feat(example): add Strands Tau2-Bench agent example#76
Merged
Conversation
luyuzhe111
reviewed
Jun 16, 2026
luyuzhe111
reviewed
Jun 17, 2026
luyuzhe111
approved these changes
Jun 17, 2026
luyuzhe111
left a comment
luyuzhe111
left a comment
Contributor
There was a problem hiding this comment.
@ChanningPing very neat PR. left a minor comment on agentcore cli installation. feel free to merge after the fix.
Adds a tau2-bench RL agent example supporting airline, retail, and telecom domains with deterministic DB / COMMUNICATE / ENV_ASSERTION / ACTION rewards. The agent runs a multi-turn conversation between a vLLM-served assistant (the model being trained) and a Bedrock-backed user simulator, with tool calls executed against fresh tau2-bench environments. Includes: - rl_app.py with multi-agent orchestration - reward.py implementing four tau-bench reward axes - utils.py with Strands tool wrapping and message-role conversion - 3 sample tasks (one per domain) sourced from tau2-bench - test_local.py smoke test (local server or deployed ACR agent) - Dockerfile with pinned tau2-bench commit - README walking developers through deployment end-to-end
Drop the legacy client-side vLLMModel wrapper in favor of standard strands.models.openai.OpenAIModel. Token IDs are now captured server-side by the rllm-model-gateway HTTP proxy used in the verl backend, so get_token_data() / rollout_data are no longer collected client-side.
Reorder Installation steps so uv venv + activation happen before any pip install — keeps every dependency (agentcore CLI, example deps, local toolkit) inside the venv instead of polluting the system Python.
ChanningPing
force-pushed
the
feat/strands-taubench-agent
branch
from
16452ba to
c96b9ae
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
2 participants
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.
Description of changes:
Adds a tau2-bench RL agent example supporting airline, retail, and telecom domains with deterministic DB / COMMUNICATE / ENV_ASSERTION / ACTION rewards. The agent runs a multi-turn conversation between a vLLM-served assistant (the model being trained) and a Bedrock-backed user simulator, with tool calls executed against fresh tau2-bench environments.
Includes:
Test plan
python3 -m py_compile)rl_app.pyserver with airline task — produces full trajectory and rewardBy submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.