[CI][ROCm][Disagg] Add GLM-5.2-FP8 to MoRIIO model catalog (#53885)
Signed-off-by: avininjamay8 <avpaul@amd.com> Co-authored-by: avininjamay8 <avpaul@amd.com> Co-authored-by: lcskrishna <lollachaitanya@gmail.com>
Avinash Paul committed
Sep 10, 2026 at 09:42 UTC
2a02f6efe319c885e3ccbcecde402e0028f9ec1e
2 files changed
+51
.buildkite/amd-disagg/models.yaml
+15
@@ -104,3 +104,18 @@ models:
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"]}'
107
+
108
+ - model: GLM-5.2-FP8
109
+ env:
110
+ VLLM_ROCM_USE_AITER: "1"
111
+ VLLM_ROCM_USE_AITER_MLA: "1"
112
+ VLLM_ROCM_USE_AITER_MOE: "1"
113
+ VLLM_ROCM_USE_AITER_RMSNORM: "1"
114
+ VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS: "0"
115
+ base_flags: "--trust-remote-code --kv-cache-dtype fp8"
116
+ prefill:
117
+ tp: "--gpu-memory-utilization 0.85"
118
+ ep: "--gpu-memory-utilization 0.85 --enforce-eager"
119
+ decode:
120
+ tp: "--gpu-memory-utilization 0.85"
121
+ ep: '--gpu-memory-utilization 0.85 --compilation-config {"cudagraph_mode":"PIECEWISE"}'
.buildkite/amd-disagg/pipeline-disagg.yaml
+36
@@ -98,6 +98,24 @@ steps:
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 bash .buildkite/amd-disagg/run-slurm-disagg-test.sh
100
101
+- label: "GLM-5.2-FP8-PD-1P1D-TP8-MoRIIO-proxy"
102
+ agents:
103
+ queue: amd_mi350_ainic
104
+ timeout_in_minutes: 180
105
+ concurrency: 2
106
+ concurrency_group: "amd-disagg/mi350/2node"
107
+ commands:
108
+ - MODEL_NAME=GLM-5.2-FP8 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
109
+
110
+- label: "GLM-5.2-FP8-PD-1P1D-TP8-MoRIIO-vllm-router"
111
+ agents:
112
+ queue: amd_mi350_ainic
113
+ timeout_in_minutes: 180
114
+ concurrency: 2
115
+ concurrency_group: "amd-disagg/mi350/2node"
116
+ commands:
117
+ - MODEL_NAME=GLM-5.2-FP8 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
118
+
119
#-------------------------------------- mi350 · disagg 2P2D (TP8, 4 nodes) · SLURM --------------------------------------#
120
121
- label: "DeepSeek-V3-PD-2P2D-TP8-MoRIIO-proxy"
@@ -281,3 +299,21 @@ steps:
299
concurrency_group: "amd-disagg/mi350/2node"
300
commands:
301
- 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
302
+
303
+- label: "GLM-5.2-FP8-PD-1P1D-EP8/DP8-WideEP-MoRIIO-proxy"
304
+ agents:
305
+ queue: amd_mi350_ainic
306
+ timeout_in_minutes: 180
307
+ concurrency: 2
308
+ concurrency_group: "amd-disagg/mi350/2node"
309
+ commands:
310
+ - MODEL_NAME=GLM-5.2-FP8 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
311
+
312
+- label: "GLM-5.2-FP8-PD-1P1D-EP8/DP8-WideEP-MoRIIO-vllm-router"
313
+ agents:
314
+ queue: amd_mi350_ainic
315
+ timeout_in_minutes: 180
316
+ concurrency: 2
317
+ concurrency_group: "amd-disagg/mi350/2node"
318
+ commands:
319
+ - MODEL_NAME=GLM-5.2-FP8 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