Instructions to use inclusionAI/LLaDA2.1-mini with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use inclusionAI/LLaDA2.1-mini with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="inclusionAI/LLaDA2.1-mini", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("inclusionAI/LLaDA2.1-mini", trust_remote_code=True, dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use inclusionAI/LLaDA2.1-mini with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "inclusionAI/LLaDA2.1-mini"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "inclusionAI/LLaDA2.1-mini",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/inclusionAI/LLaDA2.1-mini

SGLang

How to use inclusionAI/LLaDA2.1-mini with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "inclusionAI/LLaDA2.1-mini" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "inclusionAI/LLaDA2.1-mini",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "inclusionAI/LLaDA2.1-mini" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "inclusionAI/LLaDA2.1-mini",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use inclusionAI/LLaDA2.1-mini with Docker Model Runner:
```
docker model run hf.co/inclusionAI/LLaDA2.1-mini
```

kashif HF Staff commited on Mar 18

Commit

90c63c7

verified ·

1 Parent(s): 92a5aa6

Use create_bidirectional_mask for backend-agnostic attention mask handling (SDPA, FA2, flex)

Browse files

Files changed (1) hide show

modeling_llada2_moe.py +6 -14

modeling_llada2_moe.py CHANGED Viewed

@@ -28,9 +28,7 @@ from torch.nn import CrossEntropyLoss
 from transformers.activations import ACT2FN
 from transformers.cache_utils import Cache, DynamicCache
-from transformers.modeling_attn_mask_utils import (
-    _prepare_4d_causal_attention_mask_for_sdpa,
-)
 from transformers.modeling_outputs import (
     MoeModelOutputWithPast,
     MoeCausalLMOutputWithPast,
@@ -876,17 +874,11 @@ class LLaDA2MoeModel(LLaDA2MoePreTrainedModel):
                 device=inputs_embeds.device,
             )
             position_ids = position_ids.unsqueeze(0)
-        if attention_mask.size() == (batch_size, 1, seq_length, seq_length):
-            attention_mask = _prepare_4d_causal_attention_mask_for_sdpa(
-                attention_mask,
-                (batch_size, seq_length),
-                inputs_embeds,
-                past_seen_tokens,
-            )
-        else:
-            raise ValueError(
-                f"LLaDA2.0 only support block attention mask with shape: {(batch_size, 1, seq_length, seq_length)}, the input attention with shape {attention_mask.size()=}!"
-            )
         # embed positions
         hidden_states = inputs_embeds

 from transformers.activations import ACT2FN
 from transformers.cache_utils import Cache, DynamicCache
+from transformers.masking_utils import create_bidirectional_mask
 from transformers.modeling_outputs import (
     MoeModelOutputWithPast,
     MoeCausalLMOutputWithPast,
                 device=inputs_embeds.device,
             )
             position_ids = position_ids.unsqueeze(0)
+        attention_mask = create_bidirectional_mask(
+            config=self.config,
+            inputs_embeds=inputs_embeds,
+            attention_mask=attention_mask,
+        )
         # embed positions
         hidden_states = inputs_embeds