mock: cache-aware prefill/decode latency so disaggregated TTFT/TPS tests are measurable#4246
Draft
ddjukicTT wants to merge 11 commits into
Draft
mock: cache-aware prefill/decode latency so disaggregated TTFT/TPS tests are measurable#4246ddjukicTT wants to merge 11 commits into
ddjukicTT wants to merge 11 commits into
Conversation
dmadicTT
reviewed
Jun 16, 2026
dmadicTT
reviewed
Jun 16, 2026
dmadicTT
reviewed
Jun 16, 2026
| echo "DYNAMO_ENDPOINT_NAME=generate" | ||
| } | ||
|
|
||
| start_frontend() { |
Collaborator
There was a problem hiding this comment.
Can we run deploy.sh instead of run_stack.sh?
dmadicTT
reviewed
Jun 16, 2026
| // active sequence and takes MOCK_DECODE_SLEEP_US, modeling inter-token latency | ||
| // so the decode tokens-per-second (TPS ≈ 1e6 / MOCK_DECODE_SLEEP_US) is | ||
| // measurable on the mock. Default 0 (tokens emitted as fast as the loop runs). | ||
| std::chrono::microseconds mockDecodeDelay() { |
Collaborator
There was a problem hiding this comment.
LLM_DEVICE_BACKEND=mock_pipeline already has realistic model latency https://github.com/tenstorrent/tt-inference-server/blob/main/tt-media-server/cpp_server/include/runtime/runners/blaze_runner/blaze_utils.hpp#L191
dmadicTT
reviewed
Jun 16, 2026
Collaborator
There was a problem hiding this comment.
In general, we should think switching to docker-compose to deploy multiple related docker containers instead of multiple bash scripts
…rver into ddjukic/prefill-functional-requirements-test # Conflicts: # run.py # workflows/runtime_config.py # workflows/v2_bridge.py # workflows/validate_setup.py
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Make the prefill/decode disaggregation smoke suite runnable end-to-end from the top-level
run.py, against a self-contained mock stack that's faithful enough to measure prefix-cache behaviorCommands:
test_07fails on the mock simulator: the prefix cache is correct , but its TTFT-ratio assertion trips because simulator TTFT (~40 ms) is dominated by prompt transport/tokenization, not on-device prefill. Sim artifact, not a regression; tunable via TTFT_MEANINGFUL_S / TTFT_HIT_MAX_FRACTION.