ISTA-DASLab

/

Mixtral-8x7B-Instruct-v0_1-AQLM-2Bit-1x16-hf

like 16

something to do with flash-attn
https://github.com/Dao-AILab/flash-attention/issues/784

even w 2.4.2 nogo

Hi. Without the stack trace, I have zero idea what that might mean.

Hi, I'm having the same issue. Here's the stack trace:

ImportError                               Traceback (most recent call last)
File /local_disk0/.ephemeral_nfs/envs/pythonEnv-76870aa0-982b-4bf0-966d-a0e6e689516d/lib/python3.10/site-packages/transformers/utils/import_utils.py:1472, in _LazyModule._get_module(self, module_name)
   1471 try:
-> 1472     return importlib.import_module("." + module_name, self.__name__)
   1473 except Exception as e:

File /usr/lib/python3.10/importlib/__init__.py:126, in import_module(name, package)
    125         level += 1
--> 126 return _bootstrap._gcd_import(name[level:], package, level)

File <frozen importlib._bootstrap>:1050, in _gcd_import(name, package, level)

File <frozen importlib._bootstrap>:1027, in _find_and_load(name, import_)

File <frozen importlib._bootstrap>:1006, in _find_and_load_unlocked(name, import_)

File <frozen importlib._bootstrap>:688, in _load_unlocked(spec)

File <frozen importlib._bootstrap_external>:883, in exec_module(self, module)

File <frozen importlib._bootstrap>:241, in _call_with_frames_removed(f, *args, **kwds)

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-76870aa0-982b-4bf0-966d-a0e6e689516d/lib/python3.10/site-packages/transformers/models/mixtral/modeling_mixtral.py:58
     57 if is_flash_attn_2_available():
---> 58     from flash_attn import flash_attn_func, flash_attn_varlen_func
     59     from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input  # noqa

File /databricks/python/lib/python3.10/site-packages/flash_attn/__init__.py:3
      1 __version__ = "2.3.2"
----> 3 from flash_attn.flash_attn_interface import (
      4     flash_attn_func,
      5     flash_attn_kvpacked_func,
      6     flash_attn_qkvpacked_func,
      7     flash_attn_varlen_func,
      8     flash_attn_varlen_kvpacked_func,
      9     flash_attn_varlen_qkvpacked_func,
     10     flash_attn_with_kvcache,
     11 )

File /databricks/python/lib/python3.10/site-packages/flash_attn/flash_attn_interface.py:8
      6 # isort: off
      7 # We need to import the CUDA kernels after importing torch
----> 8 import flash_attn_2_cuda as flash_attn_cuda
     10 # isort: on

ImportError: /databricks/python/lib/python3.10/site-packages/flash_attn_2_cuda.cpython-310-x86_64-linux-gnu.so: undefined symbol: _ZN2at4_ops9_pad_enum4callERKNS_6TensorEN3c108ArrayRefINS5_6SymIntEEElNS5_8optionalIdEE

The above exception was the direct cause of the following exception:

RuntimeError                              Traceback (most recent call last)
File <command-1921129686235558>, line 8
      2 from peft import LoraConfig, prepare_model_for_kbit_training, get_peft_model
      3 from transformers import (
      4     AutoModelForCausalLM,
      5     AutoTokenizer,
      6     TrainingArguments,
      7 )
----> 8 model = AutoModelForCausalLM.from_pretrained(
      9     "ISTA-DASLab/Mixtral-8x7B-Instruct-v0_1-AQLM-2Bit-1x16-hf",
     10     trust_remote_code=True, torch_dtype="auto", device_map="cuda", low_cpu_mem_usage=True
     11 
     12 )
     13 tokenizer = AutoTokenizer.from_pretrained("mistralai/Mixtral-8x7B-v0.1")
     14 tokenizer.pad_token = tokenizer.eos_token

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-76870aa0-982b-4bf0-966d-a0e6e689516d/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py:562, in _BaseAutoModelClass.from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)
    558     return model_class.from_pretrained(
    559         pretrained_model_name_or_path, *model_args, config=config, **hub_kwargs, **kwargs
    560     )
    561 elif type(config) in cls._model_mapping.keys():
--> 562     model_class = _get_model_class(config, cls._model_mapping)
    563     return model_class.from_pretrained(
    564         pretrained_model_name_or_path, *model_args, config=config, **hub_kwargs, **kwargs
    565     )
    566 raise ValueError(
    567     f"Unrecognized configuration class {config.__class__} for this kind of AutoModel: {cls.__name__}.\n"
    568     f"Model type should be one of {', '.join(c.__name__ for c in cls._model_mapping.keys())}."
    569 )

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-76870aa0-982b-4bf0-966d-a0e6e689516d/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py:383, in _get_model_class(config, model_mapping)
    382 def _get_model_class(config, model_mapping):
--> 383     supported_models = model_mapping[type(config)]
    384     if not isinstance(supported_models, (list, tuple)):
    385         return supported_models

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-76870aa0-982b-4bf0-966d-a0e6e689516d/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py:734, in _LazyAutoMapping.__getitem__(self, key)
    732 if model_type in self._model_mapping:
    733     model_name = self._model_mapping[model_type]
--> 734     return self._load_attr_from_module(model_type, model_name)
    736 # Maybe there was several model types associated with this config.
    737 model_types = [k for k, v in self._config_mapping.items() if v == key.__name__]

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-76870aa0-982b-4bf0-966d-a0e6e689516d/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py:748, in _LazyAutoMapping._load_attr_from_module(self, model_type, attr)
    746 if module_name not in self._modules:
    747     self._modules[module_name] = importlib.import_module(f".{module_name}", "transformers.models")
--> 748 return getattribute_from_module(self._modules[module_name], attr)

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-76870aa0-982b-4bf0-966d-a0e6e689516d/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py:692, in getattribute_from_module(module, attr)
    690 if isinstance(attr, tuple):
    691     return tuple(getattribute_from_module(module, a) for a in attr)
--> 692 if hasattr(module, attr):
    693     return getattr(module, attr)
    694 # Some of the mappings have entries model_type -> object of another model type. In that case we try to grab the
    695 # object at the top level.

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-76870aa0-982b-4bf0-966d-a0e6e689516d/lib/python3.10/site-packages/transformers/utils/import_utils.py:1462, in _LazyModule.__getattr__(self, name)
   1460     value = self._get_module(name)
   1461 elif name in self._class_to_module.keys():
-> 1462     module = self._get_module(self._class_to_module[name])
   1463     value = getattr(module, name)
   1464 else:

File /local_disk0/.ephemeral_nfs/envs/pythonEnv-76870aa0-982b-4bf0-966d-a0e6e689516d/lib/python3.10/site-packages/transformers/utils/import_utils.py:1474, in _LazyModule._get_module(self, module_name)
   1472     return importlib.import_module("." + module_name, self.__name__)
   1473 except Exception as e:
-> 1474     raise RuntimeError(
   1475         f"Failed to import {self.__name__}.{module_name} because of the following error (look up to see its"
   1476         f" traceback):\n{e}"
   1477     ) from e

RuntimeError: Failed to import transformers.models.mixtral.modeling_mixtral because of the following error (look up to see its traceback):
/databricks/python/lib/python3.10/site-packages/flash_attn_2_cuda.cpython-310-x86_64-linux-gnu.so: undefined symbol: _ZN2at4_ops9_pad_enum4callERKNS_6TensorEN3c108ArrayRefINS5_6SymIntEEElNS5_8optionalIdEE

@amrothemich Hi!
It's obviously a flash_attn_2 import error. Looks like a CUDA version mismatch of sorts.

Thanks! what's weird is I don't encounter it on your other one, BlackSamorez/Mixtral-8x7b-AQLM-2Bit-1x16-hf

Okay it looks like flash attention is not supported by V100s, which is what I'm using. Still confused as to why the other one works though.

It's weird that it's using flash attention at all. SDPA should be the default implementation.

I know, it looks like you've got the configs set up right. 🤷