---
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
base_model: dllm-collection/Qwen3-0.6B-diffusion-bd3lm-v0.1
tags:
- diffusion
- dllm
- bd3lm
- distillation
- arxiv:2604.26951
---

<center> <div style="text-align: center;"> <img src="logo.gif" width="300" />
 </div> </center>

# distill-LLaDA2-CALM

This model was introduced in the paper [Turning the TIDE: Cross-Architecture Distillation for Diffusion Large Language Models](https://huggingface.co/papers/2604.26951).

`distill-LLaDA2-CALM` is a 0.6B diffusion language model distilled from LLaDA2.0-mini (16B MoE) into the [`Qwen3-0.6B-diffusion-bd3lm-v0.1`](https://huggingface.co/dllm-hub/Qwen3-0.6B-diffusion-bd3lm-v0.1) student in the **Cross-Tokenizer (Pipeline A)** of the TIDE framework. Forward CALM (chunk-level approximate likelihood matching) baseline.

## Model Overview

- **Method**: TIDE — [Reverse CALM / TIDAL / CompDemo](https://arxiv.org/abs/2604.26951) (cross-architecture distillation for diffusion LMs)
- **Framework**: [TIDE / dLLM](https://github.com/PKU-YuanGroup/TIDE)
- **Student (initialization)**: [`Qwen3-0.6B-diffusion-bd3lm-v0.1`](https://huggingface.co/dllm-hub/Qwen3-0.6B-diffusion-bd3lm-v0.1) (BD3LM, block_size=32)
- **Teacher**: [`inclusionAI/LLaDA2.0-mini`](https://huggingface.co/inclusionAI/LLaDA2.0-mini)
- **Distillation mode**: `--distill_mode alm`
- **Datasets**: [tulu-3-sft-mixture](https://huggingface.co/datasets/allenai/tulu-3-sft-mixture), [smoltalk](https://huggingface.co/datasets/HuggingFaceTB/smoltalk), [opc-sft-stage1](https://huggingface.co/datasets/OpenCoder-LLM/opc-sft-stage1) and [opc-sft-stage2](https://huggingface.co/datasets/OpenCoder-LLM/opc-sft-stage2) — same composition as the [`Qwen3-0.6B-diffusion-bd3lm-v0.1`](https://huggingface.co/dllm-hub/Qwen3-0.6B-diffusion-bd3lm-v0.1) base. Pre-tokenized for this teacher in [`TIDE-dllm/distill_llada2_sft`](https://huggingface.co/datasets/TIDE-dllm/distill_llada2_sft).

## Installation

```shell
pip install torch transformers accelerate
```

## Quick Start

> [!NOTE]
> This checkpoint is fully compatible with the BD3LM `generate(...)` routine published with [`dllm-hub/Qwen3-0.6B-diffusion-bd3lm-v0.1`](https://huggingface.co/dllm-hub/Qwen3-0.6B-diffusion-bd3lm-v0.1) — only the model name changes.

```python
import torch
from transformers import AutoModelForMaskedLM, AutoTokenizer

repo = "TIDE-dllm/distill-LLaDA2-CALM"
device = "cuda" if torch.cuda.is_available() else "cpu"

model = AutoModelForMaskedLM.from_pretrained(
    repo, dtype=torch.bfloat16, trust_remote_code=True,
).to(device).eval()
tokenizer = AutoTokenizer.from_pretrained(repo, trust_remote_code=True)

prompts = [
    [
        {"role": "system", "content": "You are a helpful AI assistant."},
        {"role": "user", "content": "Implement a DFS traversal in Python with clear inline comments."},
    ],
]
encoded = [tokenizer.apply_chat_template(m, add_generation_prompt=True, tokenize=True, enable_thinking=False) for m in prompts]
# ... use the same `generate()` function as in dllm-hub/Qwen3-0.6B-diffusion-bd3lm-v0.1.
```

## Command-Line Interface

For an interactive demo (visualised iterative denoising), use the script in the [TIDE / dLLM repo](https://github.com/PKU-YuanGroup/TIDE):

```shell
python -u examples/a2d/bd3lm/chat.py \
    --model_name_or_path TIDE-dllm/distill-LLaDA2-CALM \
    --chat_template True --block_size 32 --remasking low_confidence \
    --steps 256 --max_new_tokens 256
```

## Reproducing this checkpoint

```shell
git clone https://github.com/PKU-YuanGroup/TIDE && cd TIDE
pip install -e . && git submodule update --init --recursive
pip install -e "lm-evaluation-harness[ifeval,math]" && pip install -e "tokenkit[full]"

# Download the pre-tokenized SFT mixture for this teacher
huggingface-cli download TIDE-dllm/distill_llada2_sft --repo-type dataset \
    --local-dir data/distill_llada2_sft

bash scripts/distill_llada2.sh \
    --data_path data/distill_llada2_sft \
    --distill_mode alm \
    --num_gpus 8
```

## Citation

```bibtex
@misc{zhang2026turningtidecrossarchitecturedistillation,
      title={Turning the TIDE: Cross-Architecture Distillation for Diffusion Large Language Models},
      author={Gongbo Zhang and Wen Wang and Ye Tian and Li Yuan},
      year={2026},
      eprint={2604.26951},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2604.26951},
}
```