+            lm_lib.LMSamplingResult([lm_lib.LMSample('foo', score=-1.0)]),
+            lm_lib.LMSamplingResult([lm_lib.LMSample('bar', score=-1.0)]),
         ],
     )
     # Test override sampling_options.
@@ -143,7 +143,7 @@ class LanguageModelTest(unittest.TestCase):
     lm = MockModel(sampling_options=lm_lib.LMSamplingOptions(top_k=1))
     response = lm(prompt='foo')
     self.assertEqual(response.text, 'foo')
-    self.assertEqual(response.score, 0.0)
+    self.assertEqual(response.score, -1.0)
 
     # Test override sampling_options.
     self.assertEqual(
@@ -159,9 +159,9 @@ class LanguageModelTest(unittest.TestCase):
         lm.sample(prompts=['foo', 'bar']),
         [
             lm_lib.LMSamplingResult([lm_lib.LMSample(
-                message_lib.AIMessage('foo', cache_seed=0), score=0.0)]),
+                message_lib.AIMessage('foo', cache_seed=0), score=-1.0)]),
             lm_lib.LMSamplingResult([lm_lib.LMSample(
-                message_lib.AIMessage('bar', cache_seed=0), score=0.0)]),
+                message_lib.AIMessage('bar', cache_seed=0), score=-1.0)]),
         ])
     self.assertEqual(cache.stats.num_queries, 2)
     self.assertEqual(cache.stats.num_hits, 0)

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/llms/cache/in_memory_test.py

@@ -44,19 +44,19 @@ class InMemoryLMCacheTest(unittest.TestCase):
     self.assertEqual(
         list(cache.keys()),
         [
-            ('a', (0.0, 1024, 1, 40, None, None), 0),
-            ('a', (0.0, 1024, 1, 40, None, None), 1),
-            ('b', (0.0, 1024, 1, 40, None, None), 0),
-            ('c', (0.0, 1024, 1, 40, None, None), 0),
+            ('a', (None, None, 1, 40, None, None), 0),
+            ('a', (None, None, 1, 40, None, None), 1),
+            ('b', (None, None, 1, 40, None, None), 0),
+            ('c', (None, None, 1, 40, None, None), 0),
         ],
     )
     self.assertEqual(
         list(cache.keys('StaticSequence')),
         [
-            ('a', (0.0, 1024, 1, 40, None, None), 0),
-            ('a', (0.0, 1024, 1, 40, None, None), 1),
-            ('b', (0.0, 1024, 1, 40, None, None), 0),
-            ('c', (0.0, 1024, 1, 40, None, None), 0),
+            ('a', (None, None, 1, 40, None, None), 0),
+            ('a', (None, None, 1, 40, None, None), 1),
+            ('b', (None, None, 1, 40, None, None), 0),
+            ('c', (None, None, 1, 40, None, None), 0),
         ],
     )
 
@@ -90,19 +90,19 @@ class InMemoryLMCacheTest(unittest.TestCase):
         list(cache.items()),
         [
             (
-                ('a', (0.0, 1024, 1, 40, None, None), 0),
+                ('a', (None, None, 1, 40, None, None), 0),
                 cache_entry('1'),
             ),
             (
-                ('a', (0.0, 1024, 1, 40, None, None), 1),
+                ('a', (None, None, 1, 40, None, None), 1),
                 cache_entry('2', 1),
             ),
             (
-                ('b', (0.0, 1024, 1, 40, None, None), 0),
+                ('b', (None, None, 1, 40, None, None), 0),
                 cache_entry('3'),
             ),
             (
-                ('c', (0.0, 1024, 1, 40, None, None), 0),
+                ('c', (None, None, 1, 40, None, None), 0),
                 cache_entry('4'),
             ),
         ],
@@ -111,19 +111,19 @@ class InMemoryLMCacheTest(unittest.TestCase):
         list(cache.items('StaticSequence')),
         [
             (
-                ('a', (0.0, 1024, 1, 40, None, None), 0),
+                ('a', (None, None, 1, 40, None, None), 0),
                 cache_entry('1'),
             ),
             (
-                ('a', (0.0, 1024, 1, 40, None, None), 1),
+                ('a', (None, None, 1, 40, None, None), 1),
                 cache_entry('2', 1),
             ),
             (
-                ('b', (0.0, 1024, 1, 40, None, None), 0),
+                ('b', (None, None, 1, 40, None, None), 0),
                 cache_entry('3'),
             ),
             (
-                ('c', (0.0, 1024, 1, 40, None, None), 0),
+                ('c', (None, None, 1, 40, None, None), 0),
                 cache_entry('4'),
             ),
         ],
@@ -161,15 +161,15 @@ class InMemoryLMCacheTest(unittest.TestCase):
     self.assertEqual(
         list(cache.keys()),
         [
-            ('a', (0.0, 1024, 1, 40, None, None), 0),
-            ('a', (1.0, 1024, 1, 40, None, None), 0),
+            ('a', (None, None, 1, 40, None, None), 0),
+            ('a', (1.0, None, 1, 40, None, None), 0),
         ],
     )
 
   def test_different_model(self):
     cache = in_memory.InMemory()
-    lm1 = fake.StaticSequence(['1', '2', '3'], cache=cache)
-    lm2 = fake.Echo(cache=cache)
+    lm1 = fake.StaticSequence(['1', '2', '3'], cache=cache, temperature=0.0)
+    lm2 = fake.Echo(cache=cache, temperature=0.0)
 
     self.assertEqual(lm1('a'), '1')
     self.assertEqual(lm2('a'), 'a')
@@ -180,15 +180,15 @@ class InMemoryLMCacheTest(unittest.TestCase):
     self.assertEqual(
         list(cache.keys('StaticSequence')),
         [
-            ('a', (0.0, 1024, 1, 40, None, None), 0),
-            ('b', (0.0, 1024, 1, 40, None, None), 0),
+            ('a', (0.0, None, 1, 40, None, None), 0),
+            ('b', (0.0, None, 1, 40, None, None), 0),
         ],
     )
     self.assertEqual(
         list(cache.keys('Echo')),
         [
-            ('a', (0.0, 1024, 1, 40, None, None), 0),
-            ('b', (0.0, 1024, 1, 40, None, None), 0),
+            ('a', (0.0, None, 1, 40, None, None), 0),
+            ('b', (0.0, None, 1, 40, None, None), 0),
         ],
     )
     self.assertEqual(len(cache), 4)

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/llms/google_genai_test.py

@@ -152,10 +152,15 @@ class GenAITest(unittest.TestCase):
     )
 
   def test_model_hub(self):
+    orig_get_model = genai.get_model
+    genai.get_model = mock_get_model
+
     model = google_genai._GOOGLE_GENAI_MODEL_HUB.get('gemini-pro')
     self.assertIsNotNone(model)
     self.assertIs(google_genai._GOOGLE_GENAI_MODEL_HUB.get('gemini-pro'), model)
 
+    genai.get_model = orig_get_model
+
   def test_api_key_check(self):
     with self.assertRaisesRegex(ValueError, 'Please specify `api_key`'):
       _ = google_genai.GeminiPro()._api_initialized
@@ -167,7 +172,7 @@ class GenAITest(unittest.TestCase):
 
   def test_call(self):
     with mock.patch(
-        'google.generativeai.generative_models.GenerativeModel.generate_content'
+        'google.generativeai.GenerativeModel.generate_content',
     ) as mock_generate:
       orig_get_model = genai.get_model
       genai.get_model = mock_get_model
@@ -176,7 +181,7 @@ class GenAITest(unittest.TestCase):
       lm = google_genai.GeminiPro(api_key='test_key')
       self.maxDiff = None
       self.assertEqual(
-          lm('hello', temperature=2.0, top_k=20).text,
+          lm('hello', temperature=2.0, top_k=20, max_tokens=1024).text,
           (
               'This is a response to hello with n=1, temperature=2.0, '
               'top_p=None, top_k=20, max_tokens=1024, stop=None.'
@@ -197,7 +202,7 @@ class GenAITest(unittest.TestCase):
         (
             "hello to models/text-bison-001 with {'temperature': 2.0, "
             "'top_k': 20, 'top_p': None, 'candidate_count': 1, "
-            "'max_output_tokens': 1024, 'stop_sequences': None}"
+            "'max_output_tokens': None, 'stop_sequences': None}"
         ),
     )
     genai.get_model = orig_get_model

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/llms/llama_cpp.py

@@ -51,10 +51,12 @@ class LlamaCppRemote(lf.LanguageModel):
           data = {
               "prompt": prompt.text,
               "n_predict": self.sampling_options.max_tokens,
-              "temperature": self.sampling_options.temperature,
               "top_k": self.sampling_options.top_k or 50,
               "top_p": self.sampling_options.top_p or 0.95,
           }
+          if self.sampling_options.temperature is not None:
+            data["temperature"] = self.sampling_options.temperature
+
           response = requests.post(
               f"{self.url}/completion",
               json=data,

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/llms/openai.py

@@ -163,8 +163,6 @@ class OpenAI(lf.LanguageModel):
     # NOTE(daiyip): options.top_k is not applicable.
     args = dict(
         n=options.n,
-        temperature=options.temperature,
-        max_tokens=options.max_tokens,
         stream=False,
         timeout=self.timeout,
         logprobs=options.logprobs,
@@ -173,6 +171,10 @@ class OpenAI(lf.LanguageModel):
     # Completion and ChatCompletion uses different parameter name for model.
     args['model' if self.is_chat_model else 'engine'] = self.model
 
+    if options.temperature is not None:
+      args['temperature'] = options.temperature
+    if options.max_tokens is not None:
+      args['max_tokens'] = options.max_tokens
     if options.top_p is not None:
       args['top_p'] = options.top_p
     if options.stop:
@@ -220,6 +222,10 @@ class OpenAI(lf.LanguageModel):
         retry_on_errors=(
             openai_error.ServiceUnavailableError,
             openai_error.RateLimitError,
+            # Handling transient OpenAI server error (code 500). Check out
+            # https://platform.openai.com/docs/guides/error-codes/error-codes
+            (openai_error.APIError,
+             '.*The server had an error processing your request'),
         ),
     )[0]
 

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/llms/openai_test.py

@@ -121,7 +121,6 @@ class OpenaiTest(unittest.TestCase):
             top_logprobs=None,
             n=1,
             temperature=1.0,
-            max_tokens=1024,
             stream=False,
             timeout=120.0,
             stop=['\n'],

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/structured/prompting.py

@@ -78,7 +78,9 @@ class QueryStructurePython(QueryStructure):
 
       {{ output_title }}:
         ```python
-        Answer(final_answer=2)
+        Answer(
+          final_answer=2
+        )
         ```
       """
   protocol = 'python'

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/structured/prompting_test.py

@@ -116,12 +116,26 @@ class QueryTest(unittest.TestCase):
         y=2,
         lm=lm.clone(),
         expected_snippet=(
-            'Please respond to the last INPUT_OBJECT with OUTPUT_OBJECT'
-            ' according to OUTPUT_TYPE.\n\nINPUT_OBJECT:\n  1 + 1'
-            ' =\n\nOUTPUT_TYPE:\n  Answer\n\n  ```python\n  class Answer:\n   '
-            ' final_answer: int\n  ```\n\nOUTPUT_OBJECT:\n  ```python\n '
-            ' Answer(final_answer=2)\n  ```\n\nINPUT_OBJECT:\n  What is 1 +'
-            ' 2?\n\nOUTPUT_TYPE:\n  int\n\nOUTPUT_OBJECT:'
+            'Please respond to the last INPUT_OBJECT with OUTPUT_OBJECT '
+            'according to OUTPUT_TYPE.\n\n'
+            'INPUT_OBJECT:\n  1 + 1 =\n\n'
+            'OUTPUT_TYPE:\n'
+            '  Answer\n\n'
+            '  ```python\n'
+            '  class Answer:\n'
+            '    final_answer: int\n'
+            '  ```\n\n'
+            'OUTPUT_OBJECT:\n'
+            '  ```python\n'
+            '  Answer(\n'
+            '    final_answer=2\n'
+            '  )\n'
+            '  ```\n\n'
+            'INPUT_OBJECT:\n'
+            '  What is 1 + 2?\n\n'
+            'OUTPUT_TYPE:\n'
+            '  int\n\n'
+            'OUTPUT_OBJECT:'
         ),
     )
 
@@ -264,7 +278,9 @@ class QueryStructurePythonTest(unittest.TestCase):
 
             OUTPUT_OBJECT:
               ```python
-              Answer(final_answer=2)
+              Answer(
+                final_answer=2
+              )
               ```
 
             INPUT_OBJECT:
@@ -308,7 +324,9 @@ class QueryStructurePythonTest(unittest.TestCase):
 
             OUTPUT_OBJECT:
               ```python
-              Answer(final_answer=2)
+              Answer(
+                final_answer=2
+              )
               ```
 
             INPUT_OBJECT:

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun/core/structured/schema_test.py

@@ -192,9 +192,9 @@ class SchemaTest(unittest.TestCase):
     self.assertEqual(schema.parse('{"result": 1}'), 1)
     schema = schema_lib.Schema(dict[str, int])
     self.assertEqual(
-        schema.parse(
-            '{"result": {"_type": "Unknown", "x": 1}}}', force_dict=True),
-        dict(x=1))
+        schema.parse('{"result": {"x": 1}}}'),
+        dict(x=1)
+    )
     with self.assertRaisesRegex(
         schema_lib.SchemaError, 'Expect .* but encountered .*'):
       schema.parse('{"result": "def"}')

{langfun-0.0.2.dev20240408 → langfun-0.0.2.dev20240413}/langfun.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langfun
-Version: 0.0.2.dev20240408
+Version: 0.0.2.dev20240413
 Summary: Langfun: Language as Functions.
 Home-page: https://github.com/google/langfun
 Author: Langfun Authors