Skip to content

fix_yaml #4023

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

Merged
xuxinyi389 merged 1 commit into from
Mar 10, 2026
Merged

fix_yaml #4023

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
6 changes: 3 additions & 3 deletions tests/config/benchmark/config/sft/Qwen3-VL-30B-A3B-Instruct.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@ train_dataset_prob: "1.0"
eval_dataset_path: /root/paddlejob/share-storage/gpfs/system-public/efficient_benchmark/dataset/coco_grounding/val.jsonl
eval_dataset_prob: "1.0"
max_seq_len: 8192
packing: false
packing: true
mix_strategy: concat
template_backend: custom
template: qwen3_vl
Expand All @@ -31,7 +31,7 @@ evaluation_strategy: steps
save_steps: 500
save_strategy: steps
logging_steps: 1
gradient_accumulation_steps: 8
gradient_accumulation_steps: 16
logging_dir: ./vdl_log
output_dir: ./checkpoints/qwen3-vl-sft-full
disable_tqdm: true
Expand Down Expand Up @@ -61,7 +61,7 @@ moe_grouped_gemm: true
moe_deep_gemm: true

apply_rope_fusion: False
# moe_router_force_load_balancing: true
moe_router_force_load_balancing: false

# sharding
split_param: true
Expand Down
8 changes: 5 additions & 3 deletions tests/config/benchmark/config/sft/Qwen3-VL-8B-Instruct.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@ save_steps: 500
save_strategy: steps
logging_steps: 1
save_total_limit: 1
gradient_accumulation_steps: 8
gradient_accumulation_steps: 32
logging_dir: ./vdl_log_sft_full_tp_8B_coco
output_dir: ./checkpoints/qwen3-vl-sft-full-tp-8B_coco
disable_tqdm: true
Expand All @@ -44,7 +44,7 @@ warmup_steps: 20
learning_rate: 1.0e-5

# performance
tensor_model_parallel_size: 4
tensor_model_parallel_size: 2
pipeline_model_parallel_size: 1
sharding: stage1
recompute_granularity: full
Expand All @@ -57,4 +57,6 @@ save_checkpoint_format: "flex_checkpoint"
load_checkpoint_format: "flex_checkpoint"
freeze_config: freeze_vision freeze_aligner

benchmark: true
benchmark: true
dataloader_num_workers: 8
prefetch_factor: 8