llama-lang-adapt

/

me5-instruct-afri-HQ-hn-m3-save-steps-500-high-threshold-prompt-qe-pos-only-5000

like 0

Follow

Llama Adapt 11

• "numpy.core.multiarray._reconstruct",
• "numpy.dtype",
• "collections.OrderedDict",
• "numpy.ndarray",
• "torch.ByteStorage",
• "_codecs.encode",
• "torch._utils._rebuild_tensor_v2"

How to fix it?

• "accelerate.state.PartialState",
• "transformers.trainer_utils.IntervalStrategy",
• "FlagEmbedding.finetune.embedder.encoder_only.m3.arguments.EncoderOnlyEmbedderM3TrainingArguments",
• "transformers.trainer_utils.SchedulerType",
• "transformers.trainer_utils.HubStrategy",
• "transformers.training_args.OptimizerNames",
• "torch.device",
• "transformers.trainer_pt_utils.AcceleratorConfig",
• "accelerate.utils.dataclasses.DistributedType"

How to fix it?