[None][test] Remove all 1k1k perf-sanity cases from CI - #16535
Merged
Conversation
chenfeiz0326
requested review from
brnguyen2,
joyang-nv,
longlee0622,
mzweilz,
niukuo,
ruodil and
schetlur-nv
July 17, 2026 07:24
Contributor
📝 WalkthroughWalkthroughUpdated B200, GB200, and GB300 perf sanity test matrices by removing or replacing selected parametrizations, primarily shifting toward larger sequence variants. Corresponding perf sanity waiver entries were also revised. ChangesPerf sanity matrix updates
Estimated code review effort: 3 (Moderate) | ~20 minutes Possibly related PRs
Suggested reviewers: 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
longlee0622
approved these changes
Jul 17, 2026
mzweilz
approved these changes
Jul 17, 2026
1 task
chenfeiz0326
force-pushed
the
remove-1k1k-disagg
branch
from
July 17, 2026 09:02
66f7267 to
bf3d123
Compare
Collaborator
Author
|
Follow-up to my earlier review — reverted the AutoDeploy-cleanup commit and restored the Current state (force-pushed):
The other 3 AutoDeploy 1k1k cases in |
chenfeiz0326
enabled auto-merge (squash)
July 17, 2026 09:11
Collaborator
Author
|
/bot run --disable-fail-fast --stage-list "*PerfSanity*" |
Collaborator
|
PR_Github #59964 [ run ] triggered by Bot. Commit: |
Collaborator
|
PR_Github #59964 [ run ] completed with state
|
Drop every 1k1k case (aggregated + disaggregated, all modes: e2e /
gen_only / ctx_only) from the l0_*_perf_sanity_*.yml test-db lists and
from waives.txt. Config YAMLs under tests/scripts/perf-sanity/{aggregated,
disaggregated}/ are preserved on disk so re-adding a case is a
one-line-per-case restore.
20 files changed, 94 deletions.
Signed-off-by: chenfeiz0326 <chenfeiz@nvidia.com>
The previous commit dropped the SKIP for aggr_upload-super_ad_blackwell-super_ad_ws1_1k1k under the assumption that the case itself was being removed from CI. It isn't: the AutoDeploy 1k1k perf-sanity cases live in l0_b200.yml / l0_dgx_b200.yml and are outside the scope of this PR (they belong to the AutoDeploy team). Restore the waive for nvbug 6153575 so the still-active case stays gated until AutoDeploy owners decide how to handle it. Signed-off-by: chenfeiz0326 <chenfeiz@nvidia.com>
…moval
The prior commit deleted every 1k1k perf-sanity entry from the test-db
YAMLs but left the corresponding stages in jenkins/L0_Test.groovy
allocating GPUs for empty test lists. Four multi-node disagg YAMLs also
went completely empty and are now unreferenced.
Changes to jenkins/L0_Test.groovy:
- Delete 4 stage blocks whose YAMLs are now empty:
* l0_gb200_multi_nodes_perf_sanity_ctx1_node1_gpu4_gen1_node1_gpu4
* l0_gb200_multi_nodes_perf_sanity_ctx2_node1_gpu4_gen1_node4_gpu16
* l0_gb300_multi_nodes_perf_sanity_ctx1_node1_gpu4_gen1_node1_gpu4
* l0_gb300_multi_nodes_perf_sanity_ctx1_node1_gpu2_gen1_node1_gpu4
- Reduce aggregated stage counts to hit the 6-tests-per-stage cap:
* DGX_B200-8_GPUs-PyTorch-PerfSanity-Post-Merge: 6 -> 4
* GB200-4_GPUs-PyTorch-PerfSanity-Post-Merge: 9 -> 4
* GB300-4_GPUs-PyTorch-PerfSanity-Post-Merge: 3 -> 2
- Reduce disagg split counts to match remaining test count so we
don't schedule empty shards:
* l0_b200_multi_nodes_perf_sanity_ctx1_node1_gpu4_gen1_node1_gpu8: 6 -> 3
* l0_gb200_multi_nodes_perf_sanity_ctx1_node1_gpu1_gen1_node1_gpu2: 4 -> 1
* l0_gb200_multi_nodes_perf_sanity_ctx1_node1_gpu1_gen1_node1_gpu4: 7 -> 6
* l0_gb200_multi_nodes_perf_sanity_ctx1_node1_gpu4_gen1_node2_gpu8: 8 -> 3
* l0_gb200_multi_nodes_perf_sanity_ctx1_node1_gpu4_gen1_node8_gpu32: 12 -> 8
* l0_gb300_multi_nodes_perf_sanity_node2_gpu8: 3 -> 2
* l0_gb300_multi_nodes_perf_sanity_ctx1_node1_gpu4_gen1_node2_gpu8: 5 -> 3
* l0_gb300_multi_nodes_perf_sanity_ctx1_node1_gpu4_gen1_node8_gpu32: 2 -> 1
* l0_gb300_multi_nodes_perf_sanity_ctx1_node1_gpu2_gen1_node2_gpu8: 5 -> 3
* l0_gb300_multi_nodes_perf_sanity_ctx1_node1_gpu2_gen1_node8_gpu32: 2 -> 1
Also delete the 4 empty YAML files themselves and drop their now-dead
paths from getMultiGpuFileChanged()'s trigger list in
jenkins/L0_MergeRequest.groovy.
Signed-off-by: chenfeiz0326 <chenfeiz@nvidia.com>
chenfeiz0326
force-pushed
the
remove-1k1k-disagg
branch
from
July 17, 2026 14:56
1730e7d to
4f73835
Compare
Collaborator
Author
|
/bot skip --comment "Remove some perf tests, no need to run the whole CI pipeline" |
Collaborator
|
PR_Github #60000 [ skip ] triggered by Bot. Commit: |
Collaborator
|
PR_Github #60000 [ skip ] completed with state |
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
5 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.
Summary
Removes every 1k1k perf-sanity case (aggregated + disaggregated, all modes:
e2e,gen_only,ctx_only) registered in the perf-sanity L0 lists(
tests/integration/test_lists/test-db/l0_*_perf_sanity_*.yml), plus thecorresponding
waives.txtentries. Underlying config YAMLs undertests/scripts/perf-sanity/disaggregated/andtests/scripts/perf-sanity/aggregated/are not deleted, so a futureun-remove is a one-line-per-case re-add.
Diff shape: 22 files changed, 226 deletions, 19 insertions.
tests/integration/test_lists/test-db/l0_*_perf_sanity_*.ymlentries stripped of all
test_e2e[..._1k1k_...]andtest_e2e[...*_1k1k]lines.tests/integration/test_lists/test-db/l0_*_perf_sanity_*.ymlfully deleted because every remaining test in the file was a 1k1k
case:
l0_gb200_multi_nodes_perf_sanity_ctx1_node1_gpu4_gen1_node1_gpu4.ymll0_gb200_multi_nodes_perf_sanity_ctx2_node1_gpu4_gen1_node4_gpu16.ymll0_gb300_multi_nodes_perf_sanity_ctx1_node1_gpu2_gen1_node1_gpu4.ymll0_gb300_multi_nodes_perf_sanity_ctx1_node1_gpu4_gen1_node1_gpu4.ymltests/integration/test_lists/waives.txtSKIP lines removed(see caveat below).
jenkins/L0_Test.groovy(+19 / −61) — Jenkins stage countsupdated to match the reduced case count in every affected
l0_*_perf_sanity_*.yml, and the fourbuildStageConfigs(...)blocks that referenced the deleted yml files removed entirely.
Every updated stage count
kmatches the new active-test countin the corresponding yml (respecting the "1 case per stage for
disagg gen_only / e2e; up to 6 cases per stage for aggregated
and disagg ctx_only" rule in
launchTestJobs).jenkins/L0_MergeRequest.groovy(−4) — the four deleted ymlfile paths dropped from
getMultiGpuFileChanged'striggerable-file list.
Scope — not touched
The four
AutoDeploy(*_ad_*) 1k1k perf-sanity cases registered underthe "AutoDeploy Perf Sanity" sections of
l0_b200.ymlandl0_dgx_b200.ymlare out of scope for this PR — those cases areowned by the AutoDeploy team and should be handled separately if they
also want to drop 1k1k coverage. They remain active:
l0_b200.yml:450aggr_upload-super_ad_blackwell-super_ad_ws1_1k1k(gated by nvbug 6153575 inwaives.txt)l0_dgx_b200.yml:387aggr_upload-super_ad_blackwell-super_ad_ws4_1k1kl0_dgx_b200.yml:415aggr_upload-super_mtp_ad_blackwell-super_mtp_ad_ws4_1k1kl0_dgx_b200.yml:439aggr_upload-deepseek_r1_fp8_ad_blackwell-r1_fp8_ad_ws8_1k1kThe
qa/llm_perf_disagg.yml/qa/llm_perf_multinode.{yml,txt}weeklylists are also unchanged; those are not part of the L0 CI perf-sanity
scope this PR addresses.
Caveat — 1k1k cases with open bugs whose waives were dropped
Some 1k1k cases may still have functional issues and were previously
gated in
waives.txtunder an nvbug. Because we are removing the caseitself from CI, the corresponding SKIP line is no longer needed and has
been dropped. If any of these cases are ever re-registered without their
underlying bug being fixed first, the waive will need to be re-added.
The 9 previously-waived 1k1k cases whose waives were removed:
Disagg (7)
disagg_upload-e2e-gb200_deepseek-v32-fp4_1k1k_con2048_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXLdisagg_upload-gen_only-gb200_deepseek-v32-fp4_1k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXLdisagg_upload-gen_only-gb200_deepseek-v32-fp4_1k1k_con2048_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXLdisagg_upload-gen_only-gb200_gpt-oss-120b-fp4_1k1k_con2048_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXLdisagg_upload-gen_only-gb300_glm-5-fp4_1k1k_con1_ctx1_dep2_gen1_tep4_eplb0_mtp3_ccb-NIXLdisagg_upload-gen_only-gb300_kimi-k25-thinking-fp4_1k1k_con2048_ctx1_dep4_gen1_dep32_eplb0_mtp0_ccb-NIXLdisagg_upload-gen_only-gb300_kimi-k25-thinking-fp4_1k1k_con4096_ctx1_dep4_gen1_dep8_eplb0_mtp0_ccb-NIXLAggregated (2)
aggr_upload-deepseek_v32_fp4_grace_blackwell-v32_fp4_dep4_mtp1_1k1kaggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_tep8_1k1kTest plan
pre-commit run --files <changed>clean.l0_*_perf_sanity_*.yml) no longer scheduleany 1k1k case; only the 4 AutoDeploy 1k1k cases in
l0_b200.yml/
l0_dgx_b200.ymlremain (see Scope — not touched).(
ls tests/scripts/perf-sanity/{aggregated,disaggregated}/unchanged).Full CI perf-sanity case listing (per model × GPU)
211 unique CI perf-sanity case IDs across
l0_*_perf_sanity_*.yml+waives.txt. 83 are removed by this PR (markedremoved=yes); 128 remain in CI. The one 1k1k case counted here but not removed isaggr_upload-super_ad_blackwell-super_ad_ws1_1k1k— see Scope — not touched.Each Case value is the identifier tail — the head prefix (
aggr_upload-/disagg_upload-<mode>-<gpu>_<model>_) is factored out into the section heading + Mode column.deepseek-r1-fp4GPU:
b2001k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL1k1k_con2048_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL1k1k_con256_ctx1_dep4_gen1_dep8_eplb0_mtp3_ccb-NIXL8k1k_con1536_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL8k1k_con256_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL1k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL1k1k_con2048_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL1k1k_con256_ctx1_dep4_gen1_dep8_eplb0_mtp3_ccb-NIXL8k1k_con1536_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL8k1k_con256_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL1k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL1k1k_con2048_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL1k1k_con256_ctx1_dep4_gen1_dep8_eplb0_mtp3_ccb-NIXL8k1k_con1536_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL8k1k_con256_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXLGPU:
gb200128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL128k8k_con1_ctx1_pp8_gen1_tep8_eplb0_mtp3_ccb-NIXL128k8k_con64_ctx1_pp8_gen1_dep32_eplb0_mtp3_ccb-NIXL1k1k_con1024_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL1k1k_con1024_ctx1_dep4_gen1_dep8_eplb0_mtp0_ccb-NIXL1k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL1k1k_con3072_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXL8k1k_con1024_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL1k1k_con3072_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL128k8k_con1_ctx1_pp8_gen1_tep8_eplb0_mtp3_ccb-NIXL128k8k_con64_ctx1_pp8_gen1_dep32_eplb0_mtp3_ccb-NIXL1k1k_con1024_ctx1_dep4_gen1_dep32_eplb0_mtp0_ccb-NIXL1k1k_con1024_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL1k1k_con1024_ctx1_dep4_gen1_dep8_eplb0_mtp0_ccb-NIXL1k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL1k1k_con2048_ctx2_dep4_gen1_dep16_eplb0_mtp3_ccb-NIXL1k1k_con2048_ctx2_dep4_gen1_dep16_eplb288_mtp3_ccb-NIXL1k1k_con3072_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXL8k1k_con1024_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXLGPU:
gb300128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL1k1k_con3072_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL1k1k_con3072_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL1k1k_con3072_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXLdeepseek-v32-fp4GPU:
gb2001k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL1k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL1k1k_con2048_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXL32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL32k4k_con256_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL8k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL1k1k_con2048_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXL32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL1k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL1k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL1k1k_con2048_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXL32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL32k4k_con256_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL32k4k_con256_ctx1_dep8_gen1_dep8_eplb0_mtp0_ccb-NIXL8k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXLdeepseek_r1_fp4_v2GPU:
b200r1_fp4_v2_dep8_mtp1_1k1kr1_fp4_v2_tp4_mtp3_1k1kr1_fp4_v2_dep8_mtp1_8k1kr1_fp4_v2_tp4_mtp3_8k1kGPU:
b200 (multi-node)r1_fp4_v2_tep16_mtp3r1_fp4_v2_dep16_mtp1_1k1kr1_fp4_v2_dep16_mtp1_8k1kGPU:
gb200 (multi-node)r1_fp4_v2_tep8_mtp3r1_fp4_v2_dep8_mtp1_1k1kr1_fp4_v2_dep8_mtp1_8k1kGPU:
gb200 or gb300r1_fp4_v2_dep4_mtp1_1k1kr1_fp4_v2_tep4_mtp3_1k1kr1_fp4_v2_tp4_mtp3_1k1kr1_fp4_v2_dep4_mtp1_1k8kr1_fp4_v2_tep4_mtp3_1k8kr1_fp4_v2_tp4_mtp3_1k8kr1_fp4_v2_dep4_mtp1_8k1kr1_fp4_v2_tep4_mtp3_8k1kr1_fp4_v2_tp4_mtp3_8k1kdeepseek_r1_fp8GPU:
b200r1_fp8_dep8_mtp1_1k1kr1_fp8_tp8_mtp3_1k1kr1_fp8_tp8_6k1kr1_fp8_dep8_mtp1_8k1kr1_fp8_tp8_mtp3_8k1kdeepseek_v32_fp4GPU:
b200v32_fp4_dep8_mtp1_8k1kv32_fp4_tep8_mtp3_8k1kGPU:
gb200 or gb300v32_fp4_dep4_mtp1_1k1kv32_fp4_tep4_mtp3_1k1kv32_fp4_dep4_mtp1_8k1kv32_fp4_tep4_mtp3_8k1kdynamo_deepseek_v32_fp4GPU:
gb200 (multi-node)dsv32_fp4_dep8_trtllm_lpc_mnnvl_8k1kdynamo_gpt_oss_120b_fp4GPU:
b200gpt_oss_fp4_tep4_adp_cutlass_1k1kgpt_oss_fp4_tep4_adp_cutlass_8k1kdynamo_k25_thinking_fp4GPU:
b200k25_thinking_fp4_tep8_adp_2k1kdynamo_qwen3_235b_a22b_fp8GPU:
h200qwen3_235b_a22b_fp8_tp4_ep4_cutlass_8k1kdynamo_qwen3_32b_fp8GPU:
h200qwen3_32b_fp8_tp2_6k1kglm-5-fp4GPU:
gb3001k1k_con1_ctx1_dep2_gen1_tep4_eplb0_mtp3_ccb-NIXL1k1k_con4096_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL1k1k_con512_ctx1_dep2_gen1_dep32_eplb0_mtp3_ccb-NIXL8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL8k1k_con1_ctx1_dep2_gen1_tep8_eplb0_mtp3_ccb-NIXL8k1k_con512_ctx1_dep2_gen1_dep32_eplb0_mtp3_ccb-NIXL1k1k_con4096_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL1k1k_con1_ctx1_dep2_gen1_tep4_eplb0_mtp3_ccb-NIXL1k1k_con4096_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL1k1k_con512_ctx1_dep2_gen1_dep32_eplb0_mtp3_ccb-NIXL8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL8k1k_con1_ctx1_dep2_gen1_tep8_eplb0_mtp3_ccb-NIXL8k1k_con512_ctx1_dep2_gen1_dep32_eplb0_mtp3_ccb-NIXLglm5_fp4GPU:
b200glm5_fp4_dep8_mtp1_8k1kglm5_fp4_tep8_mtp3_8k1kGPU:
gb200 (multi-node)glm5_fp4_dep8_mtp1_8k1kglm5_fp4_tep8_mtp3_8k1kGPU:
gb200 or gb300glm5_fp4_dep4_mtp1_1k1kglm5_fp4_tep4_mtp3_1k1kgpt-oss-120b-fp4GPU:
gb2001k1k_con2048_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL1k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL1k1k_con64_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL8k1k_con128_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL8k1k_con4_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL8k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL1k1k_con2048_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL8k1k_con1024_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL1k1k_con2048_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL1k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL1k1k_con64_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL8k1k_con1024_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL8k1k_con128_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL8k1k_con4_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL8k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXLgpt_oss_120b_fp4GPU:
b200gpt_oss_fp4_tp1_mtp0_1k1kgpt_oss_fp4_tp2_mtp0_1k1kgpt_oss_fp4_tp1_mtp0_8k1kGPU:
gb200 or gb300gpt_oss_fp4_dep2_1k1kgpt_oss_fp4_tp1_mtp0_1k1kgpt_oss_fp4_tp2_mtp0_1k1kgpt_oss_fp4_tp4_eagle3_1k1kgpt_oss_fp4_dep4_1k8kgpt_oss_fp4_tep2_1k8kgpt_oss_fp4_tp2_1k8kgpt_oss_fp4_tp1_mtp0_8k1kk25_thinking_fp4GPU:
b200k25_thinking_fp4_dep8_32k8kk25_thinking_fp4_tep8_32k8kk25_thinking_fp4_dep8_8k1kk25_thinking_fp4_tep8_8k1kGPU:
gb200 (multi-node)k25_thinking_fp4_dep8_32k8kk25_thinking_fp4_tep8_32k8kGPU:
gb200 or gb300k25_thinking_fp4_dep4_8k1kk25_thinking_fp4_tep4_8k1kkimi-k25-thinking-fp4GPU:
gb2001k1k_con2048_ctx1_dep4_gen1_dep32_eplb0_mtp0_ccb-NIXL1k1k_con4096_ctx1_dep4_gen1_dep8_eplb0_mtp0_ccb-NIXL1k1k_con4_ctx1_dep4_gen1_tep4_eplb0_mtp0_ccb-NIXL8k1k_con1024_ctx1_dep4_gen1_dep32_eplb416_mtp3_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL1k1k_con4096_ctx1_dep4_gen1_dep8_eplb0_mtp0_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL1k1k_con2048_ctx1_dep4_gen1_dep32_eplb0_mtp0_ccb-NIXL1k1k_con4096_ctx1_dep4_gen1_dep8_eplb0_mtp0_ccb-NIXL1k1k_con4_ctx1_dep4_gen1_tep4_eplb0_mtp0_ccb-NIXL8k1k_con1024_ctx1_dep4_gen1_dep32_eplb416_mtp3_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXLGPU:
gb3001k1k_con2048_ctx1_dep4_gen1_dep32_eplb0_mtp0_ccb-NIXL1k1k_con4096_ctx1_dep4_gen1_dep8_eplb0_mtp0_ccb-NIXL1k1k_con4_ctx1_dep4_gen1_tep4_eplb0_mtp0_ccb-NIXL8k1k_con1024_ctx1_dep4_gen1_dep32_eplb416_mtp3_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL1k1k_con4096_ctx1_dep4_gen1_dep8_eplb0_mtp0_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL1k1k_con2048_ctx1_dep4_gen1_dep32_eplb0_mtp0_ccb-NIXL1k1k_con4096_ctx1_dep4_gen1_dep8_eplb0_mtp0_ccb-NIXL1k1k_con4_ctx1_dep4_gen1_tep4_eplb0_mtp0_ccb-NIXL8k1k_con1024_ctx1_dep4_gen1_dep32_eplb416_mtp3_ccb-NIXL8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXLllama_v3_3_70b_instruct_fp4GPU:
b200llama70b_fp4_tp4_1000_1000llama70b_fp4_tp4_512_32qwen3-235b-fp4GPU:
gb2008k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL8k1k_con1_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL8k1k_con64_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL8k1k_con1_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL8k1k_con64_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXLqwen3_5_397b_fp4GPU:
b200qwen3_5_397b_fp4_dep4_1k1kqwen3_5_397b_fp4_dep8_mtp3_1k1kqwen3_5_397b_fp4_tep8_1k1kqwen3_5_397b_fp4_tp2_mtp3_1k1kqwen3_5_397b_fp4_dep8_8k1kqwen3_5_397b_fp4_dep8_mtp3_8k1kqwen3_5_397b_fp4_tep4_mtp3_8k1kqwen3_5_397b_fp4_tp4_8k1ksuper_adGPU:
b200super_ad_ws1_1k1k