fix: add Apple Silicon (MPS) device support across lessons

[Jayantkhandebharad]

Summary

Adds Apple Silicon (MPS) support to the GPU code across the curriculum. Device selection was CUDA-only ("cuda" if torch.cuda.is_available() else "cpu"), so on M-series Macs lessons silently ran on CPU, reported "no GPU", or hard-errored — even though torch.backends.mps.is_available() is True. Fixes the inconsistency reported in #294 (197 files already handle mps; these 13 didn't).

What changed (13 lessons, one commit each)

• 6 training lessons (phase-03/11, phase-04/04, /05, /07, /09, /10): added a get_device() helper (cuda → mps → cpu) and used it.
• Phase 0 setup (verify.py, gpu_check.py, debug_tools.py): detect/benchmark/profile on MPS; the env check now passes on Apple Silicon.
• Sync sites (phase-04/15, /16): added torch.mps.synchronize() for accurate timing.
• Stable Diffusion (phase-04/11): runs on MPS instead of skipping.
• AMP capstone (phase-19/45): AmpTrainState now accepts device_type="mps" (GradScaler stays disabled — AMP scaling is CUDA-only).

Verification (on an Apple M4, torch 2.12.0)

• All 13 files compile; gpu_check.py and verify.py detect and run on MPS; debug_tools.py MPS memory demo works.
• 45-gradient-clipping-amp unit tests: 17/17 pass.
• scripts/audit_lessons.py: 503 lessons, 0 issues.
• No behavior change on CUDA or CPU machines — MPS is only chosen when CUDA is absent and MPS is available.

Closes #294.

🤖 Generated with Claude Code

[coderabbitai]

📝 Walkthrough

Walkthrough

Adds Apple Silicon MPS backend support across 13 lesson script files. The changes introduce get_device() helpers (CUDA→MPS→CPU) and _sync(device) synchronization helpers, replace CUDA-only device selection and synchronization calls, expand GPU reporting to cover MPS properties, and allow "mps" as a valid device_type in AmpTrainState.

Changes

MPS Backend Support

Layer / File(s)	Summary
Setup and GPU check tooling phases/00-setup-and-tooling/01-dev-environment/code/verify.py, phases/00-setup-and-tooling/03-gpu-setup-and-cloud/code/gpu_check.py	verify.py adds _gpu_backend() and updates GPU_CHECKS to report CUDA or MPS. gpu_check.py adds _sync(backend), changes backend selection to CUDA-or-MPS, conditionally prints CUDA device properties vs. an MPS unified-memory note, and makes the benchmark and capacity estimate backend-aware.
Debugging/profiling CUDA+MPS demos phases/00-setup-and-tooling/12-debugging-and-profiling/code/debug_tools.py	demo_device_checking switches between CUDA and MPS dynamically. demo_gpu_memory gains a backend-dispatch structure with CUDA and MPS paths, backend-specific cleanup, and a no-GPU fallback.
get_device() helper across DL core and CV lesson scripts phases/03-deep-learning-core/11-intro-to-pytorch/code/pytorch_intro.py, phases/04-computer-vision/04-image-classification/code/main.py, phases/04-computer-vision/05-transfer-learning/code/main.py, phases/04-computer-vision/07-semantic-segmentation-unet/code/main.py, phases/04-computer-vision/09-image-generation-gans/code/main.py, phases/04-computer-vision/10-image-generation-diffusion/code/main.py	Each file adds an identical get_device() function (CUDA→MPS→CPU) and replaces the previous inline CUDA-or-CPU ternary with a call to it. pytorch_intro.py also gates the GPU name print on device.type == "cuda".
Special-case MPS handling: Stable Diffusion, latency timing, capstone AMP phases/04-computer-vision/11-stable-diffusion/code/main.py, phases/04-computer-vision/15-real-time-edge/code/main.py, phases/04-computer-vision/16-vision-pipeline-capstone/code/main.py, phases/19-capstone-projects/45-gradient-clipping-amp/code/main.py	text_to_image_stub() now detects CUDA vs MPS and returns None when no GPU is present. real-time-edge introduces _sync(device) and uses it in measure_latency. Vision-pipeline-capstone sync() gains an MPS branch. AmpTrainState device_type validation now accepts "mps".

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 20.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The PR title clearly summarizes the primary change: adding Apple Silicon (MPS) device support across lessons.
Description check	✅ Passed	The description is directly related to the changeset, detailing what was changed, why it was needed, and how verification was performed.
Linked Issues check	✅ Passed	The PR fully addresses the primary objectives from #294: adds MPS device support to all 13 affected files, implements cuda→mps→cpu device selection helpers, enables MPS detection in setup tools, adds device-aware synchronization, and allows MPS in the AMP capstone.
Out of Scope Changes check	✅ Passed	All changes are in-scope: the PR systematically adds MPS support to the 13 CUDA-only files identified in #294 without introducing unrelated modifications.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@phases/00-setup-and-tooling/12-debugging-and-profiling/code/debug_tools.py`:
- Line 238: Remove the f-string prefix from four print statements in
debug_tools.py that contain no variable interpolation, as these trigger Ruff
F541 lint errors. At line 238, convert the f-string in the print statement (" 
After 10k x 10k tensor:") to a regular string by removing the f prefix. Apply
the same fix at lines 243, 252, and 257 where similar f-strings with no
interpolation exist. Each of these four locations should have the f prefix
removed from the string literal to satisfy the linter.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 33ae37c0-d44f-4e56-a2de-db2f6f8dc1dc

📥 Commits

Reviewing files that changed from the base of the PR and between 574a5d6 and 1b7402b.

📒 Files selected for processing (13)

• phases/00-setup-and-tooling/01-dev-environment/code/verify.py
• phases/00-setup-and-tooling/03-gpu-setup-and-cloud/code/gpu_check.py
• phases/00-setup-and-tooling/12-debugging-and-profiling/code/debug_tools.py
• phases/03-deep-learning-core/11-intro-to-pytorch/code/pytorch_intro.py
• phases/04-computer-vision/04-image-classification/code/main.py
• phases/04-computer-vision/05-transfer-learning/code/main.py
• phases/04-computer-vision/07-semantic-segmentation-unet/code/main.py
• phases/04-computer-vision/09-image-generation-gans/code/main.py
• phases/04-computer-vision/10-image-generation-diffusion/code/main.py
• phases/04-computer-vision/11-stable-diffusion/code/main.py
• phases/04-computer-vision/15-real-time-edge/code/main.py
• phases/04-computer-vision/16-vision-pipeline-capstone/code/main.py
• phases/19-capstone-projects/45-gradient-clipping-amp/code/main.py

[coderabbitai]

⚠️ Potential issue | 🟡 Minor | ⚡ Quick win

Remove no-op f-strings to satisfy Ruff F541.

These four strings have no interpolation and trigger lint errors.

Suggested patch

-        print(f"  After 10k x 10k tensor:")
+        print("  After 10k x 10k tensor:")
...
-        print(f"  After cleanup:")
+        print("  After cleanup:")
...
-        print(f"  After 10k x 10k tensor:")
+        print("  After 10k x 10k tensor:")
...
-        print(f"  After cleanup:")
+        print("  After cleanup:")

Also applies to: 243-243, 252-252, 257-257

🧰 Tools

🪛 Ruff (0.15.15)

[error] 238-238: f-string without any placeholders

Remove extraneous f prefix

(F541)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@phases/00-setup-and-tooling/12-debugging-and-profiling/code/debug_tools.py`
at line 238, Remove the f-string prefix from four print statements in
debug_tools.py that contain no variable interpolation, as these trigger Ruff
F541 lint errors. At line 238, convert the f-string in the print statement (" 
After 10k x 10k tensor:") to a regular string by removing the f prefix. Apply
the same fix at lines 243, 252, and 257 where similar f-strings with no
interpolation exist. Each of these four locations should have the f prefix
removed from the string literal to satisfy the linter.

Source: Linters/SAST tools