github/OpenHands
1
0
Fork 0
mirror of https://github.com/OpenHands/OpenHands.git synced 2025-12-26 05:48:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix - Set claude sonnet output limit (#11098)

Browse Source
This commit is contained in:
Ray Myers 2025-09-24 08:31:20 -05:00 committed by GitHub
parent 3977d4fdd7
commit 1e7024b60a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 18 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
openhands/llm/llm.py
View File

@ -502,11 +502,13 @@ class LLM(RetryMixin, DebugMixin):
# Set max_output_tokens from model info if not explicitly set
if self.config.max_output_tokens is None:
# Special case for Claude 3.7 Sonnet models
if any(
model in self.config.model
for model in ['claude-3-7-sonnet', 'claude-3.7-sonnet']
):
# Special case for Claude Sonnet models
sonnet_models = [
'claude-3-7-sonnet',
'claude-3.7-sonnet',
'claude-sonnet-4',
]
if any(model in self.config.model for model in sonnet_models):
self.config.max_output_tokens = 64000 # litellm set max to 128k, but that requires a header to be set
# Try to get from model info
elif self.model_info is not None:

17
tests/unit/llm/test_llm.py
View File

@ -1053,17 +1053,22 @@ def test_claude_3_7_sonnet_max_output_tokens():
assert llm.config.max_input_tokens is None
def test_claude_sonnet_4_max_output_tokens():
@patch('openhands.llm.llm.litellm.get_model_info')
def test_claude_sonnet_4_max_output_tokens(mock_get_model_info):
"""Test that Claude Sonnet 4 models get the correct max_output_tokens and max_input_tokens values."""
mock_get_model_info.return_value = {
'max_input_tokens': 100000,
'max_output_tokens': 100000,
}
# Create LLM instance with a Claude Sonnet 4 model
config = LLMConfig(model='claude-sonnet-4-20250514', api_key='test_key')
llm = LLM(config, service_id='test-service')
llm.init_model_info()
# Verify max_output_tokens is set to the expected value
assert llm.config.max_output_tokens == 64000
# Verify max_input_tokens is set to the expected value
# For Claude models, we expect a specific value from litellm
assert llm.config.max_input_tokens == 200000
assert llm.config.max_output_tokens == 64000, 'output max should be decreased'
assert llm.config.max_input_tokens == 100000, (
'input max should be the litellm value'
)
def test_sambanova_deepseek_model_max_output_tokens():