PortPilot
Benchmark · pre-generated, always available← back to leaderboard
Migration Readiness Report

karpathy/nanoGPT

https://github.com/karpathy/nanoGPT ↗
Ready to run on ROCm

karpathy/nanoGPT scored 100/100 for ROCm readiness (Ready to run on ROCm). Of 17 CUDA-related findings, 17 work as-is on ROCm PyTorch, 0 need a mechanical HIPify-style change, and 0 are manual blockers.

17Works as-is
0Mechanical change
0Manual blocker
Findings
17
Python files
15
Files scanned
26
Custom CUDA kernels
none

Findings by file

17 findings · 4 files
bench.py· 6
A
Device string "cuda"
device = 'cuda' # examples: 'cpu', 'cuda', 'cuda:0', 'cuda:1', etc.
:17

The literal device string "cuda" resolves to the active AMD GPU on ROCm PyTorch. .to("cuda") / device="cuda" need no edits.

RecommendNo change required on ROCm PyTorch.

A
torch.cuda API usage
dtype = 'bfloat16' if torch.cuda.is_available() and torch.cuda.is_bf16_supported() else 'float16' # 'float32' or 'bfloat16' or 'float16'
:18

Calls under torch.cuda.* are aliased by ROCm builds of PyTorch. The same code runs on AMD Instinct GPUs unchanged — torch.cuda.is_available(), streams, events and AMP all map onto HIP.

RecommendNo change required. Install the ROCm build of PyTorch (pip install torch --index-url https://download.pytorch.org/whl/rocm6.1).

A
torch.cuda API usage
torch.cuda.manual_seed(seed)
:25

Calls under torch.cuda.* are aliased by ROCm builds of PyTorch. The same code runs on AMD Instinct GPUs unchanged — torch.cuda.is_available(), streams, events and AMP all map onto HIP.

RecommendNo change required. Install the ROCm build of PyTorch (pip install torch --index-url https://download.pytorch.org/whl/rocm6.1).

A
Device string "cuda"
device_type = 'cuda' if 'cuda' in device else 'cpu' # for later use in torch.autocast
:28

The literal device string "cuda" resolves to the active AMD GPU on ROCm PyTorch. .to("cuda") / device="cuda" need no edits.

RecommendNo change required on ROCm PyTorch.

A
torch.cuda API usage
torch.cuda.synchronize()
:99

Calls under torch.cuda.* are aliased by ROCm builds of PyTorch. The same code runs on AMD Instinct GPUs unchanged — torch.cuda.is_available(), streams, events and AMP all map onto HIP.

RecommendNo change required. Install the ROCm build of PyTorch (pip install torch --index-url https://download.pytorch.org/whl/rocm6.1).

A
torch.cuda API usage
torch.cuda.synchronize()
:112

Calls under torch.cuda.* are aliased by ROCm builds of PyTorch. The same code runs on AMD Instinct GPUs unchanged — torch.cuda.is_available(), streams, events and AMP all map onto HIP.

RecommendNo change required. Install the ROCm build of PyTorch (pip install torch --index-url https://download.pytorch.org/whl/rocm6.1).

model.py· 1
A
Device string "cuda"
use_fused = fused_available and device_type == 'cuda'
:282

The literal device string "cuda" resolves to the active AMD GPU on ROCm PyTorch. .to("cuda") / device="cuda" need no edits.

RecommendNo change required on ROCm PyTorch.

sample.py· 4
A
Device string "cuda"
device = 'cuda' # examples: 'cpu', 'cuda', 'cuda:0', 'cuda:1', etc.
:20

The literal device string "cuda" resolves to the active AMD GPU on ROCm PyTorch. .to("cuda") / device="cuda" need no edits.

RecommendNo change required on ROCm PyTorch.

A
torch.cuda API usage
dtype = 'bfloat16' if torch.cuda.is_available() and torch.cuda.is_bf16_supported() else 'float16' # 'float32' or 'bfloat16' or 'float16'
:21

Calls under torch.cuda.* are aliased by ROCm builds of PyTorch. The same code runs on AMD Instinct GPUs unchanged — torch.cuda.is_available(), streams, events and AMP all map onto HIP.

RecommendNo change required. Install the ROCm build of PyTorch (pip install torch --index-url https://download.pytorch.org/whl/rocm6.1).

A
torch.cuda API usage
torch.cuda.manual_seed(seed)
:27

Calls under torch.cuda.* are aliased by ROCm builds of PyTorch. The same code runs on AMD Instinct GPUs unchanged — torch.cuda.is_available(), streams, events and AMP all map onto HIP.

RecommendNo change required. Install the ROCm build of PyTorch (pip install torch --index-url https://download.pytorch.org/whl/rocm6.1).

A
Device string "cuda"
device_type = 'cuda' if 'cuda' in device else 'cpu' # for later use in torch.autocast
:30

The literal device string "cuda" resolves to the active AMD GPU on ROCm PyTorch. .to("cuda") / device="cuda" need no edits.

RecommendNo change required on ROCm PyTorch.

train.py· 6
A
Device string "cuda"
device = 'cuda' # examples: 'cpu', 'cuda', 'cuda:0', 'cuda:1' etc., or try 'mps' on macbooks
:72

The literal device string "cuda" resolves to the active AMD GPU on ROCm PyTorch. .to("cuda") / device="cuda" need no edits.

RecommendNo change required on ROCm PyTorch.

A
torch.cuda API usage
dtype = 'bfloat16' if torch.cuda.is_available() and torch.cuda.is_bf16_supported() else 'float16' # 'float32', 'bfloat16', or 'float16', the latter will auto implement a GradScaler
:73

Calls under torch.cuda.* are aliased by ROCm builds of PyTorch. The same code runs on AMD Instinct GPUs unchanged — torch.cuda.is_available(), streams, events and AMP all map onto HIP.

RecommendNo change required. Install the ROCm build of PyTorch (pip install torch --index-url https://download.pytorch.org/whl/rocm6.1).

A
torch.cuda API usage
torch.cuda.set_device(device)
:89

Calls under torch.cuda.* are aliased by ROCm builds of PyTorch. The same code runs on AMD Instinct GPUs unchanged — torch.cuda.is_available(), streams, events and AMP all map onto HIP.

RecommendNo change required. Install the ROCm build of PyTorch (pip install torch --index-url https://download.pytorch.org/whl/rocm6.1).

A
Device string "cuda"
device_type = 'cuda' if 'cuda' in device else 'cpu' # for later use in torch.autocast
:109

The literal device string "cuda" resolves to the active AMD GPU on ROCm PyTorch. .to("cuda") / device="cuda" need no edits.

RecommendNo change required on ROCm PyTorch.

A
Device string "cuda"
if device_type == 'cuda':
:126

The literal device string "cuda" resolves to the active AMD GPU on ROCm PyTorch. .to("cuda") / device="cuda" need no edits.

RecommendNo change required on ROCm PyTorch.

A
torch.cuda API usage
scaler = torch.cuda.amp.GradScaler(enabled=(dtype == 'float16'))
:196

Calls under torch.cuda.* are aliased by ROCm builds of PyTorch. The same code runs on AMD Instinct GPUs unchanged — torch.cuda.is_available(), streams, events and AMP all map onto HIP.

RecommendNo change required. Install the ROCm build of PyTorch (pip install torch --index-url https://download.pytorch.org/whl/rocm6.1).