llama.cpp fails to return valid structured output

2 posts from 2 people2d activeSteady, +1 this weekunverified

Posts per day2 posts, Sep 17 to Oct 30

The posts behind it

GitHub issuesggml-org/llama.cpp · samsimsomOct 26

Misc. bug: json-schema-to-grammar copies raw control characters (e.g. NUL from "\u0000") in "pattern" verbatim into GBNF → "failed to parse grammar"

### Name and Version `common/json-schema-to-grammar.cpp` on current master (last touched in 48c9f08b7) still has the code path described below. ### Operating systems Mac ### Which llama.cpp modules do you know to be affected? llama-server (`common/json-schema-to-grammar.cpp`)...

3 replies

GitHub issuesggml-org/llama.cpp · mattepiuSep 17

Feature Request: Fast Tool Gating & Single-Pass Selection via Prefill Logit Slicing

### Prerequisites - [x] I am running the latest code. Mention the version if possible as well. - [x] I carefully followed the [README.md](https://github.com/ggml-org/llama.cpp/blob/master/README.md). - [x] I searched using keywords relevant to my issue to make sure that I am...

3 reactions10 replies

Companies and products named

llama.cpp

2 posts

GitHub

1 post

About this problem

Evidence

2 posts from 2 people in 1 place, about 1 a week over 40 days. Mostly on ggml-org/llama.cpp. Tools named alongside: GitHub.

Frustration Frustration 3 of 3· Seen on GitHub issues

How it was grouped

Posts that state a pain and match the "structured-output" rule about llama.cpp. First post Sep 17, 2026, latest Oct 26, 2026. Unverified: fewer than 3 posts, or one person or place only. Method

The same problem elsewhere

First seen with 2 posts.

Oct 30, 2026

llama.cpp fails to return valid structured output

The posts behind it

Misc. bug: json-schema-to-grammar copies raw control characters (e.g. NUL from "\u0000") in "pattern" verbatim into GBNF → "failed to parse grammar"

Feature Request: Fast Tool Gating & Single-Pass Selection via Prefill Logit Slicing

Companies and products named

llama.cpp

GitHub

Models fail to return valid structured output

Claude fails to return valid structured output

OpenAI fails to return valid structured output

LlamaIndex fails to return valid structured output

Model Context Protocol fails to return valid structured output

Codex fails to return valid structured output

Ollama fails to return valid structured output

Anthropic fails to return valid structured output

Gemini fails to return valid structured output

vLLM fails to return valid structured output

Qwen fails to return valid structured output

Mistral fails to return valid structured output

Running llama.cpp locally hits memory and speed limits

Trouble with llama.cpp misc

Trouble with llama.cpp compile

llama.cpp users keep filing bugs

llama.cpp upgrades break working setups

Fine-tuning llama.cpp is costly and unpredictable

llama.cpp runs out of context on real work

llama.cpp lock-in makes switching costly

llama.cpp error messages are cryptic

Trouble with llama.cpp flash

Trouble with llama.cpp builds

llama.cpp runs out of memory

First seen with 2 posts.