[ci] fix: delete ckpts after saving in github runner (#272)

- As titled

.github/workflows/e2e_gsm8k.yml

@@ -46,6 +46,11 @@ jobs:
         run: |
           ray stop --force
           bash tests/e2e/run_qwen_gsm8k_function_rm.sh
+      - name: Running gsm8k e2e without rmpad using function rm and load ckpt from previous step
+        run: |
+          ray stop --force
+          bash tests/e2e/run_qwen_gsm8k_function_rm_no_rmpad.sh
+          rm -rf ~/ckpt/*
       - name: Running gsm8k e2e training tests on 8 L20 GPUs with rmpad using function rm (GRPO)
         run: |
           ray stop --force
@@ -54,10 +59,6 @@ jobs:
         run: |
           ray stop --force
           bash tests/e2e/run_qwen_gsm8k_function_rm_remax.sh
-      - name: Running gsm8k e2e without rmpad using function rm and load ckpt from previous step
-        run: |
-          ray stop --force
-          bash tests/e2e/run_qwen_gsm8k_function_rm_no_rmpad.sh
       - name: Running gsm8k e2e with rmpad using model rm
         run: |
           ray stop --force

.github/workflows/e2e_lora.yml

@@ -45,4 +45,5 @@ jobs:
       - name: Running gsm8k e2e training tests with LoRA
         run: |
           ray stop --force
-          bash tests/sft/run_sft_qwen05_peft.sh 8 $HOME/ckpts/
\ No newline at end of file
+          bash tests/sft/run_sft_qwen05_peft.sh 8 $HOME/ckpts/
+          rm -rf $HOME/ckpts/*
\ No newline at end of file

.github/workflows/e2e_sft.yml

@@ -58,3 +58,4 @@ jobs:
         run: |
           ray stop --force
           bash tests/sft/run_sft_qwen05_sp2_liger.sh 8 $HOME/ckpts/
+          rm -rf $HOME/ckpts/

tests/e2e/run_qwen_gsm8k_function_rm.sh

@@ -39,4 +39,5 @@ python3 -m verl.trainer.main_ppo \
     trainer.n_gpus_per_node=8 \
     trainer.nnodes=1 \
     trainer.save_freq=1 \
+    trainer.default_local_dir=$HOME/ckpt/ \
     trainer.total_training_steps=1 $@

verl/workers/megatron_workers.py

@@ -415,15 +415,15 @@ class ActorRolloutRefWorker(MegatronWorker):
         return output
 
     @register(dispatch_mode=Dispatch.ONE_TO_ALL)
-    def load_checkpoint(self, checkpoint_path):
+    def load_checkpoint(self, checkpoint_path, **kwargs):
         pass
 
     @register(dispatch_mode=Dispatch.ONE_TO_ALL)
-    def load_pretrained_model(self, checkpoint_path):
+    def load_pretrained_model(self, checkpoint_path, **kwargs):
         pass
 
     @register(dispatch_mode=Dispatch.ONE_TO_ALL)
-    def save_checkpoint(self, checkpoint_path):
+    def save_checkpoint(self, checkpoint_path, **kwargs):
         assert self._is_actor
         pass
 
@@ -590,11 +590,11 @@ class CriticWorker(MegatronWorker):
         return output
 
     @register(dispatch_mode=Dispatch.ONE_TO_ALL)
-    def load_checkpoint(self, checkpoint_path):
+    def load_checkpoint(self, checkpoint_path, **kwargs):
         pass
 
     @register(dispatch_mode=Dispatch.ONE_TO_ALL)
-    def save_checkpoint(self, checkpoint_path):
+    def save_checkpoint(self, checkpoint_path, **kwargs):
         pass