[ROCm][DI][CI] Enable WideEP Intranode tests (#53195)
Signed-off-by: lcskrishna <lollachaitanya@gmail.com>
Chaitanya Sri Krishna Lolla committed
Sep 9, 2026 at 03:36 UTC
28a2ccee78699e90ad06c7e833dc018dea8fddbf
4 files changed
+144
-55
.buildkite/amd-disagg/models.yaml
+12
-3
@@ -37,14 +37,14 @@ models:
37
VLLM_ROCM_USE_AITER_MOE: "1"
38
VLLM_ROCM_USE_AITER_RMSNORM: "1"
39
VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS: "0"
40
+ VLLM_ROCM_USE_AITER_FP8BMM: "false"
41
base_flags: "--trust-remote-code --kv-cache-dtype fp8"
42
prefill:
43
tp: "--gpu-memory-utilization 0.85"
44
ep: "--gpu-memory-utilization 0.85 --enforce-eager"
45
decode:
46
tp: "--gpu-memory-utilization 0.85"
46
- #ep: '--gpu-memory-utilization 0.75 --enforce-eager'
47
- ep: '--gpu-memory-utilization 0.75 --compilation-config {"cudagraph_mode":"PIECEWISE","custom_ops":["+quant_fp8"]}'
47
+ ep: '--gpu-memory-utilization 0.75 --compilation-config {"cudagraph_mode":"FULL_DECODE_ONLY","custom_ops":["+quant_fp8"]}'
48
49
- model: MiniMax-M3-MXFP8
50
env:
@@ -53,11 +53,14 @@ models:
53
VLLM_ROCM_USE_AITER_RMSNORM: "1"
54
VLLM_USE_BREAKABLE_CUDAGRAPH: "0"
55
VLLM_ROCM_QUICK_REDUCE_QUANTIZATION: "INT6"
56
+ VLLM_KV_CACHE_LAYOUT: "HND"
57
base_flags: "--trust-remote-code --attention-backend TRITON_ATTN --block-size 128 --language-model-only --kv-cache-dtype fp8"
58
prefill:
59
tp: "--gpu-memory-utilization 0.85 --enforce-eager"
60
+ ep: "--gpu-memory-utilization 0.85 --enforce-eager"
61
decode:
62
tp: "--gpu-memory-utilization 0.85"
63
+ ep: '--gpu-memory-utilization 0.75 --compilation-config {"cudagraph_mode":"FULL_DECODE_ONLY","custom_ops":["+quant_fp8"]}'
64
65
- model: DeepSeek-R1-MXFP4
66
env:
@@ -65,11 +68,14 @@ models:
68
VLLM_ROCM_USE_AITER_MLA: "1"
69
VLLM_ROCM_USE_AITER_MOE: "1"
70
VLLM_ROCM_USE_AITER_RMSNORM: "1"
71
+ VLLM_ROCM_USE_AITER_FP8BMM: "false"
72
base_flags: "--trust-remote-code --kv-cache-dtype fp8"
73
prefill:
74
tp: "--gpu-memory-utilization 0.85"
75
+ ep: "--gpu-memory-utilization 0.85 --enforce-eager"
76
decode:
77
tp: "--gpu-memory-utilization 0.85"
78
+ ep: '--gpu-memory-utilization 0.75 --compilation-config {"cudagraph_mode":"FULL_DECODE_ONLY","custom_ops":["+quant_fp8"]}'
79
80
- model: Kimi-K2.5-MXFP4
81
env:
@@ -79,8 +85,10 @@ models:
85
base_flags: "--trust-remote-code"
86
prefill:
87
tp: "--gpu-memory-utilization 0.85"
88
+ ep: "--gpu-memory-utilization 0.85 --enforce-eager"
89
decode:
90
tp: "--gpu-memory-utilization 0.85"
91
+ ep: '--gpu-memory-utilization 0.75 --compilation-config {"cudagraph_mode":"FULL_DECODE_ONLY","custom_ops":["+quant_fp8"]}'
92
93
- model: Kimi-K2.6-MXFP4
94
env:
@@ -89,9 +97,10 @@ models:
97
VLLM_ROCM_USE_AITER_MLA: "1"
98
VLLM_ROCM_QUICK_REDUCE_QUANTIZATION: "INT4"
99
VLLM_ROCM_USE_SKINNY_GEMM: "0"
92
- VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS: "1"
100
base_flags: "--trust-remote-code --kv-cache-dtype fp8 --mm-encoder-tp-mode data --block-size 1 --attention-backend ROCM_AITER_MLA"
101
prefill:
102
tp: "--gpu-memory-utilization 0.9"
103
+ ep: "--gpu-memory-utilization 0.85 --enforce-eager"
104
decode:
105
tp: "--gpu-memory-utilization 0.9"
106
+ ep: '--gpu-memory-utilization 0.75 --compilation-config {"cudagraph_mode":"FULL_DECODE_ONLY","custom_ops":["+quant_fp8"]}'
.buildkite/amd-disagg/pipeline-disagg.yaml
+130
-49
@@ -11,192 +11,273 @@ steps:
11
- label: "DeepSeek-V3-PD-1P1D-TP8-MoRIIO-proxy"
12
agents:
13
queue: amd_mi350_ainic
14
- timeout_in_minutes: 120
14
+ timeout_in_minutes: 180
15
concurrency: 2
16
concurrency_group: "amd-disagg/mi350/2node"
17
commands:
18
- - MODEL_NAME=DeepSeek-V3 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
18
+ - MODEL_NAME=DeepSeek-V3 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
19
20
- label: "DeepSeek-V3-PD-1P1D-TP8-MoRIIO-vllm-router"
21
agents:
22
queue: amd_mi350_ainic
23
- timeout_in_minutes: 120
23
+ timeout_in_minutes: 180
24
concurrency: 2
25
concurrency_group: "amd-disagg/mi350/2node"
26
commands:
27
- - MODEL_NAME=DeepSeek-V3 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
27
+ - MODEL_NAME=DeepSeek-V3 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
28
29
- label: "DeepSeek-R1-MXFP4-PD-1P1D-TP8-MoRIIO-proxy"
30
agents:
31
queue: amd_mi350_ainic
32
- timeout_in_minutes: 120
32
+ timeout_in_minutes: 180
33
concurrency: 2
34
concurrency_group: "amd-disagg/mi350/2node"
35
commands:
36
- - MODEL_NAME=DeepSeek-R1-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
36
+ - MODEL_NAME=DeepSeek-R1-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
37
38
- label: "DeepSeek-R1-MXFP4-PD-1P1D-TP8-MoRIIO-vllm-router"
39
agents:
40
queue: amd_mi350_ainic
41
- timeout_in_minutes: 120
41
+ timeout_in_minutes: 180
42
concurrency: 2
43
concurrency_group: "amd-disagg/mi350/2node"
44
commands:
45
- - MODEL_NAME=DeepSeek-R1-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
45
+ - MODEL_NAME=DeepSeek-R1-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
46
47
- label: "Kimi-K2.5-MXFP4-PD-1P1D-TP8-MoRIIO-proxy"
48
agents:
49
queue: amd_mi350_ainic
50
- timeout_in_minutes: 120
50
+ timeout_in_minutes: 180
51
concurrency: 2
52
concurrency_group: "amd-disagg/mi350/2node"
53
commands:
54
- - MODEL_NAME=Kimi-K2.5-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
54
+ - MODEL_NAME=Kimi-K2.5-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
55
56
- label: "Kimi-K2.5-MXFP4-PD-1P1D-TP8-MoRIIO-vllm-router"
57
agents:
58
queue: amd_mi350_ainic
59
- timeout_in_minutes: 120
59
+ timeout_in_minutes: 180
60
concurrency: 2
61
concurrency_group: "amd-disagg/mi350/2node"
62
commands:
63
- - MODEL_NAME=Kimi-K2.5-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
63
+ - MODEL_NAME=Kimi-K2.5-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
64
65
- label: "Kimi-K2.6-MXFP4-PD-1P1D-TP8-MoRIIO-proxy"
66
agents:
67
queue: amd_mi350_ainic
68
- timeout_in_minutes: 120
68
+ timeout_in_minutes: 180
69
concurrency: 2
70
concurrency_group: "amd-disagg/mi350/2node"
71
commands:
72
- - MODEL_NAME=Kimi-K2.6-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
72
+ - MODEL_NAME=Kimi-K2.6-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
73
74
- label: "Kimi-K2.6-MXFP4-PD-1P1D-TP8-MoRIIO-vllm-router"
75
agents:
76
queue: amd_mi350_ainic
77
- timeout_in_minutes: 120
77
+ timeout_in_minutes: 180
78
concurrency: 2
79
concurrency_group: "amd-disagg/mi350/2node"
80
commands:
81
- - MODEL_NAME=Kimi-K2.6-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
81
+ - MODEL_NAME=Kimi-K2.6-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
82
83
- label: "MiniMax-M3-MXFP8-PD-1P1D-TP8-MoRIIO-proxy"
84
agents:
85
queue: amd_mi350_ainic
86
- timeout_in_minutes: 120
86
+ timeout_in_minutes: 180
87
concurrency: 2
88
concurrency_group: "amd-disagg/mi350/2node"
89
commands:
90
- - MODEL_NAME=MiniMax-M3-MXFP8 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
90
+ - MODEL_NAME=MiniMax-M3-MXFP8 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
91
92
- label: "MiniMax-M3-MXFP8-PD-1P1D-TP8-MoRIIO-vllm-router"
93
agents:
94
queue: amd_mi350_ainic
95
- timeout_in_minutes: 120
95
+ timeout_in_minutes: 180
96
concurrency: 2
97
concurrency_group: "amd-disagg/mi350/2node"
98
commands:
99
- - MODEL_NAME=MiniMax-M3-MXFP8 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
99
+ - MODEL_NAME=MiniMax-M3-MXFP8 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
100
101
#-------------------------------------- mi350 · disagg 2P2D (TP8, 4 nodes) · SLURM --------------------------------------#
102
103
- label: "DeepSeek-V3-PD-2P2D-TP8-MoRIIO-proxy"
104
agents:
105
queue: amd_mi350_ainic
106
- timeout_in_minutes: 120
106
+ timeout_in_minutes: 180
107
concurrency: 2
108
concurrency_group: "amd-disagg/mi350/4node"
109
commands:
110
- - MODEL_NAME=DeepSeek-V3 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
110
+ - MODEL_NAME=DeepSeek-V3 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
111
112
- label: "DeepSeek-V3-PD-2P2D-TP8-MoRIIO-vllm-router"
113
agents:
114
queue: amd_mi350_ainic
115
- timeout_in_minutes: 120
115
+ timeout_in_minutes: 180
116
concurrency: 2
117
concurrency_group: "amd-disagg/mi350/4node"
118
commands:
119
- - MODEL_NAME=DeepSeek-V3 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
119
+ - MODEL_NAME=DeepSeek-V3 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
120
121
- label: "DeepSeek-R1-MXFP4-PD-2P2D-TP8-MoRIIO-proxy"
122
agents:
123
queue: amd_mi350_ainic
124
- timeout_in_minutes: 120
124
+ timeout_in_minutes: 180
125
concurrency: 2
126
concurrency_group: "amd-disagg/mi350/4node"
127
commands:
128
- - MODEL_NAME=DeepSeek-R1-MXFP4 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
128
+ - MODEL_NAME=DeepSeek-R1-MXFP4 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
129
130
- label: "DeepSeek-R1-MXFP4-PD-2P2D-TP8-MoRIIO-vllm-router"
131
agents:
132
queue: amd_mi350_ainic
133
- timeout_in_minutes: 120
133
+ timeout_in_minutes: 180
134
concurrency: 2
135
concurrency_group: "amd-disagg/mi350/4node"
136
commands:
137
- - MODEL_NAME=DeepSeek-R1-MXFP4 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
137
+ - MODEL_NAME=DeepSeek-R1-MXFP4 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
138
139
- label: "Kimi-K2.5-MXFP4-PD-2P2D-TP8-MoRIIO-proxy"
140
agents:
141
queue: amd_mi350_ainic
142
- timeout_in_minutes: 120
142
+ timeout_in_minutes: 180
143
concurrency: 2
144
concurrency_group: "amd-disagg/mi350/4node"
145
commands:
146
- - MODEL_NAME=Kimi-K2.5-MXFP4 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
146
+ - MODEL_NAME=Kimi-K2.5-MXFP4 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
147
148
- label: "Kimi-K2.5-MXFP4-PD-2P2D-TP8-MoRIIO-vllm-router"
149
agents:
150
queue: amd_mi350_ainic
151
- timeout_in_minutes: 120
151
+ timeout_in_minutes: 180
152
concurrency: 2
153
concurrency_group: "amd-disagg/mi350/4node"
154
commands:
155
- - MODEL_NAME=Kimi-K2.5-MXFP4 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
155
+ - MODEL_NAME=Kimi-K2.5-MXFP4 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
156
157
- label: "Kimi-K2.6-MXFP4-PD-2P2D-TP8-MoRIIO-proxy"
158
agents:
159
queue: amd_mi350_ainic
160
- timeout_in_minutes: 120
160
+ timeout_in_minutes: 180
161
concurrency: 2
162
concurrency_group: "amd-disagg/mi350/4node"
163
commands:
164
- - MODEL_NAME=Kimi-K2.6-MXFP4 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
164
+ - MODEL_NAME=Kimi-K2.6-MXFP4 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
165
166
- label: "Kimi-K2.6-MXFP4-PD-2P2D-TP8-MoRIIO-vllm-router"
167
agents:
168
queue: amd_mi350_ainic
169
- timeout_in_minutes: 120
169
+ timeout_in_minutes: 180
170
concurrency: 2
171
concurrency_group: "amd-disagg/mi350/4node"
172
commands:
173
- - MODEL_NAME=Kimi-K2.6-MXFP4 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
173
+ - MODEL_NAME=Kimi-K2.6-MXFP4 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
174
175
- label: "MiniMax-M3-MXFP8-PD-2P2D-TP8-MoRIIO-proxy"
176
agents:
177
queue: amd_mi350_ainic
178
- timeout_in_minutes: 120
178
+ timeout_in_minutes: 180
179
concurrency: 2
180
concurrency_group: "amd-disagg/mi350/4node"
181
commands:
182
- - MODEL_NAME=MiniMax-M3-MXFP8 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
182
+ - MODEL_NAME=MiniMax-M3-MXFP8 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
183
184
- label: "MiniMax-M3-MXFP8-PD-2P2D-TP8-MoRIIO-vllm-router"
185
agents:
186
queue: amd_mi350_ainic
187
- timeout_in_minutes: 120
187
+ timeout_in_minutes: 180
188
concurrency: 2
189
concurrency_group: "amd-disagg/mi350/4node"
190
commands:
191
- - MODEL_NAME=MiniMax-M3-MXFP8 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
191
+ - MODEL_NAME=MiniMax-M3-MXFP8 xP=2 yD=2 NODES=4 GPUS_PER_NODE=8 WIDE_EP_MODE=0 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
192
193
-#-------------------------------------- mi350 · disagg wide-EP (disabled) --------------------------------------#
193
+#-------------------------------------- mi350 · disagg wide-EP intranode (EP8/DP8, 2 nodes) · SLURM --------------------------------------#
194
195
-# - label: "DeepSeek-V3-PD-1P1D-EP8/DP8-WideEP-MoRIIO-proxy"
196
-# agents:
197
-# queue: amd_mi350_ainic
198
-# timeout_in_minutes: 120
199
-# concurrency: 2
200
-# concurrency_group: "amd-disagg/mi350/2node"
201
-# commands:
202
-# - IMAGE=vllm/vllm-openai-rocm:nightly MODEL_NAME=DeepSeek-V3 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 SLURM_TIME_LIMIT=02:00:00 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
195
+- label: "DeepSeek-V3-PD-1P1D-EP8/DP8-WideEP-MoRIIO-proxy"
196
+ agents:
197
+ queue: amd_mi350_ainic
198
+ timeout_in_minutes: 180
199
+ concurrency: 2
200
+ concurrency_group: "amd-disagg/mi350/2node"
201
+ commands:
202
+ - MODEL_NAME=DeepSeek-V3 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
203
+
204
+- label: "DeepSeek-V3-PD-1P1D-EP8/DP8-WideEP-MoRIIO-vllm-router"
205
+ agents:
206
+ queue: amd_mi350_ainic
207
+ timeout_in_minutes: 180
208
+ concurrency: 2
209
+ concurrency_group: "amd-disagg/mi350/2node"
210
+ commands:
211
+ - MODEL_NAME=DeepSeek-V3 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
212
+
213
+- label: "DeepSeek-R1-MXFP4-PD-1P1D-EP8/DP8-WideEP-MoRIIO-proxy"
214
+ agents:
215
+ queue: amd_mi350_ainic
216
+ timeout_in_minutes: 180
217
+ concurrency: 2
218
+ concurrency_group: "amd-disagg/mi350/2node"
219
+ commands:
220
+ - MODEL_NAME=DeepSeek-R1-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
221
+
222
+- label: "DeepSeek-R1-MXFP4-PD-1P1D-EP8/DP8-WideEP-MoRIIO-vllm-router"
223
+ agents:
224
+ queue: amd_mi350_ainic
225
+ timeout_in_minutes: 180
226
+ concurrency: 2
227
+ concurrency_group: "amd-disagg/mi350/2node"
228
+ commands:
229
+ - MODEL_NAME=DeepSeek-R1-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
230
+
231
+- label: "Kimi-K2.5-MXFP4-PD-1P1D-EP8/DP8-WideEP-MoRIIO-proxy"
232
+ agents:
233
+ queue: amd_mi350_ainic
234
+ timeout_in_minutes: 180
235
+ concurrency: 2
236
+ concurrency_group: "amd-disagg/mi350/2node"
237
+ commands:
238
+ - MODEL_NAME=Kimi-K2.5-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
239
+
240
+- label: "Kimi-K2.5-MXFP4-PD-1P1D-EP8/DP8-WideEP-MoRIIO-vllm-router"
241
+ agents:
242
+ queue: amd_mi350_ainic
243
+ timeout_in_minutes: 180
244
+ concurrency: 2
245
+ concurrency_group: "amd-disagg/mi350/2node"
246
+ commands:
247
+ - MODEL_NAME=Kimi-K2.5-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
248
+
249
+- label: "Kimi-K2.6-MXFP4-PD-1P1D-EP8/DP8-WideEP-MoRIIO-proxy"
250
+ agents:
251
+ queue: amd_mi350_ainic
252
+ timeout_in_minutes: 180
253
+ concurrency: 2
254
+ concurrency_group: "amd-disagg/mi350/2node"
255
+ commands:
256
+ - MODEL_NAME=Kimi-K2.6-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
257
+
258
+- label: "Kimi-K2.6-MXFP4-PD-1P1D-EP8/DP8-WideEP-MoRIIO-vllm-router"
259
+ agents:
260
+ queue: amd_mi350_ainic
261
+ timeout_in_minutes: 180
262
+ concurrency: 2
263
+ concurrency_group: "amd-disagg/mi350/2node"
264
+ commands:
265
+ - MODEL_NAME=Kimi-K2.6-MXFP4 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
266
+
267
+- label: "MiniMax-M3-MXFP8-PD-1P1D-EP8/DP8-WideEP-MoRIIO-proxy"
268
+ agents:
269
+ queue: amd_mi350_ainic
270
+ timeout_in_minutes: 180
271
+ concurrency: 2
272
+ concurrency_group: "amd-disagg/mi350/2node"
273
+ commands:
274
+ - MODEL_NAME=MiniMax-M3-MXFP8 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=proxy WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
275
+
276
+- label: "MiniMax-M3-MXFP8-PD-1P1D-EP8/DP8-WideEP-MoRIIO-vllm-router"
277
+ agents:
278
+ queue: amd_mi350_ainic
279
+ timeout_in_minutes: 180
280
+ concurrency: 2
281
+ concurrency_group: "amd-disagg/mi350/2node"
282
+ commands:
283
+ - MODEL_NAME=MiniMax-M3-MXFP8 NODES=2 GPUS_PER_NODE=8 WIDE_EP_MODE=1 MORIIO_READ_MODE=0 RUN_AFTER_HEALTH=accuracy ROUTER_TYPE=vllm-router WAIT=1 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
.buildkite/amd-disagg/run-slurm-disagg-test.sh
+2
-2
@@ -37,12 +37,12 @@ IMAGE="${IMAGE:-vllm/vllm-openai-rocm:nightly}"
37
NODES="${NODES:-2}"
38
GPUS_PER_NODE="${GPUS_PER_NODE:-8}"
39
PARTITION="${SLURM_PARTITION:-}"
40
-TIME_LIMIT="${SLURM_TIME_LIMIT:-02:00:00}"
40
+TIME_LIMIT="${SLURM_TIME_LIMIT:-03:00:00}"
41
WIDE_EP_MODE="${WIDE_EP_MODE:-0}" # 0 -> 1P1D TP8 (default); 1 -> wide-EP
42
xP="${xP:-1}"
43
yD="${yD:-1}"
44
RUN_AFTER_HEALTH="${RUN_AFTER_HEALTH:-accuracy}"
45
-HEALTH_TIMEOUT_S="${HEALTH_TIMEOUT_S:-3600}" # P/D bring-up budget; +900s grace must fit the 2h wall
45
+HEALTH_TIMEOUT_S="${HEALTH_TIMEOUT_S:-3600}" # P/D bring-up budget; +900s grace must fit the CI step timeout
46
SHARED_MOUNT="${SHARED_MOUNT:-/data}"
47
LOG_ROOT="${LOG_ROOT:-${SHARED_MOUNT}/${USER:-$(whoami)}/disagg_logs}"
48
DRY_RUN="${DRY_RUN:-0}"
.buildkite/amd-disagg/run_xPyD_disagg.slurm
-1
@@ -4,7 +4,6 @@
4
#SBATCH --ntasks-per-node=1
5
#SBATCH --gres=gpu:8
6
#SBATCH --exclusive
7
-#SBATCH --time=02:00:00
7
#SBATCH --chdir=/tmp
8
#SBATCH --output=/tmp/spur-%j.out
9
#SBATCH --error=/tmp/spur-%j.err