Topic - Edition
Anthropic’s best AI model struggles to attract users as cheaper tools thrive A few interesting numbers in this FT story gathered from "people with knowledge of the matter": Anthropic's "annualized revenue" for July is…
Prior to Fable, it felt silly to waste too much time improving your coding harness or context strategies. A new model would arrive at the same price (or cheaper!) and paper over most of your problems. But then Fable…
Follow-up to the recent typo-fix passes, covering comments and docstrings under torch/ that were missed: subject-verb agreement, misspellings such as "decret"/"cames"/"3th", its/it's and lose/loss confusions, "Pytorch"…
[Test] Refactor test_optim to improve test reuse and add `HardwareCla…
What will Artificial Intelligence look like in 2026? 🤖 AI is rapidly moving beyond simple chatbots and one-size-fits-all solutions. The next generation of AI is becoming more specialized, intelligent, secure, automated,…
This PR adds a regression test for the scan Autograd correctness bug reported in #153679. The reported problem was ultimately an Inductor-only issue that surfaced through scan: backward donated-buffer indices are…
This PR adds a regression test for the scan Autograd correctness bug reported in #153679. The reported problem was ultimately an Inductor-only issue that surfaced through scan: backward donated-buffer indices are…
The checked-in file is two lines saying the real content is generated by cmake. It is: cmake/Codegen.cmake:47 writes it to ${CMAKE_BINARY_DIR}, build.bzl:16 to the Bazel output tree, and only the generated one is…
[FSDP2] Guard _unsharded_param access in to_accumulated_grad_if_neede…
Impact: CI only (single flex_attention FLASH test) Risk: low What Broaden the assertRaisesRegex in test_captured_float_fails_with_dynamic to accept either of the two valid rejection messages a captured Python float can…
Impact: CI only (single flex_attention FLASH test) Risk: low What Broaden the assertRaisesRegex in test_captured_float_fails_with_dynamic to accept either of the two valid rejection messages a captured Python float can…
Highlights Ray Serve LLM: In this release we've completed KV cache and token aware request routing, which was previewed in 2.57. Tokenization now happens in-process on the LLMRouter ingress replica, the routing decision…
This PR is auto-generated nightly by this action. Update the pinned vision hash. Pull Request resolved: #194006 Approved by: https://github.com/pytorchbot
This PR is auto-generated nightly by this action. Update the pinned torchcomms hash. Pull Request resolved: #194475 Approved by: https://github.com/pytorchbot
This PR is auto-generated nightly by this action. Update the pinned vllm hash. Pull Request resolved: #194476 Approved by: https://github.com/pytorchbot
Sweep of pre-existing grammatical slips in comments and docstrings across inductor, the JIT frontend/passes/runtime, the C++ data API, FX, export, and the headeronly float8 helpers. Most are missing or misplaced…
Sweep of pre-existing grammatical slips in comments and docstrings across inductor, the JIT frontend/passes/runtime, the C++ data API, FX, export, and the headeronly float8 helpers. Most are missing or misplaced…
Sweep of small grammatical corrections across READMEs, tutorials, and Python docstrings under torch/: subject-verb agreement in docstring summary lines (Average ... and returns -> and return), possessive and plural…
Fix indefinite article and grammar typos in comments and docstrings (…
Corrects subject-verb agreement, past-participle forms ("ran" -> "run"), and a few wrong words ("Mobile" -> "Module", "pickable" -> "picklable", "number elements" -> "number of elements") across comments and docstrings.…
31 items available
