mirror of
https://github.com/volcengine/verl.git
synced 2025-10-20 13:43:50 +08:00
[BREAKING] [rollout] chore: remove default rollout selection (#2757)
### What does this PR do? As title ### Checklist Before Starting - [ ] Search for similar PRs. Paste at least one query link here: ... - [ ] Format the PR title as `[{modules}] {type}: {description}` (This will be checked by the CI) - `{modules}` include `fsdp`, `megatron`, `sglang`, `vllm`, `rollout`, `trainer`, `ci`, `training_utils`, `recipe`, `hardware`, `deployment`, `ray`, `worker`, `single_controller`, `misc`, `perf`, `model`, `algo`, `env`, `tool`, `ckpt`, `doc`, `data` - If this PR involves multiple modules, separate them with `,` like `[megatron, fsdp, doc]` - `{type}` is in `feat`, `fix`, `refactor`, `chore`, `test` - If this PR breaks any API (CLI arguments, config, function signature, etc.), add `[BREAKING]` to the beginning of the title. - Example: `[BREAKING][fsdp, megatron] feat: dynamic batching` ### Test > For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc. ### API and Usage Example > Demonstrate how the API changes if any, and provide usage example(s) if possible. ```python # Add code snippet or script demonstrating how to use this ``` ### Design & Code Changes > Demonstrate the high-level design if this PR is complex, and list the specific changes. ### Checklist Before Submitting > [!IMPORTANT] > Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review. - [ ] Read the [Contribute Guide](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md). - [ ] Apply [pre-commit checks](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md#code-linting-and-formatting): `pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always` - [ ] Add / Update [the documentation](https://github.com/volcengine/verl/tree/main/docs). - [ ] Add unit or end-to-end test(s) to [the CI workflow](https://github.com/volcengine/verl/tree/main/.github/workflows) to cover all the code. If not feasible, explain why: ... - [ ] Once your PR is ready for CI, send a message in [the `ci-request` channel](https://verl-project.slack.com/archives/C091TCESWB1) in [the `verl` Slack workspace](https://join.slack.com/t/verl-project/shared_invite/zt-3855yhg8g-CTkqXu~hKojPCmo7k_yXTQ). (If not accessible, please try [the Feishu group (飞书群)](https://applink.larkoffice.com/client/chat/chatter/add_by_link?link_token=772jd4f1-cd91-441e-a820-498c6614126a).)
This commit is contained in:
@ -111,6 +111,7 @@ python3 -m verl.trainer.main_ppo \
|
|||||||
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
||||||
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
||||||
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.ref.megatron.pipeline_model_parallel_size=${train_pp} \
|
actor_rollout_ref.ref.megatron.pipeline_model_parallel_size=${train_pp} \
|
||||||
actor_rollout_ref.ref.megatron.tensor_model_parallel_size=${train_tp} \
|
actor_rollout_ref.ref.megatron.tensor_model_parallel_size=${train_tp} \
|
||||||
actor_rollout_ref.ref.megatron.expert_model_parallel_size=${train_ep} \
|
actor_rollout_ref.ref.megatron.expert_model_parallel_size=${train_ep} \
|
||||||
|
@ -108,6 +108,7 @@ ray job submit --no-wait --runtime-env="${RUNTIME_ENV}" \
|
|||||||
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
||||||
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
||||||
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.ref.fsdp_config.param_offload=${offload} \
|
actor_rollout_ref.ref.fsdp_config.param_offload=${offload} \
|
||||||
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
||||||
actor_rollout_ref.actor.fsdp_config.fsdp_size=-1 \
|
actor_rollout_ref.actor.fsdp_config.fsdp_size=-1 \
|
||||||
|
@ -110,6 +110,7 @@ ray job submit --no-wait --runtime-env="${RUNTIME_ENV}" \
|
|||||||
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
||||||
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
||||||
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.ref.fsdp_config.param_offload=${offload} \
|
actor_rollout_ref.ref.fsdp_config.param_offload=${offload} \
|
||||||
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
||||||
actor_rollout_ref.actor.fsdp_config.fsdp_size=-1 \
|
actor_rollout_ref.actor.fsdp_config.fsdp_size=-1 \
|
||||||
|
@ -105,6 +105,7 @@ ray job submit --no-wait --runtime-env="${RUNTIME_ENV}" \
|
|||||||
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
||||||
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
||||||
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.ref.fsdp_config.param_offload=${offload} \
|
actor_rollout_ref.ref.fsdp_config.param_offload=${offload} \
|
||||||
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
||||||
actor_rollout_ref.actor.fsdp_config.fsdp_size=-1 \
|
actor_rollout_ref.actor.fsdp_config.fsdp_size=-1 \
|
||||||
|
@ -85,6 +85,7 @@ ray job submit --no-wait --runtime-env="${RUNTIME_ENV}" \
|
|||||||
actor_rollout_ref.rollout.log_prob_max_token_len_per_gpu=$((max_prompt_length + max_response_length)) \
|
actor_rollout_ref.rollout.log_prob_max_token_len_per_gpu=$((max_prompt_length + max_response_length)) \
|
||||||
actor_rollout_ref.model.path="${MODEL_PATH}" \
|
actor_rollout_ref.model.path="${MODEL_PATH}" \
|
||||||
actor_rollout_ref.model.enable_gradient_checkpointing=True \
|
actor_rollout_ref.model.enable_gradient_checkpointing=True \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.actor.optim.lr=1e-6 \
|
actor_rollout_ref.actor.optim.lr=1e-6 \
|
||||||
actor_rollout_ref.actor.optim.lr_warmup_steps=10 \
|
actor_rollout_ref.actor.optim.lr_warmup_steps=10 \
|
||||||
actor_rollout_ref.actor.optim.weight_decay=0.1 \
|
actor_rollout_ref.actor.optim.weight_decay=0.1 \
|
||||||
|
@ -82,6 +82,7 @@ python3 -m verl.trainer.main_ppo \
|
|||||||
actor_rollout_ref.actor.ppo_max_token_len_per_gpu=${actor_ppo_max_token_len} \
|
actor_rollout_ref.actor.ppo_max_token_len_per_gpu=${actor_ppo_max_token_len} \
|
||||||
actor_rollout_ref.ref.log_prob_max_token_len_per_gpu=${infer_ppo_max_token_len} \
|
actor_rollout_ref.ref.log_prob_max_token_len_per_gpu=${infer_ppo_max_token_len} \
|
||||||
actor_rollout_ref.rollout.log_prob_max_token_len_per_gpu=${infer_ppo_max_token_len} \
|
actor_rollout_ref.rollout.log_prob_max_token_len_per_gpu=${infer_ppo_max_token_len} \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.model.path="${MODEL_PATH}" \
|
actor_rollout_ref.model.path="${MODEL_PATH}" \
|
||||||
actor_rollout_ref.model.enable_gradient_checkpointing=True \
|
actor_rollout_ref.model.enable_gradient_checkpointing=True \
|
||||||
actor_rollout_ref.actor.optim.lr=1e-6 \
|
actor_rollout_ref.actor.optim.lr=1e-6 \
|
||||||
|
@ -81,6 +81,7 @@ python3 -m verl.trainer.main_ppo \
|
|||||||
actor_rollout_ref.actor.ppo_max_token_len_per_gpu=${actor_ppo_max_token_len} \
|
actor_rollout_ref.actor.ppo_max_token_len_per_gpu=${actor_ppo_max_token_len} \
|
||||||
actor_rollout_ref.ref.log_prob_max_token_len_per_gpu=${infer_ppo_max_token_len} \
|
actor_rollout_ref.ref.log_prob_max_token_len_per_gpu=${infer_ppo_max_token_len} \
|
||||||
actor_rollout_ref.rollout.log_prob_max_token_len_per_gpu=${infer_ppo_max_token_len} \
|
actor_rollout_ref.rollout.log_prob_max_token_len_per_gpu=${infer_ppo_max_token_len} \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.model.path="${MODEL_PATH}" \
|
actor_rollout_ref.model.path="${MODEL_PATH}" \
|
||||||
actor_rollout_ref.model.enable_gradient_checkpointing=True \
|
actor_rollout_ref.model.enable_gradient_checkpointing=True \
|
||||||
actor_rollout_ref.model.lora_rank=8 \
|
actor_rollout_ref.model.lora_rank=8 \
|
||||||
|
@ -108,6 +108,7 @@ python3 -m verl.trainer.main_ppo \
|
|||||||
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
||||||
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
||||||
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.ref.megatron.pipeline_model_parallel_size=${train_pp} \
|
actor_rollout_ref.ref.megatron.pipeline_model_parallel_size=${train_pp} \
|
||||||
actor_rollout_ref.ref.megatron.tensor_model_parallel_size=${train_tp} \
|
actor_rollout_ref.ref.megatron.tensor_model_parallel_size=${train_tp} \
|
||||||
actor_rollout_ref.ref.megatron.param_offload=${offload} \
|
actor_rollout_ref.ref.megatron.param_offload=${offload} \
|
||||||
|
@ -115,6 +115,7 @@ python3 -m verl.trainer.main_ppo \
|
|||||||
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
||||||
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
||||||
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.ref.megatron.pipeline_model_parallel_size=${train_pp} \
|
actor_rollout_ref.ref.megatron.pipeline_model_parallel_size=${train_pp} \
|
||||||
actor_rollout_ref.ref.megatron.tensor_model_parallel_size=${train_tp} \
|
actor_rollout_ref.ref.megatron.tensor_model_parallel_size=${train_tp} \
|
||||||
actor_rollout_ref.ref.megatron.expert_model_parallel_size=${train_ep} \
|
actor_rollout_ref.ref.megatron.expert_model_parallel_size=${train_ep} \
|
||||||
|
@ -102,6 +102,7 @@ python3 -m verl.trainer.main_ppo \
|
|||||||
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
||||||
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
||||||
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.ref.fsdp_config.param_offload=${offload} \
|
actor_rollout_ref.ref.fsdp_config.param_offload=${offload} \
|
||||||
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
||||||
actor_rollout_ref.actor.fsdp_config.fsdp_size=${fsdp_size} \
|
actor_rollout_ref.actor.fsdp_config.fsdp_size=${fsdp_size} \
|
||||||
|
@ -102,6 +102,7 @@ python3 -m verl.trainer.main_ppo \
|
|||||||
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
||||||
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
||||||
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.ref.fsdp_config.param_offload=${offload} \
|
actor_rollout_ref.ref.fsdp_config.param_offload=${offload} \
|
||||||
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
||||||
actor_rollout_ref.actor.fsdp_config.fsdp_size=${fsdp_size} \
|
actor_rollout_ref.actor.fsdp_config.fsdp_size=${fsdp_size} \
|
||||||
|
@ -85,6 +85,7 @@ HYDRA_FULL_ERROR=1 python -m recipe.entropy.main_entropy \
|
|||||||
actor_rollout_ref.actor.policy_loss.clip_cov_ub=${clip_cov_ub} \
|
actor_rollout_ref.actor.policy_loss.clip_cov_ub=${clip_cov_ub} \
|
||||||
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
|
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
|
||||||
actor_rollout_ref.rollout.mode=sync \
|
actor_rollout_ref.rollout.mode=sync \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
algorithm.adv_estimator=${adv_estimator} \
|
algorithm.adv_estimator=${adv_estimator} \
|
||||||
algorithm.use_kl_in_reward=${use_kl_in_reward} \
|
algorithm.use_kl_in_reward=${use_kl_in_reward} \
|
||||||
algorithm.kl_ctrl.kl_coef=${kl_coef} \
|
algorithm.kl_ctrl.kl_coef=${kl_coef} \
|
||||||
|
@ -82,6 +82,7 @@ HYDRA_FULL_ERROR=1 python -m recipe.entropy.main_entropy \
|
|||||||
actor_rollout_ref.actor.policy_loss.ppo_kl_coef=${ppo_kl_coef} \
|
actor_rollout_ref.actor.policy_loss.ppo_kl_coef=${ppo_kl_coef} \
|
||||||
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
|
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
|
||||||
actor_rollout_ref.rollout.mode=sync \
|
actor_rollout_ref.rollout.mode=sync \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
algorithm.adv_estimator=${adv_estimator} \
|
algorithm.adv_estimator=${adv_estimator} \
|
||||||
algorithm.use_kl_in_reward=${use_kl_in_reward} \
|
algorithm.use_kl_in_reward=${use_kl_in_reward} \
|
||||||
algorithm.kl_ctrl.kl_coef=${kl_coef} \
|
algorithm.kl_ctrl.kl_coef=${kl_coef} \
|
||||||
|
@ -81,6 +81,7 @@ HYDRA_FULL_ERROR=1 python -m recipe.entropy.main_entropy \
|
|||||||
actor_rollout_ref.actor.policy_loss.ppo_kl_coef=${ppo_kl_coef} \
|
actor_rollout_ref.actor.policy_loss.ppo_kl_coef=${ppo_kl_coef} \
|
||||||
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
|
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
|
||||||
actor_rollout_ref.rollout.mode=sync \
|
actor_rollout_ref.rollout.mode=sync \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
algorithm.adv_estimator=${adv_estimator} \
|
algorithm.adv_estimator=${adv_estimator} \
|
||||||
algorithm.use_kl_in_reward=${use_kl_in_reward} \
|
algorithm.use_kl_in_reward=${use_kl_in_reward} \
|
||||||
algorithm.kl_ctrl.kl_coef=${kl_coef} \
|
algorithm.kl_ctrl.kl_coef=${kl_coef} \
|
||||||
|
@ -85,6 +85,7 @@ HYDRA_FULL_ERROR=1 python -m recipe.entropy.main_entropy \
|
|||||||
actor_rollout_ref.actor.policy_loss.clip_cov_ub=${clip_cov_ub} \
|
actor_rollout_ref.actor.policy_loss.clip_cov_ub=${clip_cov_ub} \
|
||||||
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
|
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
|
||||||
actor_rollout_ref.rollout.mode=sync \
|
actor_rollout_ref.rollout.mode=sync \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
algorithm.adv_estimator=${adv_estimator} \
|
algorithm.adv_estimator=${adv_estimator} \
|
||||||
algorithm.use_kl_in_reward=${use_kl_in_reward} \
|
algorithm.use_kl_in_reward=${use_kl_in_reward} \
|
||||||
algorithm.kl_ctrl.kl_coef=${kl_coef} \
|
algorithm.kl_ctrl.kl_coef=${kl_coef} \
|
||||||
|
@ -81,6 +81,7 @@ HYDRA_FULL_ERROR=1 python -m recipe.entropy.main_entropy \
|
|||||||
actor_rollout_ref.actor.policy_loss.ppo_kl_coef=${ppo_kl_coef} \
|
actor_rollout_ref.actor.policy_loss.ppo_kl_coef=${ppo_kl_coef} \
|
||||||
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
|
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
|
||||||
actor_rollout_ref.rollout.mode=sync \
|
actor_rollout_ref.rollout.mode=sync \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
algorithm.adv_estimator=${adv_estimator} \
|
algorithm.adv_estimator=${adv_estimator} \
|
||||||
algorithm.use_kl_in_reward=${use_kl_in_reward} \
|
algorithm.use_kl_in_reward=${use_kl_in_reward} \
|
||||||
algorithm.kl_ctrl.kl_coef=${kl_coef} \
|
algorithm.kl_ctrl.kl_coef=${kl_coef} \
|
||||||
|
@ -112,6 +112,7 @@ python3 -m recipe.one_step_off_policy.main_ppo \
|
|||||||
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
||||||
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
||||||
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.ref.fsdp_config.param_offload=${ref_offload} \
|
actor_rollout_ref.ref.fsdp_config.param_offload=${ref_offload} \
|
||||||
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
||||||
actor_rollout_ref.actor.fsdp_config.fsdp_size=${fsdp_size} \
|
actor_rollout_ref.actor.fsdp_config.fsdp_size=${fsdp_size} \
|
||||||
|
@ -107,6 +107,7 @@ python3 -m verl.trainer.main_ppo \
|
|||||||
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
||||||
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
||||||
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.ref.fsdp_config.param_offload=${offload} \
|
actor_rollout_ref.ref.fsdp_config.param_offload=${offload} \
|
||||||
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
actor_rollout_ref.ref.ulysses_sequence_parallel_size=${sp_size} \
|
||||||
actor_rollout_ref.actor.fsdp_config.fsdp_size=${fsdp_size} \
|
actor_rollout_ref.actor.fsdp_config.fsdp_size=${fsdp_size} \
|
||||||
|
@ -119,6 +119,7 @@ python3 -m recipe.one_step_off_policy.main_ppo \
|
|||||||
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
||||||
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
||||||
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.ref.megatron.pipeline_model_parallel_size=${train_pp} \
|
actor_rollout_ref.ref.megatron.pipeline_model_parallel_size=${train_pp} \
|
||||||
actor_rollout_ref.ref.megatron.tensor_model_parallel_size=${train_tp} \
|
actor_rollout_ref.ref.megatron.tensor_model_parallel_size=${train_tp} \
|
||||||
actor_rollout_ref.ref.megatron.param_offload=${ref_offload} \
|
actor_rollout_ref.ref.megatron.param_offload=${ref_offload} \
|
||||||
|
@ -113,6 +113,7 @@ python3 -m verl.trainer.main_ppo \
|
|||||||
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
actor_rollout_ref.rollout.val_kwargs.top_k=${top_k} \
|
||||||
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
actor_rollout_ref.rollout.val_kwargs.do_sample=True \
|
||||||
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
actor_rollout_ref.rollout.val_kwargs.n=1 \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
actor_rollout_ref.ref.megatron.pipeline_model_parallel_size=${train_pp} \
|
actor_rollout_ref.ref.megatron.pipeline_model_parallel_size=${train_pp} \
|
||||||
actor_rollout_ref.ref.megatron.tensor_model_parallel_size=${train_tp} \
|
actor_rollout_ref.ref.megatron.tensor_model_parallel_size=${train_tp} \
|
||||||
actor_rollout_ref.ref.megatron.param_offload=${offload} \
|
actor_rollout_ref.ref.megatron.param_offload=${offload} \
|
||||||
|
@ -89,6 +89,7 @@ common_params=(
|
|||||||
actor_rollout_ref.rollout.val_kwargs.do_sample=True
|
actor_rollout_ref.rollout.val_kwargs.do_sample=True
|
||||||
actor_rollout_ref.rollout.val_kwargs.n=1
|
actor_rollout_ref.rollout.val_kwargs.n=1
|
||||||
actor_rollout_ref.rollout.enable_chunked_prefill=True \
|
actor_rollout_ref.rollout.enable_chunked_prefill=True \
|
||||||
|
actor_rollout_ref.rollout.name=vllm \
|
||||||
reward_model.reward_manager=dapo
|
reward_model.reward_manager=dapo
|
||||||
+reward_model.reward_kwargs.overlong_buffer_cfg.enable=${enable_overlong_buffer}
|
+reward_model.reward_kwargs.overlong_buffer_cfg.enable=${enable_overlong_buffer}
|
||||||
+reward_model.reward_kwargs.overlong_buffer_cfg.len=${overlong_buffer_len}
|
+reward_model.reward_kwargs.overlong_buffer_cfg.len=${overlong_buffer_len}
|
||||||
|
@ -121,7 +121,7 @@ actor_rollout_ref:
|
|||||||
save_path: null
|
save_path: null
|
||||||
load_weight: true
|
load_weight: true
|
||||||
rollout:
|
rollout:
|
||||||
name: vllm
|
name: ???
|
||||||
mode: sync
|
mode: sync
|
||||||
temperature: 1.0
|
temperature: 1.0
|
||||||
top_k: -1
|
top_k: -1
|
||||||
|
@ -84,7 +84,7 @@ actor_rollout_ref:
|
|||||||
entropy_from_logits_with_chunking: false
|
entropy_from_logits_with_chunking: false
|
||||||
entropy_checkpointing: false
|
entropy_checkpointing: false
|
||||||
rollout:
|
rollout:
|
||||||
name: vllm
|
name: ???
|
||||||
mode: sync
|
mode: sync
|
||||||
temperature: 1.0
|
temperature: 1.0
|
||||||
top_k: -1
|
top_k: -1
|
||||||
|
@ -1,5 +1,5 @@
|
|||||||
# actor_rollout_ref.rollout.name: hf/vllm/sglang. The default value will be removed in the future
|
# actor_rollout_ref.rollout.name: hf/vllm/sglang. The default value will be removed in the future
|
||||||
name: vllm
|
name: ???
|
||||||
|
|
||||||
# sync: LLM, async: AsyncLLM
|
# sync: LLM, async: AsyncLLM
|
||||||
mode: sync
|
mode: sync
|
||||||
|
Reference in New Issue
Block a user