fix: specify T4 GPU in kernel metadata

[nvandessel]

Summary

• Add machine_shape: "Gpu T4 x2" to kernel-metadata.json
• Default Gpu gives P100 (CUDA 6.0) which is incompatible with PyTorch/unsloth (requires 7.0+)

Test plan

• Previous run confirmed P100 assignment and CUDA error
• Next run with T4 specified should succeed

🤖 Generated with Claude Code

[greptile-apps]

Greptile Summary

This PR fixes a GPU compatibility issue in the Kaggle kernel metadata by explicitly specifying NvidiaTeslaT4Highmem as the machine_shape, replacing the implicit default which assigns a P100 GPU (CUDA compute capability 6.0). PyTorch and unsloth require compute capability 7.0+, so this change is necessary for the notebook to run correctly.

• Adds "machine_shape": "NvidiaTeslaT4Highmem" to notebooks/kernel-metadata.json
• T4 GPU has CUDA compute capability 7.5, satisfying the PyTorch/unsloth >= 7.0 requirement
• Note: the PR description references "Gpu T4 x2" (dual-T4 accelerator string) while the actual committed value is "NvidiaTeslaT4Highmem" (single T4 high-memory variant) — both are T4-based and meet the CUDA requirement, but they differ in the number of GPUs provided; the committed value appears correct for Kaggle's API

Confidence Score: 5/5

Safe to merge — single-field config change with a clear, verified rationale

The change is minimal (one line added to a JSON config file), the reasoning is well-documented, and NvidiaTeslaT4Highmem is a known valid Kaggle machine_shape value that maps to a T4 GPU satisfying the CUDA 7.5 requirement. No code logic is affected.

No files require special attention

Important Files Changed

Filename	Overview
notebooks/kernel-metadata.json	Adds machine_shape 'NvidiaTeslaT4Highmem' to select T4 GPU (CUDA 7.5) over the default P100 (CUDA 6.0), fixing PyTorch/unsloth compatibility

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[kaggle kernel push] --> B{machine_shape set?}
    B -- No --> C[Default: P100 CUDA 6.0]
    B -- Yes: NvidiaTeslaT4Highmem --> D[T4 GPU CUDA 7.5]
    C --> E[CUDA Error: unsloth requires >= 7.0]
    D --> F[PyTorch / unsloth compatible ✓]

Loading

_{Reviews (1): Last reviewed commit: "fix: specify T4 GPU in kernel metadata" | Re-trigger Greptile}