nvidia

/

Nemotron-4-340B-Reward

like 113

Follow

NVIDIA 6.71k

Thanks for sharing such a great reward model!

I am trying to reproduce the result on RewardBench and serve this checkpoint with the official docker built from the Dockerfile:

nemo_aligner                  0.4.0.dev0
nemo_toolkit                  2.0.0rc0
megatron_core                 0.7.0
transformer-engine            1.7.0.dev0+a51ff54
transformers                  4.40.2

However, I run into this checkpoint missing error:

load
    sharded_objects, sharded_state_dict = load_sharded_objects(
  File "/home/data/lukeming.lkm/NeMo-Aligner/build_env/Megatron-LM/megatron/core/dist_checkpointing/serialization.py", line 227, in load_sharded_objects
    return dict_list_map_inplace(load_sharded_object, sharded_objects), sharded_state_dict
  File "/home/data/lukeming.lkm/NeMo-Aligner/build_env/Megatron-LM/megatron/core/dist_checkpointing/dict_utils.py", line 180, in dict_list_map_inplace
    x[k] = dict_list_map_inplace(f, v)
  File "/home/data/lukeming.lkm/NeMo-Aligner/build_env/Megatron-LM/megatron/core/dist_checkpointing/dict_utils.py", line 180, in dict_list_map_inplace
    x[k] = dict_list_map_inplace(f, v)
  File "/home/data/lukeming.lkm/NeMo-Aligner/build_env/Megatron-LM/megatron/core/dist_checkpointing/dict_utils.py", line 184, in dict_list_map_inplace
    return f(x)
  File "/home/data/lukeming.lkm/NeMo-Aligner/build_env/Megatron-LM/megatron/core/dist_checkpointing/serialization.py", line 224, in load_sharded_object
    raise CheckpointingException(err_msg) from e
megatron.core.dist_checkpointing.core.CheckpointingException: Object shard /cpfs_01/296bu4pgyxye1ubw3fj/data/shared/Group-m6/lukeming.lkm/ckpts/public/Nemotron-4-340B-Reward/model_weights/model.rm_head._extra_state/shard_0_1.pt not found

It looks like there is no model.rm_head._extra_state part in the checkpoint.