-
Notifications
You must be signed in to change notification settings - Fork 51
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
[Feature] Add models/deepseek/v4-pro-w8a8: DeepSeek-V4-Pro A8A8 decode at 128k / EP=128 / B=2 S=4 (MTP=3) and benchmark the main operators
enhancementNew feature or requestNew feature or requestStatus: Open.#873 In hw-native-sys/pypto-lib;[Docs] PyPTO 性能优化案例集:Qwen3-14B Decode 语法图鉴与 DeepSeek-V4 Swimlane 调优实录
documentationImprovements or additions to documentationImprovements or additions to documentationStatus: Open.#828 In hw-native-sys/pypto-lib;- Status: Open.#798 In hw-native-sys/pypto-lib;
- Status: Open.#766 In hw-native-sys/pypto-lib;
[Feature] Add serving contract support for lib-serving decoupling
enhancementNew feature or requestNew feature or requestStatus: Open.#752 In hw-native-sys/pypto-lib;[Feature] DSv4 routed/shared experts: adopt MXFP4 weights to halve HBM bandwidth
enhancementNew feature or requestNew feature or requestStatus: Open.#723 In hw-native-sys/pypto-lib;- Status: Open.#713 In hw-native-sys/pypto-lib;
[Performance] DeepSeek V4 hc_pre Sinkhorn layout underutilizes vector sublanes
enhancementNew feature or requestNew feature or requestStatus: Open.#700 In hw-native-sys/pypto-lib;- Status: Open.#696 In hw-native-sys/pypto-lib;
- Status: Open.#665 In hw-native-sys/pypto-lib;
[Profiling Report] Quantifying the pypto ↔ CCE paged-attention gap (follow-up to the 6× in simpler/pull: #986)
enhancementNew feature or requestNew feature or requestStatus: Open.#622 In hw-native-sys/pypto-lib;