langfun 0.1.2.dev202504150804__tar.gz → 0.1.2.dev202504170804__tar.gz

{langfun-0.1.2.dev202504150804 → langfun-0.1.2.dev202504170804}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: langfun
-Version: 0.1.2.dev202504150804
+Version: 0.1.2.dev202504170804
 Summary: Langfun: Language as Functions.
 Home-page: https://github.com/google/langfun
 Author: Langfun Authors

{langfun-0.1.2.dev202504150804 → langfun-0.1.2.dev202504170804}/langfun/core/__init__.py

@@ -111,6 +111,8 @@ from langfun.core.language_model import LMCache
 from langfun.core.language_model import LMDebugMode
 
 from langfun.core.language_model import LMError
+from langfun.core.language_model import LMInputError
+from langfun.core.language_model import ContextLimitError
 from langfun.core.language_model import RetryableLMError
 from langfun.core.language_model import RateLimitError
 from langfun.core.language_model import TemporaryLMError

{langfun-0.1.2.dev202504150804 → langfun-0.1.2.dev202504170804}/langfun/core/language_model.py

@@ -40,6 +40,14 @@ class LMError(RuntimeError):
   """Base class for language model errors."""
 
 
+class LMInputError(LMError):
+  """Base class for errors with bad input."""
+
+
+class ContextLimitError(LMInputError):
+  """Error for context limit exceeded."""
+
+
 class RetryableLMError(LMError):
   """Base class for LLM errors that can be solved by retrying."""
 

{langfun-0.1.2.dev202504150804 → langfun-0.1.2.dev202504170804}/langfun/core/llms/anthropic.py

@@ -573,6 +573,11 @@ class Anthropic(rest.REST):
         ),
     )
 
+  def _error(self, status_code: int, content: str) -> lf.LMError:
+    if status_code == 413 and b'Prompt is too long' in content:
+      return lf.ContextLimitError(f'{status_code}: {content}')
+    return super()._error(status_code, content)
+
 
 class Claude37(Anthropic):
   """Base class for Claude 3.7 models."""

{langfun-0.1.2.dev202504150804 → langfun-0.1.2.dev202504170804}/langfun/core/llms/anthropic_test.py

@@ -217,6 +217,17 @@ class AnthropicTest(unittest.TestCase):
         ):
           lm('hello', max_attempts=1)
 
+  def test_call_with_context_limit_error(self):
+    with mock.patch('requests.Session.post') as mock_request:
+      mock_request.side_effect = mock_requests_post_error(
+          413, 'bad_request', 'Prompt is too long.'
+      )
+      lm = anthropic.Claude3Haiku(api_key='fake_key')
+      with self.assertRaisesRegex(
+          lf.ContextLimitError, 'Prompt is too long.'
+      ):
+        lm('hello', max_attempts=1)
+
   def test_lm_get(self):
     self.assertIsInstance(
         lf.LanguageModel.get('claude-3-5-sonnet-latest'),

{langfun-0.1.2.dev202504150804 → langfun-0.1.2.dev202504170804}/langfun/core/llms/gemini.py

@@ -645,3 +645,9 @@ class Gemini(rest.REST):
             total_tokens=input_tokens + output_tokens,
         ),
     )
+
+  def _error(self, status_code: int, content: str) -> lf.LMError:
+    if (status_code == 400
+        and b'exceeds the maximum number of tokens' in content):
+      return lf.ContextLimitError(f'{status_code}: {content}')
+    return super()._error(status_code, content)

{langfun-0.1.2.dev202504150804 → langfun-0.1.2.dev202504170804}/langfun/core/llms/gemini_test.py

@@ -178,6 +178,22 @@ class GeminiTest(unittest.TestCase):
       self.assertEqual(r.metadata.usage.prompt_tokens, 3)
       self.assertEqual(r.metadata.usage.completion_tokens, 4)
 
+  def test_call_model_with_context_limit_error(self):
+    def mock_requests_post_error(*args, **kwargs):
+      del args, kwargs
+      response = requests.Response()
+      response.status_code = 400
+      response._content = b'exceeds the maximum number of tokens.'
+      return response
+
+    with mock.patch('requests.Session.post') as mock_generate:
+      mock_generate.side_effect = mock_requests_post_error
+      lm = gemini.Gemini('gemini-1.5-pro', api_endpoint='')
+      with self.assertRaisesRegex(
+          lf.ContextLimitError, 'exceeds the maximum number of tokens.'
+      ):
+        lm('hello')
+
   def test_call_model_with_system_message(self):
     with mock.patch('requests.Session.post') as mock_generate:
       mock_generate.side_effect = mock_requests_post

{langfun-0.1.2.dev202504150804 → langfun-0.1.2.dev202504170804}/langfun/core/llms/openai_compatible.py

@@ -159,3 +159,9 @@ class OpenAICompatible(rest.REST):
 
         ),
     )
+
+  def _error(self, status_code: int, content: str) -> lf.LMError:
+    if (status_code == 413
+        or (status_code == 400 and b'string_above_max_length' in content)):
+      return lf.ContextLimitError(f'{status_code}: {content}')
+    return super()._error(status_code, content)

{langfun-0.1.2.dev202504150804 → langfun-0.1.2.dev202504170804}/langfun/core/llms/openai_compatible_test.py

@@ -505,6 +505,24 @@ class OpenAIComptibleTest(unittest.TestCase):
     ):
       lm(lf.UserMessage('hello', json_schema={}))
 
+  def test_call_with_context_limit_error(self):
+    def mock_context_limit_error(*args, **kwargs):
+      del args, kwargs
+      response = requests.Response()
+      response.status_code = 400
+      response._content = b'string_above_max_length'
+      return response
+
+    with mock.patch('requests.Session.post') as mock_request:
+      mock_request.side_effect = mock_context_limit_error
+      lm = openai_compatible.OpenAICompatible(
+          api_endpoint='https://test-server', model='test-model'
+      )
+      with self.assertRaisesRegex(
+          lf.ContextLimitError, 'string_above_max_length'
+      ):
+        lm(lf.UserMessage('hello'))
+
 
 if __name__ == '__main__':
   unittest.main()

{langfun-0.1.2.dev202504150804 → langfun-0.1.2.dev202504170804}/langfun.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: langfun
-Version: 0.1.2.dev202504150804
+Version: 0.1.2.dev202504170804
 Summary: Langfun: Language as Functions.
 Home-page: https://github.com/google/langfun
 Author: Langfun Authors