1  # Ollama / LLM Troubleshooting
  2  
  3  Common Ollama and LLM troubleshooting steps for ECHO.
  4  
  5  ## Ollama Not Running
  6  
  7  **Symptom:** `Connection refused` or `Failed to get response from Ollama`
  8  
  9  **Solution:**
 10  ```bash
 11  # Check if Ollama is running
 12  curl http://localhost:11434/api/tags
 13  
 14  # If not running (Docker):
 15  docker ps | grep ollama
 16  docker start ollama
 17  
 18  # If not running (systemd):
 19  systemctl status ollama
 20  systemctl start ollama
 21  ```
 22  
 23  ## Model Not Found
 24  
 25  **Symptom:** `Error: model qwen2.5:14b not found`
 26  
 27  **Solution:**
 28  ```bash
 29  # List installed models
 30  ollama list
 31  
 32  # Pull missing model
 33  ollama pull qwen2.5:14b
 34  
 35  # Or install all ECHO models
 36  ./setup_llms.sh
 37  ```
 38  
 39  ## Slow Responses / Timeouts
 40  
 41  **Symptom:** LLM queries take >60 seconds or timeout
 42  
 43  **Possible Causes:**
 44  - CPU bottleneck (other processes using CPU)
 45  - Memory pressure (not enough RAM for model)
 46  - Model too large for hardware
 47  
 48  **Solutions:**
 49  ```bash
 50  # Check system resources
 51  top  # Linux
 52  Activity Monitor  # macOS
 53  
 54  # Use smaller model (if applicable)
 55  export CEO_MODEL=qwen2.5:7b  # Instead of 14b
 56  
 57  # Increase timeout
 58  export LLM_TIMEOUT=300  # 5 minutes
 59  
 60  # Restart Ollama
 61  docker restart ollama  # If using Docker
 62  ```
 63  
 64  ## Model Inference Errors
 65  
 66  **Symptom:** `(RuntimeError) model inference failed`
 67  
 68  **Debug:**
 69  ```bash
 70  # Test Ollama directly
 71  curl http://localhost:11434/api/generate -d '{
 72    "model": "qwen2.5:14b",
 73    "prompt": "Hello, world!",
 74    "stream": false
 75  }'
 76  
 77  # Check Ollama logs
 78  docker logs ollama  # If using Docker
 79  journalctl -u ollama -f  # If using systemd
 80  ```
 81  
 82  ## Out of Memory
 83  
 84  **Symptom:** `Out of memory` or Ollama crashes
 85  
 86  **Solution:**
 87  ```bash
 88  # Check available memory
 89  free -h  # Linux
 90  vm_stat  # macOS
 91  
 92  # Unload unused models
 93  ollama unload qwen2.5:14b
 94  
 95  # Use smaller models
 96  # See benchmark_models/claude.md for model comparison
 97  ```
 98  
 99  ## GPU Issues
100  
101  **Symptom:** `CUDA error` or `GPU not found`
102  
103  **Solution:**
104  ```bash
105  # Check GPU availability
106  nvidia-smi  # NVIDIA GPUs
107  
108  # Verify Ollama GPU support
109  docker run --gpus all ollama/ollama:latest nvidia-smi
110  
111  # Force CPU mode (if GPU unavailable)
112  export OLLAMA_GPU=false
113  ```
114  
115  **Used in:**
116  - CLAUDE.md (troubleshooting section)
117  - apps/claude.md (agent development)
118  - scripts/claude.md (LocalCode troubleshooting)
119  - benchmark_models/claude.md