[DYNAMO] wire TITO client to Dynamo nvext transport#2446
Draft
AmeenP wants to merge 4 commits into
Draft
Conversation
cfe5428 to
4a2503d
Compare
9cad934 to
360f0c1
Compare
4a2503d to
3f89cf5
Compare
setup_clients() previously only routed renderer-mode clients onto the
dynamo_chat_nvext wire shape when client.backend = "dynamo". The TITO
client (openai_chat_completions_token) was left on the legacy
prime_vllm_generate path, so it kept POSTing to /v1/chat/completions/
tokens and /tokenize -- both 404 on Dynamo bis/dynamo-rl, which forced
multi-turn TITO to silently degrade to MITO via the verifiers fallback.
Extend the selector to cover both token-aware client types so that
use_token_client = true + client.backend = "dynamo" picks the
dynamo_chat_nvext shape automatically. Pairs with the verifiers
feat/tito-dynamo-transport patch which actually consumes
ClientConfig.renderer_transport in OpenAIChatCompletionsTokenClient.
Validated end-to-end via:
- bis-dev/4-02/tito-multiturn (multi-turn TITO + full-FT)
- bis-dev/5/sft (multi-turn TITO + full-FT, with
inline Probe 3 + SIDECAR-SKIP-TOKENIZE)
- bis-dev/5/lora (multi-turn TITO + LoRA)
3f89cf5 to
058755e
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Ports the missing prime-rl client-side TITO wiring from Biswa's tested
biswapanda/prime-rl@bis/prime-rl-mergedbranch on top of #2420 (feat/dynamo-renderer-transport-config).renderersrepo pin from [DYNAMO] pass renderer transport to verifiers #2420 (PrimeIntellect-ai/renderers@7ca1ab3, PR vLLM weight save/load and (mostly) end to end. (Don't merge) #11)4bb3da64, verifiers fix gradio #1313)openai_chat_completions_tokenas token-aware forclient.backend = "dynamo"renderer_transport='dynamo_chat_nvext'for both renderer-mode and TITO clients when using DynamoContext
This pairs with the verifiers follow-up PR that teaches
OpenAIChatCompletionsTokenClientto consumerenderer_transport='dynamo_chat_nvext': PrimeIntellect-ai/verifiers#1313Validation
uv lock --checkuvx ruff==0.13.0 check src/prime_rl/utils/client.py src/prime_rl/utils/elastic.py packages/prime-rl-configs/src/prime_rl/configs/shared.py tests/unit/utils/test_client.py tests/unit/utils/test_elastic.pyuvx ruff==0.13.0 format --check src/prime_rl/utils/client.py src/prime_rl/utils/elastic.py packages/prime-rl-configs/src/prime_rl/configs/shared.py tests/unit/utils/test_client.py tests/unit/utils/test_elastic.pypython -m py_compile src/prime_rl/utils/client.py src/prime_rl/utils/elastic.py packages/prime-rl-configs/src/prime_rl/configs/shared.py tests/unit/utils/test_client.py tests/unit/utils/test_elastic.pyFull pytest is left to Linux CI for this stack because the local checkout is macOS while the lockfile targets Linux environments.