Skip to content

Commit 879bb37

Browse files
authored
[VideoGenX] sweep Wan2.2 B200 on 1, 2, 4 and 8 GPUs / [VideoGenX] 在 B200 上以 1、2、4、8 卡扫描 Wan2.2 (#5084)
Sweep Wan2.2 vLLM-Omni on B200 with 1, 2, 4 and 8 GPUs (reuses run 43669876970).
1 parent 3370376 commit 879bb37

5 files changed

Lines changed: 204 additions & 0 deletions

File tree

Lines changed: 61 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,61 @@
1+
schema: 2
2+
name: wan22-vllm-omni-serving-client-g1
3+
4+
resources:
5+
gpu_type: b200
6+
7+
services:
8+
- name: video-server
9+
type: generic
10+
nodes: 1
11+
critical: true
12+
start: before_workers
13+
command: [python3, -m, infx.bench, video, serve]
14+
args:
15+
- --model-view
16+
- snapshot
17+
- --
18+
- vllm
19+
- serve
20+
- "{model_path}"
21+
- --served-model-name
22+
- "{model_id}"
23+
- --omni
24+
- --host
25+
- 0.0.0.0
26+
- --port
27+
- "30283"
28+
- --num-gpus
29+
- "{gpus_per_node}"
30+
# Deployment only: one GPU runs both CFG branches with no sequence parallelism.
31+
- --tensor-parallel-size
32+
- "1"
33+
- --cfg-parallel-size
34+
- "1"
35+
- --usp
36+
- "1"
37+
- --ring
38+
- "1"
39+
- --vae-use-tiling
40+
- --init-timeout
41+
- "3600"
42+
- --stage-init-timeout
43+
- "3600"
44+
env:
45+
PYTHONPATH: /video-package:/infmax-workspace
46+
PYTHONUNBUFFERED: "1"
47+
VIDEO_SERVER_PORT: "30283"
48+
VIDEO_RUNTIME_REVISION: 5917730a4b02aedb44d374173cdc1b10484ccd52
49+
HF_HUB_OFFLINE: "1"
50+
TRANSFORMERS_OFFLINE: "1"
51+
VLLM_WORKER_MULTIPROC_METHOD: spawn
52+
# The release set reuses each prompt across five seeds; cached text embeddings would skip work.
53+
OMNI_DIFFUSION_PROMPT_EMBED_CACHE: "0"
54+
VLLM_OMNI_VIDEO_SYNC_TIMEOUT: "14400"
55+
VLLM_OMNI_SERVER_STORAGE__PATH: /logs/video/server/storage
56+
readiness:
57+
http:
58+
port: 30283
59+
path: /health
60+
timeout_seconds: 3600
61+
interval_seconds: 5
Lines changed: 63 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,63 @@
1+
schema: 2
2+
name: wan22-vllm-omni-serving-client-g2
3+
4+
resources:
5+
gpu_type: b200
6+
7+
services:
8+
- name: video-server
9+
type: generic
10+
nodes: 1
11+
critical: true
12+
start: before_workers
13+
command: [python3, -m, infx.bench, video, serve]
14+
args:
15+
- --model-view
16+
- snapshot
17+
- --
18+
- vllm
19+
- serve
20+
- "{model_path}"
21+
- --served-model-name
22+
- "{model_id}"
23+
- --omni
24+
- --host
25+
- 0.0.0.0
26+
- --port
27+
- "30283"
28+
- --num-gpus
29+
- "{gpus_per_node}"
30+
# Deployment only: CFG branches on two GPUs, no sequence parallelism.
31+
- --tensor-parallel-size
32+
- "1"
33+
- --cfg-parallel-size
34+
- "2"
35+
- --usp
36+
- "1"
37+
- --ring
38+
- "1"
39+
- --vae-patch-parallel-size
40+
- "{gpus_per_node}"
41+
- --vae-use-tiling
42+
- --init-timeout
43+
- "3600"
44+
- --stage-init-timeout
45+
- "3600"
46+
env:
47+
PYTHONPATH: /video-package:/infmax-workspace
48+
PYTHONUNBUFFERED: "1"
49+
VIDEO_SERVER_PORT: "30283"
50+
VIDEO_RUNTIME_REVISION: 5917730a4b02aedb44d374173cdc1b10484ccd52
51+
HF_HUB_OFFLINE: "1"
52+
TRANSFORMERS_OFFLINE: "1"
53+
VLLM_WORKER_MULTIPROC_METHOD: spawn
54+
# The release set reuses each prompt across five seeds; cached text embeddings would skip work.
55+
OMNI_DIFFUSION_PROMPT_EMBED_CACHE: "0"
56+
VLLM_OMNI_VIDEO_SYNC_TIMEOUT: "14400"
57+
VLLM_OMNI_SERVER_STORAGE__PATH: /logs/video/server/storage
58+
readiness:
59+
http:
60+
port: 30283
61+
path: /health
62+
timeout_seconds: 3600
63+
interval_seconds: 5
Lines changed: 63 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,63 @@
1+
schema: 2
2+
name: wan22-vllm-omni-serving-client-g4
3+
4+
resources:
5+
gpu_type: b200
6+
7+
services:
8+
- name: video-server
9+
type: generic
10+
nodes: 1
11+
critical: true
12+
start: before_workers
13+
command: [python3, -m, infx.bench, video, serve]
14+
args:
15+
- --model-view
16+
- snapshot
17+
- --
18+
- vllm
19+
- serve
20+
- "{model_path}"
21+
- --served-model-name
22+
- "{model_id}"
23+
- --omni
24+
- --host
25+
- 0.0.0.0
26+
- --port
27+
- "30283"
28+
- --num-gpus
29+
- "{gpus_per_node}"
30+
# Deployment only: CFG branches on two GPU groups, Ulysses sequence parallelism within each.
31+
- --tensor-parallel-size
32+
- "1"
33+
- --cfg-parallel-size
34+
- "2"
35+
- --usp
36+
- "2"
37+
- --ring
38+
- "1"
39+
- --vae-patch-parallel-size
40+
- "{gpus_per_node}"
41+
- --vae-use-tiling
42+
- --init-timeout
43+
- "3600"
44+
- --stage-init-timeout
45+
- "3600"
46+
env:
47+
PYTHONPATH: /video-package:/infmax-workspace
48+
PYTHONUNBUFFERED: "1"
49+
VIDEO_SERVER_PORT: "30283"
50+
VIDEO_RUNTIME_REVISION: 5917730a4b02aedb44d374173cdc1b10484ccd52
51+
HF_HUB_OFFLINE: "1"
52+
TRANSFORMERS_OFFLINE: "1"
53+
VLLM_WORKER_MULTIPROC_METHOD: spawn
54+
# The release set reuses each prompt across five seeds; cached text embeddings would skip work.
55+
OMNI_DIFFUSION_PROMPT_EMBED_CACHE: "0"
56+
VLLM_OMNI_VIDEO_SYNC_TIMEOUT: "14400"
57+
VLLM_OMNI_SERVER_STORAGE__PATH: /logs/video/server/storage
58+
readiness:
59+
http:
60+
port: 30283
61+
path: /health
62+
timeout_seconds: 3600
63+
interval_seconds: 5

‎inferencex-e2e/configs/nvidia-master.yaml‎

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7812,6 +7812,15 @@ wan22-bf16-b200-vllm-omni-video:
78127812
- spec: wan22.yaml
78137813
set: release-v1
78147814
search-space:
7815+
- gpus: 1
7816+
srt-recipe: serving-client-g1.yaml
7817+
conc-list: [1]
7818+
- gpus: 2
7819+
srt-recipe: serving-client-g2.yaml
7820+
conc-list: [1]
7821+
- gpus: 4
7822+
srt-recipe: serving-client-g4.yaml
7823+
conc-list: [1]
78157824
- gpus: 8
78167825
srt-recipe: serving-client.yaml
78177826
conc-list: [1]

‎inferencex-e2e/perf-changelog.yaml‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -12890,3 +12890,11 @@
1289012890
- "Add a B200 Wan2.2-T2V-A14B baseline on vLLM-Omni v0.32.0, pinned by its linux/amd64 digest: wan22.yaml release-v1 (four prompts by five seeds, one warmup; 832x480, 81 frames at 16 fps, 40 UniPC steps, guidance 4.0 high-noise / 3.0 low-noise, boundary 0.875, flow_shift 12.0, empty negative prompt, BF16), C1 on 8 GPUs."
1289112891
- "Deployment: CFG parallel 2 x Ulysses 4, tiled VAE patch parallelism 8, prompt-embedding cache disabled."
1289212892
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/5081
12893+
12894+
- config-keys:
12895+
- wan22-bf16-b200-vllm-omni-video
12896+
scenario-type: [video-generation]
12897+
description:
12898+
- "Sweep the B200 Wan2.2-T2V-A14B vLLM-Omni baseline on 1, 2, 4 and 8 GPUs at C1 on the same wan22.yaml release-v1 workload, so the dashboard shows a GPU-count curve."
12899+
- "Deployment per count: 1 GPU runs both CFG branches with Ulysses 1 and no VAE patch parallelism; 2 GPUs CFG parallel 2 x Ulysses 1; 4 GPUs CFG parallel 2 x Ulysses 2; 8 GPUs keeps CFG parallel 2 x Ulysses 4. Tiled VAE patch parallelism spans every GPU from 2 GPUs up."
12900+
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/5084

0 commit comments

Comments
 (0)