Update for open LLMs

[HyunjoonCho]

Update for open LLMs

• attach function descriptions to the prompt
• append messages into a single prompt

Analyze function call patterns

• mimicked the figure Function Call Distribution at Each Step from the original AutoFL paper
• need further analysis on syntactic failures

run first few experiments

• llama3-8B
• codellama-7B, 13B
• Caution: results for first runs, i.e., those without update for open LLMs, are lost - only combined FL results are available as *_baseline.json