atirna opened a new pull request, #20130: URL: https://github.com/apache/tvm/pull/20130
## Summary - retain ShapeTuples evicted from the bounded LRU until `CacheState.dispose()` - dispose each retained or still-cached tuple exactly once at TVM-instance teardown - add a regression test for the live borrowed-reference lifecycle This prevents an LRU miss from disposing a tuple which a pending GPU dispatch may still use. ## Validation - `npx jest --config jest.config.js --runTestsByPath tests/node/test_cache_state.js --runInBand` - `npx eslint -c .eslintrc.cjs src/cache_state.ts tests/node/test_cache_state.js` - `npx tsc --noEmit -p tsconfig.json` The full `npm test` requires generated `dist/` WASM artifacts not present in the sparse checkout; the existing integration suites cannot resolve `../../dist` in that environment. Relates to mlc-ai/web-llm#844. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
