unslothai/unsloth

[Bug] Error Patching SFTTrainer

Open

#2,597 opened on May 21, 2025

View on GitHub
 (4 comments) (0 reactions) (0 assignees)Python (64,271 stars) (5,658 forks)batch import
help wantedinactive

Description

Platform: RunPod GPU: 1x A100

Steps to Reproduce

from unsloth import FastLanguageModel, GRPOTrainer, QLoRAConfig
from datasets import load_dataset
import torch
Observed Error
Flash Attention 2 issues detected, tried to fall back to Xformers. Then encountered:
RuntimeError: Unsloth: Please file a bug report! Error patching SFTTrainer

Full Error Traceback

---------------------------------------------------------------------------
SyntaxError                               Traceback (most recent call last)
File ~/miniconda3/envs/py3.11/lib/python3.11/site-packages/unsloth/tokenizer_utils.py:1037
   1036 try:
-> 1037     exec(trainer_text, globals())
   1038 except:

SyntaxError: invalid syntax (<string>, line 4)

During handling of the above exception, another exception occurred:

RuntimeError                              Traceback (most recent call last)
Cell In[1], line 6
      3 get_ipython().system('apt-get update && apt-get install -y iverilog')
      5 # Import libraries
----> 6 from unsloth import FastLanguageModel, GRPOTrainer, QLoRAConfig
      7 from datasets import load_dataset
      8 import torch

File ~/miniconda3/envs/py3.11/lib/python3.11/site-packages/unsloth/__init__.py:174
    171     raise ImportError("Unsloth: Please install unsloth_zoo via `pip install unsloth-zoo`")
    172 pass
--> 174 from .models import *
    175 from .save import *
    176 from .chat_templates import *

File ~/miniconda3/envs/py3.11/lib/python3.11/site-packages/unsloth/models/__init__.py:16
      1 # Copyright 2023-present Daniel Han-Chen & the Unsloth team. All rights reserved.
      2 #
      3 # Licensed under the Apache License, Version 2.0 (the "License");
   (...)     12 # See the License for the specific language governing permissions and
     13 # limitations under the License.
---> 16 from .granite import FastGraniteModel
     17 from .loader  import FastLanguageModel, FastVisionModel
     18 from .llama   import FastLlamaModel

File ~/miniconda3/envs/py3.11/lib/python3.11/site-packages/unsloth/models/granite.py:15
      1 # Copyright 2023-present Daniel Han-Chen & the Unsloth team. All rights reserved.
      2 #
      3 # Licensed under the Apache License, Version 2.0 (the "License");
   (...)     12 # See the License for the specific language governing permissions and
     13 # limitations under the License.
---> 15 from .llama import *
     16 import os
     17 from ._utils import __version__

File ~/miniconda3/envs/py3.11/lib/python3.11/site-packages/unsloth/models/llama.py:32
     28 from transformers.modeling_attn_mask_utils import (
     29     _prepare_4d_causal_attention_mask_for_sdpa,
     30 )
     31 from ..kernels import *
---> 32 from ..tokenizer_utils import *
     33 if HAS_FLASH_ATTENTION:
     34     from flash_attn import flash_attn_func

File ~/miniconda3/envs/py3.11/lib/python3.11/site-packages/unsloth/tokenizer_utils.py:1039
   1037         exec(trainer_text, globals())
   1038     except:
-> 1039         raise RuntimeError(f"Unsloth: Please file a bug report! Error patching {trainer_name}")
   1040     exec(f"trl.trainer.{trainer_name} = Unsloth{trainer_name}", globals())
   1041 pass

RuntimeError: Unsloth: Please file a bug report! Error patching SFTTrainer

Unsloth Output

🦥 Unsloth: Will patch your computer to enable 2x faster free finetuning.
Unsloth: Your Flash Attention 2 installation seems to be broken?
A possible explanation is you have a new CUDA version which isn't
yet compatible with FA2? Please file a ticket to Unsloth or FA2.
We shall now use Xformers instead, which does not have any performance hits!
We found this negligible impact by benchmarking on 1x A100.
🦥 Unsloth Zoo will now patch everything to make training faster!

Additional Information

  • Using GRPOTrainer module for training
  • The error occurs at the import stage

Contributor guide