Troubleshooting

This page provides quick access to troubleshooting resources and solutions for common DNALLM issues.

Quick Links

🔧 Comprehensive FAQ

For detailed solutions to common problems, see our Frequently Asked Questions (FAQ) page, which covers:

• Installation Issues: Package installation, dependency conflicts, network problems
• Training Issues: CUDA out of memory, loss instability, optimization problems
• Model Loading: Custom architectures, tokenizer mismatches, cache issues
• Performance Issues: Memory optimization, speed improvements, hardware requirements
• Task-Specific Issues: Model-task mismatches, label encoding problems

📚 Related Resources

• Model Selection Guide: Choose the right model for your task
• Model Troubleshooting: Model-specific issues and solutions
• Performance Optimization: Speed and memory optimization guides
• Installation Guide: Complete installation instructions

Common Quick Fixes

Installation Problems

# For Mamba models
uv pip install -e '.[mamba]' --no-cache-dir --no-build-isolation

# For EVO models
uv pip install evo-model  # EVO-1
uv pip install evo2       # EVO-2

# For network issues
export HTTP_PROXY="http://your.proxy.server:port"
export HTTPS_PROXY="http://your.proxy.server:port"

Memory Issues

# In your config file
finetune:
  gradient_accumulation_steps: 4
  per_device_train_batch_size: 2
  fp16: true

Model Loading

# trust_remote_code is applied internally where needed — just pass a TaskConfig
from dnallm.configuration.configs import TaskConfig

task_config = TaskConfig(task_type="binary", num_labels=2)
model, tokenizer = load_model_and_tokenizer("model_name", task_config, source="huggingface")

Still Need Help?

If you can't find the answer to your question:

1. Check the FAQ for comprehensive solutions
2. Search GitHub Issues for similar problems
3. Create a new issue with detailed information about your problem
4. Join community discussions on GitHub

---

For the most up-to-date troubleshooting information, always refer to the FAQ page.