v0.53.1-rc8

Note

If you are installing from a release, please refer to the README, INSTALLATION instructions, and any other documentation packaged with the release, not on the main branch. There may be differences between the latest main and the previous release.

The changelog will now follow, showing the changes from last release.

This release was generated by the CI workflow https://github.com/tenstorrent/tt-metal/actions/runs/12131243392

🚀 Features

• #15540: support bfp8_b 1d tensor in ttnn
  • PR: #15541
• #15542: Add dtype to to_torch
  • PR: #15543

📦 Uncategorized

• Fix the test for whether to install the wheel, and also exit the script on the first error
  • PR: #15480
• Use the post-merge commit of UMD
  • PR: #15522
• [skip ci] Disable clang-format in pre-commit and move version ahead
  • PR: #15521
• Support for new matmul1d op with gather_in0
  • PR: #14964
• Remove dead includes that break without proper include paths
  • PR: #15530
• More formatting of ttnn
  • PR: #15519
• Make MEM_LOCAL_BASE accessible behind Hal
  • PR: #15315
• Add support for rank-n tensors to tilize and untilize
  • PR: #15520
• Add reduce scatter perf to tg
  • PR: #15160
• Fix Circular Buffer Allocation in untilize_with_halo
  • PR: #15492
• [skip ci] use git-clang-format in CI
  • PR: #15528
• #13676: i1 op kernel implementation and improve i0_bw pcc
  • PR: #15325
• Increase perf margin on unet
  • PR: #15532
• #12558: TTNN implementation of MNIST model
  • PR: #12647
• Remove unsupported shapes to make pipeline green
  • PR: #15531
• #13401: Add data parallel support for Bert-Tiny model
  • PR: #14033
• #15297: Allow MeshDevice to be initialized for chips without eth coordinates
  • PR: #15475
• #0: Disable clang-format precommit check once again due to errors
  • PR: #15556
• #15337: Fix incorrectly sized cb in remote cb microbenchmark
  • PR: #15506
• [skip ci] Update CONTRIBUTING.md with pre-commit info
  • PR: #15537
• Remove ClusterDescriptor path from constructor
  • PR: #15554
• Add performance and accuracy configurations to Llama 3
  • PR: #15545
• Disable upblock 3 and 4 unet unit tests
  • PR: #15568
• Re-enable git-clang-format for pre-commit again
  • PR: #15562
• Fix race condition in DRAM sharded MM
  • PR: #15569
• Fix to concat support for tensors with tile padding
  • PR: #15513
• Adjust perf test targets for Falcon7b-t3k-decode-noasync to account for CI instability
  • PR: #15573
• Format Subset of tests/ directory
  • PR: #15576
• [skip ci] Update git-blame-ignore-revs
  • PR: #15577
• #14982: Update Unary examples docs
  • PR: #15417
• fix graph_processor buffer_alloc_id calculus
  • PR: #15426
• #13415: Update left shift
  • PR: #15538
• #14982: Update Unary examples
  • PR: #15421
• #14982: Update Unary example and docs
  • PR: #15453
• #5893: Fix ops with low PCC
  • PR: #14652
• #13127: Clean up layout conversion
  • PR: #15393
• Sharded sweep tests
  • PR: #15246
• #0: Fix typos in ttnn docs
  • PR: #15517
• #0: Make sub-device bank ids for cores the same as global bank ids
  • PR: #15579
• #12151: Generalize max_pool2d code into generic pool op
  • PR: #15561
• Update INSTALLING.md
  • PR: #15512
• #13944: Prohibit copying memory objects
  • PR: #15434
• #15267: Merge erisc kernel data and bss sections
  • PR: #15457
• Update perf and latest features for llm models (Dec 2)
  • PR: #15600
• Update llms.md
  • PR: #15612
• #11795: Update pgm dispatch golden file
  • PR: #15614
• [tt-train] Add kahan summation in AdamW
  • PR: #15518
• Remove core_config.h from host code inclusion in dev_msgs.h
  • PR: #15584
• Add new Hal API "valid_reg_addr"
  • PR: #15559
• #0: Fix DPRINT_DATA macro bug
  • PR: #15617