The mellea test test/formatters/granite/test_intrinsics_formatters.py::test_canned_input[uncertainty] is having issues after the update to granitelib-core.
It appears that on mellea/formatters/granite/intrinsics/input.py:439 this line, the edits.update(self.parameters) applies all parameters from the YAML directly onto the chat completion request. Since temperature: 0.0 is now in parameters, the rewriter adds "temperature": 0.0 to the transformed request, which isn't in the snapshot.
Recommend either:
- copying the io.yaml files into testdata/input_yaml/ (I think the other tests do this already) and updating the snapshots eliminating the network dependency entirely for test_canned_input
- pinning the revision in the affected YamlJsonCombo entries to the specific HF commit that matches the snapshot if we want to test the round trip (which also seems reasonable).
The mellea test
test/formatters/granite/test_intrinsics_formatters.py::test_canned_input[uncertainty]is having issues after the update to granitelib-core.It appears that on
mellea/formatters/granite/intrinsics/input.py:439this line, theedits.update(self.parameters)applies all parameters from the YAML directly onto the chat completion request. Since temperature: 0.0 is now in parameters, the rewriter adds "temperature": 0.0 to the transformed request, which isn't in the snapshot.Recommend either: