diffusion_language_models.md

Diffusion Language Models

Diffusion language models have shown promise for non-autoregressive text generation with parallel decoding capabilities. Unlike auto-regressive language models, different diffusion language models require different decoding strategies.

Example Launch Command

SGLang supports different DLLM algorithms such as LowConfidence, JointThreshold, and FullAttnMultiBlock.

python3 -m sglang.launch_server \
  --model-path inclusionAI/LLaDA2.0-mini \ # example HF/local path
  --dllm-algorithm LowConfidence \
  --dllm-algorithm-config ./config.yaml \ # Optional. Uses the algorithm's default if not set.
  --host 0.0.0.0 \
  --port 30000

Example Configuration File

Depending on the algorithm selected, the configuration parameters vary.

LowConfidence Config:

# Confidence threshold for accepting predicted tokens
# - Higher values: More conservative, better quality but slower
# - Lower values: More aggressive, faster but potentially lower quality
# Range: 0.0 - 1.0
threshold: 0.95

# Default: 32, for LLaDA2MoeModelLM
block_size: 32

JointThreshold Config:

# Decoding threshold for Mask-to-Token (M2T) phase
# - Higher values: More conservative, better quality but slower
# - Lower values: More aggressive, faster but potentially lower quality
# Range: 0.0 - 1.0
threshold: 0.5
# Decoding threshold for Token-to-Token (T2T) phase
# Range: 0.0 - 1.0
# Setting to 0.0 allows full editing (recommended for most cases).
edit_threshold: 0.0
# Max extra T2T steps after all masks are removed. Prevents infinite loops.
max_post_edit_steps: 16
# 2-gram repetition penalty (default 0).
# An empirical value of 3 is often sufficient to mitigate most repetitions.
penalty_lambda: 0

FullAttnMultiBlock Config (for bidirectional models like LLaDA and DREAM):

# Confidence threshold for accepting predicted tokens
# Range: 0.0 - 1.0
threshold: 0.5
# Additional threshold increment per decoding step
block_add: 0.1
# Threshold for considering a token as "decoded"
decoded_thresh: 0.95
# Sub-block size for parallel decoding
sub_block_size: 32
# Number of iterations to delay before caching
cache_delay_iter: 2
# Interval for refreshing the attention cache
refresh_interval: 10000

Example Client Code Snippet

Just like other supported models, diffusion language models can be used via the REST API or Python client.

Python client example for making a generation request to the launched server:

import sglang as sgl

def main():
    llm = sgl.Engine(model_path="inclusionAI/LLaDA2.0-mini",
                     dllm_algorithm="LowConfidence",
                     max_running_requests=1,
                     trust_remote_code=True)

    prompts = [
        "<role>SYSTEM</role>detailed thinking off<|role_end|><role>HUMAN</role> Write a brief introduction of the great wall <|role_end|><role>ASSISTANT</role>"
    ]

    sampling_params = {
        "temperature": 0,
        "max_new_tokens": 1024,
    }

    outputs = llm.generate(prompts, sampling_params)
    print(outputs)

if __name__ == '__main__':
    main()

Curl example for making a generation request to the launched server:

curl -X POST "http://127.0.0.1:30000/generate" \
     -H "Content-Type: application/json" \
     -d '{
        "text": [
            "<role>SYSTEM</role>detailed thinking off<|role_end|><role>HUMAN</role> Write the number from 1 to 128 <|role_end|><role>ASSISTANT</role>",
            "<role>SYSTEM</role>detailed thinking off<|role_end|><role>HUMAN</role> Write a brief introduction of the great wall <|role_end|><role>ASSISTANT</role>"
        ],
        "stream": true,
        "sampling_params": {
            "temperature": 0,
            "max_new_tokens": 1024
        }
    }'

Supported Models

Below the supported models are summarized in a table.

Model Family	Example Model	Algorithm	Description
LLaDA2.0 (mini, flash)	inclusionAI/LLaDA2.0-mini	LowConfidence	LLaDA2.0-mini is a diffusion language model with dense architecture.
LLaDA2.0 (mini, flash)	inclusionAI/LLaDA2.0-flash	LowConfidence	LLaDA2.0-flash is a diffusion language model featuring a 100B Mixture-of-Experts (MoE) architecture.
LLaDA2.1-mini	inclusionAI/LLaDA2.1-mini	LowConfidence	LLaDA2.1-mini is an improved version of LLaDA2.0-mini with better performance.
d3LLM-LLaDA	d3LLM/d3LLM_LLaDA	FullAttnMultiBlock	Bidirectional diffusion LLM based on LLaDA architecture with full attention.
d3LLM-Dream	d3LLM/d3LLM_Dream	FullAttnMultiBlock	Bidirectional diffusion LLM based on DREAM architecture with full attention.
SDAR (JetLM)	JetLM/SDAR-8B-Chat	JointThreshold	SDAR series diffusion language model (Chat), dense architecture.
SDAR (JetLM)	JetLM/SDAR-30B-A3B-Chat	JointThreshold	SDAR series diffusion language model (Chat), MoE architecture.