v3.0.0

ascend-robot[openlibing.ci] auto fixes by pre-commit

64913abd创建于 5月11日历史提交

文件	最后提交记录	最后更新时间
__init__.py	[Feature][dummy_run] Add Qwen-Image and FLUX.1-dev dummy run with unified CLI and ms time Co-authored-by: blian6<bin.lian@outlook.com> Co-authored-by: blian<lianbin@huawei.com> # message auto-generated for no-merge-commit merge: !275 merge dev into dev [Feature][dummy_run] Add Qwen-Image and FLUX.1-dev dummy run with unified CLI and ms time Created-by: blian Commit-by: blian6;blian Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes > Fixes #待补充 # Purpose 在现有 Wan2.2 dummy run 基础上，新增 Qwen-Image 和 FLUX.1-dev 两个模型的空权重运行时验证： - 统一 CLI 参数（--num_layers --compile --profile --skip-vae） - 统一 NPU 直连加载（移除 CPU offload） - 统一毫秒计时（2 位小数），profiling 仅覆盖 timed step - 精简日志输出（减少 ~30 条冗余 warning） - README 从 354 行精简到 182 行 # Test Plan 1. 安装依赖：pip install -r examples/dummy_run/requirements.txt 2. 远端 NPU 验证（910B, 64GB HBM）： shell python wan_infer.py --device_id 0 python qwen_image_infer.py --device_id 0 python flux_infer.py --device_id 0 --config_cache /path/to/flux_configs 3. 检查输出 PASSED + 各组件毫秒计时 + Summary 表 # Test Report Model Build (ms) Timed (ms) Peak Mem Status Wan2.2 1,200 7,000 10.18 GB PASSED Qwen-Image 7,000 100 6.26 GB PASSED FLUX.1-dev 20,500 900 24.20 GB PASSED See merge request: Ascend/MindIE-SD!275	2 个月前
flux_model.py	[Feature][dummy_run] Add Qwen-Image and FLUX.1-dev dummy run with unified CLI and ms time Co-authored-by: blian6<bin.lian@outlook.com> Co-authored-by: blian<lianbin@huawei.com> # message auto-generated for no-merge-commit merge: !275 merge dev into dev [Feature][dummy_run] Add Qwen-Image and FLUX.1-dev dummy run with unified CLI and ms time Created-by: blian Commit-by: blian6;blian Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes > Fixes #待补充 # Purpose 在现有 Wan2.2 dummy run 基础上，新增 Qwen-Image 和 FLUX.1-dev 两个模型的空权重运行时验证： - 统一 CLI 参数（--num_layers --compile --profile --skip-vae） - 统一 NPU 直连加载（移除 CPU offload） - 统一毫秒计时（2 位小数），profiling 仅覆盖 timed step - 精简日志输出（减少 ~30 条冗余 warning） - README 从 354 行精简到 182 行 # Test Plan 1. 安装依赖：pip install -r examples/dummy_run/requirements.txt 2. 远端 NPU 验证（910B, 64GB HBM）： shell python wan_infer.py --device_id 0 python qwen_image_infer.py --device_id 0 python flux_infer.py --device_id 0 --config_cache /path/to/flux_configs 3. 检查输出 PASSED + 各组件毫秒计时 + Summary 表 # Test Report Model Build (ms) Timed (ms) Peak Mem Status Wan2.2 1,200 7,000 10.18 GB PASSED Qwen-Image 7,000 100 6.26 GB PASSED FLUX.1-dev 20,500 900 24.20 GB PASSED See merge request: Ascend/MindIE-SD!275	2 个月前
hunyuan_image3_model.py	[openlibing.ci] auto fixes by pre-commit	2 个月前
qwen_image_model.py	[Feature][dummy_run] Add Qwen-Image and FLUX.1-dev dummy run with unified CLI and ms time Co-authored-by: blian6<bin.lian@outlook.com> Co-authored-by: blian<lianbin@huawei.com> # message auto-generated for no-merge-commit merge: !275 merge dev into dev [Feature][dummy_run] Add Qwen-Image and FLUX.1-dev dummy run with unified CLI and ms time Created-by: blian Commit-by: blian6;blian Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes > Fixes #待补充 # Purpose 在现有 Wan2.2 dummy run 基础上，新增 Qwen-Image 和 FLUX.1-dev 两个模型的空权重运行时验证： - 统一 CLI 参数（--num_layers --compile --profile --skip-vae） - 统一 NPU 直连加载（移除 CPU offload） - 统一毫秒计时（2 位小数），profiling 仅覆盖 timed step - 精简日志输出（减少 ~30 条冗余 warning） - README 从 354 行精简到 182 行 # Test Plan 1. 安装依赖：pip install -r examples/dummy_run/requirements.txt 2. 远端 NPU 验证（910B, 64GB HBM）： shell python wan_infer.py --device_id 0 python qwen_image_infer.py --device_id 0 python flux_infer.py --device_id 0 --config_cache /path/to/flux_configs 3. 检查输出 PASSED + 各组件毫秒计时 + Summary 表 # Test Report Model Build (ms) Timed (ms) Peak Mem Status Wan2.2 1,200 7,000 10.18 GB PASSED Qwen-Image 7,000 100 6.26 GB PASSED FLUX.1-dev 20,500 900 24.20 GB PASSED See merge request: Ascend/MindIE-SD!275	2 个月前
wan_model.py	[Feature][dummy_run] Add Qwen-Image and FLUX.1-dev dummy run with unified CLI and ms time Co-authored-by: blian6<bin.lian@outlook.com> Co-authored-by: blian<lianbin@huawei.com> # message auto-generated for no-merge-commit merge: !275 merge dev into dev [Feature][dummy_run] Add Qwen-Image and FLUX.1-dev dummy run with unified CLI and ms time Created-by: blian Commit-by: blian6;blian Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes > Fixes #待补充 # Purpose 在现有 Wan2.2 dummy run 基础上，新增 Qwen-Image 和 FLUX.1-dev 两个模型的空权重运行时验证： - 统一 CLI 参数（--num_layers --compile --profile --skip-vae） - 统一 NPU 直连加载（移除 CPU offload） - 统一毫秒计时（2 位小数），profiling 仅覆盖 timed step - 精简日志输出（减少 ~30 条冗余 warning） - README 从 354 行精简到 182 行 # Test Plan 1. 安装依赖：pip install -r examples/dummy_run/requirements.txt 2. 远端 NPU 验证（910B, 64GB HBM）： shell python wan_infer.py --device_id 0 python qwen_image_infer.py --device_id 0 python flux_infer.py --device_id 0 --config_cache /path/to/flux_configs 3. 检查输出 PASSED + 各组件毫秒计时 + Summary 表 # Test Report Model Build (ms) Timed (ms) Peak Mem Status Wan2.2 1,200 7,000 10.18 GB PASSED Qwen-Image 7,000 100 6.26 GB PASSED FLUX.1-dev 20,500 900 24.20 GB PASSED See merge request: Ascend/MindIE-SD!275	2 个月前