gpt-oss · Issues· 149 open

Duplicate call_id silently rebinds function_call_output to the wrong function in the rendered prompt

GPQA debug mode ignores an explicit --examples sample count

Eval CLI --examples crashes non-debug AIME and GPQA because n_repeats stays 8

Browser HTML conversion can leak global html2text monkey patches after an exception

AIME and GPQA treat num_examples=0 as the full dataset

HealthBench debug mode ignores an explicit --examples override

Distributed chat executes each tool call independently on every model-parallel rank

Eval progress helper crashes on an empty sample set

Local Harmony tokenizer assigns token ID 200018 to two special tokens

Nullable Responses request fields can crash the local server

Responses stub backend does not reset fake output on new requests

Responses Ollama backend injects token 0 when no streamed token is ready

Responses API default checkpoint path is not expanded

Responses vLLM backend passes TP environment value as a string

Responses API Triton backend binds CUDA devices using global rank