akaashrp opened a new pull request, #20121: URL: https://github.com/apache/tvm/pull/20121
Extend `PagedKVCache` for models whose logical attention layers reuse K/V cached by another physical layer. - Adds non-mutating `attention_with_shared_kv` support. - Makes per-layer sliding-window size configurable. - Corrects sliding-window masking during chunked prefill. This is intended to enable downstream Gemma 4 support in MLC-LLM and WebLLM. -- 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]
