According to our error analysis, we observe that open-source LLMs often face difficulty in 1. API selection
2. API argument population,
3. generating legitimate and executable code. (6/10)

→ View original post on X — @sambanovaai,