feat(clients): add dynamo_chat renderer transport (TITO over Dynamo)#1574

Open

biswapanda wants to merge 19 commits into

PrimeIntellect-ai:mainfrom

biswapanda:rl-sdk-4

Commits on Jun 9, 2026

feat(types): add RendererTransport literal + ClientConfig.renderer_transport
biswapanda
committed
feat(clients): graft nvext.engine_data onto OpenAI response in parse_tokens
biswapanda
committed
feat(tito): add dynamo_chat_nvext transport + local bridge tokenize
biswapanda
committed
feat(clients): graft top-level nvext.completion_token_ids + prompt_token_ids (plan B3)
biswapanda
committed
feat(clients): thread renderer_transport from ClientConfig to renderers.generate()
biswapanda
committed
fix(clients): address PR review R1-R5 (guard transport kwarg, import ChatCompletion, scrub return_token_ids, forward sampling args, graft engine_data logprobs) + rename to dynamo_chat
biswapanda
committed
fix(clients): graft engine_data logprobs even when choice logprobs is content-less; trim test comments
biswapanda
committed
fix(clients): dynamo_chat forwards full normalized sampling_args (drop fixed allowlist) for vLLM-path parity
biswapanda
committed
fix(clients): centralize Dynamo denylist scrub (MITO+TITO), guard logprob length, tokenizer override, drop dead renderer field
biswapanda
committed
fix(clients): enforce logprobs/ids length invariant in parse_tokens (all paths)
biswapanda
committed
fix(clients): centralize tokenizer override in _get_local_tokenizer; route dynamo TITO through routed-experts sidecar helper
biswapanda
committed
fix(clients): load HF tokenizer inside worker thread (cache-miss from_pretrained must not block the event loop)
biswapanda
committed

Commits on Jun 10, 2026

Commits on Jun 11, 2026

docs(clients): drop PR-number and branch/plan references from dynamo_chat comments
biswapanda
committed

Commits on Jun 12, 2026