chore(genai): comfyui-qwen requirements aligned with prod container + rollback state doc

docker-configurations/services/comfyui-qwen/ROLLBACK_STATE_20260620.md

@@ -0,0 +1,43 @@
+# ComfyUI-Qwen rollback state (before nunchaku fix + ComfyUI upgrade, 2026-06-20)
+
+Recorded BEFORE applying fixes per ai-01 greenlight.
+
+## Pre-fix state (DEGRADED: gen hangs, CPU offload)
+
+### ComfyUI core
+- Repository: https://github.com/comfyanonymous/ComfyUI.git
+- HEAD: `f8b981ae` (2025-11-30, 1187 commits behind master as of 2026-06-20)
+- Path (bind-mounted): `docker-configurations/services/comfyui-qwen/workspace`
+
+### pip packages
+- torch: 2.6.0+cu124
+- nunchaku: `1.0.1+torch2.6` (GitHub wheel, not PyPI)
+- transformers: (4.50.3+ per requirements)
+
+### Custom node ComfyUI-nunchaku
+- Repository: https://github.com/nunchaku-ai/ComfyUI-nunchaku.git
+- SHA: `930bc2266a4f3277e16be75e67c11875d20a01c6` (tag v1.1.0)
+
+### Known bugs (root causes — CORRECTED diagnosis 2026-06-20)
+1. **getpwuid blocker** (RESOLVED by USER=bonsai/LOGNAME=bonsai env compose):
+   `torch._inductor` → `getpass.getuser()` → `pwd.getpwuid(1000)` KeyError (UID 1000
+   not in python:3.11 /etc/passwd). Aborts transformers→hqq→inductor chain → 0 Nunchaku loaders.
+2. **VRAM integer-overflow** (UNRESOLVED — the reason for ComfyUI upgrade attempt):
+   `comfy/model_patcher.py:797` logs `lowvram_model_memory = 9.5e19 MB usable` →
+   `full_load: True` but KSampler offloads to CPU (GPU 0%, CPU 176%). Bug is in
+   ComfyUI core memory accumulator, NOT in mem_get_info (returns correct free=23GB).
+   NOT fixed by --disable-cuda-malloc or --normalvram (tested). Hypothesis: fixed
+   upstream in the 1187 commits between f8b981ae and master.
+
+### Container
+- Image: `python:3.11`
+- user: "1000:1000"
+
+## Fixes applied on branch fix/comfyui-qwen-nunchaku
+- docker-compose.yml: USER=bonsai + LOGNAME=bonsai (env), memory 12G→28G, shm_size 16g
+
+## Rollback procedure
+1. Revert docker-compose.yml: `git checkout origin/main -- docker-configurations/services/comfyui-qwen/docker-compose.yml`
+2. Revert ComfyUI core (if upgraded): `docker exec -u 0 comfyui-qwen bash -c "cd /workspace/ComfyUI && git config --global --add safe.directory '*' && git checkout f8b981ae"`
+3. `docker compose -f docker-configurations/services/comfyui-qwen/docker-compose.yml --env-file .env up -d --force-recreate comfyui-qwen`
+4. Re-ping ai-01 on dashboard.

docker-configurations/services/comfyui-qwen/workspace/requirements.txt

@@ -1,6 +1,6 @@
-comfyui-frontend-package==1.45.19
-comfyui-workflow-templates==0.10.0
-comfyui-embedded-docs==0.5.4
+comfyui-frontend-package==1.52.7
+comfyui-workflow-templates==0.11.62
+comfyui-embedded-docs==0.5.11
 torch
 torchsde
 torchvision
@@ -21,9 +21,9 @@ psutil
 alembic
 SQLAlchemy>=2.0.0
 filelock
-av>=16.0.0
-comfy-kitchen==0.2.10
-comfy-aimdo==0.4.10
+av>=17.0.0
+comfy-kitchen==0.2.34
+comfy-aimdo==0.5.5
 requests
 simpleeval>=1.0.0
 blake3
@@ -33,5 +33,5 @@ kornia>=0.7.1
 spandrel
 pydantic~=2.0
 pydantic-settings~=2.0
-PyOpenGL
-glfw
+PyOpenGL>=3.1.8
+comfy-angle