philschmid

/

flan-t5-xxl-sharded-fp16

like 52

Hello, I'm trying to load the following model using 8 bit precision

from transformers import AutoModelForSeq2SeqLM

model = AutoModelForSeq2SeqLM.from_pretrained("philschmid/flan-t5-xxl-sharded-fp16", load_in_8bit=True,device_map='auto')

but whenever I try loading using load_in_8bit I run into the traceback below. I'm currently using a ml.p3.2xlarge instance via Sagemaker. Would greatly appreciate any help on this @philschmid

ValueError                                Traceback (most recent call last)
Cell In[134], line 3
      1 from transformers import AutoModelForSeq2SeqLM
----> 3 model = AutoModelForSeq2SeqLM.from_pretrained("philschmid/flan-t5-xxl-sharded-fp16", load_in_8bit=True,device_map='auto')

File ~/anaconda3/envs/python3/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py:471, in _BaseAutoModelClass.from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)
    469 elif type(config) in cls._model_mapping.keys():
    470     model_class = _get_model_class(config, cls._model_mapping)
--> 471     return model_class.from_pretrained(
    472         pretrained_model_name_or_path, *model_args, config=config, **hub_kwargs, **kwargs
    473     )
    474 raise ValueError(
    475     f"Unrecognized configuration class {config.__class__} for this kind of AutoModel: {cls.__name__}.\n"
    476     f"Model type should be one of {', '.join(c.__name__ for c in cls._model_mapping.keys())}."
    477 )

File ~/anaconda3/envs/python3/lib/python3.10/site-packages/transformers/modeling_utils.py:2591, in PreTrainedModel.from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)
   2587         device_map_without_lm_head = {
   2588             key: device_map[key] for key in device_map.keys() if key not in modules_to_not_convert
   2589         }
   2590         if "cpu" in device_map_without_lm_head.values() or "disk" in device_map_without_lm_head.values():
-> 2591             raise ValueError(
   2592                 """
   2593                 Some modules are dispatched on the CPU or the disk. Make sure you have enough GPU RAM to fit
   2594                 the quantized model. If you want to dispatch the model on the CPU or the disk while keeping
   2595                 these modules in 32-bit, you need to set `load_in_8bit_fp32_cpu_offload=True` and pass a custom
   2596                 `device_map` to `from_pretrained`. Check
   2597                 https://huggingface.co/docs/transformers/main/en/main_classes/quantization#offload-between-cpu-and-gpu
   2598                 for more details.
   2599                 """
   2600             )
   2601         del device_map_without_lm_head
   2603 if from_tf:

ValueError: 
                        Some modules are dispatched on the CPU or the disk. Make sure you have enough GPU RAM to fit
                        the quantized model. If you want to dispatch the model on the CPU or the disk while keeping
                        these modules in 32-bit, you need to set `load_in_8bit_fp32_cpu_offload=True` and pass a custom
                        `device_map` to `from_pretrained`. Check
                        https://huggingface.co/docs/transformers/main/en/main_classes/quantization#offload-between-cpu-and-gpu
                        for more details.