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

[IMJONEZZ]

Running the newer GRPO examples unchanged on WSL with Pixi for package management. For anyone unfamiliar with Pixi, it's just uv + conda.

Here's the pixi.toml for all the packages:

[project]
channels = ["https://prefix.dev/conda-forge", "nvidia", "pytorch", "xformers"]
description = "Add a short description here"
name = "Unsloth Demo"
platforms = ["linux-64", "win-64"]
version = "0.1.0"

[tasks]
checkxform = 'python -m xformers.info'
cudacheck = { cmd = 'python -c "import torch; print(torch.cuda.is_available()); print(torch.__version__)"', depends-on = ["checkxform"] }
start_small = { cmd = 'minimal_grpo.py', depends-on = ["cudacheck"] }
start_big = { cmd = 'python llama3_1_\(8b\)_grpo.py', depends-on = ["cudacheck"] }


[system-requirements]
cuda = "12.4"

[dependencies]
python = "3.11.9*"
black = ">=25.1.0,<26"
cuda-version = "==12.4"

[pypi-dependencies]
vllm="==0.7.2"
diffusers="==0.32.2"
peft="==0.14.0"
accelerate="==1.3.0"
bitsandbytes="==0.45.2"
trl = { git = "git+https://github.com/huggingface/trl.git@e95f9fb74a3c3647b86f251b7e230ec51c64b72b" }
xformers = ">=0.0.28.post3, <0.0.30"
torch = { version = "==2.5.1", index = "https://download.pytorch.org/whl/cu124" }
torchvision = { version = "==0.20.1", index = "https://download.pytorch.org/whl/cu124" }
setuptools = ">=75.8.0, <76"
unsloth = ">=2025.2.5, <2026"
unsloth-zoo = ">=2025.2.3, <2026"

Please let me know if there's an issue with one of these packages.

Here's the output of xformers.info and torch.cuda.is_available as well as version:

xFormers 0.0.28.post3                                                                                                                                                                                                                                                                                                        
memory_efficient_attention.ckF:                    unavailable                                                                                                                                                                                                                                                               
memory_efficient_attention.ckB:                    unavailable
memory_efficient_attention.ck_decoderF:            unavailable
memory_efficient_attention.ck_splitKF:             unavailable
memory_efficient_attention.cutlassF-pt:            available
memory_efficient_attention.cutlassB-pt:            available
[email protected]:         available
[email protected]:         available
[email protected]:             unavailable
[email protected]:             unavailable
memory_efficient_attention.triton_splitKF:         available
indexing.scaled_index_addF:                        available
indexing.scaled_index_addB:                        available
indexing.index_select:                             available
sequence_parallel_fused.write_values:              available
sequence_parallel_fused.wait_values:               available
sequence_parallel_fused.cuda_memset_32b_async:     available
sp24.sparse24_sparsify_both_ways:                  available
sp24.sparse24_apply:                               available
sp24.sparse24_apply_dense_output:                  available
sp24._sparse24_gemm:                               available
[email protected]:                 available
[email protected]:                        available
swiglu.dual_gemm_silu:                             available
swiglu.gemm_fused_operand_sum:                     available
swiglu.fused.p.cpp:                                available
is_triton_available:                               True
pytorch.version:                                   2.5.1+cu124
pytorch.cuda:                                      available
gpu.compute_capability:                            8.6
gpu.name:                                          NVIDIA GeForce RTX 3090
dcgm_profiler:                                     unavailable
build.info:                                        available
build.cuda_version:                                1201
build.hip_version:                                 None
build.python_version:                              3.11.10
build.torch_version:                               2.5.1+cu121
build.env.TORCH_CUDA_ARCH_LIST:                    6.0+PTX 7.0 7.5 8.0+PTX 9.0a
build.env.PYTORCH_ROCM_ARCH:                       None
build.env.XFORMERS_BUILD_TYPE:                     Release
build.env.XFORMERS_ENABLE_DEBUG_ASSERTIONS:        None
build.env.NVCC_FLAGS:                              -allow-unsupported-compiler
build.env.XFORMERS_PACKAGE_FROM:                   wheel-v0.0.28.post3
build.nvcc_version:                                12.1.66
source.privacy:                                    open source

True
2.5.1+cu124

And finally, here's the output of running the minimal GRPO example:

$ pixi run python minimal_grpo.py
🦥 Unsloth: Will patch your computer to enable 2x faster free finetuning.                                                                                                                                                                                                                                                    
🦥 Unsloth Zoo will now patch everything to make training faster!                                                                                                                                                                                                                                                            
Traceback (most recent call last):
  File "/mnt/c/Users/me/Unsloth Demo/.pixi/envs/default/lib/python3.11/site-packages/unsloth/tokenizer_utils.py", line 1061, in <module>
    exec(trainer_text, globals())
  File "<string>", line 4
    model = <class 'inspect._empty'>,
            ^
SyntaxError: invalid syntax

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/mnt/c/Users/me/Unsloth Demo/minimal_grpo.py", line 1, in <module>
    from unsloth import FastLanguageModel, PatchFastRL
  File "/mnt/c/Users/me/Unsloth Demo/.pixi/envs/default/lib/python3.11/site-packages/unsloth/__init__.py", line 212, in <module>
    from .models import *
  File "/mnt/c/Users/me/Unsloth Demo/.pixi/envs/default/lib/python3.11/site-packages/unsloth/models/__init__.py", line 16, in <module>
    from .granite import FastGraniteModel
  File "/mnt/c/Users/me/Unsloth Demo/.pixi/envs/default/lib/python3.11/site-packages/unsloth/models/granite.py", line 15, in <module>
    from .llama import *
  File "/mnt/c/Users/me/Unsloth Demo/.pixi/envs/default/lib/python3.11/site-packages/unsloth/models/llama.py", line 36, in <module>
    from ..tokenizer_utils import *
  File "/mnt/c/Users/me/Unsloth Demo/.pixi/envs/default/lib/python3.11/site-packages/unsloth/tokenizer_utils.py", line 1063, in <module>
    raise RuntimeError(f"Unsloth: Please file a bug report! Error patching {trainer_name}")
RuntimeError: Unsloth: Please file a bug report! Error patching SFTTrainer

Is this because of the WSL mounted file system? Or maybe something else?

[Yazooliu]

I installed trl-0.14.0 to fix . previse version is 0.15.0.dev0
And my other key package and version is:
unsloth 2025.2.4
unsloth_zoo 2025.2.3
torch 2.5.1
torchaudio 2.5.1
torchvision 0.20.1
vllm 0.7.2
xformers 0.0.28.post3
xgrammar 0.1.11

[shimmyshimmer]

Thanks we'll investigate

[danielhanchen]

Oh yep working on multiple fixes to resolve this!

[mbx10br]

I installed trl-0.14.0 to fix . previse version is 0.15.0.dev0 And my other key package and version is: unsloth 2025.2.4 unsloth_zoo 2025.2.3 torch 2.5.1 torchaudio 2.5.1 torchvision 0.20.1 vllm 0.7.2 xformers 0.0.28.post3 xgrammar 0.1.11

TY, it worked !

[ZILECAO]

I installed trl-0.14.0 to fix . previse version is 0.15.0.dev0 And my other key package and version is: unsloth 2025.2.4 unsloth_zoo 2025.2.3 torch 2.5.1 torchaudio 2.5.1 torchvision 0.20.1 vllm 0.7.2 xformers 0.0.28.post3 xgrammar 0.1.11

legend, this fixed the dependency nightmare

[kkailaasa]

This issue is also being tracked here #1699 (comment) with a patch pushed, but I think it has opened up another issue