2.3.0

ascend-robot[Modify][MindSpore] Optimize data preprocessing performance for Qwen3vl

1c97c395创建于 2025年12月11日历史提交

文件	最后提交记录	最后更新时间
data	【mindspore】Adapted Qwen25-Omni/Qwen3vl Co-authored-by: AACAES<liuyan7892@163.com> # message auto-generated for no-merge-commit merge: !1682 merge master into master 【mindspore】Adapted Qwen25-Omni/Qwen3vl Created-by: AACAES Commit-by: AACAES Merged-by: ascend-robot Description: ## Motivation Adapted Qwen25-Omni/Qwen3vl ## Modification Adapted Qwen25-Omni/Qwen3vl ## Self-test (Optional) If modifications to this PR may cause/fix function/accuracy/performance DTSs/issues, a self-inspection record needs to be attached. Qwen25-Omni:  Qwen3vl 数据集处理部分精度对齐:  ## BC-breaking (Optional) If there are compatibility issues, such as dependencies on cann/torch_npu versions, they need to be explained in the PR. ## Checklist **Before PR**: - [x] The new code needs to comply with the Clean Code specification. - [x] The PR content is self-checked, and the expression can be clear and the writing standardized **After PR**: - [x] CLA has been signed and all committers have signed the CLA in this PR. - [x] The ci-pipeline is passed, Code Check is passed. See merge request: Ascend/MindSpeed-MM!1682	6 个月前
models	[master][Modify]Add _gather patch support for InternVL3 Co-authored-by: iquoyuw<wuyouqi1@h-partners.com> # message auto-generated for no-merge-commit merge: !1697 merge internvl3-7b into master [master][Modify]Add _gather patch support for InternVL3 Created-by: weixin_47897441 Commit-by: iquoyuw Merged-by: ascend-robot Description: ## Motivation Add _gather patch support for InternVL3 model. ## Modification This modification extends the device type check to include the "Ascend" device type in addition to "cuda" and "npu". ## Self-test (Optional) This patch enables accurate training of Internvl3 on the MS framework by aligning its 100-iteration, 16-bit loss/gradient with the PT framework, and no new accuracy issues are caused.  ## BC-breaking (Optional) If there are compatibility issues, such as dependencies on cann/torch_npu versions, they need to be explained in the PR. ## Checklist **Before PR**: - [ ] The new code needs to comply with the Clean Code specification. - [ ] The PR content is self-checked, and the expression can be clear and the writing standardized **After PR**: - [ ] CLA has been signed and all committers have signed the CLA in this PR. - [ ] The ci-pipeline is passed, Code Check is passed. See merge request: Ascend/MindSpeed-MM!1697	6 个月前
third_party	[Modify][MindSpore] Optimize data preprocessing performance for Qwen3vl Co-authored-by: weixin_47897441<wuyouqi1@h-partners.com> # message auto-generated for no-merge-commit merge: !1908 merge 2.3.0-qwenvl3 into 2.3.0 [Modify][MindSpore] Optimize data preprocessing performance for Qwen3vl Created-by: weixin_47897441 Commit-by: weixin_47897441 Merged-by: ascend-robot Description: ## Motivation This PR aims to optimize the performance of qwenvl3 data preprocessing by replacing partial tensor operations with tensor-to-numpy conversions, reducing unnecessary tensor memory copies and improving the overall efficiency of the preprocessing pipeline. ## Modification Replace tensor-based data operations that cause redundant memory copies with tensor-to-numpy conversion operations. ## Self-test (Optional) 1.After modification, the output of qwenvl3 data preprocessing is consistent with the original version. 2.The preprocessing time is reduced from 10xPTA to 5.2xPTA.  ## BC-breaking (Optional) If there are compatibility issues, such as dependencies on cann/torch_npu versions, they need to be explained in the PR. ## Checklist **Before PR**: - [x] The new code needs to comply with the Clean Code specification. - [x] The PR content is self-checked, and the expression can be clear and the writing standardized **After PR**: - [x] CLA has been signed and all committers have signed the CLA in this PR. - [x] The ci-pipeline is passed, Code Check is passed. See merge request: Ascend/MindSpeed-MM!1908	5 个月前
utils	[mindspore] support qwen3vl finetune for megatron backend Co-authored-by: ffmh<fengminghao2@huawei.com> # message auto-generated for no-merge-commit merge: !1785 merge qwen3vl_megatron into master [mindspore] support qwen3vl finetune for megatron backend Created-by: ffmh Commit-by: ffmh Merged-by: ascend-robot Description: ## Motivation Please describe the motivation of this PR and the goal you want to achieve through this PR. 适配 qwen3vl 8B/30B megatron后端 ## Modification Please briefly describe what modification is made in this PR. 添加 layer_spec修改、text模块transformer block/layer添加deepstack相关适配 ## Self-test (Optional) If modifications to this PR may cause/fix function/accuracy/performance DTSs/issues, a self-inspection record needs to be attached. ## BC-breaking (Optional) If there are compatibility issues, such as dependencies on cann/torch_npu versions, they need to be explained in the PR. ## Checklist **Before PR**: - [ ] The new code needs to comply with the Clean Code specification. - [ ] The PR content is self-checked, and the expression can be clear and the writing standardized **After PR**: - [ ] CLA has been signed and all committers have signed the CLA in this PR. - [ ] The ci-pipeline is passed, Code Check is passed. See merge request: Ascend/MindSpeed-MM!1785	6 个月前
mindspore_adaptor.py	[Modify][MindSpore] Optimize data preprocessing performance for Qwen3vl Co-authored-by: weixin_47897441<wuyouqi1@h-partners.com> # message auto-generated for no-merge-commit merge: !1908 merge 2.3.0-qwenvl3 into 2.3.0 [Modify][MindSpore] Optimize data preprocessing performance for Qwen3vl Created-by: weixin_47897441 Commit-by: weixin_47897441 Merged-by: ascend-robot Description: ## Motivation This PR aims to optimize the performance of qwenvl3 data preprocessing by replacing partial tensor operations with tensor-to-numpy conversions, reducing unnecessary tensor memory copies and improving the overall efficiency of the preprocessing pipeline. ## Modification Replace tensor-based data operations that cause redundant memory copies with tensor-to-numpy conversion operations. ## Self-test (Optional) 1.After modification, the output of qwenvl3 data preprocessing is consistent with the original version. 2.The preprocessing time is reduced from 10xPTA to 5.2xPTA.  ## BC-breaking (Optional) If there are compatibility issues, such as dependencies on cann/torch_npu versions, they need to be explained in the PR. ## Checklist **Before PR**: - [x] The new code needs to comply with the Clean Code specification. - [x] The PR content is self-checked, and the expression can be clear and the writing standardized **After PR**: - [x] CLA has been signed and all committers have signed the CLA in this PR. - [x] The ci-pipeline is passed, Code Check is passed. See merge request: Ascend/MindSpeed-MM!1908	5 个月前
pretrain_qwen3vl.py	[Bugfix][2.3.0][MindSpore] fix EP validate in vision for mindspore qwen3vl Co-authored-by: ffmh<fengminghao2@huawei.com> # message auto-generated for no-merge-commit merge: !1844 merge fix_ep_2.3.0 into 2.3.0 [Bugfix][2.3.0][MindSpore] fix EP validate in vision for mindspore qwen3vl Created-by: ffmh Commit-by: ffmh Merged-by: ascend-robot Description: ## Motivation Fix TranformerConfig expert_model_parallel_size validation in vision encoder/projecter ## Modification Set vision encoder/projector expert_model_parallel_size to 1 ## Self-test (Optional) If modifications to this PR may cause/fix function/accuracy/performance DTSs/issues, a self-inspection record needs to be attached. fix: ``` [rank6]: model = model_provider_func(*args, **kwargs) [rank6]: File "/home/f00913055/mm/code1124_qwen3vl/MindSpeed-MM/mindspeed_mm/mindspore/pretrain_qwen3vl.py", line 46, in model_provider [rank6]: _configure_modules(vlm_config, modules) [rank6]: File "/home/f00913055/mm/code1124_qwen3vl/MindSpeed-MM/mindspeed_mm/mindspore/pretrain_qwen3vl.py", line 69, in _configure_modules [rank6]: config_func(vlm_config) [rank6]: File "/home/f00913055/mm/code1124_qwen3vl/MindSpeed-MM/mindspeed_mm/mindspore/pretrain_qwen3vl.py", line 83, in _configure_image_encoder [rank6]: vlm_config.image_encoder.vision_encoder = get_model_config(vlm_config.image_encoder.vision_encoder) [rank6]: File "/home/f00913055/mm/code1124_qwen3vl/MindSpeed-MM/mindspeed_mm/mindspore/utils/transformer_model_config.py", line 69, in get_model_config [rank6]: trans_config = TransformerConfig(**t_config) [rank6]: File "/home/f00913055/mm/code1124_qwen3vl/MindSpeed/mindspeed/core/megatron_basic/arguments_basic.py", line 111, in wrapper [rank6]: fn(self, *args, **known_config) [rank6]: File "<string>", line 177, in __init__ [rank6]: File "/home/f00913055/mm/code1124_qwen3vl/MindSpeed-MM/mindspeed_mm/utils/transformer_model_config.py", line 133, in wrapper [rank6]: fn(self) [rank6]: File "/home/f00913055/mm/code1124_qwen3vl/MindSpeed/mindspeed/core/fusions/fused_rope.py", line 83, in wrapper [rank6]: fn(self) [rank6]: File "/home/f00913055/mm/code1124_qwen3vl/MindSpeed/mindspeed/core/megatron_basic/arguments_basic.py", line 79, in wrapper [rank6]: fn(self) [rank6]: File "/home/f00913055/mm/code1124_qwen3vl/Megatron-LM/megatron/core/transformer/transformer_config.py", line 621, in __post_init__ [rank6]: raise ValueError('num_moe_experts must be non None to use expert-parallel.') [rank6]: ValueError: num_moe_experts must be non None to use expert-parallel. ``` ## BC-breaking (Optional) If there are compatibility issues, such as dependencies on cann/torch_npu versions, they need to be explained in the PR. ## Checklist **Before PR**: - [ ] The new code needs to comply with the Clean Code specification. - [ ] The PR content is self-checked, and the expression can be clear and the writing standardized **After PR**: - [ ] CLA has been signed and all committers have signed the CLA in this PR. - [ ] The ci-pipeline is passed, Code Check is passed. See merge request: Ascend/MindSpeed-MM!1844	6 个月前