Skip to content

Commit f316a0e

Browse files
nlevin-uicquil11
nlevin-ui
and
cquil11
authored
Fix nvidia-master.yaml metadata to align with recipe files (#919)
* Fix nvidia-master.yaml metadata to align with recipe files Audit found 34 metadata mismatches across 412 entries in nvidia-master.yaml where declared TP/EP/DP-attn values didn't match the actual recipe files. Fixes applied: - B200 FP8 TRT 8K/1K: 15 prefill_ep (8->1), 1 prefill_dp_attn (true->false) - H200 FP8 TRT: 17 prefill_dp_attn mismatches - 1K/1K: false->true (9 entries) - 8K/1K: true->false (8 entries) - GB200 FP4 TRT 8K/1K: 2 decode_dp_attn (true->false) All changes are metadata-only in nvidia-master.yaml to reflect actual recipe values. No recipe files were modified. Made-with: Cursor * Update perf-changelog.yaml with metadata fix entry Made-with: Cursor --------- Co-authored-by: Cameron Quilici <cjquilici@gmail.com>
1 parent 27c28e5 commit f316a0e

2 files changed

Lines changed: 48 additions & 35 deletions

File tree

‎.github/configs/nvidia-master.yaml‎

Lines changed: 35 additions & 35 deletions
Original file line numberDiff line numberDiff line change
@@ -615,7 +615,7 @@ dsr1-fp8-b200-dynamo-trt:
615615
prefill:
616616
num-worker: 1
617617
tp: 8
618-
ep: 8
618+
ep: 1
619619
dp-attn: true
620620
additional-settings:
621621
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/mtp/ctx1_gen6_tp8_batch8_eplb0_mtp3_8.yaml
@@ -630,7 +630,7 @@ dsr1-fp8-b200-dynamo-trt:
630630
prefill:
631631
num-worker: 1
632632
tp: 8
633-
ep: 8
633+
ep: 1
634634
dp-attn: true
635635
additional-settings:
636636
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/mtp/ctx1_gen2_tp8_batch32_eplb0_mtp3_8.yaml
@@ -645,7 +645,7 @@ dsr1-fp8-b200-dynamo-trt:
645645
prefill:
646646
num-worker: 1
647647
tp: 8
648-
ep: 8
648+
ep: 1
649649
dp-attn: true
650650
additional-settings:
651651
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/mtp/ctx1_gen6_tp8_batch8_eplb0_mtp3_48.yaml
@@ -660,7 +660,7 @@ dsr1-fp8-b200-dynamo-trt:
660660
prefill:
661661
num-worker: 1
662662
tp: 8
663-
ep: 8
663+
ep: 1
664664
dp-attn: true
665665
additional-settings:
666666
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/mtp/ctx1_gen4_tp8_batch16_eplb0_mtp3_64.yaml
@@ -676,7 +676,7 @@ dsr1-fp8-b200-dynamo-trt:
676676
prefill:
677677
num-worker: 2
678678
tp: 8
679-
ep: 8
679+
ep: 1
680680
dp-attn: true
681681
additional-settings:
682682
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/mtp/ctx2_gen3_dep8_batch8_eplb0_mtp3_224.yaml
@@ -691,7 +691,7 @@ dsr1-fp8-b200-dynamo-trt:
691691
prefill:
692692
num-worker: 2
693693
tp: 8
694-
ep: 8
694+
ep: 1
695695
dp-attn: true
696696
additional-settings:
697697
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/mtp/ctx2_gen1_dep8_batch32_eplb0_mtp3_288.yaml
@@ -706,7 +706,7 @@ dsr1-fp8-b200-dynamo-trt:
706706
prefill:
707707
num-worker: 4
708708
tp: 8
709-
ep: 8
709+
ep: 1
710710
dp-attn: true
711711
additional-settings:
712712
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/mtp/ctx4_gen1_dep8_batch128_eplb0_mtp2_1088.yaml
@@ -722,8 +722,8 @@ dsr1-fp8-b200-dynamo-trt:
722722
prefill:
723723
num-worker: 1
724724
tp: 8
725-
ep: 8
726-
dp-attn: true
725+
ep: 1
726+
dp-attn: false
727727
additional-settings:
728728
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/stp/ctx1_gen1_tp8_batch1_eplb0_mtp0_1.yaml
729729
- "CONFIG_FILE=recipes/trtllm/b200-fp8/8k1k/stp/ctx1_gen1_tp8_batch1_eplb0_mtp0_1.yaml"
@@ -736,7 +736,7 @@ dsr1-fp8-b200-dynamo-trt:
736736
prefill:
737737
num-worker: 1
738738
tp: 8
739-
ep: 8
739+
ep: 1
740740
dp-attn: true
741741
additional-settings:
742742
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/stp/ctx1_gen4_tp8_batch32_eplb0_mtp0_32.yaml
@@ -750,7 +750,7 @@ dsr1-fp8-b200-dynamo-trt:
750750
prefill:
751751
num-worker: 1
752752
tp: 8
753-
ep: 8
753+
ep: 1
754754
dp-attn: true
755755
additional-settings:
756756
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/stp/ctx1_gen4_tp8_batch32_eplb0_mtp0_128.yaml
@@ -764,7 +764,7 @@ dsr1-fp8-b200-dynamo-trt:
764764
prefill:
765765
num-worker: 1
766766
tp: 8
767-
ep: 8
767+
ep: 1
768768
dp-attn: true
769769
additional-settings:
770770
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/stp/ctx1_gen6_tp8_batch16_eplb0_mtp0_96.yaml
@@ -779,7 +779,7 @@ dsr1-fp8-b200-dynamo-trt:
779779
prefill:
780780
num-worker: 1
781781
tp: 8
782-
ep: 8
782+
ep: 1
783783
dp-attn: true
784784
additional-settings:
785785
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/stp/ctx1_gen1_dep8_batch128_eplb0_mtp0_128.yaml
@@ -793,7 +793,7 @@ dsr1-fp8-b200-dynamo-trt:
793793
prefill:
794794
num-worker: 1
795795
tp: 8
796-
ep: 8
796+
ep: 1
797797
dp-attn: true
798798
additional-settings:
799799
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/stp/ctx1_gen2_dep8_batch64_eplb0_mtp0_128.yaml
@@ -807,7 +807,7 @@ dsr1-fp8-b200-dynamo-trt:
807807
prefill:
808808
num-worker: 1
809809
tp: 8
810-
ep: 8
810+
ep: 1
811811
dp-attn: true
812812
additional-settings:
813813
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/stp/ctx1_gen1_dep8_batch256_eplb0_mtp0_256.yaml
@@ -821,7 +821,7 @@ dsr1-fp8-b200-dynamo-trt:
821821
prefill:
822822
num-worker: 2
823823
tp: 8
824-
ep: 8
824+
ep: 1
825825
dp-attn: true
826826
additional-settings:
827827
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/b200-fp8/8k1k/stp/ctx2_gen1_dep8_batch640_eplb0_mtp0_640.yaml
@@ -2177,7 +2177,7 @@ dsr1-fp8-h200-dynamo-trt:
21772177
num-worker: 1
21782178
tp: 8
21792179
ep: 8
2180-
dp-attn: false
2180+
dp-attn: true
21812181
additional-settings:
21822182
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/1k1k/mtp/c4_ctx1_gen11_tep8_batch128_eplb0_mtp3.yaml
21832183
- "CONFIG_FILE=recipes/trtllm/h200/1k1k/mtp/c4_ctx1_gen11_tep8_batch128_eplb0_mtp3.yaml"
@@ -2192,7 +2192,7 @@ dsr1-fp8-h200-dynamo-trt:
21922192
num-worker: 1
21932193
tp: 8
21942194
ep: 8
2195-
dp-attn: false
2195+
dp-attn: true
21962196
additional-settings:
21972197
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/1k1k/mtp/c8_ctx1_gen11_tep8_batch128_eplb0_mtp3.yaml
21982198
- "CONFIG_FILE=recipes/trtllm/h200/1k1k/mtp/c8_ctx1_gen11_tep8_batch128_eplb0_mtp3.yaml"
@@ -2207,7 +2207,7 @@ dsr1-fp8-h200-dynamo-trt:
22072207
num-worker: 1
22082208
tp: 8
22092209
ep: 8
2210-
dp-attn: false
2210+
dp-attn: true
22112211
additional-settings:
22122212
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/1k1k/mtp/c16_ctx1_gen9_tep8_batch128_eplb0_mtp3.yaml
22132213
- "CONFIG_FILE=recipes/trtllm/h200/1k1k/mtp/c16_ctx1_gen9_tep8_batch128_eplb0_mtp3.yaml"
@@ -2222,7 +2222,7 @@ dsr1-fp8-h200-dynamo-trt:
22222222
num-worker: 1
22232223
tp: 8
22242224
ep: 8
2225-
dp-attn: false
2225+
dp-attn: true
22262226
additional-settings:
22272227
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/1k1k/mtp/c32_ctx1_gen11_tep8_batch128_eplb0_mtp3.yaml
22282228
- "CONFIG_FILE=recipes/trtllm/h200/1k1k/mtp/c32_ctx1_gen11_tep8_batch128_eplb0_mtp3.yaml"
@@ -2311,7 +2311,7 @@ dsr1-fp8-h200-dynamo-trt:
23112311
num-worker: 1
23122312
tp: 8
23132313
ep: 8
2314-
dp-attn: false
2314+
dp-attn: true
23152315
additional-settings:
23162316
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/1k1k/stp/c4_ctx1_gen9_tep8_batch256_eplb0_mtp0.yaml
23172317
- "CONFIG_FILE=recipes/trtllm/h200/1k1k/stp/c4_ctx1_gen9_tep8_batch256_eplb0_mtp0.yaml"
@@ -2325,7 +2325,7 @@ dsr1-fp8-h200-dynamo-trt:
23252325
num-worker: 1
23262326
tp: 8
23272327
ep: 8
2328-
dp-attn: false
2328+
dp-attn: true
23292329
additional-settings:
23302330
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/1k1k/stp/c8_ctx1_gen9_tep8_batch256_eplb0_mtp0.yaml
23312331
- "CONFIG_FILE=recipes/trtllm/h200/1k1k/stp/c8_ctx1_gen9_tep8_batch256_eplb0_mtp0.yaml"
@@ -2339,7 +2339,7 @@ dsr1-fp8-h200-dynamo-trt:
23392339
num-worker: 1
23402340
tp: 8
23412341
ep: 8
2342-
dp-attn: false
2342+
dp-attn: true
23432343
additional-settings:
23442344
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/1k1k/stp/c16_ctx1_gen9_tep8_batch256_eplb0_mtp0.yaml
23452345
- "CONFIG_FILE=recipes/trtllm/h200/1k1k/stp/c16_ctx1_gen9_tep8_batch256_eplb0_mtp0.yaml"
@@ -2353,7 +2353,7 @@ dsr1-fp8-h200-dynamo-trt:
23532353
num-worker: 1
23542354
tp: 8
23552355
ep: 8
2356-
dp-attn: false
2356+
dp-attn: true
23572357
additional-settings:
23582358
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/1k1k/stp/c32_ctx1_gen9_tep8_batch256_eplb0_mtp0.yaml
23592359
- "CONFIG_FILE=recipes/trtllm/h200/1k1k/stp/c32_ctx1_gen9_tep8_batch256_eplb0_mtp0.yaml"
@@ -2367,7 +2367,7 @@ dsr1-fp8-h200-dynamo-trt:
23672367
num-worker: 1
23682368
tp: 8
23692369
ep: 8
2370-
dp-attn: false
2370+
dp-attn: true
23712371
additional-settings:
23722372
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/1k1k/stp/c64_ctx1_gen9_tep8_batch256_eplb0_mtp0.yaml
23732373
- "CONFIG_FILE=recipes/trtllm/h200/1k1k/stp/c64_ctx1_gen9_tep8_batch256_eplb0_mtp0.yaml"
@@ -2503,7 +2503,7 @@ dsr1-fp8-h200-dynamo-trt:
25032503
num-worker: 1
25042504
tp: 8
25052505
ep: 8
2506-
dp-attn: true
2506+
dp-attn: false
25072507
additional-settings:
25082508
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/8k1k/mtp/c64_ctx1_gen1_dep8_batch32_eplb0_mtp2.yaml
25092509
- "CONFIG_FILE=recipes/trtllm/h200/8k1k/mtp/c64_ctx1_gen1_dep8_batch32_eplb0_mtp2.yaml"
@@ -2518,7 +2518,7 @@ dsr1-fp8-h200-dynamo-trt:
25182518
num-worker: 2
25192519
tp: 8
25202520
ep: 8
2521-
dp-attn: true
2521+
dp-attn: false
25222522
additional-settings:
25232523
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/8k1k/mtp/c128_ctx2_gen1_dep8_batch32_eplb0_mtp2.yaml
25242524
- "CONFIG_FILE=recipes/trtllm/h200/8k1k/mtp/c128_ctx2_gen1_dep8_batch32_eplb0_mtp2.yaml"
@@ -2533,7 +2533,7 @@ dsr1-fp8-h200-dynamo-trt:
25332533
num-worker: 3
25342534
tp: 8
25352535
ep: 8
2536-
dp-attn: true
2536+
dp-attn: false
25372537
additional-settings:
25382538
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/8k1k/mtp/c256_ctx3_gen1_dep8_batch32_eplb0_mtp2.yaml
25392539
- "CONFIG_FILE=recipes/trtllm/h200/8k1k/mtp/c256_ctx3_gen1_dep8_batch32_eplb0_mtp2.yaml"
@@ -2548,7 +2548,7 @@ dsr1-fp8-h200-dynamo-trt:
25482548
num-worker: 3
25492549
tp: 8
25502550
ep: 8
2551-
dp-attn: true
2551+
dp-attn: false
25522552
additional-settings:
25532553
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/8k1k/mtp/c512_ctx3_gen1_dep8_batch64_eplb0_mtp1.yaml
25542554
- "CONFIG_FILE=recipes/trtllm/h200/8k1k/mtp/c512_ctx3_gen1_dep8_batch64_eplb0_mtp1.yaml"
@@ -2633,7 +2633,7 @@ dsr1-fp8-h200-dynamo-trt:
26332633
num-worker: 2
26342634
tp: 8
26352635
ep: 8
2636-
dp-attn: true
2636+
dp-attn: false
26372637
additional-settings:
26382638
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/8k1k/stp/c64_ctx2_gen3_dep8_batch128_eplb0_mtp0.yaml
26392639
- "CONFIG_FILE=recipes/trtllm/h200/8k1k/stp/c64_ctx2_gen3_dep8_batch128_eplb0_mtp0.yaml"
@@ -2647,7 +2647,7 @@ dsr1-fp8-h200-dynamo-trt:
26472647
num-worker: 1
26482648
tp: 8
26492649
ep: 8
2650-
dp-attn: true
2650+
dp-attn: false
26512651
additional-settings:
26522652
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/8k1k/stp/c128_ctx1_gen1_dep8_batch256_eplb0_mtp0.yaml
26532653
- "CONFIG_FILE=recipes/trtllm/h200/8k1k/stp/c128_ctx1_gen1_dep8_batch256_eplb0_mtp0.yaml"
@@ -2661,7 +2661,7 @@ dsr1-fp8-h200-dynamo-trt:
26612661
num-worker: 5
26622662
tp: 8
26632663
ep: 8
2664-
dp-attn: true
2664+
dp-attn: false
26652665
additional-settings:
26662666
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/8k1k/stp/c256_ctx5_gen3_dep8_batch256_eplb0_mtp0.yaml
26672667
- "CONFIG_FILE=recipes/trtllm/h200/8k1k/stp/c256_ctx5_gen3_dep8_batch256_eplb0_mtp0.yaml"
@@ -2675,7 +2675,7 @@ dsr1-fp8-h200-dynamo-trt:
26752675
num-worker: 3
26762676
tp: 8
26772677
ep: 8
2678-
dp-attn: true
2678+
dp-attn: false
26792679
additional-settings:
26802680
# https://github.com/ishandhanani/srt-slurm/blob/sa-submission-q1-2026/recipes/trtllm/h200/8k1k/stp/c512_ctx3_gen1_dep8_batch512_eplb0_mtp0.yaml
26812681
- "CONFIG_FILE=recipes/trtllm/h200/8k1k/stp/c512_ctx3_gen1_dep8_batch512_eplb0_mtp0.yaml"
@@ -4048,7 +4048,7 @@ dsr1-fp4-gb200-dynamo-trt:
40484048
num-worker: 4
40494049
tp: 8
40504050
ep: 8
4051-
dp-attn: true
4051+
dp-attn: false
40524052
- conc-list: [ 5 ]
40534053
prefill:
40544054
num-worker: 1
@@ -4062,7 +4062,7 @@ dsr1-fp4-gb200-dynamo-trt:
40624062
num-worker: 4
40634063
tp: 8
40644064
ep: 8
4065-
dp-attn: true
4065+
dp-attn: false
40664066
- conc-list: [ 333 ]
40674067
prefill:
40684068
num-worker: 2

‎perf-changelog.yaml‎

Lines changed: 13 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,16 @@
1+
- config-keys:
2+
- dsr1-fp8-b200-dynamo-trt
3+
- dsr1-fp8-h200-dynamo-trt
4+
- dsr1-fp4-gb200-dynamo-trt
5+
description:
6+
- "Fix metadata inconsistencies in nvidia-master.yaml - TP/EP/DP-attn values now match actual recipe files"
7+
- "B200 FP8 TRT 8K/1K: prefill_ep 8→1 (15 entries), prefill_dp_attn true→false (1 entry)"
8+
- "H200 FP8 TRT 1K/1K: prefill_dp_attn false→true (9 entries)"
9+
- "H200 FP8 TRT 8K/1K: prefill_dp_attn true→false (8 entries)"
10+
- "GB200 FP4 TRT 8K/1K: decode_dp_attn true→false (2 entries)"
11+
- "All fixes are metadata-only; no recipe files were modified"
12+
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/919
13+
114
- config-keys:
215
- kimik2.5-int4-mi325x-vllm
316
description:

0 commit comments

Comments
 (0)