open-thought/reasoning-gym
1
0
Fork 0
mirror of https://github.com/open-thought/reasoning-gym.git synced 2026-05-03 17:53:26 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

cfg

Browse source
This commit is contained in:
Oliver 2025-04-28 23:17:41 +01:00
parent 82ac4b27ba
commit f404b2e603
1 changed files with 4 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

8
training/configs/external_generalisation/math_curriculum_qwen_7b.yaml
View file

@ -64,8 +64,8 @@ data:
prompt_key: prompt prompt_key: prompt
max_prompt_length: 4096 max_prompt_length: 4096
max_response_length: 2048 max_response_length: 2048
train_batch_size: 32 train_batch_size: 16
val_batch_size: 64 val_batch_size: 8
return_raw_chat: True return_raw_chat: True
return_raw_input_ids: True return_raw_input_ids: True
@ -79,9 +79,9 @@ actor_rollout_ref:
use_remove_padding: True use_remove_padding: True
actor: actor:
strategy: fsdp # This is for backward-compatibility strategy: fsdp # This is for backward-compatibility
ppo_mini_batch_size: 32 ppo_mini_batch_size: 4
ppo_micro_batch_size: null # will be deprecated, use ppo_micro_batch_size_per_gpu ppo_micro_batch_size: null # will be deprecated, use ppo_micro_batch_size_per_gpu
ppo_micro_batch_size_per_gpu: 8 ppo_micro_batch_size_per_gpu: 4
use_dynamic_bsz: False use_dynamic_bsz: False
ppo_max_token_len_per_gpu: 49152 # n * ${data.max_prompt_length} + ${data.max_response_length} ppo_max_token_len_per_gpu: 49152 # n * ${data.max_prompt_length} + ${data.max_response_length}
grad_clip: 1.0 grad_clip: 1.0