fix megatron config

volcengine · Feb 25, 2025 · b3425b5 · b3425b5
1 parent ceff369
commit b3425b5
Showing 1 changed file with 1 addition and 0 deletions.
diff --git a/verl/trainer/config/ppo_megatron_trainer.yaml b/verl/trainer/config/ppo_megatron_trainer.yaml
@@ -69,6 +69,7 @@ actor_rollout_ref:
     load_format: dummy_megatron
     tensor_model_parallel_size: 2
     max_num_batched_tokens: 8192
+    max_model_len: None
     max_num_seqs: 1024
     log_prob_micro_batch_size: null # will be deprecated, use log_prob_micro_batch_size_per_gpu
     log_prob_micro_batch_size_per_gpu: null