Misc. bug: JSON schema that defines array with 0 elements generates un-parseable GBNF

[rick-github]

Name and Version

$ ./build/bin/llama-cli --version
version: 5188 (514c456)
built with cc (Ubuntu 13.2.0-23ubuntu4) 13.2.0 for x86_64-linux-gnu

Operating systems

Linux

Which llama.cpp modules do you know to be affected?

llama-cli

Command line

./build/bin/llama-cli -m ./tmp/mnt/models/quantize/gemma-1.1-2b-it.Q8_0.gguf -j "$(python -c $'import json, pydantic\nclass Result(pydantic.BaseModel):  colours:list[str]=pydantic.Field(max_length=0)\nprint(json.dumps(Result.model_json_schema()))')" --no-display-prompt  -p "Here are some colours: " -no-cnv

Problem description & steps to reproduce

A syntactically correct (albeit nonsensical) JSON schema that defines an element as a 0-length array creates a GBNF grammar that causes llama-cli to error with a failed to parse grammar error.

For example, this code:

import json, pydantic
class Result(pydantic.BaseModel):  
  colours:list[str]=pydantic.Field(max_length=0)
print(json.dumps(Result.model_json_schema()))

creates this schema:

{
    "properties": {
        "colours": {
            "items": {
                "type": "string"
            },
            "maxItems": 0,
            "title": "Colours",
            "type": "array"
        }
    },
    "required": [
        "colours"
    ],
    "title": "Result",
    "type": "object"
}

and passing to llama-cli results in an error:

$ ./build/bin/llama-cli -m ./tmp/mnt/models/quantize/gemma-1.1-2b-it.Q8_0.gguf -j "$(python -c $'import json, pydantic\nclass Result(pydantic.BaseModel):  colours:list[str]=pydantic.Field(max_length=0)\nprint(json.dumps(Result.model_json_schema()))')" --no-display-prompt  -p "Here are some colours: " -no-cnv
...

system_info: n_threads = 8 (n_threads_batch = 8) / 24 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | LLAMAFILE = 1 | OPENMP = 1 | AARCH64_REPACK = 1 | 

parse: error parsing grammar: expecting '}' at -1})? "]" space
colours-kv ::= "\"colours\"" space ":" space colours
root ::= "{" space colours-kv "}" space
space ::= | " " | "\n"{1,2} [ \t]{0,20}
string ::= "\"" char* "\"" space


char ::= [^"\\\x7F\x00-\x1F] | [\\] (["\\bfnrt] | "u" [0-9a-fA-F]{4})
colours ::= "[" space (string ("," space string){0,-1})? "]" space
colours-kv ::= "\"colours\"" space ":" space colours
root ::= "{" space colours-kv "}" space
space ::= | " " | "\n"{1,2} [ \t]{0,20}
string ::= "\"" char* "\"" space

llama_grammar_init_impl: failed to parse grammar
main: failed to initialize sampling subsystem

First Bad Commit

55b2d08

Relevant log output

common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
main: llama threadpool init, n_threads = 8

system_info: n_threads = 8 (n_threads_batch = 8) / 24 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | LLAMAFILE = 1 | OPENMP = 1 | AARCH64_REPACK = 1 | 

parse: error parsing grammar: expecting '}' at -1})? "]" space
colours-kv ::= "\"colours\"" space ":" space colours
root ::= "{" space colours-kv "}" space
space ::= | " " | "\n"{1,2} [ \t]{0,20}
string ::= "\"" char* "\"" space


char ::= [^"\\\x7F\x00-\x1F] | [\\] (["\\bfnrt] | "u" [0-9a-fA-F]{4})
colours ::= "[" space (string ("," space string){0,-1})? "]" space
colours-kv ::= "\"colours\"" space ":" space colours
root ::= "{" space colours-kv "}" space
space ::= | " " | "\n"{1,2} [ \t]{0,20}
string ::= "\"" char* "\"" space

llama_grammar_init_impl: failed to parse grammar
main: failed to initialize sampling subsystem