AI Coding BATTLE | Which Open Source Model is BEST?

---

---

AI Summary

Summary of Video Transcript

Coding Models Compared

• Models Tested: Deep Sea Coder V2, Yoder 9B Chat, Quen 2.5 Coder 7B
• Criteria: Local coding without internet
• Hardware: Dell Precision 5860 with two RTX A6000 GPUs (48 GB VRAM each)

Test 1: Snake Game in Python

• Deep Sea Coder V2:
  • Speed: 30 tokens per second
  • Used tkinter library
  • Game loaded but had issues
• Yoder 9B Chat:
  • Speed: 50 tokens per second
  • Used turtle library
  • Game worked better but had minor issues
• Quen 2.5 Coder 7B:
  • Speed: 70 tokens per second
  • Used pygame library
  • Best performance, game worked correctly

Test 2: Tetris Game in Python

• Deep Sea Coder V2:
  • Used sys, time, and random libraries
  • Game was laggy and incomplete
• Yoder 9B Chat:
  • Used turtle library
  • Game did not work and froze
• Quen 2.5 Coder 7B:
  • Used pygame library
  • Game had errors and did not work

Test 3: Coding Challenges on Code Wars

• Challenge 1: Move 10 Letters
  • All models passed the challenge
• Challenge 2: 1K Kai
  • Deep Sea Coder V2: Failed with 166 passes and 52 fails
  • Yoder 9B Chat: Slightly better but still failed
  • Quen 2.5 Coder 7B: Failed with 120 passes and 98 fails
• Challenge 3: Prime Streaming PG-13
  • All models timed out and failed the challenge

Test 4: Median of Two Sorted Arrays

• All models successfully completed the challenge

Conclusion

• Quen 2.5 Coder 7B was the overall winner, with the best speed and performance
• The Dell Precision 5860 with Nvidia RTX A6000 GPUs was capable of handling all models simultaneously

Additional Notes

• The video was sponsored by Dell and Nvidia
• The tests were run through LM Studio
• The video creator is open to suggestions for further tests

Detailed Instructions and URLs

• No specific CLI commands, website URLs, or detailed instructions were provided in the transcript.