rope.frequencies missing exception

[kc334]

I tried to use rope with this command:

dora --verbose run  \
    solver=musicgen/musicgen_base_32khz_test \
    model/lm/model_scale=xsmall \
    conditioner=none \
    transformer_lm.positional_embedding=sin_rope \
    --clear

this is the configuration file musicgen_base_32khz_test:

# @package __global__

# This is the training loop solver
# for the base MusicGen model (text-to-music)
# on monophonic audio sampled at 32 kHz
defaults:
  - musicgen/default
  - /model: lm/musicgen_lm
  # - override /dset: audio/default
  # - override /model/lm/model_scale: xsmall
  - override /dset: audio/example
  - _self_

autocast: true
autocast_dtype: float16

# EnCodec large trained on mono-channel music audio sampled at 32khz
# with a total stride of 640 leading to 50 frames/s.
# rvq.n_q=4, rvq.bins=2048, no quantization dropout
# (transformer_lm card and n_q must be compatible)
compression_model_checkpoint: //pretrained/facebook/encodec_32khz

channels: 1
sample_rate: 32000

deadlock:
  use: true  # deadlock detection

dataset:
  batch_size: 2  # 32 GPUs
  # segment_duration: 1
  sample_on_weight: false  # Uniform sampling all the way
  sample_on_duration: false  # Uniform sampling all the way
  valid:
    num_samples: 1
  generate:
    num_samples: 1
  evaluate:
    num_samples: 0
  num_workers: 2


evaluate:
  metrics:
    kld: false

metrics:
  kld:
    use_gt: false
    model: passt
    passt:
      pretrained_length: 20

generate:
  lm:
    use_sampling: true
    top_k: 250
    top_p: 0.0
  

optim:
  epochs: 1
  updates_per_epoch: 1
  optimizer: dadam
  lr: 1e-6
  ema:
    use: true
    updates: 10
    device: cpu

logging:
  log_tensorboard: true

schedule:
  lr_scheduler: cosine
  cosine:
    warmup: 100
    lr_min_ratio: 0.0
    cycle_length: 1.0

tensorboard:
  name: debug_test

and got following exception after finishing training epoch

Exception has occurred: RuntimeError       (note: full exception trace is shown but execution is paused at: _run_module_as_main)
Error(s) in loading state_dict for LMModel:
	Missing key(s) in state_dict: "transformer.layers.0.self_attn.rope.frequencies", "transformer.layers.1.self_attn.rope.frequencies". 

I guess a similar issue could occur after turning xpos on.