Skip to content

Inference Sparse Optimization of Wan2.2 on Ascend Platform #1301

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Watebear wants to merge 6 commits into
base: main
Choose a base branch
from
Open

Inference Sparse Optimization of Wan2.2 on Ascend Platform #1301

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
90fff81
feat: support INT8 Ulysses communication
Watebear Jul 27, 2026
47e699b
feat: support INT8 sparse attntion optimization of Wan22
Watebear Jul 27, 2026
51d81f9
chore: clean up Ascend sparse integration
Watebear Jul 27, 2026
3ee6b94
chore: relocate Ascend RainFusion distill example
Watebear Jul 27, 2026
5d902f2
Revert "chore: relocate Ascend RainFusion distill example"
Watebear Jul 27, 2026
ebbfeba
chore: relocate Ascend INT8 distill example
Watebear Jul 27, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
0 ...e_i2v_distill_int8_4step_ulysses_npu.json → ...e_i2v_distill_int8_4step_ulysses_npu.json
View file
Open in desktop
File renamed without changes.
47 changes: 47 additions & 0 deletions configs/platforms/ascend_npu/extreme/wan_moe_i2v_distill_int8_dynamic_sparse_attn.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,47 @@
{
"infer_steps": 4,
"target_video_length": 81,
"text_len": 512,
"target_height": 720,
"target_width": 1280,
"self_attn_1_type": "npu_dynamic_sparse_attn",
"dynamic_sparse_attn_setting": {
"sparsity_ratio": 0.9,
"q_block_size": 128,
"kv_block_size": 128,
"backend": "mindie_rf_v2",
"allow_v2_fallback": true,
"v3_inner_precise": 0,
"v2_inner_precise": 0
},
"cross_attn_1_type": "npu_flash_attn",
"cross_attn_2_type": "npu_flash_attn",
"sample_guide_scale": [
3.5,
3.5
],
"sample_shift": 7.0,
"enable_cfg": false,
"cpu_offload": false,
"t5_cpu_offload": true,
"vae_cpu_offload": false,
"use_image_encoder": false,
"boundary_step_index": 2,
"denoising_step_list": [
1000,
750,
500,
250
],
"dit_quantized": true,
"dit_quant_scheme": "int8-npu",
"mxfp8_fuse_enable": false,
"modulate_type": "torch",
"rope_type": "npu_rope",
"layer_norm_type": "npu_layer_norm",
"rms_norm_type": "npu_rms_norm",
"high_noise_quantized_ckpt": "Wan2.2-Int8-Sparse/Wan2.2-I2V-A14B_INT8_Sparse_high.safetensors",
"low_noise_quantized_ckpt": "Wan2.2-Int8-Sparse/Wan2.2-I2V-A14B_INT8_Sparse_low.safetensors",
"high_noise_original_ckpt": null,
"low_noise_original_ckpt": null
}
56 changes: 56 additions & 0 deletions ...latforms/ascend_npu/extreme/wan_moe_i2v_distill_int8_dynamic_sparse_attn_sp_parallel.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,56 @@
{
"infer_steps": 4,
"target_video_length": 81,
"text_len": 512,
"target_height": 720,
"target_width": 1280,
"self_attn_1_type": "npu_dynamic_sparse_attn",
"dynamic_sparse_attn_setting": {
"sparsity_ratio": 0.9,
"q_block_size": 128,
"kv_block_size": 128,
"backend": "mindie_rf_v2",
"allow_v2_fallback": true,
"v3_inner_precise": 0,
"v2_inner_precise": 0
},
"cross_attn_1_type": "npu_flash_attn",
"cross_attn_2_type": "npu_flash_attn",
"sample_guide_scale": [
3.5,
3.5
],
"sample_shift": 7.0,
"enable_cfg": false,
"cpu_offload": false,
"t5_cpu_offload": true,
"vae_cpu_offload": false,
"use_image_encoder": false,
"boundary_step_index": 2,
"denoising_step_list": [
1000,
750,
500,
250
],
"dit_quantized": true,
"dit_quant_scheme": "int8-npu",
"mxfp8_fuse_enable": false,
"modulate_type": "torch",
"rope_type": "npu_rope",
"layer_norm_type": "npu_layer_norm",
"rms_norm_type": "npu_rms_norm",
"high_noise_quantized_ckpt": "Wan2.2-Int8-Sparse/Wan2.2-I2V-A14B_INT8_Sparse_high.safetensors",
"low_noise_quantized_ckpt": "Wan2.2-Int8-Sparse/Wan2.2-I2V-A14B_INT8_Sparse_low.safetensors",
"high_noise_original_ckpt": null,
"low_noise_original_ckpt": null,
"parallel": {
"seq_p_size": 8,
"seq_p_quant_scheme": "int8",
"seq_p_head_parallel": true,
"seq_p_tensor_fusion": true,
"seq_p_attn_type": "ulysses",
"seq_p_prepost_backend": "torch",
"seq_p_a2a_backend": "torch"
}
}
47 changes: 47 additions & 0 deletions configs/platforms/ascend_npu/extreme/wan_moe_t2v_distill_int8_dynamic_sparse_attn.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,47 @@
{
"infer_steps": 4,
"target_video_length": 81,
"text_len": 512,
"target_height": 720,
"target_width": 1280,
"self_attn_1_type": "npu_dynamic_sparse_attn",
"dynamic_sparse_attn_setting": {
"sparsity_ratio": 0.9,
"q_block_size": 128,
"kv_block_size": 128,
"backend": "mindie_rf_v2",
"allow_v2_fallback": true,
"v3_inner_precise": 0,
"v2_inner_precise": 0
},
"cross_attn_1_type": "npu_flash_attn",
"cross_attn_2_type": "npu_flash_attn",
"sample_guide_scale": [
4.0,
3.0
],
"sample_shift": 5.0,
"enable_cfg": false,
"cpu_offload": false,
"t5_cpu_offload": true,
"vae_cpu_offload": false,
"use_image_encoder": false,
"boundary_step_index": 2,
"denoising_step_list": [
1000,
750,
500,
250
],
"dit_quantized": true,
"dit_quant_scheme": "int8-npu",
"mxfp8_fuse_enable": false,
"modulate_type": "torch",
"rope_type": "npu_rope",
"layer_norm_type": "npu_layer_norm",
"rms_norm_type": "npu_rms_norm",
"high_noise_quantized_ckpt": "Wan2.2-Int8-Sparse/Wan2.2-T2V-A14B_INT8_Sparse_high.safetensors",
"low_noise_quantized_ckpt": "Wan2.2-Int8-Sparse/Wan2.2-T2V-A14B_INT8_Sparse_low.safetensors",
"high_noise_original_ckpt": null,
"low_noise_original_ckpt": null
}
56 changes: 56 additions & 0 deletions ...latforms/ascend_npu/extreme/wan_moe_t2v_distill_int8_dynamic_sparse_attn_sp_parallel.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,56 @@
{
"infer_steps": 4,
"target_video_length": 81,
"text_len": 512,
"target_height": 720,
"target_width": 1280,
"self_attn_1_type": "npu_dynamic_sparse_attn",
"dynamic_sparse_attn_setting": {
"sparsity_ratio": 0.9,
"q_block_size": 128,
"kv_block_size": 128,
"backend": "mindie_rf_v2",
"allow_v2_fallback": true,
"v3_inner_precise": 0,
"v2_inner_precise": 0
},
"cross_attn_1_type": "npu_flash_attn",
"cross_attn_2_type": "npu_flash_attn",
"sample_guide_scale": [
4.0,
3.0
],
"sample_shift": 5.0,
"enable_cfg": false,
"cpu_offload": false,
"t5_cpu_offload": true,
"vae_cpu_offload": false,
"use_image_encoder": false,
"boundary_step_index": 2,
"denoising_step_list": [
1000,
750,
500,
250
],
"dit_quantized": true,
"dit_quant_scheme": "int8-npu",
"mxfp8_fuse_enable": false,
"modulate_type": "torch",
"rope_type": "npu_rope",
"layer_norm_type": "npu_layer_norm",
"rms_norm_type": "npu_rms_norm",
"high_noise_quantized_ckpt": "Wan2.2-Int8-Sparse/Wan2.2-T2V-A14B_INT8_Sparse_high.safetensors",
"low_noise_quantized_ckpt": "Wan2.2-Int8-Sparse/Wan2.2-T2V-A14B_INT8_Sparse_low.safetensors",
"high_noise_original_ckpt": null,
"low_noise_original_ckpt": null,
"parallel": {
"seq_p_size": 8,
"seq_p_quant_scheme": "int8",
"seq_p_head_parallel": true,
"seq_p_tensor_fusion": true,
"seq_p_attn_type": "ulysses",
"seq_p_prepost_backend": "torch",
"seq_p_a2a_backend": "torch"
}
}
Loading
Loading