●Change8

Migrating to Unsloth December-2025

Version December-2025 introduces 5 breaking changes. This guide details how to update your code.

Released: 12/18/2025

5
Breaking Changes
5
Migration Steps
11
Affected Symbols

⚠️ Check Your Code

If you use any of these symbols, you need to read this guide:

llama tokenizerQwen3MoeForCausalLMmodel.generateTorchAOQATDeepSeek OCRmodel.for_training()vllmGRPO trainerinit_empty_weightsunsloth_tiled_mlp

Breaking Changes

●Issue #1

The internal parameter `rope_theta` has been changed to `rope_parameters['rope_theta']`.

✓Solution

Update code referencing `rope_theta` to use `rope_parameters['rope_theta']`.

●Issue #2

The argument `unsloth_tiled_mlp` must now be passed as a `from_pretrained` argument.

✓Solution

Ensure `unsloth_tiled_mlp` is passed correctly during model loading via `from_pretrained`.

●Issue #3

The `load_in_fp8` kwarg is now prevented from reaching the `Qwen3MoeForCausalLM` constructor.

✓Solution

If you were explicitly passing `load_in_fp8` to this constructor, you may need to adjust how you initialize the model, though this change is likely a fix for internal incompatibility.

●Issue #4

The `reload_weights` rpc call has been removed from the GRPO trainer.

✓Solution

No action required unless you were relying on this specific internal RPC call.

●Issue #5

The `include_buffers` argument was removed from `init_empty_weights` in `unsloth_zoo`.

✓Solution

Remove `include_buffers` argument when calling `init_empty_weights`.

Migration Steps

  1. 1
    Update Unsloth and Docker to the latest versions.
  2. 2
    To upgrade Unsloth: `pip install --upgrade --force-reinstall --no-cache-dir --no-deps unsloth unsloth_zoo`
  3. 3
    If you specifically want PyTorch 2.9: `pip install --upgrade unsloth unsloth_zoo`
  4. 4
    If you encounter issues related to `rope_theta`, change usages to `rope_parameters['rope_theta']`.
  5. 5
    Ensure `unsloth_tiled_mlp` is passed as an argument to `from_pretrained`.

Release Summary

This release introduces massive performance gains with 3x faster training via new Triton kernels and enables 500K context length fine-tuning. It also adds support for Transformers v5, preliminary multi-GPU training, and several new model guides.

Need More Details?

View the full release notes and all changes for Unsloth December-2025.

View Full Changelog