fix: nightlies using v1 can't use model_save_format=safetensors (#1226)

Signed-off-by: Terry Kong <terryk@nvidia.com>

Author: terrykong

examples/configs/recipes/llm/grpo-deepscaler-1.5b-8K.yaml

@@ -8,6 +8,7 @@ loss_fn:
   reference_policy_kl_penalty: 0.0
 checkpointing:
   keep_top_k: 10
+  model_save_format: null
 policy:
   model_name: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
   train_global_batch_size: 64

examples/configs/recipes/llm/grpo-gemma3-27b-it-8n8g-fsdp2tp8-actckpt-long.yaml

@@ -5,6 +5,7 @@ grpo:
   max_num_steps: 20
 checkpointing:
   checkpoint_dir: results/grpo-gemma3-27b-it-8n8g-fsdp2tp8sp-actckpt-long
+  model_save_format: null
 policy:
   model_name: google/gemma-3-27b-it
   tokenizer:

examples/configs/recipes/llm/grpo-gspo-deepscaler-1.5b-8K.yaml

@@ -10,6 +10,7 @@ loss_fn:
   token_level_loss: false
 checkpointing:
   keep_top_k: 10
+  model_save_format: null
 policy:
   model_name: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
   train_global_batch_size: 64