attention backend sweep#29
Draft
stacyroberts wants to merge 63 commits into
Draft
Conversation
stacyroberts
force-pushed
the
sroberts-backend-test
branch
from
7c0f99b to
97251af
Compare
Add a new `bfcl:` recipe block for running Berkeley Function Calling Leaderboard categories against a live vLLM server. Results are transformed to lm_eval-compatible JSON so the existing ingestion pipeline and dashboard auto-discover scores as `bfcl_<category>` tasks — zero changes to ingest.py or the dashboard. - lib/parse_workload.py: parse/validate bfcl block, emit BFCL_TSV - lib/run_bfcl.py: register model, run generate+evaluate, transform output - lib/run.sh: dispatch loop for bfcl categories after lm_eval - .buildkite/generate_pipeline.py: conditional bfcl-eval install - workloads/deepseek_v3_2_h200.yaml: enable 4 BFCL categories - README.md: document bfcl block schema This PR was authored with assistance from Claude Code. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Signed-off-by: dougbtv <dosmith@redhat.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
BFCL's OpenAI client was hitting /chat/completions instead of /v1/chat/completions, causing 404s from vLLM. Also fix the JSONL fallback glob to match BFCL V4 filenames (BFCL_v4_<cat>_score.json). Co-Authored-By: Claude <noreply@anthropic.com> Signed-off-by: dougbtv <dosmith@redhat.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
BFCL output was hidden inside the last lm_eval collapsed group. Each category now gets its own `--- ☎️ bfcl <category>` header. Co-Authored-By: Claude <noreply@anthropic.com> Signed-off-by: dougbtv <dosmith@redhat.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
TP=2 × DP=4 + EP with deep_gemm_mega_moe backend and MTP speculative decoding, matching vllm/vllm#42111. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
The speed_bench dataset depends on an external URL (opencompass.openxlab.space) that's unreachable from the B200 k8s pods, causing the entire run to fail before lm_eval runs. Drop vllm_bench for now; can re-add once the dataset is cached. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Commercially critical model with an active perf regression (vllm#40838, ~16% throughput drop v0.16→v0.19). Adding nightly tracking so future regressions are caught at the commit that introduces them. Config: 8xH200 TP=8, gsm8k 5-shot eval, 8k-in/1k-out throughput bench. Signed-off-by: dougbtv <dosmith@redhat.com> Co-Authored-By: Claude <noreply@anthropic.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Adds simple_python, multiple, parallel, parallel_multiple categories with --tool-call-parser openai and --enable-auto-tool-choice serve args. Signed-off-by: dougbtv <dosmith@redhat.com> Co-Authored-By: Claude <noreply@anthropic.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Matches the regression benchmark config from vllm#40838 (TP=4 on H200). Signed-off-by: dougbtv <dosmith@redhat.com> Co-Authored-By: Claude <noreply@anthropic.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Adds simple_python, multiple, parallel, parallel_multiple BFCL categories to DSv4 Pro, DSv4 Flash, GLM 5.1, Kimi K2.5, and MiniMax M2.5. All already had --tool-call-parser and --enable-auto-tool-choice in their serve_args. Signed-off-by: dougbtv <dosmith@redhat.com> Co-Authored-By: Claude <noreply@anthropic.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
B200 jobs pulling from public.ecr.aws hit 429 rate limits, causing ImagePullBackOff failures. Rewrite public ECR URLs to the private pull-through cache (same pattern CI already uses for K8s jobs). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Kevin H. Luu <khluu000@gmail.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Micah Williamson <micah.williamson@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Micah Williamson <micah.williamson@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Micah Williamson <micah.williamson@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Micah Williamson <micah.williamson@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Divin Honnappa <divin.honnappa@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Divin Honnappa <divin.honnappa@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: dhonnappa-amd <dhonnapp@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: dhonnappa-amd <dhonnapp@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Divin Honnappa <divin.honnappa@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Divin Honnappa <divin.honnappa@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Micah Williamson <micah.williamson@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
* add mi300x Signed-off-by: Micah Williamson <micah.williamson@amd.com> * ensure correct k8s plugin is used Signed-off-by: Micah Williamson <micah.williamson@amd.com> * hardcode builder for testing Signed-off-by: Micah Williamson <micah.williamson@amd.com> * revert hardcoded builder Signed-off-by: Micah Williamson <micah.williamson@amd.com> --------- Signed-off-by: Micah Williamson <micah.williamson@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Micah Williamson <micah.williamson@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Micah Williamson <micah.williamson@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Micah Williamson <micah.williamson@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Micah Williamson <micah.williamson@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Stacy Roberts <sroberts@amd.com>
…ng the default Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Stacy Roberts <sroberts@amd.com>
…running on various models (not all complete yet) Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Stacy Roberts <sroberts@amd.com>
…er viewing. All current attn_sweep yamls updated Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Tarun Kumar <takumar@redhat.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
…ect#15) (vllm-project#16) Co-authored-by: Claude Fable 5 <noreply@anthropic.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Tarun Kumar <takumar@redhat.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
…ct#20) Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
…ject#21) Signed-off-by: Tarun Kumar <takumar@redhat.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
stacyroberts
force-pushed
the
sroberts-backend-test
branch
from
e981a1d to
5e229a6
Compare
Signed-off-by: sroberts-amd <sroberts@amd.com>
Signed-off-by: sroberts-amd <sroberts@amd.com>
updated gen_report.py to reflect movement of results files. Signed-off-by: sroberts-amd <sroberts@amd.com>
Signed-off-by: sroberts-amd <sroberts@amd.com>
Signed-off-by: sroberts-amd <sroberts@amd.com>
Signed-off-by: sroberts-amd <sroberts@amd.com>
Signed-off-by: sroberts-amd <sroberts@amd.com>
Signed-off-by: sroberts-amd <sroberts@amd.com>
Signed-off-by: sroberts-amd <sroberts@amd.com>
Signed-off-by: sroberts-amd <sroberts@amd.com>
Signed-off-by: Stacy Roberts <sroberts@amd.com>
…rom AFO-LLM port. Signed-off-by: Stacy Roberts <sroberts@amd.com> Signed-off-by: Stacy Roberts <sroberts@amd.com>
Signed-off-by: Stacy Roberts <sroberts@amd.com>
…processes at end of workflow) Signed-off-by: Stacy Roberts <sroberts@amd.com>
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
6 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.
runs attn_sweep for backends of gpt and deepseek