Repository navigation

Tmp - #24226

Closed

Erik-Lundell wants to merge 25 commits into

pytorch:codex/arm-docgen-github-checkfrom

Erik-Lundell:tmp

Closed

Tmp#24226
Erik-Lundell wants to merge 25 commits into
pytorch:codex/arm-docgen-github-checkfrom
Erik-Lundell:tmp

Commits on Oct 28, 2026
- Let a backend share one delegate across calls of the same lowered module (#24127)
  
  shoumikhin
  authored and
  Erik-Lundell
  committed
  b19d241
- [ET-VK][embedding] Support per-row Q4 scales
  
  SS-JIA
  authored and
  Erik-Lundell
  committed
  89d55e7
- [ET-VK][SDPA] Preserve shared KV cache semantics
  
  SS-JIA
  authored and
  Erik-Lundell
  committed
  f2e0a43
- [ET-VK][gemma4] Add ETDump profiling
  
  SS-JIA
  authored and
  Erik-Lundell
  committed
  5ea48f4
- Fix device placement of constants, shared inputs and delegate outputs (#24088)
  
  shoumikhin
  authored and
  Erik-Lundell
  committed
  bb28668
- Update the TensorRT backend docs page (#24136)
  
  shoumikhin
  authored and
  Erik-Lundell
  committed
  3448326
- Add CMake CI for native runtime, GGML, and Vulkan (#23930)
  
  JakeStevens
  authored and
  Erik-Lundell
  committed
  eed639a
- Support integer and boolean CPU tensors (#24191)
  
  JakeStevens
  authored and
  Erik-Lundell
  committed
  cc7bfb9
- Add shared linear-batch-norm fusion pass (#24192)
  
  JakeStevens
  authored and
  Erik-Lundell
  committed
  d272337
- Serialize only a view constant's own bytes in XNNPACK and Vulkan (#24181)
  
  shoumikhin
  authored and
  Erik-Lundell
  committed
  5f08567
- Fold evaluation BatchNorm during Native AOT export (#24068)
  
  JakeStevens
  authored and
  Erik-Lundell
  committed
  deb3225
- Add torchvision export workflow to `CpuBackend` (#23934)
  
  JakeStevens
  authored and
  Erik-Lundell
  committed
  b388c7f
- Wire batch-norm fusion into CpuBackend.preprocess (#24193) (#24193)
  
  JakeStevens
  authored and
  Erik-Lundell
  committed
  3f7b9a6
- Route _neg_view_copy to ET neg kernel (#24194) (#24194)
  
  JakeStevens
  authored and
  Erik-Lundell
  committed
  619574a
- Qualcomm AI Engine Direct - Tag multimodal encoder quantized I/O as uint16 (#22924)
  
  DannyYuyang-quic
  authored and
  Erik-Lundell
  committed
  7b23ff6
- Remove duplicate fuse_batch_norm_with_linear target (#24213)
  
  JakeStevens
  authored and
  Erik-Lundell
  committed
  5d31b25
- Qualcomm AI Engine Direct - Keep a non-delegated op's weight dequantize on CPU (#23915)
  
  msluszniak
  authored and
  Erik-Lundell
  committed
  da02aad
- Migrate runtime consumers to `std::string_view` (#24139)
  
  r-barnes
  authored and
  Erik-Lundell
  committed
  809f731
- Accept empty tensor bindings in CPUBackend execute (#24195)
  
  JakeStevens
  authored and
  Erik-Lundell
  committed
  6caa2ab
- Add CMake targets for HF LLM runners (#23899)
  
  JakeStevens
  authored and
  Erik-Lundell
  committed
  be9681c
- Namespace immutable constants per method in `to_native` (#23891)
  
  JakeStevens
  authored and
  Erik-Lundell
  committed
  0617f2d
- Add Method.set_output and write unplanned input mutations in place in pybindings (#23183)
  
  authored and
  Erik-Lundell
  committed
  8dc3b94
- Drop the "newer than 2.15" marks from the TensorRT backend page (#24214)
  
  shoumikhin
  authored and
  Erik-Lundell
  committed
  a43171b
- Fix two crashes caused by constant_prop_pass (#24182)
  
  shoumikhin
  authored and
  Erik-Lundell
  committed
  d3ea0c9
- TEST commit
  Erik-Lundell
  committed
  a492daf

Commits

Commits on Oct 28, 2026