xinyin025/LangBot
1
0
Fork 0
mirror of https://github.com/langbot-app/LangBot.git synced 2026-06-14 17:56:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(provider): use LiteLLM input window for context length (#2243)

Browse Source
This commit is contained in:
huanghuoguoguo
2026-06-13 21:27:47 +08:00
committed by GitHub
parent b6fde30aa7
commit 7fe3eedeea
2 changed files with 45 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
tests/unit_tests/provider/test_litellmchat.py
View File

@@ -1034,11 +1034,28 @@ class TestScanModels:
},
)
with patch.object(litellmchat.litellm, 'get_max_tokens') as mock_get_max_tokens:
mock_get_max_tokens.side_effect = lambda model: 131072 if model == 'moonshot/moonshot-v1-128k' else None
with patch.object(litellmchat.litellm, 'get_model_info') as mock_get_model_info:
mock_get_model_info.side_effect = (
lambda model: {'max_input_tokens': 131072}
if model == 'moonshot/moonshot-v1-128k'
else {}
)
assert requester._safe_context_length('moonshot-v1-128k') == 131072
def test_safe_context_length_uses_litellm_max_input_tokens(self):
"""LiteLLM max_output_tokens must not be treated as the context window."""
requester = litellmchat.LiteLLMRequester(ap=Mock(), config={})
with patch.object(litellmchat.litellm, 'get_model_info') as mock_get_model_info:
mock_get_model_info.return_value = {
'max_input_tokens': 128000,
'max_output_tokens': 16384,
'max_tokens': 16384,
}
assert requester._safe_context_length('gpt-4o') == 128000
def test_litellm_bool_helper_tries_moonshot_metadata_alias(self):
"""OpenAI-compatible Moonshot endpoints still use Moonshot metadata for abilities."""
requester = litellmchat.LiteLLMRequester(
@@ -1102,7 +1119,7 @@ class TestScanModels:
},
)
with patch.object(litellmchat.litellm, 'get_max_tokens', side_effect=Exception('not mapped')):
with patch.object(litellmchat.litellm, 'get_model_info', side_effect=Exception('not mapped')):
assert requester._safe_context_length('deepseek-v4-pro') == 1_000_000
assert requester._safe_context_length('deepseek-v4-flash') == 1_000_000