tanlong Du

icon company 北京邮电大学

Results 7 comments of tanlong Du

Could not override 'actor_rollout_ref.actor.dtype'. To append to your config use +actor_rollout_ref.actor.dtype=float16

我也遇到相同的错误，尝试使用+actor_rollout_ref.actor.fsdp_config.mixed_precision.param_dtype=${dtype} \ 但是发现训练很不稳定

ValueError: No dot product attention backend is available for the provided inputs.

same error

Fail to save megatron distributed checkpoints when using megatron as a back-end, and distributed checkpoint as actor model

我也遇到相同的问题

megatron optimizer precision setting

fsdp上的混合精度是设置是这样的： if mixed_precision_config is not None: param_dtype = PrecisionType.to_dtype(mixed_precision_config.get("param_dtype", "bf16")) reduce_dtype = PrecisionType.to_dtype(mixed_precision_config.get("reduce_dtype", "fp32")) buffer_dtype = PrecisionType.to_dtype(mixed_precision_config.get("buffer_dtype", "fp32")) else: param_dtype = torch.bfloat16 reduce_dtype = torch.float32 buffer_dtype = torch.float32 mixed_precision =...

dapo reward_manager

Thank you for your response. However, I didn't understand why Verl moved this part of the code to this specific path. Was the code in the "recipe" path no longer...

dapo reward_manager

ok，thanks

[Bug] Megatron model merger for Qwen3 MoE models

@vermouth1992 Hello, the saved HF model lacks the weight file. How can this be resolved?