ci: lint autofix on pull requests

.cursor/archive/markr-handoff-2026-06-22.mdc

@@ -0,0 +1,140 @@
+---
+description: Archived Markr session handoff (stale — do not apply)
+alwaysApply: false
+---
+
+<!-- Archived 2026-06-22. Superseded by AGENTS.md. Regenerate via Markr when starting a new handoff. -->
+
+# Handoff from Cursor (archived)
+> 10 messages | ~344 tokens | Projects/VideoTuna | branch `main`
+>
+> Conditional Residual Handoff — transmits what the repo can't tell you (decisions, dead-ends, constraints, uncommitted diff), not the code itself.
+
+## ⚡ Paste this first
+
+Continue in Cursor Chat. Use the project files as source of truth and keep changes scoped.
+
+```text
+I'm resuming a previous Cursor session on Projects/VideoTuna. You have the repository — read it for anything not stated here. This handoff carries only what the code itself cannot tell you.
+
+TASK
+Pin SimpleTuner upstream SHA on next sync
+Migrate third_party/flux/ → videotuna/vendor/simpletuner/ via submodule
+Remove cogvideo_sat after SAT deprecation
+First-party Flux LoRA trainer to drop the 71-file snapshot
+Original task: @/home/menes/.cursor/projects/home-menes-Projects-VideoTuna/terminals/10.txt:9-239
+
+STATE
+- Branch `main` · 40 uncommitted file(s)
+UNCOMMITTED (in-flight — not on HEAD, you can't see this by reading committed code)
+- M README.md
+- D configs/005_cogvideox1.5/cogvideox1.5_5b.yaml
+- M docs/MODEL_VERSIONS.md
+- M docs/checkpoints.md
+- M poetry.lock
+- M pyproject.toml
+- M scripts/__init__.py
+- D scripts/inference_cogVideo_sat_refactor.py
+- M tests/conftest.py
+- M tests/test_import_smoke.py
+- M uv.lock
+- D videotuna/models/cogvideo_sat/arguments.py
+- D videotuna/models/cogvideo_sat/data_video.py
+- D videotuna/models/cogvideo_sat/diffusion_video.py
+- D videotuna/models/cogvideo_sat/dit_video_concat.py
+- D videotuna/models/cogvideo_sat/sgm/__init__.py
+- D videotuna/models/cogvideo_sat/sgm/lr_scheduler.py
+- D videotuna/models/cogvideo_sat/sgm/models/__init__.py
+- D videotuna/models/cogvideo_sat/sgm/models/autoencoder.py
+- D videotuna/models/cogvideo_sat/sgm/modules/__init__.py
+NEXT
+- Run the relevant build, lint, or test command before calling the handoff complete.
+- Preserve existing user changes and avoid reverting unrelated work.
+VERIFY
+- No verification command was captured — run the project build/lint/test before finishing.
+
+SYNTHESIS — before you change anything, restate in one line: (a) the task, and (b) the one constraint you must not break. Then proceed.
+```
+
+---
+
+## 🧠 Decision log
+
+_No explicit decisions were captured in the transcript._
+
+## 🛑 Dead-ends — do not redo
+
+_None captured._
+
+## 📌 Constraints
+
+_None explicitly stated._
+
+## 🔀 In-flight (uncommitted) state
+
+Branch: `main`
+
+Uncommitted changes:
+- `M README.md`
+- `D configs/005_cogvideox1.5/cogvideox1.5_5b.yaml`
+- `M docs/MODEL_VERSIONS.md`
+- `M docs/checkpoints.md`
+- `M poetry.lock`
+- `M pyproject.toml`
+- `M scripts/__init__.py`
+- `D scripts/inference_cogVideo_sat_refactor.py`
+- `M tests/conftest.py`
+- `M tests/test_import_smoke.py`
+- `M uv.lock`
+- `D videotuna/models/cogvideo_sat/arguments.py`
+- `D videotuna/models/cogvideo_sat/data_video.py`
+- `D videotuna/models/cogvideo_sat/diffusion_video.py`
+- `D videotuna/models/cogvideo_sat/dit_video_concat.py`
+- `D videotuna/models/cogvideo_sat/sgm/__init__.py`
+- `D videotuna/models/cogvideo_sat/sgm/lr_scheduler.py`
+- `D videotuna/models/cogvideo_sat/sgm/models/__init__.py`
+- `D videotuna/models/cogvideo_sat/sgm/models/autoencoder.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/__init__.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/attention.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/__init__.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/losses/__init__.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/losses/discriminator_loss.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/losses/lpips.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/losses/video_loss.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/lpips/__init__.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/lpips/loss/.gitignore`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/lpips/loss/LICENSE`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/lpips/loss/__init__.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/lpips/loss/lpips.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/lpips/model/LICENSE`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/lpips/model/__init__.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/lpips/model/model.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/lpips/util.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/lpips/vqperceptual.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/magvit2_pytorch.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/regularizers/__init__.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/regularizers/base.py`
+- `D videotuna/models/cogvideo_sat/sgm/modules/autoencoding/regularizers/finite_scalar_quantization.py`
+
+## 🎯 Task
+
+**Continue:** Pin SimpleTuner upstream SHA on next sync
+Migrate third_party/flux/ → videotuna/vendor/simpletuner/ via submodule
+Remove cogvideo_sat after SAT deprecation
+First-party Flux LoRA trainer to drop the 71-file snapshot
+
+**Original request:** @/home/menes/.cursor/projects/home-menes-Projects-VideoTuna/terminals/10.txt:9-239
+
+## 💬 Recent exchange (tail)
+
+**You**: Provide me with 3 comprehensive prompts, to run in plan model to setup amdu rocm support, imrpove nvidia support and use cpu. Also, be thorough on how to improve integration with the current system.
+
+**You**: This is too slow poetry run pytest tests/test_diffusers_video_flow.py
+
+**You**: @videotuna/third_party Is there a better way than doing this in our repo ? Provide me with a prompt to re-organize and improve the dependencies, management, etc
+
+**You**: Consume this article https://bitmovin.com/blog/ai-video-research/ , suggest me 10 improvements you would do on this codebase based on the information.
+
+**You**: Provide me with 3 comprehensive prompts, to run in plan mode to setup amdu rocm support, imrpove nvidia support and use cpu. Also, be thorough on how to improve integration with the current system.
+
+**You**: Pin SimpleTuner upstream SHA on next sync Migrate third_party/flux/ → videotuna/vendor/simpletuner/ via submodule Remove cogvideo_sat after SAT deprecation First-party Flux LoRA trainer to drop the 71-file snapshot

.cursor/mcp.json

@@ -0,0 +1,3 @@
+{
+  "mcpServers": {}
+}

.cursor/rules/privtune.mdc

@@ -0,0 +1,15 @@
+---
+description: PrivTune project conventions and agent workflow
+alwaysApply: true
+---
+
+# PrivTune
+
+**Role:** Private-domain LoRA training platform (Flux T2I + Wan 2.1 T2V train, Wan 2.2 Diffusers validate). Optimize for correct training behavior, portable CUDA/ROCm/CPU handling, and minimal scoped diffs.
+
+Primary instructions: [`AGENTS.md`](../AGENTS.md) at the repo root.
+
+- Python 3.11+ · Poetry default (`poetry run …`) · optional uv
+- **Before finishing (required):** `poetry run lint`, `poetry run format-check`, `poetry run coverage-gate`
+- Env vars: [`.env.example`](../.env.example) (`VIDEOTUNA_*` retained) · Vendor policy: [`docs/vendor-policy.md`](../docs/vendor-policy.md)
+- Never commit `.env`, checkpoints, `outputs/`, weights, or secrets

.cursor/settings.json

@@ -0,0 +1,7 @@
+{
+  "plugins": {
+    "context7-plugin": {
+      "enabled": true
+    }
+  }
+}

.env.example

@@ -0,0 +1,54 @@
+# PrivTune environment variables (VIDEOTUNA_* prefix retained for compatibility)
+# Copy to .env and export, or set in your shell profile.
+# Do not commit .env — it may contain secrets.
+
+# --- Compute backend ---
+# auto | cuda | rocm | cpu
+VIDEOTUNA_COMPUTE_BACKEND=auto
+
+# --- Attention backend ---
+# auto | flash | sdpa | eager
+# ROCm: use sdpa (flash is not supported)
+VIDEOTUNA_ATTN_BACKEND=auto
+
+# Fail when flash is requested but flash-attn is not installed (0 | 1)
+VIDEOTUNA_ATTN_BACKEND_STRICT=0
+
+# --- torch.compile (denoiser only) ---
+VIDEOTUNA_TORCH_COMPILE=0
+VIDEOTUNA_TORCH_COMPILE_MODE=reduce-overhead
+
+# --- Metrics ---
+# Inference metrics.json ownership (not training experiment tracking): script | flow
+VIDEOTUNA_METRICS_OWNER=script
+
+# Training experiment tracking (Flux Phase 1): tensorboard | trackio
+# trackio requires: poetry install -E trackio
+# VIDEOTUNA_METRICS_BACKEND=tensorboard
+# Optional Trackio remote dashboard (private Space recommended for domain content):
+# VIDEOTUNA_TRACKIO_SPACE_ID=username/privtune-trackio
+# VIDEOTUNA_TRACKIO_PROJECT=flux-domain-lora
+
+# --- CPU inference mode ---
+# off | smoke | force
+VIDEOTUNA_CPU_MODE=off
+# Deprecated — use VIDEOTUNA_CPU_MODE=force
+# VIDEOTUNA_ALLOW_CPU_INFERENCE=0
+
+# --- Training / benchmarks (optional) ---
+# VIDEOTUNA_LOG_LEVEL=INFO
+# VIDEOTUNA_BENCH_MODEL=
+
+# --- GPU selection ---
+# CUDA_VISIBLE_DEVICES=0
+# HIP_VISIBLE_DEVICES=0
+
+# --- Hugging Face (gated models, higher rate limits) ---
+# HF_TOKEN=
+# HF_HOME=~/.cache/huggingface
+
+# --- DashScope (Wan prompt extension via dashscope method) ---
+# DASH_API_KEY=
+
+# --- Test harness ---
+# ENV=test

.gemini/settings.json

@@ -0,0 +1,15 @@
+{
+	"hooks": {
+		"AfterAgent": [
+			{
+				"hooks": [
+					{
+						"type": "command",
+						"command": "\"$HOME/.jolli/jollimemory/run-hook\" gemini-after-agent",
+						"name": "jolli-session-tracker"
+					}
+				]
+			}
+		]
+	}
+}

.github/workflows/cpu.yml

@@ -0,0 +1,37 @@
+name: CPU
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+
+      - name: Install Poetry
+        run: pip install poetry
+
+      - name: Install dependencies (CPU extra + dev + training)
+        run: poetry install -E cpu --with dev --with training
+
+      - name: Install CPU PyTorch wheels
+        run: poetry run install-cpu-torch
+
+      - name: Verify CPU torch
+        run: poetry run verify-cpu-torch
+
+      - name: Lint
+        run: poetry run lint
+
+      - name: Format check
+        run: poetry run format-check
+
+      - name: Smoke tests + coverage gate
+        run: poetry run coverage-gate

.github/workflows/lint-autofix.yml

@@ -0,0 +1,43 @@
+name: Lint autofix
+
+on:
+  pull_request:
+    types: [opened, synchronize, reopened]
+
+permissions:
+  contents: write
+
+concurrency:
+  group: lint-autofix-${{ github.event.pull_request.number }}
+  cancel-in-progress: true
+
+jobs:
+  autofix:
+    if: github.event.pull_request.head.repo.full_name == github.repository
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.ref }}
+          repository: ${{ github.event.pull_request.head.repo.full_name }}
+          token: ${{ secrets.GITHUB_TOKEN }}
+          persist-credentials: true
+
+      # Install ruff only (no Poetry / torch). Pin matches poetry.lock (0.6.9).
+      # Mirrors scripts/__init__.py code_format() targets and command order.
+      - name: Install ruff
+        uses: astral-sh/ruff-action@v4.0.0
+        with:
+          args: "--version"
+          version: "0.6.9"
+
+      - name: Apply lint and format fixes
+        run: |
+          ruff check --fix videotuna tests scripts tools
+          ruff check --select I --fix .
+          ruff format .
+
+      - uses: stefanzweifel/git-auto-commit-action@v7
+        with:
+          commit_message: "style: apply ruff lint and format fixes"
+          commit_author: "github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>"

.gitignore

@@ -19,6 +19,10 @@ Dataset/
 output/
 outputs/
 /data
+data/t2i/domain/
+data/t2v/domain/
+
+cloud/vast/.env
 
 HPSv2/
 SwissArmyTransformer/
@@ -28,3 +32,5 @@ temp
 
 
 *.outputs
+.jolli/jollimemory
+.jolli/jollimemory/debug.log

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "videotuna/vendor/simpletuner"]
+	path = videotuna/vendor/simpletuner
+	url = https://github.com/bghira/SimpleTuner.git

.markr/config-tests.json

@@ -0,0 +1,18 @@
+{
+  "version": 1,
+  "suites": [
+    {
+      "configPath": ".agents/skills/jolli-recall/SKILL.md",
+      "tests": [
+        {
+          "id": "3266f96c-3940-4c08-a8f9-0334be69b6e4",
+          "name": "New test",
+          "prompt": "",
+          "expectedBehavior": "",
+          "mustInclude": [],
+          "mustNotInclude": []
+        }
+      ]
+    }
+  ]
+}