Skip to content
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

[misc] feat: support offload parameter and optimizer during rollout #284

Merged
merged 4 commits into from
Feb 17, 2025
Merged

[misc] feat: support offload parameter and optimizer during rollout #284

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
1b13d15
fix fsdp actor critic load/offload
PeterSH6 Feb 15, 2025
d6b1f6e
delete grad_offload option
PeterSH6 Feb 15, 2025
1c8dbaa
add offload when rollout
PeterSH6 Feb 15, 2025
ee00b59
lint
PeterSH6 Feb 15, 2025
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
1 change: 0 additions & 1 deletion docs/examples/config.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,7 +93,6 @@ Actor/Rollout/Reference Policy
# transformer_layer_cls_to_wrap: None
min_num_params: 0
param_offload: False
grad_offload: False
optimizer_offload: False
fsdp_size: -1
ref:
Expand Down
2 changes: 0 additions & 2 deletions docs/examples/gsm8k_example.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -139,7 +139,6 @@ The script of run_deepseek7b_llm.sh
actor_rollout_ref.actor.ppo_mini_batch_size=256 \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=16 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=32 \
Expand All @@ -154,7 +153,6 @@ The script of run_deepseek7b_llm.sh
critic.model.enable_gradient_checkpointing=True \
critic.ppo_micro_batch_size_per_gpu=32 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
algorithm.kl_ctrl.kl_coef=0.001 \
trainer.critic_warmup=0 \
Expand Down
1 change: 0 additions & 1 deletion examples/grpo_trainer/run_deepseek7b_llm.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.kl_loss_type=low_var_kl \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=160 \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
Expand Down
1 change: 0 additions & 1 deletion examples/grpo_trainer/run_deepseek7b_llm_seq_balance.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.kl_loss_type=low_var_kl \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
actor_rollout_ref.rollout.name=vllm \
Expand Down
1 change: 0 additions & 1 deletion examples/grpo_trainer/run_qwen2-7b.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.kl_loss_type=low_var_kl \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=160 \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
Expand Down
1 change: 0 additions & 1 deletion examples/grpo_trainer/run_qwen2-7b_seq_balance.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.kl_loss_type=low_var_kl \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
actor_rollout_ref.rollout.name=vllm \
Expand Down
2 changes: 0 additions & 2 deletions examples/ppo_trainer/run_deepseek7b_llm.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ppo_mini_batch_size=256 \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=16 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=32 \
Expand All @@ -28,7 +27,6 @@ python3 -m verl.trainer.main_ppo \
critic.model.enable_gradient_checkpointing=True \
critic.ppo_micro_batch_size_per_gpu=32 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
algorithm.kl_ctrl.kl_coef=0.001 \
trainer.critic_warmup=0 \
Expand Down
2 changes: 0 additions & 2 deletions examples/ppo_trainer/run_deepseek7b_llm_sp2.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ulysses_sequence_parallel_size=2 \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64 \
actor_rollout_ref.rollout.tensor_model_parallel_size=4 \
Expand All @@ -30,7 +29,6 @@ python3 -m verl.trainer.main_ppo \
critic.model.enable_gradient_checkpointing=True \
critic.ppo_micro_batch_size_per_gpu=64 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
algorithm.kl_ctrl.kl_coef=0.001 \
trainer.critic_warmup=0 \
Expand Down
2 changes: 0 additions & 2 deletions examples/ppo_trainer/run_gemma.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ppo_mini_batch_size=128 \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=4 \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
Expand All @@ -27,7 +26,6 @@ python3 -m verl.trainer.main_ppo \
critic.model.enable_gradient_checkpointing=False \
critic.ppo_micro_batch_size_per_gpu=4 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
algorithm.kl_ctrl.kl_coef=0.001 \
trainer.critic_warmup=0 \
Expand Down
2 changes: 0 additions & 2 deletions examples/ppo_trainer/run_qwen2-7b.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=16 \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=32 \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
Expand All @@ -36,7 +35,6 @@ python3 -m verl.trainer.main_ppo \
critic.model.enable_gradient_checkpointing=True \
critic.ppo_micro_batch_size_per_gpu=32 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
algorithm.kl_ctrl.kl_coef=0.001 \
trainer.critic_warmup=0 \
Expand Down
2 changes: 0 additions & 2 deletions examples/ppo_trainer/run_qwen2-7b_rm.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,7 +38,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=16 \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 \
actor_rollout_ref.rollout.tensor_model_parallel_size=1 \
Expand All @@ -53,7 +52,6 @@ python3 -m verl.trainer.main_ppo \
critic.model.enable_gradient_checkpointing=True \
critic.ppo_micro_batch_size_per_gpu=32 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
reward_model.enable=True \
reward_model.model.path="$HOME/models/FsfairX-LLaMA3-RM-v0.1" \
Expand Down
2 changes: 0 additions & 2 deletions examples/ppo_trainer/run_qwen2-7b_rm_seq_balance.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.use_dynamic_bsz=True \
actor_rollout_ref.actor.ppo_max_token_len_per_gpu=24000 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
actor_rollout_ref.rollout.name=vllm \
Expand All @@ -39,7 +38,6 @@ python3 -m verl.trainer.main_ppo \
critic.use_dynamic_bsz=True \
critic.ppo_max_token_len_per_gpu=98304 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
reward_model.enable=True \
reward_model.model.path=sfairXC/FsfairX-LLaMA3-RM-v0.1\
Expand Down
2 changes: 0 additions & 2 deletions examples/ppo_trainer/run_qwen2-7b_seq_balance.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.use_dynamic_bsz=True \
actor_rollout_ref.actor.ppo_max_token_len_per_gpu=24000 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
actor_rollout_ref.rollout.name=vllm \
Expand All @@ -37,7 +36,6 @@ python3 -m verl.trainer.main_ppo \
critic.model.enable_gradient_checkpointing=True \
critic.ppo_max_token_len_per_gpu=98304 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
algorithm.kl_ctrl.kl_coef=0.001 \
trainer.critic_warmup=0 \
Expand Down
2 changes: 0 additions & 2 deletions examples/ppo_trainer/run_qwen2.5-32b.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 \
actor_rollout_ref.rollout.tensor_model_parallel_size=4 \
Expand All @@ -37,7 +36,6 @@ python3 -m verl.trainer.main_ppo \
critic.model.enable_gradient_checkpointing=False \
critic.ppo_micro_batch_size_per_gpu=8 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
algorithm.kl_ctrl.kl_coef=0.0001 \
trainer.critic_warmup=0 \
Expand Down
1 change: 0 additions & 1 deletion examples/remax_trainer/run_qwen2.5-3b_seq_balance.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.kl_loss_type=low_var_kl \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
actor_rollout_ref.rollout.name=vllm \
Expand Down
1 change: 0 additions & 1 deletion examples/remax_trainer/run_qwen2.5-7b_seq_balance.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.kl_loss_type=low_var_kl \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
actor_rollout_ref.rollout.name=vllm \
Expand Down
2 changes: 0 additions & 2 deletions examples/split_placement/config/ppo_trainer_split.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,7 +38,6 @@ actor_rollout_ref:
# transformer_layer_cls_to_wrap: None
min_num_params: 0
param_offload: False
grad_offload: False
optimizer_offload: False
ref:
fsdp_config:
Expand Down Expand Up @@ -88,7 +87,6 @@ critic:
enable_gradient_checkpointing: False
fsdp_config:
param_offload: False
grad_offload: False
optimizer_offload: False
wrap_policy:
# transformer_layer_cls_to_wrap: None
Expand Down
2 changes: 0 additions & 2 deletions examples/split_placement/run_deepseek7b_llm.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,6 @@ python3 main_ppo_split.py \
actor_rollout_ref.actor.ppo_mini_batch_size=256 \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=8 \
actor_rollout_ref.rollout.tensor_model_parallel_size=4 \
Expand All @@ -25,7 +24,6 @@ python3 main_ppo_split.py \
critic.model.enable_gradient_checkpointing=False \
critic.ppo_micro_batch_size_per_gpu=8 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
algorithm.kl_ctrl.kl_coef=0.001 \
trainer.critic_warmup=0 \
Expand Down
2 changes: 0 additions & 2 deletions tests/e2e/run_qwen_gsm8k_function_rm.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ppo_mini_batch_size=256 \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
Expand All @@ -29,7 +28,6 @@ python3 -m verl.trainer.main_ppo \
critic.model.enable_gradient_checkpointing=False \
critic.ppo_micro_batch_size_per_gpu=4 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
algorithm.kl_ctrl.kl_coef=0.001 \
trainer.critic_warmup=0 \
Expand Down
1 change: 0 additions & 1 deletion tests/e2e/run_qwen_gsm8k_function_rm_grpo.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ppo_mini_batch_size=256 \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
Expand Down
2 changes: 0 additions & 2 deletions tests/e2e/run_qwen_gsm8k_function_rm_no_rmpad.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ppo_mini_batch_size=256 \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
Expand All @@ -29,7 +28,6 @@ python3 -m verl.trainer.main_ppo \
critic.model.enable_gradient_checkpointing=False \
critic.ppo_micro_batch_size_per_gpu=4 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
algorithm.kl_ctrl.kl_coef=0.001 \
trainer.critic_warmup=0 \
Expand Down
1 change: 0 additions & 1 deletion tests/e2e/run_qwen_gsm8k_function_rm_remax.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ppo_mini_batch_size=256 \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
Expand Down
2 changes: 0 additions & 2 deletions tests/e2e/run_qwen_gsm8k_model_rm.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ppo_mini_batch_size=256 \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
Expand All @@ -32,7 +31,6 @@ python3 -m verl.trainer.main_ppo \
critic.model.enable_gradient_checkpointing=False \
critic.ppo_micro_batch_size_per_gpu=4 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
reward_model.enable=True \
reward_model.model.path=Qwen/Qwen2.5-0.5B\
Expand Down
2 changes: 0 additions & 2 deletions tests/e2e/run_qwen_gsm8k_model_rm_liger_kernel.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ppo_mini_batch_size=256 \
actor_rollout_ref.actor.ppo_micro_batch_size=32 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size=128 \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
Expand All @@ -33,7 +32,6 @@ python3 -m verl.trainer.main_ppo \
critic.model.enable_gradient_checkpointing=False \
critic.ppo_micro_batch_size=32 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
reward_model.enable=True \
reward_model.model.path=Qwen/Qwen2.5-0.5B\
Expand Down
2 changes: 0 additions & 2 deletions tests/e2e/run_qwen_gsm8k_model_rm_no_rmpad.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ppo_mini_batch_size=256 \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
Expand All @@ -32,7 +31,6 @@ python3 -m verl.trainer.main_ppo \
critic.model.enable_gradient_checkpointing=False \
critic.ppo_micro_batch_size_per_gpu=4 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
reward_model.enable=True \
reward_model.model.path=Qwen/Qwen2.5-0.5B\
Expand Down
2 changes: 0 additions & 2 deletions tests/e2e/run_qwen_gsm8k_model_rm_seq_balance.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.use_dynamic_bsz=True \
actor_rollout_ref.actor.ppo_max_token_len_per_gpu=12000 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
actor_rollout_ref.rollout.name=vllm \
Expand All @@ -34,7 +33,6 @@ python3 -m verl.trainer.main_ppo \
critic.use_dynamic_bsz=True \
critic.ppo_max_token_len_per_gpu=98304 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
reward_model.enable=True \
reward_model.model.path=Qwen/Qwen2.5-0.5B\
Expand Down
2 changes: 0 additions & 2 deletions tests/e2e/run_qwen_gsm8k_model_rm_ulysses.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,6 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
actor_rollout_ref.actor.ulysses_sequence_parallel_size=2 \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.grad_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.actor.fsdp_config.fsdp_size=-1 \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 \
Expand All @@ -35,7 +34,6 @@ python3 -m verl.trainer.main_ppo \
critic.model.enable_gradient_checkpointing=False \
critic.ppo_micro_batch_size_per_gpu=4 \
critic.model.fsdp_config.param_offload=False \
critic.model.fsdp_config.grad_offload=False \
critic.model.fsdp_config.optimizer_offload=False \
critic.model.fsdp_config.fsdp_size=-1 \
reward_model.enable=True \
Expand Down
2 changes: 0 additions & 2 deletions verl/trainer/config/ppo_trainer.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,7 +46,6 @@ actor_rollout_ref:
# transformer_layer_cls_to_wrap: None
min_num_params: 0
param_offload: False
grad_offload: False
optimizer_offload: False
fsdp_size: -1
ref:
Expand Down Expand Up @@ -105,7 +104,6 @@ critic:
use_remove_padding: False
fsdp_config:
param_offload: False
grad_offload: False
optimizer_offload: False
wrap_policy:
# transformer_layer_cls_to_wrap: None
Expand Down
Loading
Loading