Add discretionary GPU hint to base instructions

datasets/tabred-real/cooking-time/instruction.md

@@ -63,6 +63,11 @@ variant with the BEST validation score — never simply the last one trained.
 Before writing the submission, state which variant won and print its
 `VALIDATION_SCORE=<value>`.
 
+## Hardware
+A CUDA GPU may be available (check `torch.cuda.is_available()`). Use it at your
+discretion — for deep models, moving the model and batches to the GPU (`.to('cuda')`)
+speeds up training considerably.
+
 ## Submission — REQUIRED
 Write predictions to **`/workspace/submission.csv`** with EXACTLY two columns:
 - `id` — copied from `test.csv`

datasets/tabred-real/delivery-eta/instruction.md

@@ -63,6 +63,11 @@ variant with the BEST validation score — never simply the last one trained.
 Before writing the submission, state which variant won and print its
 `VALIDATION_SCORE=<value>`.
 
+## Hardware
+A CUDA GPU may be available (check `torch.cuda.is_available()`). Use it at your
+discretion — for deep models, moving the model and batches to the GPU (`.to('cuda')`)
+speeds up training considerably.
+
 ## Submission — REQUIRED
 Write predictions to **`/workspace/submission.csv`** with EXACTLY two columns:
 - `id` — copied from `test.csv`

datasets/tabred-real/ecom-offers/instruction.md

@@ -63,6 +63,11 @@ variant with the BEST validation score — never simply the last one trained.
 Before writing the submission, state which variant won and print its
 `VALIDATION_SCORE=<value>`.
 
+## Hardware
+A CUDA GPU may be available (check `torch.cuda.is_available()`). Use it at your
+discretion — for deep models, moving the model and batches to the GPU (`.to('cuda')`)
+speeds up training considerably.
+
 ## Submission — REQUIRED
 Write predictions to **`/workspace/submission.csv`** with EXACTLY two columns:
 - `id` — copied from `test.csv`

datasets/tabred-real/homecredit-default/instruction.md

@@ -63,6 +63,11 @@ variant with the BEST validation score — never simply the last one trained.
 Before writing the submission, state which variant won and print its
 `VALIDATION_SCORE=<value>`.
 
+## Hardware
+A CUDA GPU may be available (check `torch.cuda.is_available()`). Use it at your
+discretion — for deep models, moving the model and batches to the GPU (`.to('cuda')`)
+speeds up training considerably.
+
 ## Submission — REQUIRED
 Write predictions to **`/workspace/submission.csv`** with EXACTLY two columns:
 - `id` — copied from `test.csv`

datasets/tabred-real/homesite-insurance/instruction.md

@@ -63,6 +63,11 @@ variant with the BEST validation score — never simply the last one trained.
 Before writing the submission, state which variant won and print its
 `VALIDATION_SCORE=<value>`.
 
+## Hardware
+A CUDA GPU may be available (check `torch.cuda.is_available()`). Use it at your
+discretion — for deep models, moving the model and batches to the GPU (`.to('cuda')`)
+speeds up training considerably.
+
 ## Submission — REQUIRED
 Write predictions to **`/workspace/submission.csv`** with EXACTLY two columns:
 - `id` — copied from `test.csv`

datasets/tabred-real/maps-routing/instruction.md

@@ -63,6 +63,11 @@ variant with the BEST validation score — never simply the last one trained.
 Before writing the submission, state which variant won and print its
 `VALIDATION_SCORE=<value>`.
 
+## Hardware
+A CUDA GPU may be available (check `torch.cuda.is_available()`). Use it at your
+discretion — for deep models, moving the model and batches to the GPU (`.to('cuda')`)
+speeds up training considerably.
+
 ## Submission — REQUIRED
 Write predictions to **`/workspace/submission.csv`** with EXACTLY two columns:
 - `id` — copied from `test.csv`

datasets/tabred-real/sberbank-housing/instruction.md

@@ -63,6 +63,11 @@ variant with the BEST validation score — never simply the last one trained.
 Before writing the submission, state which variant won and print its
 `VALIDATION_SCORE=<value>`.
 
+## Hardware
+A CUDA GPU may be available (check `torch.cuda.is_available()`). Use it at your
+discretion — for deep models, moving the model and batches to the GPU (`.to('cuda')`)
+speeds up training considerably.
+
 ## Submission — REQUIRED
 Write predictions to **`/workspace/submission.csv`** with EXACTLY two columns:
 - `id` — copied from `test.csv`

datasets/tabred-real/weather/instruction.md

@@ -63,6 +63,11 @@ variant with the BEST validation score — never simply the last one trained.
 Before writing the submission, state which variant won and print its
 `VALIDATION_SCORE=<value>`.
 
+## Hardware
+A CUDA GPU may be available (check `torch.cuda.is_available()`). Use it at your
+discretion — for deep models, moving the model and batches to the GPU (`.to('cuda')`)
+speeds up training considerably.
+
 ## Submission — REQUIRED
 Write predictions to **`/workspace/submission.csv`** with EXACTLY two columns:
 - `id` — copied from `test.csv`