feat: add vLLM 0.16 mistral option parity for exllamav3#415

Open

lesj0610 wants to merge 19 commits intotheroyallab:mainfrom

lesj0610:feat/vllm-mistral-parity

Commits on Feb 24, 2026

feat(oai): add vLLM-style reasoning parsers and robust tool-call compatibility
lesj0610
committed
fix(oai): map top-level thinking flags into chat template kwargs
lesj0610
committed
feat(oai): add vllm-compatible parser options and reasoning/tool parsing
lesj0610
committed
fix(reasoning): make exaone4 parser independent and add tests
lesj0610
committed
feat(oai): add vLLM-style tool parser registry and native parser flows
lesj0610
committed
Fix exaone4 streaming reasoning split across think/tool boundaries
lesj0610
committed
Full tool-calling support: XML parsing, streaming compliance, Pydantic fix, inference abort fix

devnen
authored and
lesj0610
committed
Broader model compatibility, tool_choice support, bug fixes and cleanup

devnen
authored and
lesj0610
committed
fix(oai): preserve parser_key parser dispatch after stacking PRs
lesj0610
committed
feat(oai): add mistral tool-call parser compatibility
lesj0610
committed
feat(config): add tokenizer_mode and mistral-safe tool ID handling
lesj0610
committed
test(reasoning): align mistral parser token handling with mistral_common
lesj0610
committed
test(model): skip exllamav2 model tests when dependency is missing
lesj0610
committed

Commits on Feb 25, 2026

Align ExLlama tokenizer modes with vLLM and enforce flashinfer path
lesj0610
committed

Commits on Feb 26, 2026

Commits on Mar 2, 2026