initial: OpenClaw + custom 70B IQ2_XS + Telegram

Dockerfile

@@ -0,0 +1,18 @@
+FROM python:3.11-slim-bookworm
+
+RUN apt-get update && apt-get install -y \
+    curl nginx cron procps git \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN curl -fsSL https://ollama.com/install.sh | sh
+
+RUN pip install --no-cache-dir openclaw huggingface_hub
+
+RUN printf 'server {\n    listen 7860;\n    location / {\n        proxy_pass http://127.0.0.1:8080;\n        proxy_http_version 1.1;\n        proxy_set_header Upgrade $http_upgrade;\n        proxy_set_header Connection "upgrade";\n        proxy_set_header Host $host;\n    }\n}\n' > /etc/nginx/sites-available/default
+
+WORKDIR /app
+COPY . .
+RUN chmod +x start.sh
+
+EXPOSE 7860
+CMD ["/app/start.sh"]

SOUL.md

@@ -0,0 +1,27 @@
+# YT Pipeline Assistant
+
+You are the AI assistant for a YouTube Shorts automation pipeline. You run on a Hugging Face Space with DeepSeek R1 Distill Llama 70B via Ollama.
+
+## Your Environment
+- **You**: OpenClaw on HF Space `yt-claw`
+- **Pipeline**: HF Space `yt-flow` (automation.py orchestrates Kaggle kernels)
+- **Codebase for editing**: `/app/yt-flow/` — git-enabled, you can commit and push
+- **State backup**: `~/.openclaw/` synced to HF Dataset every 15 min
+
+## Pipeline Steps
+1. CONTENT — Kaggle kernel with DeepSeek-R1-32B GGUF
+2. FLUX — Kaggle kernel generates images
+3. TTS — Kaggle kernel with StyleTTS2 + Whisper
+
+## Your Capabilities
+- Read pipeline logs at `https://arshitmalik-yt-flow.hf.space`
+- Edit code in `/app/yt-flow/` and push (git add → commit → push)
+- HF auto-rebuilds yt-flow on push
+- Debug Kaggle kernel failures by analyzing kernel output
+- Monitor topic/fact history to avoid repetition
+
+## Common Bugs You Can Fix
+- Kaggle 429 → increase time.sleep between kernel pushes
+- llama-cpp build failure → use --only-binary :all: in pip install
+- Space paused → check Ollama memory, reduce context if needed
+- YouTube auth expired → remind user to run auth_youtube.py from Mac

start.sh

@@ -0,0 +1,157 @@
+#!/bin/bash
+set -e
+
+echo "[boot] Creating 10 GB swap..."
+fallocate -l 10G /swapfile 2>/dev/null || dd if=/dev/zero of=/swapfile bs=1M count=10240 2>/dev/null
+chmod 600 /swapfile && mkswap /swapfile && swapon /swapfile
+
+echo "[boot] Pulling OpenClaw state..."
+python3 -c "
+import sys, os; sys.path.insert(0, '/app')
+HF_TOKEN = os.environ.get('HF_TOKEN', '')
+DATASET_REPO = os.environ.get('OPENCLAW_DATASET_REPO', '')
+if HF_TOKEN and DATASET_REPO:
+    from huggingface_hub import hf_hub_download
+    from huggingface_hub.utils import EntryNotFoundError
+    for fname in ['openclaw.json', 'SOUL.md']:
+        try:
+            hf_hub_download(repo_id=DATASET_REPO, filename=f'openclaw/{fname}',
+                          repo_type='dataset', token=HF_TOKEN,
+                          local_dir='/root', local_dir_use_symlinks=False)
+        except EntryNotFoundError: pass
+        except Exception as e: print(f'  {e}')
+"
+
+echo "[boot] Configuring OpenClaw..."
+mkdir -p ~/.openclaw
+cat > ~/.openclaw/openclaw.json << JSONEOF
+{
+  "gateway": {"port": 8080, "bind": "0.0.0.0"},
+  "identity": {"name": "YT Pipeline Assistant"},
+  "channels": {
+    "telegram": {
+      "enabled": true,
+      "botToken": "${TELEGRAM_BOT_TOKEN}",
+      "dmPolicy": "pairing"
+    }
+  },
+  "models": {
+    "providers": {
+      "ollama": {
+        "baseUrl": "http://127.0.0.1:11434/v1",
+        "apiKey": "ollama-local",
+        "api": "openai-chat",
+        "models": [
+          {
+            "id": "deepseek-r1-70b",
+            "name": "DeepSeek R1 Distill Llama 70B IQ2_XS",
+            "contextWindow": 131072,
+            "maxTokens": 8192
+          }
+        ]
+      }
+    }
+  },
+  "agents": {
+    "defaults": {
+      "model": {"primary": "ollama/deepseek-r1-70b"},
+      "workspace": "/app/yt-flow"
+    }
+  }
+}
+JSONEOF
+
+[ -f /app/SOUL.md ] && cp /app/SOUL.md ~/.openclaw/SOUL.md
+
+git config --global user.name "YT Pipeline Bot"
+git config --global user.email "yt-pipe-bot@hf.space"
+git clone https://arshitmalik:${HF_TOKEN}@huggingface.co/spaces/arshitmalik/yt-flow /app/yt-flow 2>/dev/null || true
+
+# Backup cron
+cat > /tmp/backup.sh << 'CRONEOF'
+#!/bin/bash
+python3 -c "
+import os, sys; sys.path.insert(0, '/app')
+HF_TOKEN = os.environ.get('HF_TOKEN', '')
+DATASET_REPO = os.environ.get('OPENCLAW_DATASET_REPO', '')
+if not HF_TOKEN or not DATASET_REPO: sys.exit(0)
+from huggingface_hub import HfApi
+api = HfApi(token=HF_TOKEN)
+from pathlib import Path
+for f in Path('/root/.openclaw').glob('*'):
+    if f.is_file():
+        try:
+            api.upload_file(
+                path_or_fileobj=str(f),
+                path_in_repo=f'openclaw/{f.name}',
+                repo_id=DATASET_REPO,
+                repo_type='dataset'
+            )
+        except Exception as e: pass
+" 2>/dev/null
+CRONEOF
+chmod +x /tmp/backup.sh
+echo "*/15 * * * * /tmp/backup.sh" | crontab -
+
+# Temporary health server (keeps Space alive during setup)
+python3 -c "
+from http.server import HTTPServer, BaseHTTPRequestHandler
+class H(BaseHTTPRequestHandler):
+    def do_GET(self):
+        self.send_response(200); self.end_headers(); self.wfile.write(b'ok')
+HTTPServer(('0.0.0.0', 8080), H).serve_forever()
+" &
+HEALTH_PID=$!
+
+echo "[boot] Starting nginx..."
+nginx
+
+echo "[boot] Starting Ollama..."
+export OLLAMA_FLASH_ATTENTION=1
+export OLLAMA_KV_CACHE_TYPE=q4_0
+OLLAMA_HOST=127.0.0.1 OLLAMA_NUM_PARALLEL=1 OLLAMA_MAX_LOADED_MODELS=1 ollama serve &
+
+for i in $(seq 1 30); do
+    if curl -sf http://127.0.0.1:11434/api/tags > /dev/null 2>&1; then
+        echo "[boot] Ollama ready after ${i}s"; break
+    fi
+    sleep 2
+done
+
+echo "[boot] Building custom 70B IQ2_XS model (~3 min)..."
+python3 << PYEOF
+import subprocess, sys, os
+from pathlib import Path
+from huggingface_hub import hf_hub_download
+
+gguf_path = hf_hub_download(
+    repo_id="bartowski/DeepSeek-R1-Distill-Llama-70B-GGUF",
+    filename="DeepSeek-R1-Distill-Llama-70B-IQ2_XS.gguf",
+    local_dir="/tmp/gguf",
+    resume_download=True,
+    token=os.environ.get("HF_TOKEN", None)
+)
+modelfile = f"""FROM {gguf_path}
+TEMPLATE \"\"\"<｜begin▁of▁sentence｜>{{ .System }}
+
+{{ .Prompt }}<｜end▁of▁sentence｜>
+<｜begin▁of▁sentence｜>assistant
+\"\"\"
+PARAMETER stop "<｜begin▁of▁sentence｜>"
+PARAMETER stop "<｜end▁of▁sentence｜>"
+PARAMETER num_ctx 131072
+PARAMETER num_predict 8192
+"""
+(Path("/tmp") / "Modelfile").write_text(modelfile)
+subprocess.run(["ollama", "create", "deepseek-r1-70b", "-f", "/tmp/Modelfile"], check=True)
+Path(gguf_path).unlink()
+PYEOF
+
+# Kill temp health server, start OpenClaw
+kill $HEALTH_PID 2>/dev/null
+echo "[boot] Starting OpenClaw gateway..."
+openclaw gateway &
+OPENCLAW_PID=$!
+echo "[boot] Deployment complete. PID: $OPENCLAW_PID"
+wait $OPENCLAW_PID 2>/dev/null
+sleep infinity