Skip to content

Commit 574f38e

Browse files
peterjunparkkTorp
peterjunpark
and
kTorp
committed
Update docs for xDiT 26.5 release (ROCm#6375)
* archive previous version * Update xDiT docs yaml * Update xDiT docs rst * Update whats new section * add links to rocm 7.13 --------- Co-authored-by: Kristoffer <kristoffer.torp@amd.com> (cherry picked from commit aad1b8b)
1 parent c56b973 commit 574f38e

5 files changed

Lines changed: 708 additions & 24 deletions

File tree

‎docs/data/how-to/rocm-for-ai/inference/previous-versions/xdit_26.5-inference-models.yaml‎

Lines changed: 363 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,363 @@
1+
docker:
2+
pull_tag: rocm/pytorch-xdit:v26.5
3+
docker_hub_url: https://hub.docker.com/layers/rocm/pytorch-xdit/v26.5/images/sha256-b8ad9fd4b41bc116ac2aff07c1066bf369cf7fc110b1a323f6302191985a51fd
4+
ROCm: 7.13.0
5+
whats_new:
6+
- "Hunyuan Video 1.5 sparse attention (SSTA) support"
7+
- "Support fp8 MLA for MI355"
8+
- "Block wise sparsity support for AMD triton FAv3 Sage attention"
9+
components:
10+
TheRock:
11+
version: cbff3d1
12+
url: https://github.com/ROCm/TheRock
13+
rocm-libraries:
14+
version: a668483b
15+
url: https://github.com/ROCm/rocm-libraries
16+
rocm-systems:
17+
version: c76140fa
18+
url: https://github.com/ROCm/rocm-systems
19+
torch:
20+
version: ff65f5b
21+
url: https://github.com/ROCm/pytorch
22+
torchaudio:
23+
version: e3c6ee2
24+
url: https://github.com/pytorch/audio
25+
torchvision:
26+
version: b919bd0
27+
url: https://github.com/pytorch/vision
28+
triton:
29+
version: a272dfa
30+
url: https://github.com/ROCm/triton
31+
accelerate:
32+
version: 46ba481
33+
url: https://github.com/huggingface/accelerate
34+
aiter:
35+
version: bc5ea32c
36+
url: https://github.com/ROCm/aiter
37+
diffusers:
38+
version: 447e571a
39+
url: https://github.com/huggingface/diffusers
40+
distvae:
41+
version: 5a0fcbb
42+
url: https://github.com/xdit-project/DistVAE
43+
xfuser:
44+
version: 051db68f
45+
url: https://github.com/xdit-project/xDiT
46+
yunchang:
47+
version: 631bdfd
48+
url: https://github.com/feifeibear/long-context-attention
49+
supported_models:
50+
- group: Hunyuan Video
51+
js_tag: hunyuan
52+
models:
53+
- model: Hunyuan Video
54+
model_repo: tencent/HunyuanVideo
55+
revision: refs/pr/18
56+
url: https://huggingface.co/tencent/HunyuanVideo
57+
github: https://github.com/Tencent-Hunyuan/HunyuanVideo
58+
mad_tag: pyt_xdit_hunyuanvideo
59+
js_tag: hunyuan_tag
60+
benchmark_command:
61+
- mkdir results
62+
- 'xdit \'
63+
- '--model {model_repo} \'
64+
- '--prompt "In the large cage, two puppies were wagging their tails at each other." \'
65+
- '--batch_size 1 \'
66+
- '--height 720 --width 1280 \'
67+
- '--seed 1168860793 \'
68+
- '--num_frames 129 \'
69+
- '--num_inference_steps 50 \'
70+
- '--warmup_calls 1 \'
71+
- '--num_iterations 1 \'
72+
- '--ulysses_degree 8 \'
73+
- '--enable_tiling --enable_slicing \'
74+
- '--guidance_scale 6.0 \'
75+
- '--use_torch_compile \'
76+
- '--attention_backend aiter \'
77+
- '--output_directory results'
78+
- model: Hunyuan Video 1.5
79+
model_repo: hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-720p_t2v
80+
url: https://huggingface.co/hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-720p_t2v
81+
github: https://github.com/Tencent-Hunyuan/HunyuanVideo-1.5
82+
mad_tag: pyt_xdit_hunyuanvideo_1_5
83+
js_tag: hunyuan_1_5_tag
84+
benchmark_command:
85+
- mkdir results
86+
- 'xdit \'
87+
- '--model {model_repo} \'
88+
- '--prompt "In the large cage, two puppies were wagging their tails at each other." \'
89+
- '--task t2v \'
90+
- '--height 720 --width 1280 \'
91+
- '--seed 1168860793 \'
92+
- '--num_frames 129 \'
93+
- '--num_inference_steps 50 \'
94+
- '--num_iterations 1 \'
95+
- '--ulysses_degree 8 \'
96+
- '--enable_tiling --enable_slicing \'
97+
- '--use_torch_compile \'
98+
- '--attention_backend aiter \'
99+
- '--output_directory results'
100+
- group: Wan-AI
101+
js_tag: wan
102+
models:
103+
- model: Wan2.1
104+
model_repo: Wan-AI/Wan2.1-I2V-14B-720P-Diffusers
105+
url: https://huggingface.co/Wan-AI/Wan2.1-I2V-14B-720P-Diffusers
106+
github: https://github.com/Wan-Video/Wan2.1
107+
mad_tag: pyt_xdit_wan_2_1
108+
js_tag: wan_21_tag
109+
benchmark_command:
110+
- mkdir results
111+
- 'xdit \'
112+
- '--model {model_repo} \'
113+
- '--prompt "Summer beach vacation style, a white cat wearing sunglasses sits on a surfboard. The fluffy-furred feline gazes directly at the camera with a relaxed expression. Blurred beach scenery forms the background featuring crystal-clear waters, distant green hills, and a blue sky dotted with white clouds. The cat assumes a naturally relaxed posture, as if savoring the sea breeze and warm sunlight. A close-up shot highlights the feline''s intricate details and the refreshing atmosphere of the seaside." \'
114+
- '--height 720 \'
115+
- '--width 1280 \'
116+
- '--input_images /app/data/wan_input.jpg \'
117+
- '--num_frames 81 \'
118+
- '--ulysses_degree 8 \'
119+
- '--use_parallel_vae \'
120+
- '--seed 42 \'
121+
- '--guidance_scale 3.0 \'
122+
- '--num_iterations 1 \'
123+
- '--num_inference_steps 40 \'
124+
- '--use_torch_compile \'
125+
- '--attention_backend aiter \'
126+
- '--output_directory results'
127+
- model: Wan2.2
128+
model_repo: Wan-AI/Wan2.2-I2V-A14B-Diffusers
129+
url: https://huggingface.co/Wan-AI/Wan2.2-I2V-A14B-Diffusers
130+
github: https://github.com/Wan-Video/Wan2.2
131+
mad_tag: pyt_xdit_wan_2_2
132+
js_tag: wan_22_tag
133+
benchmark_command:
134+
- mkdir results
135+
- 'xdit \'
136+
- '--model {model_repo} \'
137+
- '--prompt "Summer beach vacation style, a white cat wearing sunglasses sits on a surfboard. The fluffy-furred feline gazes directly at the camera with a relaxed expression. Blurred beach scenery forms the background featuring crystal-clear waters, distant green hills, and a blue sky dotted with white clouds. The cat assumes a naturally relaxed posture, as if savoring the sea breeze and warm sunlight. A close-up shot highlights the feline''s intricate details and the refreshing atmosphere of the seaside." \'
138+
- '--height 720 \'
139+
- '--width 1280 \'
140+
- '--input_images /app/data/wan_input.jpg \'
141+
- '--num_frames 81 \'
142+
- '--ulysses_degree 8 \'
143+
- '--use_parallel_vae \'
144+
- '--seed 42 \'
145+
- '--num_iterations 1 \'
146+
- '--num_inference_steps 40 \'
147+
- '--use_torch_compile \'
148+
- '--attention_backend aiter \'
149+
- '--output_directory results'
150+
- group: FLUX
151+
js_tag: flux
152+
models:
153+
- model: FLUX.1
154+
model_repo: black-forest-labs/FLUX.1-dev
155+
url: https://huggingface.co/black-forest-labs/FLUX.1-dev
156+
github: https://github.com/black-forest-labs/flux
157+
mad_tag: pyt_xdit_flux
158+
js_tag: flux_1_tag
159+
benchmark_command:
160+
- mkdir results
161+
- 'xdit \'
162+
- '--model {model_repo} \'
163+
- '--seed 42 \'
164+
- '--prompt "A small cat" \'
165+
- '--height 1024 \'
166+
- '--width 1024 \'
167+
- '--num_inference_steps 25 \'
168+
- '--max_sequence_length 256 \'
169+
- '--warmup_calls 5 \'
170+
- '--ulysses_degree 8 \'
171+
- '--use_torch_compile \'
172+
- '--guidance_scale 0.0 \'
173+
- '--num_iterations 50 \'
174+
- '--attention_backend aiter \'
175+
- '--output_directory results'
176+
- model: FLUX.1 Kontext
177+
model_repo: black-forest-labs/FLUX.1-Kontext-dev
178+
url: https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev
179+
github: https://github.com/black-forest-labs/flux
180+
mad_tag: pyt_xdit_flux_kontext
181+
js_tag: flux_1_kontext_tag
182+
benchmark_command:
183+
- mkdir results
184+
- 'xdit \'
185+
- '--model {model_repo} \'
186+
- '--seed 42 \'
187+
- '--prompt "Add a cool hat to the cat" \'
188+
- '--height 1024 \'
189+
- '--width 1024 \'
190+
- '--num_inference_steps 30 \'
191+
- '--max_sequence_length 512 \'
192+
- '--warmup_calls 5 \'
193+
- '--ulysses_degree 8 \'
194+
- '--use_torch_compile \'
195+
- '--input_images /app/data/flux_cat.png \'
196+
- '--guidance_scale 2.5 \'
197+
- '--num_iterations 25 \'
198+
- '--attention_backend aiter \'
199+
- '--output_directory results'
200+
- model: FLUX.2
201+
model_repo: black-forest-labs/FLUX.2-dev
202+
url: https://huggingface.co/black-forest-labs/FLUX.2-dev
203+
github: https://github.com/black-forest-labs/flux2
204+
mad_tag: pyt_xdit_flux_2
205+
js_tag: flux_2_tag
206+
benchmark_command:
207+
- mkdir results
208+
- 'xdit \'
209+
- '--model {model_repo} \'
210+
- '--seed 42 \'
211+
- '--prompt "Add a cool hat to the cat" \'
212+
- '--height 1024 \'
213+
- '--width 1024 \'
214+
- '--num_inference_steps 50 \'
215+
- '--max_sequence_length 512 \'
216+
- '--warmup_calls 5 \'
217+
- '--ulysses_degree 8 \'
218+
- '--use_torch_compile \'
219+
- '--input_images /app/data/flux_cat.png \'
220+
- '--guidance_scale 4.0 \'
221+
- '--num_iterations 25 \'
222+
- '--attention_backend aiter \'
223+
- '--output_directory results'
224+
- model: FLUX.2 Klein
225+
model_repo: black-forest-labs/FLUX.2-klein-9B
226+
url: https://huggingface.co/black-forest-labs/FLUX.2-klein-9B
227+
github: https://github.com/black-forest-labs/flux2
228+
mad_tag: pyt_xdit_flux_2_klein
229+
js_tag: flux_2_klein_tag
230+
benchmark_command:
231+
- mkdir results
232+
- 'xdit \'
233+
- '--model {model_repo} \'
234+
- '--seed 42 \'
235+
- '--prompt "A spectacular sunset over the ocean" \'
236+
- '--height 2048 \'
237+
- '--width 2048 \'
238+
- '--num_inference_steps 4 \'
239+
- '--warmup_calls 5 \'
240+
- '--ulysses_degree 8 \'
241+
- '--use_torch_compile \'
242+
- '--guidance_scale 1.0 \'
243+
- '--num_iterations 25 \'
244+
- '--attention_backend aiter \'
245+
- '--output_directory results'
246+
- group: StableDiffusion
247+
js_tag: stablediffusion
248+
models:
249+
- model: stable-diffusion-3.5-large
250+
model_repo: stabilityai/stable-diffusion-3.5-large
251+
url: https://huggingface.co/stabilityai/stable-diffusion-3.5-large
252+
github: https://github.com/Stability-AI/sd3.5
253+
mad_tag: pyt_xdit_sd_3_5
254+
js_tag: stable_diffusion_3_5_large_tag
255+
benchmark_command:
256+
- mkdir results
257+
- 'xdit \'
258+
- '--model {model_repo} \'
259+
- '--prompt "A capybara holding a sign that reads Hello World" \'
260+
- '--num_iterations 50 \'
261+
- '--num_inference_steps 28 \'
262+
- '--pipefusion_parallel_degree 4 \'
263+
- '--use_cfg_parallel \'
264+
- '--use_torch_compile \'
265+
- '--attention_backend aiter \'
266+
- '--output_directory results'
267+
- group: Z-Image
268+
js_tag: z_image
269+
models:
270+
- model: Z-Image
271+
model_repo: Tongyi-MAI/Z-Image
272+
url: https://huggingface.co/Tongyi-MAI/Z-Image
273+
github: https://github.com/Tongyi-MAI/Z-Image
274+
mad_tag: pyt_xdit_z_image
275+
js_tag: z_image_tag
276+
benchmark_command:
277+
- mkdir results
278+
- 'xdit \'
279+
- '--model {model_repo} \'
280+
- '--seed 42 \'
281+
- '--prompt "A crowded beach" \'
282+
- '--height 1088 \'
283+
- '--width 1920 \'
284+
- '--num_inference_steps 50 \'
285+
- '--ulysses_degree 2 \'
286+
- '--ring_degree 2 \'
287+
- '--use_cfg_parallel \'
288+
- '--use_torch_compile \'
289+
- '--guidance_scale 4.0 \'
290+
- '--num_iterations 25 \'
291+
- '--attention_backend aiter \'
292+
- '--output_directory results'
293+
- group: LTX
294+
js_tag: ltx
295+
models:
296+
- model: LTX-2
297+
model_repo: Lightricks/LTX-2
298+
url: https://huggingface.co/Lightricks/LTX-2
299+
github: https://github.com/Lightricks/LTX-2
300+
mad_tag: pyt_xdit_ltx2
301+
js_tag: ltx2_tag
302+
benchmark_command:
303+
- mkdir results
304+
- 'xdit \'
305+
- '--model {model_repo} \'
306+
- '--seed 42 \'
307+
- '--prompt "Cinematic action packed shot. The man says silently: \"We need to run.\". The camera zooms in on his mouth then immediately screams: \"NOW!\". The camera zooms back out, he turns around and bolts it." \'
308+
- '--height 1088 \'
309+
- '--width 1920 \'
310+
- '--num_inference_steps 40 \'
311+
- '--ulysses_degree 8 \'
312+
- '--use_torch_compile \'
313+
- '--guidance_scale 4.0 \'
314+
- '--num_iterations 1 \'
315+
- '--attention_backend aiter \'
316+
- '--output_directory results'
317+
- group: Qwen-Image
318+
js_tag: qwen_image
319+
models:
320+
- model: Qwen-Image
321+
model_repo: Qwen/Qwen-Image-2512
322+
url: https://huggingface.co/Qwen/Qwen-Image-2512
323+
github: https://github.com/QwenLM/Qwen-Image
324+
mad_tag: pyt_xdit_qwen_image
325+
js_tag: qwen_image_tag
326+
benchmark_command:
327+
- mkdir results
328+
- 'xdit \'
329+
- '--model {model_repo} \'
330+
- '--seed 42 \'
331+
- '--prompt "A cat holding a sign that says hello world" \'
332+
- '--height 2048 \'
333+
- '--width 2048 \'
334+
- '--num_inference_steps 50 \'
335+
- '--ulysses_degree 8 \'
336+
- '--use_torch_compile \'
337+
- '--guidance_scale 0.0 \'
338+
- '--num_iterations 25 \'
339+
- '--attention_backend aiter \'
340+
- '--output_directory results'
341+
- model: Qwen-Image-Edit
342+
model_repo: Qwen/Qwen-Image-Edit
343+
url: https://huggingface.co/Qwen/Qwen-Image-Edit
344+
github: https://github.com/QwenLM/Qwen-Image
345+
mad_tag: pyt_xdit_qwen_image_edit
346+
js_tag: qwen_image_edit_tag
347+
benchmark_command:
348+
- mkdir results
349+
- 'xdit \'
350+
- '--model {model_repo} \'
351+
- '--seed 42 \'
352+
- '--prompt "Add a cool hat to the cat." \'
353+
- '--negative_prompt "" \'
354+
- '--input_images /app/data/flux_cat.png \'
355+
- '--height 2048 \'
356+
- '--width 2048 \'
357+
- '--num_inference_steps 50 \'
358+
- '--ulysses_degree 8 \'
359+
- '--use_torch_compile \'
360+
- '--guidance_scale 4.0 \'
361+
- '--num_iterations 25 \'
362+
- '--attention_backend aiter \'
363+
- '--output_directory results'

0 commit comments

Comments
 (0)