| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| langroid-0.67.5-py3-none-any.whl | 2026-08-31 | 484.7 kB | |
| langroid-0.67.5.tar.gz | 2026-08-31 | 432.8 kB | |
| 0.67.5 source code.tar.gz | 2026-08-31 | 59.2 MB | |
| 0.67.5 source code.zip | 2026-08-31 | 59.7 MB | |
| README.md | 2026-08-31 | 1.2 kB | |
| Totals: 5 Items | 119.9 MB | 0 | |
Bug fixes
Task instances no longer share the default TaskConfig (#1124)
The default TaskConfig was a shared mutable object, so mutating one
task's config (e.g. its done-sequences or termination settings) could
silently change the behavior of every other Task constructed with
the default. Each Task now gets its own copy. Fixes [#1123]. Thanks
@Starchaser1024.
Test improvements
Table chat agent tests are deterministic (#1125, superseding [#1121])
tests/main/test_table_chat_agent.py no longer depends on a live
LLM: a scripted MockLM subclass drives the agent, and the previously
permissive assertions (which accepted empty or unfinished results) are
now load-bearing — a mutation probe forcing a wrong pandas_eval
result fails all 11 tests. On top of the original contribution by
@SiluPanda, the merged version also exercises the configured
OpenAI tool-call wire format (oai_tool_calls, with role/id
assertions on the fed-back tool result) rather than the legacy
function_call shape, and asserts the scripted sequence is fully
consumed so early task termination can't silently skip checks.
Closes [#288].
Full Changelog: https://github.com/langroid/langroid/compare/0.67.4...0.67.5