Daily edition · AI Tools
TILens turns technical updates into a focused daily brief: official releases, trusted reporting, and practitioner analysis, deduplicated and organized by topic.
port setup-build.ps1 to setup_sdk.py, to facilitate installation of Hexagon and OpenCL SDKs on Windows rename setup_sdk.py -> setup-sdk.py flake8 fix: print() -> logger.info() Co-authored-by: Kristopher Urquhart…
What's Changed Ollama's app now follows the system appearance again, restoring dark mode support Fixed the macOS app to properly hand off to an already-running instance instead of starting a second one The Claude…
Summary Integrates the torch.distributed._watchdog timeout machinery into symmetric memory's blocking and stream-based operations so that hangs in multi-process rendezvous or one-sided NCCL ops are detected and surfaced…
bench: add --tensor-read-lazy (#27881) bench: add --tensor-read-lazy rm the alias rename to LLAMA_LAZY_MODE_* Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/43739963…
Set USE_CMAKE_INSTRUMENTATION before including cmake/Dependencies.cma…
Address the Python wheel tree by name, not through the install prefix…
Alpha preview of langchain.mcp — a first-party adapter that turns any MCP server into LangChain tools you can hand straight to create_agent. Connection handling is FastMCP's, so its client features are available as-is…
Pull Request resolved: #195117 Approved by: https://github.com/Skylion007
Per-provider model catalog files. Updated weekly by CI.
Summary Fixes #84886. There was a race in the post-local SGD checkpoint reload test. Rank 0 created the temporary checkpoint and all ranks synchronized after it was written. However, there was no synchronization after…
This PR, in short, does: expose and publicize functional APIs (they already existed) document this more uniformly add big examples in optim.md for using them (and also step hooks), notably a bf16 adamw example. Pull…
The CK Tile FMHA generator already compiles working kernels for gfx12 (--targets=gfx1201 passes through to the kernel template), but two guards prevented them from being used on RDNA4 hardware: launch_kernel_pt.hpp:…
Fix Inductor codegen for user-defined Triton kernels that pass non-builtin Python objects as tl.constexpr values. When such constants are serialized with constructor-style repr output, for example…
Issue Fixes #193933 Summary Dynamo models a tensor subclass without __torch_dispatch__ as a TensorWithTFOverrideVariable holding the Python type in class_type, while the fake tensor behind the proxy stays a plain…
Fixes #194442. baddbmm/addbmm/addmm document that when beta == 0 the input/bias is ignored, so nan/inf in it must not propagate. On MPS, real dtypes already honor this (the MPSGraph path drops the bias node when beta ==…
Fixes #194442. baddbmm/addbmm/addmm document that when beta == 0 the input/bias is ignored, so nan/inf in it must not propagate. On MPS, real dtypes already honor this (the MPSGraph path drops the bias node when beta ==…
periodic.yml still asked for pytorch-linux-jammy-py3.10-clang18, which #191108 retired on 2026-07-27. docker-builds.yml never builds that name, so the tag is absent from ECR, the pod hangs in ImagePullBackOff, and ARC…
at::divup(x, y) computes (x + y - 1) / y. The addition overflows for large x or y, and the result then silently rounds down instead of up — divup(2, INT64_MAX) returns 0, not 1. Guarded with c10::add_overflows, falling…
[XPU][Test]Port device agnostic cases in 3 test files to Intel GPU (#…
Revert "Use C++20 instead of c10::llvm bit helpers in c10 (#187…
Showing 1 day · 50 items available
