Enable native Kev inference with the WebGPU backend
维护者通常 1 天内回复
@akdybala 已经在做这个了。
开始于 2026年9月25日。
评估
- 难度
- 5/5
- 预计耗时
- 一周以上
- 新手友好度
- 35/100
- Issue 类型
- 功能
- 描述清晰度
- 基本清楚
- 活跃度
- 活跃
调研方向
Start with the WebGPU backend guide and operator tests in the repository. Examine the existing C++ Module API and native Dawn build. Review the Kev example's prefill and score graphs, and the Qwen3.5 MoE model reference for GDN math. The work involves auditing operator support, adding WGSL kernels, and integrating with the kev_runner and kev_benchmark. Success is measured by matching logits and probabilities against the PyTorch implementation.
由索引模型根据 Issue 内容生成。
描述
Enable the native Kev example on the WebGPU backend, extending the XNNPACK and MLX support added in #23023.
Kev runs prefill followed by pointer-head scoring, with an immutable prefix snapshot reused across question batches. Start with the existing C++ Module API and a native Dawn build, without Python at inference.
Use the WebGPU backend guide and operator tests as starting points. Qwen3.5 MoE's GatedDeltaNet implementation provides a reference for GDN math and state layouts. Coordinate shared export/lowering work with the Vulkan issue, #23098.
Work to cover:
- Audit Kev's
prefillandscoregraphs against WebGPU runtime support. Add missing WGSL kernels and lowering support, including GDN, with focused backend regression tests. Export success alone is insufficient:WebGPUPartitionershares Vulkan's frontend, whose operator coverage is broader. - Add a WebGPU export option and CMake linkage for the existing
kev_runnerandkev_benchmark. Respect the current mutually exclusive Vulkan/WebGPU runtime registration when linking. - Preserve
system_one, explicitprefill/evaluate, configurable token limits, variable question/option counts, and batching beyond eight questions. Repeated evaluations must leave the prefix unchanged. - Start with an unquantized FP32 baseline, retaining the fitted temperature and checkpoint metadata. Compare logits, probabilities, and reused state against upstream Kev's PyTorch implementation and the existing FP32 path, with documented tolerances.
- Verify the backbone, including GDN, executes on WebGPU and report any CPU fallback. Document a tested adapter, device/buffer limits, export/build/run commands, and measurements using
kev_benchmark.
Keep the integration in the existing example and reuse the runner and benchmark. Document browser/WebAssembly constraints separately from the initial native Dawn support.
cc @iseeyuan @lucylq @helunwencser @tarun292 @kimishpatel @jackzhxng
- 主要语言
- Python
- 星标
- 5.1k
- 派生
- 1.2k
- 平均合并
- 2 天 14 分钟
- 30 天内合并 PR
- 704
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
pytorch/executorch 的其他 Issue
-
Portable remainder kernel: wrong sign for integers and float64 rounded to float可能已有人在做 @JakeStevens 于 4 天前认领。 未关闭bug module: kernels triaged
难度 2/5 1-3 小时 新手友好度 82/100
pytorch/executorch#23682 · 已指派 2 人 ·
维护者通常 1 天内回复
-
enhancement triaged
难度 2/5 半天 新手友好度 68/100
pytorch/executorch#21640 ·
维护者通常 1 天内回复
-
[QNN] 16a4w_block stride-2 conv + ReLU fails HTP graph prepare on SM8850 (Relu16bParamsPlaceholder)可能已有人在做 @psiddh 今天认领。 未关闭bug module: qnn partner: qualcomm
pytorch/executorch#23922 · 已指派 1 人 ·
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 40/100
pytorch/executorch#23913 ·
维护者通常 1 天内回复
-
bug module: microcontrollers partner: arm
难度 4/5 3-5 天 新手友好度 38/100
pytorch/executorch#23812 · 2 条评论 · 1 个 reaction ·
维护者通常 1 天内回复
查看 pytorch/executorch 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 72/100
PolicyEngine/policyengine-us#10262 ·
维护者通常 2 天内回复
-
难度 2/5 1-3 小时 新手友好度 74/100
kellyjonbrazil/jc#788 ·
维护者通常 2 天内回复
-
bug
难度 2/5 1-3 小时 新手友好度 72/100
维护者通常 1 天内回复
-
`/search.json` ignores the documented `editions.sort` parameter可能已有人在做 @PranavD2905 今天认领。 未关闭Affects: Server Good First Issue Module: Solr Needs: Triage Theme: Public APIs Theme: Search
难度 2/5 1-3 小时 新手友好度 74/100
internetarchive/openlibrary#13989 · 1 条评论 ·
维护者通常 1 天内回复
-
bug P2: misleading query
难度 2/5 1-3 小时 新手友好度 62/100
jeffknupp/association#478 ·