[Tracking] Fix the context length fetching in local mode #1013
Description
We compile different libraries using separate max-context-length parameter to limit the amount of memory used. After the recent distributed support we switched to read max-ctx-len from mlc-chat-config.json.
While this change is necessary for distributed (since it is harder to fetch meta-data in host), it is still useful for local-mode where some truncated limit be applied in the libraries.
When there is inconsistency in the config, the app can crash instead of truncating input correctly
Action Items
- Update the FunctionTable to obtain the max-ctx-length optionally from library only in local mode Fix the max window length fetching in local mode #1016
- Update the android apps with this change.