Uplift third_party/tt-metal to a6ec1c0a02c3b9dc55e769618639e0439ef9c06b 2025-01-13

[vmilosevic]

This PR uplifts the third_party/tt-metal to the a6ec1c0a02c3b9dc55e769618639e0439ef9c06b

• This brings 1D tensor fix for forge fails blocking uplifts recently, and increased matmul precision fixes in some caes
• Remove halo=false on ShardSpec, was removed in tt-metal at df59448e

[kmabeeTT]

Not clean - this uplift to tt-metal at ca2c8677eb1ff3e2531c030070fa28554edb382c brought 35 tt-metal commits:

git l 9c5a0b2e0a0edba747613d2afb3e56dacb5776ac..ca2c8677eb1ff3e2531c030070fa28554edb382c
2025-01-10 ca2c8677eb by GitHub (Author [email protected]) : Revert "#16621: Add barriers at end of cq_dispatch_slave.cpp" (#16645)
2025-01-10 93cba40d4e by John Bauman (Author [email protected]) : #16503: Optimize CoreRangeSets for CBs and semaphores
2025-01-10 b4709be227 by Mo Memarian (Author [email protected]) : #16590: profiler trace detection fix
2025-01-10 84fef0a65e by GitHub (Author [email protected]) : Build wheels in ttnn unit tests workflow because the tests need it and we forgot to put it in [skip ci] (#16605)
2025-01-10 c20ec92da9 by GitHub (Author [email protected]) : Support subcoregrids in concat_heads (#16223)
2025-01-10 aa494d461f by Mo Memarian (Author [email protected]) : #0: Fix T3K profiler CI
2025-01-10 a55ce407a6 by Mo Memarian (Author [email protected]) : #10234: Ethernet dispatch profiling tests
2025-01-10 243dd549c9 by Mo Memarian (Author [email protected]) : #10234: Ethernet dispatch profiling support
2025-01-10 43e641da1b by GitHub (Author [email protected]) : Build wheels in models unit tests workflow (#16615)
2025-01-10 ea4e35184f by John Bauman (Author [email protected]) : #16621: Add barriers at end of cq_dispatch_slave.cpp
2025-01-10 652490d827 by GitHub (Author [email protected]) : #16366: Changed default kernal_config_val for 32bit matmul (#16567)
2025-01-10 f479c4f0e1 by Brian Liu (Author [email protected]) : #16312: Fix full op to properly handle 1D input shapes - Fix full op to query physical shape for buffer volume - Fix ttnn.to_torch to handle 1D tensors   * Add logical_shape to tensor attributes (in order to query original tensor rank)   * Unsqueeze to original tensor rank after stripping padding - Add unit tests for ttnn.full op   * Copy and rename existing test_full.py to test_moreh_full.py   * Update test_full.py to test ttnn.full op instead   * Add 1D input shape to test ttnn.full op
2025-01-10 4a4150b46d by GitHub (Author [email protected]) : [tt-train] Add nanogpt ddp mode (#16614)
2025-01-10 182627ae57 by GitHub (Author [email protected]) : Remove ARCH_NAME from host library code (#16616)
2025-01-10 a6789b1d13 by GitHub (Author [email protected]) : Composite binary sweeps: gcd and lcm (#16423)
2025-01-10 f627ca8514 by GitHub (Author [email protected]) : lower fabric erisc datamover eth context switching frequency when workload is running (#16610)
2025-01-10 0ad9bd16a1 by GitHub (Author [email protected]) : Reorganize Print Pages Infrastructure (#16463)
2025-01-10 eba0c3a5ca by Almeet Bhullar (Author [email protected]) : Skipping over active BH eth cores when assigning idle eth cores for dispatch
2025-01-10 b68374ee20 by Almeet Bhullar (Author [email protected]) : Add missing bank arrays in idle erisc1 fw
2025-01-10 cf82e3d7bf by Almeet Bhullar (Author [email protected]) : Update eth kernel hash to include processor
2025-01-10 afaa94ea1a by Almeet Bhullar (Author [email protected]) : Update waypoint test to account for both eriscs on BH
2025-01-10 37526a9d95 by Almeet Bhullar (Author [email protected]) : Allow Host Watcher to allow writing to erisc pc registers on BH
2025-01-10 931f691217 by Almeet Bhullar (Author [email protected]) : Add address and value that will launch fw on a programmable core into the HalJitBuildConfig
2025-01-10 a686bf93af by GitHub (Author [email protected]) : Enable multi-core and fixing bfloat8 for untilize with unpadding  (#16555)
2025-01-10 5a0f881d5e by GitHub (Author [email protected]) : Update CODEOWNERS (#16604)
2025-01-10 e292cd62da by GitHub (Author [email protected]) : #11512: Refactor bitwise sweeps, add bitwise sharded sweeps, modify t<E2><80><A6> (#15704)
2025-01-10 8393ab96df by GitHub (Author [email protected]) : Resubmit #16339: parameterize dispatch_constants (#16478)
2025-01-10 4549ef0f63 by GitHub (Author [email protected]) : Remove noc_parameters.h inclusion from ttnn (#16593)
2025-01-10 700a52340b by Keerthana (Author [email protected]) : #6344: Port RoBERTa model to n300
2025-01-10 ae6180278f by Austin Ho (Author [email protected]) : #16492: Remove sub_device_ids apis from various read/write functions throughout the stack
2025-01-10 8bfef1387d by Raymond Kim (Author [email protected]) : #0: Revert "Remove halo from shard spec (#15900)"
2025-01-10 4af7ff5380 by GitHub (Author [email protected]) : Address issues of  var & std (#16545)
2025-01-10 ab2e869f26 by GitHub (Author [email protected]) : Remove halo from shard spec (#15900)
2025-01-10 786664210d by Pavle Josipovi<C4><87> (Author [email protected]) : Short list failing conv2d for forge sweeps
2025-01-10 b2912feff4 by GitHub (Author [email protected]) : Fix ttnn.from_torch for 0D/1D tensors with tile layout (#16484)

Failing a test in tt-torch (https://github.com/tenstorrent/tt-torch/actions/runs/12722170639/job/35466363484):

FAILED tests/torch/test_basic.py::test_div_zero - AssertionError: PCC of output 0: nan, threshold: 0.99 ❌
ATOL of output 0: nan, threshold: 0.01 ❌
= 1 failed, 149 passed, 208 skipped, 5 xfailed, 1 xpassed, 4 warnings in 154.18s (0:02:34) =

And bunch in tt-forge-fe (https://github.com/tenstorrent/tt-forge-fe/actions/runs/12722170869) - here are 9 fails in runner1 job (2 xpass):

2025-01-11T07:43:18.0002412Z ℹ️ - TT-Forge-FE Tests - 692 tests run, 417 passed, 266 skipped, 9 failed.
2025-01-11T07:43:18.0004075Z    🧪 - forge/test/mlir/test_ops.py | ValueError: Data mismatch -> AutomaticValueChecker (compare_with_golden): framework_model=tensor([0.49626, 0.76822, 0.08848, 0.13203]), compiled_model=tensor([0.49626, 0.00000, 0.00000, 0.00000])
2025-01-11T07:43:18.0010562Z    🧪 - forge/test/mlir/test_ops.py | ValueError: Data mismatch -> AutomaticValueChecker (compare_with_golden): framework_model=tensor([-0.70066, -0.26368, -2.42501, -2.02472, -1.17953, -0.45558, -0.71316, -0.10932, -0.78608, -0.45838, -1.05299, -0.91201, -3.80201, -1.77869]), compiled_model=tensor([-0.71420,     -inf,     -inf,     -inf,     -inf,     -inf,     -inf,     -inf,     -inf,     -inf,     -inf,     -inf,     -inf,     -inf])
2025-01-11T07:43:18.0015514Z    🧪 - forge/test/mlir/test_ops.py | ValueError: Data mismatch -> AutomaticValueChecker (compare_with_golden): framework_model=tensor([-7.00662e-01, -2.63677e-01, -2.42501e+00, -2.02472e+00, -1.17953e+00, -4.55582e-01, -7.13159e-01, -1.09319e-01, -7.86079e-01, -4.58381e-01, -1.05299e+00, -9.12007e-01, -3.80201e+00, -1.77869e+00, -1.22456e+00, -6.56773e-01, -3.60013e-01, -2.23129e-01, -1.82617e+00, -1.26490e+00, -3.83300e-01, -8.86192e-02,
2025-01-11T07:43:18.0019642Z    🧪 - forge/test/mlir/test_ops.py | ValueError: Data mismatch -> AutomaticValueChecker (compare_with_golden): framework_model=tensor([-0.70066, -0.26368, -2.42501, -2.02472, -1.17953, -0.45558, -0.71316]), compiled_model=tensor([-0.71420,     -inf,     -inf,     -inf,     -inf,     -inf,     -inf])
2025-01-11T07:43:18.0025761Z    🧪 - forge/test/mlir/test_ops.py | ValueError: Data mismatch -> AutomaticValueChecker (compare_with_golden): framework_model=tensor([1.67713, 1.15178, 5.30568, 4.30978, 2.45445, 1.36225, 1.69465, 1.00357, 1.79977, 1.36555, 2.22770, 1.99315, 8.17489, 3.71942, 2.53879, 1.61676, 1.25365, 1.11110, 3.83099, 2.61596, 1.27941, 0.98503, 2.01167, 1.02653, 1.92527, 1.53161, 0.94990, 7.34404, 3.50593, 2.11230, 2.46853, 0.96899]), compiled_model=tensor([1.67713e+00, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38, 1.70135e+38,
2025-01-11T07:43:18.0030661Z    🧪 - forge/test/mlir/test_ops.py | ValueError: Data mismatch -> AutomaticValueChecker (compare_with_golden): framework_model=tensor([0.62158, 0.68314, 0.52210, 0.53296, 0.57626, 0.65341, 0.62013]), compiled_model=tensor([0.61364, 0.50000, 0.50000, 0.50000, 0.50000, 0.50000, 0.50000])
2025-01-11T07:43:18.0032418Z    🧪 - forge/test/mlir/llama/tests/test_specific_ops_llama32.py | [XPASS(strict)] pcc ~ 0.65
2025-01-11T07:43:18.0033570Z    🧪 - forge/test/mlir/mnist/training/test_training.py | assert False
2025-01-11T07:43:18.0035061Z    🧪 - forge/test/mlir/resnet/test_resnet_unique_ops.py | [XPASS(strict)] Tensor mismatch. PCC = 0.9425581505871167, but required = 0.99. Tracking on: https://github.com/tenstorrent/tt-mlir/issues/1576
2025-01-11T07:43:18.0060707Z ℹ️ - TT-Forge-FE Tests - Creating check (Annotations: 9)

quick look at tt-metal tree shows they had bad day yesterday with lots of regressions. The latest commit when this job was kicked off still had fails. There was one commit that went in few hours later at tenstorrent/tt-metal@323a5d7 that might be worth picking up, it improves their pass rate. In the interest of low effort debug, going to push that to this branch to kickoff CI (here and downstream) again and cross fingers.

[kmabeeTT]

Same results with that run. Several bisects on CI in the background show all the new fails are due to this commit which happened to be the very first commit in the range.

2025-01-10 b2912feff4 by GitHub (Author [email protected]) : Fix ttnn.from_torch for 0D/1D tensors with tile layout (#16484)

I pinged Stas in slack metal pipelines thread, but probably need someone to help followup.

[sdjordjevicTT]

Managed to reproduce the failure with the following TTNN test on latest tt-metal main:

import torch
import ttnn

with ttnn.manage_device(0) as device:
    torch_input = torch.rand(14)
    torch_output = torch.log(torch_input)

    ttnn_input = ttnn.from_torch(torch_input, dtype=ttnn.float32)
    ttnn_input = ttnn.to_layout(ttnn_input, ttnn.TILE_LAYOUT)
    ttnn_input = ttnn.to_device(ttnn_input, device, memory_config=ttnn.DRAM_MEMORY_CONFIG)

    ttnn_output = ttnn.log(ttnn_input)

    ttnn_output = ttnn.from_device(ttnn_output)
    ttnn_output = ttnn.to_layout(ttnn_output, ttnn.ROW_MAJOR_LAYOUT)
    ttnn_output = ttnn.to_torch(ttnn_output)

    print("Torch Output: ")
    print(torch_output)
    print("TTNN Output: ")
    print(ttnn_output)

    torch.testing.assert_close(ttnn_output, torch_output, rtol=1e-5, atol=1e-5)

The failure log:

2025-01-13 11:32:06.525 | INFO     | SiliconDriver   - Opened PCI device 1; KMD version: 1.31.0, IOMMU: disabled
2025-01-13 11:32:06.526 | INFO     | SiliconDriver   - Detected PCI devices: [1]
2025-01-13 11:32:06.526 | INFO     | SiliconDriver   - Using local chip ids: {0} and remote chip ids {1}
2025-01-13 11:32:06.535 | INFO     | SiliconDriver   - Software version 6.0.0, Ethernet FW version 6.10.0 (Device 0)
2025-01-13 11:32:06.536 | INFO     | SiliconDriver   - Software version 6.0.0, Ethernet FW version 6.10.0 (Device 1)
                  Metal | INFO     | Initializing device 0. Program cache is NOT enabled
                  Metal | INFO     | AI CLK for device 0 is:   1000 MHz
                  Metal | WARNING  | Unable to bind worker thread to CPU Core. May see performance degradation. Error Code: 22
Torch Output: 
tensor([-0.4576, -0.2158, -2.0279, -1.0105, -1.3695, -1.0895, -0.9112, -0.1052,
        -0.5727, -1.2189, -1.0405, -0.0320, -2.3473, -1.0702])
TTNN Output: 
TorchTensor([-0.4600,    -inf,    -inf,    -inf,    -inf,    -inf,    -inf,
                -inf,    -inf,    -inf,    -inf,    -inf,    -inf,    -inf])
                  Metal | INFO     | Closing device 0
                  Metal | INFO     | Disabling and clearing program cache on device 0
Traceback (most recent call last):
  File "/localdev/sdjordjevic/src/tt-metal/test.py", line 23, in <module>
    torch.testing.assert_close(ttnn_output, torch_output, rtol=1e-5, atol=1e-5)
  File "/localdev/sdjordjevic/src/tt-metal/python_env/lib/python3.10/site-packages/torch/testing/_comparison.py", line 1520, in assert_close
    raise error_metas[0].to_error(msg)
AssertionError: Tensor-likes are not close!

Mismatched elements: 14 / 14 (100.0%)
Greatest absolute difference: inf at index (1,) (up to 1e-05 allowed)
Greatest relative difference: inf at index (1,) (up to 1e-05 allowed)
                 Device | INFO     | Closing user mode device drivers

[sdjordjevicTT]

Created the following issue on the metal side to track the issue:
tenstorrent/tt-metal#16657

[kmabeeTT]

Going to merge. Pulled in Stas' bug fix, which brought some more small issues (ShardSpec interface changed to remove halo, made updates for it here).

• It passes tt-mlir
• passes tt-torch (link)
• This brings 2 xpass fails test_specific_ops_llama32.test_matmul[shapes6] and test_resnet_unique_ops.test_matmul_resnet[1-1000-2048] in tt-forge-fe CI (link) for increased PCC (nice), but brings one slight failure where training test test_training.test_mnist_training_with_grad_accumulation fails with rtol 10.03% which exceeds threshold 10% (we expected it to drop with tt-metal fix tenstorrent/tt-metal@652490d, not increase, so it needs some debug). Borys is suggesting we try to force Hifi4 data-format as experiment, and they would consider making it happen automatically for float32 data format if we like it and it helps.

Since we are falling behind in tt-metal, and above is only fail, I have to force this through, will flag to folks in slack.

This brings 70 tt-metal commits now.

git l 9c5a0b2e0a0edba747613d2afb3e56dacb5776ac..a6ec1c0a02c3b9dc55e769618639e0439ef9c06b
2025-01-13 a6ec1c0a02 by GitHub (Author [email protected]) : #16657: Fix to_layout conversion into row major for 1D tensors (#16684)
2025-01-13 ca145b4b99 by GitHub (Author [email protected]) : Update README.md (#16702)
2025-01-13 8c9694509d by GitHub (Author [email protected]) : [skip ci] Update perf and latest features for llm models (Jan 13) (#16677)
2025-01-13 f9345aa3f1 by GitHub (Author [email protected]) : #0: Clean up confusing refs to Greyskull from ttnn.copy error messages. (#16647)
2025-01-13 30f0824cfb by Paul Keller (Author [email protected]) : Tweak BH csrrs init code
2025-01-13 520c1dce50 by GitHub (Author [email protected]) : [skip ci] Update .clang-format-ignore (#16681)
2025-01-13 07aa1881b9 by Tapasvi Patel (Author [email protected]) : #16367: Added support to enable dram and l1 memory collection without saving to disk
2025-01-13 e5f21f6c61 by GitHub (Author [email protected]) : Update README.md (#16676)
2025-01-13 549669a8fb by GitHub (Author [email protected]) : Create README.md (#16675)
2025-01-13 ac0481238d by Patrick Roberts (Author [email protected]) : #16353: Allow device ops to skip launch
2025-01-13 f3574912d0 by Patrick Roberts (Author [email protected]) : #16353: Skipping the operation if the output tensor has not volume
2025-01-13 b028321387 by GitHub (Author [email protected]) : Fix Pre-allgather Layernorm bad PCC when use 1D reduction (#16622)
2025-01-13 919cf54686 by GitHub (Author [email protected]) : Fix resnet large on GS (#16665)
2025-01-13 7f292010c0 by GitHub (Author [email protected]) : Read from and write to partial buffer regions for interleaved buffers where offset and size of specified buffer region are divisible by buffer page size (#16102)
2025-01-13 06f15a18cd by GitHub (Author [email protected]) : [skip ci] Update subdevice doc (#16669)
2025-01-13 3738051226 by Paul Keller (Author [email protected]) : Stress NOC mcast test
2025-01-13 9162a0cec4 by GitHub (Author [email protected]) : #16495: reduce grid for falcon7b mlp matmul (#16569)
2025-01-13 80623a428e by Raymond Kim (Author [email protected]) : [skip ci] Put in missed is_grayskull import
2025-01-13 01053788fe by Pavle Josipovi<C4><87> (Author [email protected]) : Logical sharding for input tensor and halo output
2025-01-13 df59448e54 by Raymond Kim (Author [email protected]) : Re-landing: Revert "#0: Revert "Remove halo from shard spec (#15900)""
2025-01-13 ceb03b8631 by Raymond Kim (Author [email protected]) : #0: [skip ci] Skip resnet50 and resnet50 xlarge ttnn integration tests because it fails because of a recent re-calculation change
2025-01-13 649b795fae by Raymond Kim (Author [email protected]) : #0: Use tenstorrent docker run action for workflows that missed the initial change as now our builders are re-pulling
2025-01-13 42616404a1 by GitHub (Author [email protected]) : Fix build break (#16656)
2025-01-13 f69fbc0a08 by GitHub (Author [email protected]) : #15246: Add sweep tests for addcdiv, addcmul, rdiv, rsub, ceil (#15998)
2025-01-12 e258476ec5 by GitHub (Author [email protected]) : Port all data movements ops to compute_output_specs (#16652)
2025-01-12 3ab3739f10 by GitHub (Author [email protected]) : Enable to/from torch tests for 0D/1D tensors (#16653)
2025-01-12 6b4ee69bc4 by GitHub (Author [email protected]) : #16443: Add a programming example of vecadd_multi_core and gtest (#16446)
2025-01-12 1a7e545e01 by GitHub (Author [email protected]) : Port all experimental ops to compute_output_specs (#16595)
2025-01-12 35c7145ace by GitHub (Author [email protected]) : #12253: Implement Batch norm operation for inference mode (#16432)
2025-01-12 880f700437 by GitHub (Author [email protected]) : #15088: Create Infrastructure to exactly calculate L1 Memory Usage for Conv2D (#15455)
2025-01-11 b19de42b0b by GitHub (Author [email protected]) : support i/p tensors of all dimensions/rank for prod operation (#16301)
2025-01-11 b68365f3ed by GitHub (Author [email protected]) : fix reduce scatter multi-link support bug (#16636)
2025-01-11 4d8660a030 by GitHub (Author [email protected]) : Sweeps: fixed abs, added acos and acosh sharded and non sharded (#16381)
2025-01-11 195e74a586 by GitHub (Author [email protected]) : #0: Used github team for conv files (#16563)
2025-01-11 323a5d786d by GitHub (Author [email protected]) : Fix nightly stable diffusion tests (#16629)
2025-01-10 ca2c8677eb by GitHub (Author [email protected]) : Revert "#16621: Add barriers at end of cq_dispatch_slave.cpp" (#16645)
2025-01-10 93cba40d4e by John Bauman (Author [email protected]) : #16503: Optimize CoreRangeSets for CBs and semaphores
2025-01-10 b4709be227 by Mo Memarian (Author [email protected]) : #16590: profiler trace detection fix
2025-01-10 84fef0a65e by GitHub (Author [email protected]) : Build wheels in ttnn unit tests workflow because the tests need it and we forgot to put it in [skip ci] (#16605)
2025-01-10 c20ec92da9 by GitHub (Author [email protected]) : Support subcoregrids in concat_heads (#16223)
2025-01-10 aa494d461f by Mo Memarian (Author [email protected]) : #0: Fix T3K profiler CI
2025-01-10 a55ce407a6 by Mo Memarian (Author [email protected]) : #10234: Ethernet dispatch profiling tests
2025-01-10 243dd549c9 by Mo Memarian (Author [email protected]) : #10234: Ethernet dispatch profiling support
2025-01-10 43e641da1b by GitHub (Author [email protected]) : Build wheels in models unit tests workflow (#16615)
2025-01-10 ea4e35184f by John Bauman (Author [email protected]) : #16621: Add barriers at end of cq_dispatch_slave.cpp
2025-01-10 652490d827 by GitHub (Author [email protected]) : #16366: Changed default kernal_config_val for 32bit matmul (#16567)
2025-01-10 f479c4f0e1 by Brian Liu (Author [email protected]) : #16312: Fix full op to properly handle 1D input shapes - Fix full op to query physical shape for buffer volume - Fix ttnn.to_torch to handle 1D tensors   * Add logical_shape to tensor attributes (in order to query original tensor rank)   * Unsqueeze to original tensor rank after stripping padding - Add unit tests for ttnn.full op   * Copy and rename existing test_full.py to test_moreh_full.py   * Update test_full.py to test ttnn.full op instead   * Add 1D input shape to test ttnn.full op
2025-01-10 4a4150b46d by GitHub (Author [email protected]) : [tt-train] Add nanogpt ddp mode (#16614)
2025-01-10 182627ae57 by GitHub (Author [email protected]) : Remove ARCH_NAME from host library code (#16616)
2025-01-10 a6789b1d13 by GitHub (Author [email protected]) : Composite binary sweeps: gcd and lcm (#16423)
2025-01-10 f627ca8514 by GitHub (Author [email protected]) : lower fabric erisc datamover eth context switching frequency when workload is running (#16610)
2025-01-10 0ad9bd16a1 by GitHub (Author [email protected]) : Reorganize Print Pages Infrastructure (#16463)
2025-01-10 eba0c3a5ca by Almeet Bhullar (Author [email protected]) : Skipping over active BH eth cores when assigning idle eth cores for dispatch
2025-01-10 b68374ee20 by Almeet Bhullar (Author [email protected]) : Add missing bank arrays in idle erisc1 fw
2025-01-10 cf82e3d7bf by Almeet Bhullar (Author [email protected]) : Update eth kernel hash to include processor
2025-01-10 afaa94ea1a by Almeet Bhullar (Author [email protected]) : Update waypoint test to account for both eriscs on BH
2025-01-10 37526a9d95 by Almeet Bhullar (Author [email protected]) : Allow Host Watcher to allow writing to erisc pc registers on BH
2025-01-10 931f691217 by Almeet Bhullar (Author [email protected]) : Add address and value that will launch fw on a programmable core into the HalJitBuildConfig
2025-01-10 a686bf93af by GitHub (Author [email protected]) : Enable multi-core and fixing bfloat8 for untilize with unpadding  (#16555)
2025-01-10 5a0f881d5e by GitHub (Author [email protected]) : Update CODEOWNERS (#16604)
2025-01-10 e292cd62da by GitHub (Author [email protected]) : #11512: Refactor bitwise sweeps, add bitwise sharded sweeps, modify t<E2><80><A6> (#15704)
2025-01-10 8393ab96df by GitHub (Author [email protected]) : Resubmit #16339: parameterize dispatch_constants (#16478)
2025-01-10 4549ef0f63 by GitHub (Author [email protected]) : Remove noc_parameters.h inclusion from ttnn (#16593)
2025-01-10 700a52340b by Keerthana (Author [email protected]) : #6344: Port RoBERTa model to n300
2025-01-10 ae6180278f by Austin Ho (Author [email protected]) : #16492: Remove sub_device_ids apis from various read/write functions throughout the stack
2025-01-10 8bfef1387d by Raymond Kim (Author [email protected]) : #0: Revert "Remove halo from shard spec (#15900)"
2025-01-10 4af7ff5380 by GitHub (Author [email protected]) : Address issues of  var & std (#16545)
2025-01-10 ab2e869f26 by GitHub (Author [email protected]) : Remove halo from shard spec (#15900)
2025-01-10 786664210d by Pavle Josipovi<C4><87> (Author [email protected]) : Short list failing conv2d for forge sweeps
2025-01-10 b2912feff4 by GitHub (Author [email protected]) : Fix ttnn.from_torch for 0D/1D tensors with tile layout (#16484)