chore: import upstream snapshot with attribution

examples/megatron/multimodal/dense/dpo.sh

@@ -0,0 +1,41 @@
+# 4 * 50GiB 14s/it
+PYTORCH_CUDA_ALLOC_CONF='expandable_segments:True' \
+NPROC_PER_NODE=4 \
+MAX_PIXELS=1003520 \
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+megatron rlhf \
+    --rlhf_type dpo \
+    --model Qwen/Qwen2.5-VL-7B-Instruct \
+    --save_safetensors true \
+    --dataset 'swift/RLAIF-V-Dataset#20000' \
+    --load_from_cache_file true \
+    --tuner_type full \
+    --tensor_model_parallel_size 4 \
+    --sequence_parallel true \
+    --freeze_llm false \
+    --freeze_vit true \
+    --freeze_aligner true \
+    --packing true \
+    --split_dataset_ratio 0.01 \
+    --micro_batch_size 1 \
+    --global_batch_size 4 \
+    --recompute_granularity full \
+    --recompute_method uniform \
+    --recompute_num_layers 1 \
+    --finetune true \
+    --cross_entropy_loss_fusion true \
+    --lr 1e-5 \
+    --lr_warmup_fraction 0.05 \
+    --min_lr 1e-6 \
+    --num_train_epochs 1 \
+    --output_dir megatron_output/Qwen2.5-VL-7B-Instruct \
+    --save_steps 200 \
+    --max_length 8192 \
+    --dataloader_num_workers 4 \
+    --no_save_optim true \
+    --no_save_rng true \
+    --dataset_num_proc 8 \
+    --attention_backend flash \
+    --rpo_alpha 0.1 \
+    --beta 0.1 \
+    --loss_type sigmoid

examples/megatron/multimodal/dense/full.sh

@@ -0,0 +1,35 @@
+# 2 * 72GiB; 4.1s/it
+PYTORCH_CUDA_ALLOC_CONF='expandable_segments:True' \
+NPROC_PER_NODE=2 \
+MAX_PIXELS=1003520 \
+CUDA_VISIBLE_DEVICES=0,1 \
+megatron sft \
+    --model Qwen/Qwen2.5-VL-7B-Instruct \
+    --save_safetensors true \
+    --dataset 'AI-ModelScope/LaTeX_OCR:human_handwrite#5000' \
+    --load_from_cache_file true \
+    --tensor_model_parallel_size 2 \
+    --sequence_parallel true \
+    --packing true \
+    --freeze_llm false \
+    --freeze_vit true \
+    --freeze_aligner true \
+    --split_dataset_ratio 0.01 \
+    --micro_batch_size 1 \
+    --global_batch_size 4 \
+    --recompute_granularity full \
+    --recompute_method uniform \
+    --recompute_num_layers 1 \
+    --finetune true \
+    --cross_entropy_loss_fusion true \
+    --lr 1e-5 \
+    --lr_warmup_fraction 0.05 \
+    --min_lr 1e-6 \
+    --num_train_epochs 1 \
+    --output_dir megatron_output/Qwen2.5-VL-7B-Instruct \
+    --save_steps 200 \
+    --max_length 2048 \
+    --dataloader_num_workers 4 \
+    --no_save_optim true \
+    --no_save_rng true \
+    --dataset_num_proc 8

examples/megatron/multimodal/dense/lora.sh

@@ -0,0 +1,40 @@
+# 2 * 23GiB; 2.3s/it
+PYTORCH_CUDA_ALLOC_CONF='expandable_segments:True' \
+NPROC_PER_NODE=2 \
+MAX_PIXELS=1003520 \
+CUDA_VISIBLE_DEVICES=0,1 \
+megatron sft \
+    --model Qwen/Qwen2.5-VL-7B-Instruct \
+    --save_safetensors true \
+    --merge_lora false \
+    --dataset 'AI-ModelScope/LaTeX_OCR:human_handwrite#5000' \
+    --load_from_cache_file true \
+    --tuner_type lora \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --target_modules all-linear \
+    --tensor_model_parallel_size 1 \
+    --sequence_parallel true \
+    --freeze_llm false \
+    --freeze_vit true \
+    --freeze_aligner true \
+    --packing true \
+    --split_dataset_ratio 0.01 \
+    --micro_batch_size 1 \
+    --global_batch_size 4 \
+    --recompute_granularity full \
+    --recompute_method uniform \
+    --recompute_num_layers 1 \
+    --finetune true \
+    --cross_entropy_loss_fusion true \
+    --lr 1e-4 \
+    --lr_warmup_fraction 0.05 \
+    --min_lr 1e-5 \
+    --num_train_epochs 1 \
+    --output_dir megatron_output/Qwen2.5-VL-7B-Instruct \
+    --save_steps 200 \
+    --max_length 2048 \
+    --dataloader_num_workers 4 \
+    --no_save_optim true \
+    --no_save_rng true \
+    --dataset_num_proc 8

examples/megatron/multimodal/lora_llm_vit_full/sft.sh

@@ -0,0 +1,69 @@
+# Currently, 'lora_llm' must set `--merge_lora true` and `--no_save_optim true`
+
+PYTORCH_CUDA_ALLOC_CONF='expandable_segments:True' \
+NPROC_PER_NODE=4 \
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+IMAGE_MAX_TOKEN_NUM=1024 \
+VIDEO_MAX_TOKEN_NUM=128 \
+FPS_MAX_FRAMES=12 \
+megatron sft \
+    --model Qwen/Qwen3.5-35B-A3B \
+    --save_safetensors true \
+    --merge_lora true \
+    --dataset 'AI-ModelScope/alpaca-gpt4-data-zh#500' \
+              'AI-ModelScope/alpaca-gpt4-data-en#500' \
+              'swift/self-cognition#500' \
+              'AI-ModelScope/LaTeX_OCR:human_handwrite#2000' \
+    --load_from_cache_file true \
+    --add_non_thinking_prefix true \
+    --loss_scale ignore_empty_think \
+    --split_dataset_ratio 0.01 \
+    --tuner_type lora_llm \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --target_modules all-linear \
+    --expert_model_parallel_size 4 \
+    --moe_permute_fusion true \
+    --moe_grouped_gemm true \
+    --moe_shared_expert_overlap true \
+    --moe_aux_loss_coeff 1e-6 \
+    --micro_batch_size 1 \
+    --global_batch_size 4 \
+    --recompute_granularity full \
+    --recompute_method uniform \
+    --recompute_num_layers 1 \
+    --num_train_epochs 2 \
+    --finetune true \
+    --cross_entropy_loss_fusion true \
+    --lr 1e-4 \
+    --vit_lr 1e-5 \
+    --aligner_lr 2e-5 \
+    --lr_warmup_fraction 0.05 \
+    --min_lr 1e-5 \
+    --output_dir megatron_output/Qwen3.5-35B-A3B \
+    --eval_steps 200 \
+    --save_steps 200 \
+    --max_length 2048 \
+    --dataloader_num_workers 8 \
+    --dataset_num_proc 8 \
+    --no_save_optim true \
+    --no_save_rng true \
+    --sequence_parallel true \
+    --attention_backend flash \
+    --padding_free true \
+    --packing true \
+    --model_author swift \
+    --model_name swift-robot
+
+
+# PYTORCH_CUDA_ALLOC_CONF='expandable_segments:True' \
+# CUDA_VISIBLE_DEVICES=0,1,2,3 \
+# IMAGE_MAX_TOKEN_NUM=1024 \
+# VIDEO_MAX_TOKEN_NUM=128 \
+# FPS_MAX_FRAMES=12 \
+# swift infer \
+#     --model megatron_output/Qwen3.5-35B-A3B/vx-xxx/checkpoint-xxx-merged \
+#     --stream true \
+#     --experts_impl grouped_mm \
+#     --enable_thinking false \
+#     --load_data_args true

examples/megatron/multimodal/moe/full_dpo_offload.sh

@@ -0,0 +1,52 @@
+# 16s/it; 8 * 65GiB
+PYTORCH_CUDA_ALLOC_CONF='expandable_segments:True' \
+NPROC_PER_NODE=8 \
+CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
+megatron rlhf \
+    --rlhf_type dpo \
+    --model OpenGVLab/InternVL3_5-30B-A3B \
+    --save_safetensors true \
+    --dataset 'swift/RLAIF-V-Dataset#20000' \
+    --load_from_cache_file true \
+    --tuner_type full \
+    --tensor_model_parallel_size 4 \
+    --expert_tensor_parallel_size 1 \
+    --pipeline_model_parallel_size 2 \
+    --decoder_first_pipeline_num_layers 23 \
+    --split_dataset_ratio 0.01 \
+    --expert_model_parallel_size 4 \
+    --moe_permute_fusion true \
+    --moe_grouped_gemm true \
+    --moe_shared_expert_overlap true \
+    --moe_aux_loss_coeff 1e-3 \
+    --micro_batch_size 1 \
+    --global_batch_size 4 \
+    --packing true \
+    --recompute_granularity full \
+    --recompute_method uniform \
+    --recompute_num_layers 1 \
+    --finetune true \
+    --cross_entropy_loss_fusion true \
+    --lr 1e-5 \
+    --lr_warmup_fraction 0.05 \
+    --min_lr 1e-6 \
+    --output_dir megatron_output/InternVL3_5-30B-A3B \
+    --eval_steps 500 \
+    --save_steps 500 \
+    --max_length 16384 \
+    --num_train_epochs 1 \
+    --dataloader_num_workers 8 \
+    --dataset_num_proc 8 \
+    --no_save_optim true \
+    --no_save_rng true \
+    --sequence_parallel true \
+    --freeze_llm false \
+    --freeze_vit true \
+    --freeze_aligner true \
+    --optimizer_cpu_offload true \
+    --use_precision_aware_optimizer true \
+    --optimizer_offload_fraction 0.65 \
+    --attention_backend flash \
+    --rpo_alpha 0.1 \
+    --beta 0.1 \
+    --loss_type sigmoid

examples/megatron/multimodal/moe/lora.sh

@@ -0,0 +1,45 @@
+# 2 * 43GiB, 8s/it
+PYTORCH_CUDA_ALLOC_CONF='expandable_segments:True' \
+NPROC_PER_NODE=2 \
+CUDA_VISIBLE_DEVICES=0,1 \
+megatron sft \
+    --model OpenGVLab/InternVL3_5-30B-A3B \
+    --save_safetensors true \
+    --merge_lora false \
+    --dataset 'AI-ModelScope/LaTeX_OCR:human_handwrite#5000' \
+    --load_from_cache_file true \
+    --tuner_type lora \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --target_modules all-linear \
+    --sequence_parallel true \
+    --freeze_llm false \
+    --freeze_vit true \
+    --freeze_aligner true \
+    --packing true \
+    --split_dataset_ratio 0.01 \
+    --expert_model_parallel_size 2 \
+    --moe_permute_fusion true \
+    --moe_grouped_gemm true \
+    --moe_shared_expert_overlap true \
+    --moe_aux_loss_coeff 1e-3 \
+    --micro_batch_size 1 \
+    --global_batch_size 4 \
+    --recompute_granularity full \
+    --recompute_method uniform \
+    --recompute_num_layers 1 \
+    --finetune true \
+    --cross_entropy_loss_fusion true \
+    --lr 1e-4 \
+    --lr_warmup_fraction 0.05 \
+    --min_lr 1e-5 \
+    --num_train_epochs 1 \
+    --output_dir megatron_output/InternVL3_5-30B-A3B \
+    --eval_steps 200 \
+    --save_steps 200 \
+    --max_length 2048 \
+    --dataloader_num_workers 8 \
+    --dataset_num_proc 8 \
+    --no_save_optim true \
+    --no_save_rng true \
+    --attention_backend flash

examples/megatron/multimodal/omni/dense.sh

@@ -0,0 +1,41 @@
+# 2 * 75GiB; 7s/it
+# Supports mixed modalities
+PYTORCH_CUDA_ALLOC_CONF='expandable_segments:True' \
+ENABLE_AUDIO_OUTPUT=0 \
+NPROC_PER_NODE=2 \
+MAX_PIXELS=1003520 \
+VIDEO_MAX_PIXELS=50176 \
+FPS_MAX_FRAMES=12 \
+CUDA_VISIBLE_DEVICES=0,1 \
+megatron sft \
+    --model Qwen/Qwen2.5-Omni-7B \
+    --save_safetensors true \
+    --dataset 'AI-ModelScope/alpaca-gpt4-data-zh#10000' \
+              'AI-ModelScope/LaTeX_OCR:human_handwrite#5000' \
+              'speech_asr/speech_asr_aishell1_trainsets:validation#5000' \
+    --load_from_cache_file true \
+    --tensor_model_parallel_size 2 \
+    --sequence_parallel true \
+    --packing true \
+    --freeze_llm false \
+    --freeze_vit true \
+    --freeze_aligner true \
+    --split_dataset_ratio 0.01 \
+    --micro_batch_size 1 \
+    --global_batch_size 4 \
+    --recompute_granularity full \
+    --recompute_method uniform \
+    --recompute_num_layers 1 \
+    --finetune true \
+    --cross_entropy_loss_fusion true \
+    --lr 1e-5 \
+    --lr_warmup_fraction 0.05 \
+    --min_lr 1e-6 \
+    --num_train_epochs 1 \
+    --output_dir megatron_output/Qwen2.5-Omni-7B \
+    --save_steps 100 \
+    --max_length 4096 \
+    --dataloader_num_workers 4 \
+    --no_save_optim true \
+    --no_save_rng true \
+    --dataset_num_proc 8

examples/megatron/multimodal/omni/moe.sh

@@ -0,0 +1,51 @@
+# 2 * 50GiB
+PYTORCH_CUDA_ALLOC_CONF='expandable_segments:True' \
+ENABLE_AUDIO_OUTPUT=0 \
+NPROC_PER_NODE=2 \
+IMAGE_MAX_TOKEN_NUM=1024 \
+VIDEO_MAX_TOKEN_NUM=128 \
+FPS_MAX_FRAMES=12 \
+CUDA_VISIBLE_DEVICES=0,1 \
+megatron sft \
+    --model Qwen/Qwen3-Omni-30B-A3B-Instruct \
+    --save_safetensors true \
+    --merge_lora false \
+    --dataset 'AI-ModelScope/alpaca-gpt4-data-zh#10000' \
+              'AI-ModelScope/LaTeX_OCR:human_handwrite#5000' \
+              'speech_asr/speech_asr_aishell1_trainsets:validation#5000' \
+    --load_from_cache_file true \
+    --tuner_type lora \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --target_modules all-linear \
+    --sequence_parallel true \
+    --packing true \
+    --freeze_llm false \
+    --freeze_vit true \
+    --freeze_aligner true \
+    --split_dataset_ratio 0.01 \
+    --expert_model_parallel_size 2 \
+    --moe_permute_fusion true \
+    --moe_grouped_gemm true \
+    --moe_shared_expert_overlap true \
+    --moe_aux_loss_coeff 1e-3 \
+    --micro_batch_size 1 \
+    --global_batch_size 4 \
+    --recompute_granularity full \
+    --recompute_method uniform \
+    --recompute_num_layers 1 \
+    --finetune true \
+    --cross_entropy_loss_fusion true \
+    --lr 1e-4 \
+    --lr_warmup_fraction 0.05 \
+    --min_lr 1e-5 \
+    --num_train_epochs 1 \
+    --output_dir megatron_output/Qwen3-Omni-30B-A3B-Instruct \
+    --eval_steps 100 \
+    --save_steps 100 \
+    --max_length 4096 \
+    --dataloader_num_workers 8 \
+    --dataset_num_proc 8 \
+    --no_save_optim true \
+    --no_save_rng true \
+    --attention_backend flash