26.0.0

ascend-robotchore(pytorch):optimize GLM5 pretraining script

c41d0d46创建于 3月25日历史提交

文件	最后提交记录	最后更新时间
ckpt_convert_glm5_hf2mcore.sh	chore(pytorch):optimize GLM5 pretraining script Co-authored-by: HanhuiChen<chenhanhui1@h-partners.com> # message auto-generated for no-merge-commit merge: !4340 merge glm into master chore(pytorch):optimize GLM5 pretraining script Created-by: HANHU1CHEN Commit-by: HanhuiChen Merged-by: ascend-robot Description: ## What this PR does / why we need it? This PR updates the GLM5 pretraining script to improve training performance and efficiency. ## Does this PR introduce any user-facing change? No major user-facing changes, but minor adjustments to configs or launch commands may be required. ## How was this patch tested? Validated through end-to-end pretraining runs, confirming improved performance and stable training behavior. See merge request: Ascend/MindSpeed-LLM!4340	2 个月前
ckpt_convert_glm5_mcore2hf.sh	[pytorch][model]add megatron weight-converter for glm5 Co-authored-by: l00961550<liuqingyuan19@huawei.com> # message auto-generated for no-merge-commit merge: !4162 merge master into master [pytorch][model]add megatron weight-converter for glm5 Created-by: lqy16 Commit-by: l00961550 Merged-by: ascend-robot Description: add megatron weight-converter for glm5 See merge request: Ascend/MindSpeed-LLM!4162	3 个月前
data_convert_glm5_pretrain.sh	[pytorch][model]add pretrain_glm5_744b_4k_A3_ptd.sh Co-authored-by: EVA1<jingsiyu1@huawei.com> # message auto-generated for no-merge-commit merge: !4189 merge master into master [pytorch][model]add pretrain_glm5_744b_4k_A3_ptd.sh Created-by: EVA1 Commit-by: EVA1 Merged-by: ascend-robot Description: add pretrain_glm5_744b_4k_A3_ptd.sh See merge request: Ascend/MindSpeed-LLM!4189	3 个月前
generate_glm5_744b_A3_ptd.sh	[pytorch][bugfix]fix problems of glm5 scripts Co-authored-by: HanhuiChen<chenhanhui1@h-partners.com> # message auto-generated for no-merge-commit merge: !4221 merge glm into master [pytorch][bugfix]fix problems of glm5 scripts Created-by: HANHU1CHEN Commit-by: HanhuiChen Merged-by: ascend-robot Description: fix problems of glm5 scripts See merge request: Ascend/MindSpeed-LLM!4221	3 个月前
pretrain_glm5_744b_4k_A3_ptd.sh	chore(pytorch):optimize GLM5 pretraining script Co-authored-by: HanhuiChen<chenhanhui1@h-partners.com> # message auto-generated for no-merge-commit merge: !4340 merge glm into master chore(pytorch):optimize GLM5 pretraining script Created-by: HANHU1CHEN Commit-by: HanhuiChen Merged-by: ascend-robot Description: ## What this PR does / why we need it? This PR updates the GLM5 pretraining script to improve training performance and efficiency. ## Does this PR introduce any user-facing change? No major user-facing changes, but minor adjustments to configs or launch commands may be required. ## How was this patch tested? Validated through end-to-end pretraining runs, confirming improved performance and stable training behavior. See merge request: Ascend/MindSpeed-LLM!4340	2 个月前