test GB CI for Oceananigans bump

[Pangoraw]

No description provided.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 31.07%. Comparing base (36b215f) to head (fe01978).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2595      +/-   ##
==========================================
+ Coverage   28.68%   31.07%   +2.38%     
==========================================
  Files         225      225              
  Lines       44480    44512      +32     
==========================================
+ Hits        12760    13833    +1073     
+ Misses      31720    30679    -1041     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Pangoraw]

    metrics = Metrics(
        flops = 0.11671181737040262,  # [flops / (peak flops * program time)], capped at 1.0
        bandwidth_utils = [0.5740172565663073, 0.0, 0.0],
        uncapped_flops = 0.11671181737040262, 
        raw_time = 23.26307480s,  # Raw time in seconds
        raw_flops = 8.46749149821572e14,  # Total FLOPs performed
        bf16_flops = 8.46749149821572e14,  # Total FLOPs Normalized to the bf16 (default) devices peak bandwidth
        raw_bytes_accessed = [2.0767217595944e13, 0.0, 0.0],
        occurrences = 0,
        raw_flops_rate = 3.639884912468974e13,  # Raw FLOPs rate in FLOPs/seconds
        bf16_flops_rate = 3.639884912468974e13,  # BF16 FLOPs rate in FLOPs/seconds
    )

it's green again ✔️

[wsmoses]

what's the error for f64?

[Pangoraw]

f64 GPU was failing because of some CUDA error on the runner, it's good as well now.
TPU still suffers from the initialization problem that openxla/xla#45444 ought to fix 🤔

[wsmoses]

@Pangoraw I started a rerun [just merged a later libtpu to reactant, that contains its own xla and needs to be updated to accomodate updated xla]

[wsmoses]

@Pangoraw correctness here is restored tpu, seemingly?

[Pangoraw]

Correctness looks good, flops-wise not so much on tpu?

    metrics = Metrics(
        flops = 0.0006474319815451959,  # [flops / (peak flops * program time)], capped at 1.0
        bandwidth_utils = [0.804190223014755, 0.015916926572879797, 0.011573493297569806],
        uncapped_flops = 0.0006474319815451959, 
        raw_time = 27.28843157s,  # Raw time in seconds
        raw_flops = 1.672573072726e13,  # Total FLOPs performed
        bf16_flops = 1.672573072726e13,  # Total FLOPs Normalized to the bf16 (default) devices peak bandwidth
        raw_bytes_accessed = [3.5945906937972e13, 1.0118574451204e13, 5.093594964828e12],
        occurrences = 0,
        raw_flops_rate = 6.12923856928837e11,  # Raw FLOPs rate in FLOPs/seconds
        bf16_flops_rate = 6.12923856928837e11,  # BF16 FLOPs rate in FLOPs/seconds
    )