v0.12.0:4 - hotfix: torchaudio build fails without --no-build-isolation

Build was crashing inside torchaudio's setup.py with:
  ModuleNotFoundError: No module named 'torch'

PIP_CONSTRAINT was correctly pinning torch/torchvision in the install
target env, but pip's PEP 517 build isolation creates a SEPARATE fresh
Python env just for the build wheel step — and that env has no torch
in it. torchaudio's setup.py imports torch to discover CUDA flags, so
it crashes. Pip even printed a deprecation warning that this isolation
behavior is hardening, not relaxing.

Fix:
  1. Pre-install torchaudio's build deps (setuptools, wheel, ninja,
     pybind11) into the main env since we're disabling isolation.
  2. Add --no-build-isolation to the torchaudio install so the build
     uses NGC's torch directly.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

image/whisperx_container/Dockerfile

@@ -43,11 +43,24 @@ sys.stdout.write(f'torch=={torch.__version__}\ntorchvision=={torchvision.__versi
 # NGC PyTorch images don't include torchaudio (NVIDIA optimizes for
 # vision/text workloads). Stock torchaudio wheels are ABI-incompatible with
 # NGC's custom torch 2.10a, so the only working option is building from
-# source against the NGC torch already in the image. Pinning to v2.5.1 — the
+# source against the NGC torch already in the image.
-# last torchaudio tag that builds cleanly against torch 2.5–2.10 and is a
+#
-# proven compatibility target.
+# Build env knobs:
+#   USE_CUDA=1                — build CUDA kernels (we have a GPU)
+#   BUILD_SOX=0               — skip libsox (we only use audio decoding)
+#   TORCH_CUDA_ARCH_LIST=...  — build kernels for Hopper + Blackwell datacenter
+#                               + Blackwell consumer (sm_120 = GB10)
+#   --no-build-isolation      — CRITICAL: PEP 517 build isolation creates a
+#                               fresh env with no torch in it. torchaudio's
+#                               setup.py imports torch to discover the build
+#                               flags, so it crashes without this flag.
+#                               With it, the build uses NGC's torch directly.
 ENV USE_CUDA=1 BUILD_SOX=0 TORCH_CUDA_ARCH_LIST="9.0;10.0;12.0"
+# Pre-install torchaudio's build-time deps (PEP 517 would normally install
+# these in the isolated build env, but we just turned isolation off).
 RUN pip install --break-system-packages --no-cache-dir \
+    "setuptools>=61" wheel ninja "pybind11>=2.10"
+RUN pip install --break-system-packages --no-cache-dir --no-build-isolation \
     git+https://github.com/pytorch/audio.git@v2.5.1 \
  && python3 -c "import torchaudio; print('torchaudio built:', torchaudio.__version__)"
 

package/startos/versions/v0_1_0.ts

@@ -1,10 +1,10 @@
 import { VersionInfo, IMPOSSIBLE } from '@start9labs/start-sdk'
 
 export const v0_1_0 = VersionInfo.of({
-  version: '0.12.0:3',
+  version: '0.12.0:4',
   releaseNotes: {
     en_US:
-      'v0.12.0:3 — hotfix: deeper torchaudio fix. The Spark is ARM64 (Grace + GB10 Blackwell), and the NGC PyTorch container — the only base with a working torch for sm_120 ARM64 — does NOT ship torchaudio at all. Stock pip wheels are amd64-only and ABI-incompatible with NGC\'s custom torch anyway. Real fix: build torchaudio from source against NGC\'s torch (v2.5.1, the last torchaudio tag that compiles cleanly against torch 2.5–2.10) with TORCH_CUDA_ARCH_LIST set for Blackwell sm_120. Build adds ~3-5 min to the first WhisperX install (only first time — Docker layer cache reuses it after). Plus the constraints.txt approach from 0.12.0:2 to lock torch + torchvision + torchaudio against any later pip swap-out.',
+      'v0.12.0:4 — hotfix: torchaudio build was failing with "ModuleNotFoundError: No module named torch" during its setup.py. Root cause: pip\'s PEP 517 build isolation creates a fresh Python env for the build that doesn\'t see NGC\'s torch (which is what we need for ABI compat). Fix: add --no-build-isolation to the pip install so the build uses the existing torch, plus pre-install setuptools/wheel/ninja/pybind11 since pip won\'t auto-pull them when build isolation is off. Should now finally compile torchaudio v2.5.1 against NGC\'s torch 2.10 and proceed to the whisperx install.',
   },
   migrations: {
     up: async ({ effects }) => {},