langfun 0.0.2.dev20240409__tar.gz → 0.0.2.dev20240411__tar.gz

{langfun-0.0.2.dev20240409 → langfun-0.0.2.dev20240411}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langfun
-Version: 0.0.2.dev20240409
+Version: 0.0.2.dev20240411
 Summary: Langfun: Language as Functions.
 Home-page: https://github.com/google/langfun
 Author: Langfun Authors

{langfun-0.0.2.dev20240409 → langfun-0.0.2.dev20240411}/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, '4a2f9694')
     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@28030a68',
                 dir=s.dir,
                 model='StaticSequence',
                 prompt_template='{{example.question}}',
@@ -321,7 +321,7 @@ 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, 'e2a499d0')
 
     summary = s.run(verbose=True)
     self.assertEqual(len(summary.evaluations), 2)
@@ -448,7 +448,7 @@ class SuiteTest(unittest.TestCase):
         lm=lm
     )
     # Test for persistent hash.
-    self.assertEqual(s.hash, '7285e52b')
+    self.assertEqual(s.hash, '8eb9e73f')
     s.run()
     expected = {
         s.children[0].id: dict(

{langfun-0.0.2.dev20240409 → langfun-0.0.2.dev20240411}/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@9e3c5aa3',
                 dir=s.dir,
                 model='StaticSequence',
                 prompt_template='{{example.question}}',

{langfun-0.0.2.dev20240409 → langfun-0.0.2.dev20240411}/langfun/core/eval/scoring_test.py

@@ -75,13 +75,14 @@ class ScoringTest(unittest.TestCase):
     ])
 
     s = eval_set(lm=lm)
+    self.maxDiff = None
     self.assertEqual(s.avg_score, 0.0)
     s.run()
     self.assertEqual(
         s.result,
         dict(
             experiment_setup=dict(
-                id='ConstraintFollowing@9e51bb9e',
+                id='ConstraintFollowing@76285393',
                 dir=s.dir,
                 model='StaticSequence',
                 prompt_template='{{example}}',

{langfun-0.0.2.dev20240409 → langfun-0.0.2.dev20240411}/langfun/core/langfunc_test.py

@@ -93,7 +93,7 @@ class LangFuncCallTest(unittest.TestCase):
         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,'
+        ' 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.dev20240409 → langfun-0.0.2.dev20240411}/langfun/core/language_model.py

@@ -69,7 +69,14 @@ class LMSamplingOptions(component.Component):
           'OpenAI models have temperature range from 0.0 to 2.0.'
       )
   ] = 0.0
-  max_tokens: Annotated[int, 'Per example max tokens to generate.'] = 1024
+  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,

{langfun-0.0.2.dev20240409 → langfun-0.0.2.dev20240411}/langfun/core/language_model_test.py

@@ -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, (0.0, 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, (0.0, None, 1, 40, None, None))
 
 
 class LanguageModelTest(unittest.TestCase):

{langfun-0.0.2.dev20240409 → langfun-0.0.2.dev20240411}/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', (0.0, None, 1, 40, None, None), 0),
+            ('a', (0.0, None, 1, 40, None, None), 1),
+            ('b', (0.0, None, 1, 40, None, None), 0),
+            ('c', (0.0, 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', (0.0, None, 1, 40, None, None), 0),
+            ('a', (0.0, None, 1, 40, None, None), 1),
+            ('b', (0.0, None, 1, 40, None, None), 0),
+            ('c', (0.0, 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', (0.0, None, 1, 40, None, None), 0),
                 cache_entry('1'),
             ),
             (
-                ('a', (0.0, 1024, 1, 40, None, None), 1),
+                ('a', (0.0, None, 1, 40, None, None), 1),
                 cache_entry('2', 1),
             ),
             (
-                ('b', (0.0, 1024, 1, 40, None, None), 0),
+                ('b', (0.0, None, 1, 40, None, None), 0),
                 cache_entry('3'),
             ),
             (
-                ('c', (0.0, 1024, 1, 40, None, None), 0),
+                ('c', (0.0, 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', (0.0, None, 1, 40, None, None), 0),
                 cache_entry('1'),
             ),
             (
-                ('a', (0.0, 1024, 1, 40, None, None), 1),
+                ('a', (0.0, None, 1, 40, None, None), 1),
                 cache_entry('2', 1),
             ),
             (
-                ('b', (0.0, 1024, 1, 40, None, None), 0),
+                ('b', (0.0, None, 1, 40, None, None), 0),
                 cache_entry('3'),
             ),
             (
-                ('c', (0.0, 1024, 1, 40, None, None), 0),
+                ('c', (0.0, None, 1, 40, None, None), 0),
                 cache_entry('4'),
             ),
         ],
@@ -161,8 +161,8 @@ 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', (0.0, None, 1, 40, None, None), 0),
+            ('a', (1.0, None, 1, 40, None, None), 0),
         ],
     )
 
@@ -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.dev20240409 → langfun-0.0.2.dev20240411}/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.dev20240409 → langfun-0.0.2.dev20240411}/langfun/core/llms/openai.py

@@ -164,7 +164,6 @@ class OpenAI(lf.LanguageModel):
     args = dict(
         n=options.n,
         temperature=options.temperature,
-        max_tokens=options.max_tokens,
         stream=False,
         timeout=self.timeout,
         logprobs=options.logprobs,
@@ -173,6 +172,8 @@ 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.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:

{langfun-0.0.2.dev20240409 → langfun-0.0.2.dev20240411}/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.dev20240409 → langfun-0.0.2.dev20240411}/langfun.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langfun
-Version: 0.0.2.dev20240409
+Version: 0.0.2.dev20240411
 Summary: Langfun: Language as Functions.
 Home-page: https://github.com/google/langfun
 Author: Langfun Authors