HorHang

I encountered the same problems. Adjusting GPU utilization, i.e. `actor_rollout_ref.rollout.gpu_memory_utilization=0.8`, solved my case.

> > Some tips (might be helpful) > > > > 1. Decrease `actor_rollout_ref.rollout.n` > > 2. Ensure the setting `export VLLM_ATTENTION_BACKEND=XFORMERS` > > 3. Decrease `actor_rollout_ref.actor.ppo_micro_batch_size` > > 4....

> We have already implemented this feature, please check `reward_model.launch_reward_fn_async=True` argument Beside this `reward_model.launch_reward_fn_async=True` setting, do we need to self define or ray decoration for `compute_score` function? Looking at `nvtop`...

> > **Description:** I am encountering an issue while fine-tuning `qwen2.5-7b-instruct` using Unsloth. The training process frequently runs into a `ConstraintViolationError`, seemingly due to issues with tensor shapes during loss...

Another thing, how could we `dojo.run_tac()` on multiple tactics or single line contain `;`? Some seems to work while other doesn't. Example: 1. theorem `Finset.map_nsmul_piAntidiag` can run tactic `\n classical...

> The code example is not runnable. I can take a look if you have a minimal runnable example to reproduce the error. > > Usually the "Cannot find the...

> Re > > > Another thing, how could we `dojo.run_tac()` on multiple tactics or single line contain `;`? Some seems to work while other doesn't. > > Example: >...