Instructions to use leon-se/Aria-sequential_mlp-FP8-dynamic with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use leon-se/Aria-sequential_mlp-FP8-dynamic with Transformers:

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

pipe = pipeline("image-text-to-text", model="leon-se/Aria-sequential_mlp-FP8-dynamic", trust_remote_code=True)
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("leon-se/Aria-sequential_mlp-FP8-dynamic", trust_remote_code=True)
model = AutoModelForMultimodalLM.from_pretrained("leon-se/Aria-sequential_mlp-FP8-dynamic", trust_remote_code=True, device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use leon-se/Aria-sequential_mlp-FP8-dynamic with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "leon-se/Aria-sequential_mlp-FP8-dynamic"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "leon-se/Aria-sequential_mlp-FP8-dynamic",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/leon-se/Aria-sequential_mlp-FP8-dynamic

SGLang

How to use leon-se/Aria-sequential_mlp-FP8-dynamic 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 "leon-se/Aria-sequential_mlp-FP8-dynamic" \
    --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": "leon-se/Aria-sequential_mlp-FP8-dynamic",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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 "leon-se/Aria-sequential_mlp-FP8-dynamic" \
        --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": "leon-se/Aria-sequential_mlp-FP8-dynamic",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use leon-se/Aria-sequential_mlp-FP8-dynamic with Docker Model Runner:
```
docker model run hf.co/leon-se/Aria-sequential_mlp-FP8-dynamic
```

Aria-sequential_mlp-FP8-dynamic / config.json

Leon

Upload folder using huggingface_hub

e79c572 verified almost 2 years ago

Raw

History Blame Contribute Delete

13.5 kB

	{
	"_name_or_path": "/root/.cache/huggingface/hub/models--rhymes-ai--Aria-sequential_mlp/snapshots/021f586a105dd50134166e3e7b57ec7d785e319a",
	"architectures": [
	"AriaForConditionalGeneration"
	],
	"auto_map": {
	"AutoConfig": "configuration_aria.AriaConfig",
	"AutoModelForCausalLM": "modeling_aria.AriaForConditionalGeneration"
	},
	"ignore_index": -100,
	"image_token_index": 9,
	"model_type": "aria",
	"projector_patch_to_query_dict": {
	"1225": 128,
	"4900": 256
	},
	"quantization_config": {
	"config_groups": {
	"group_0": {
	"input_activations": {
	"actorder": null,
	"block_structure": null,
	"dynamic": true,
	"group_size": null,
	"num_bits": 8,
	"observer": null,
	"observer_kwargs": {},
	"strategy": "token",
	"symmetric": true,
	"type": "float"
	},
	"output_activations": null,
	"targets": [
	"Linear"
	],
	"weights": {
	"actorder": null,
	"block_structure": null,
	"dynamic": false,
	"group_size": null,
	"num_bits": 8,
	"observer": "minmax",
	"observer_kwargs": {},
	"strategy": "channel",
	"symmetric": true,
	"type": "float"
	}
	}
	},
	"format": "float-quantized",
	"global_compression_ratio": 1.5592466451256677,
	"ignore": [
	"vision_tower.vision_model.encoder.layers.0.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.0.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.0.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.0.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.0.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.0.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.1.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.1.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.1.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.1.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.1.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.1.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.2.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.2.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.2.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.2.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.2.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.2.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.3.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.3.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.3.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.3.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.3.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.3.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.4.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.4.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.4.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.4.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.4.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.4.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.5.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.5.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.5.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.5.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.5.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.5.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.6.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.6.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.6.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.6.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.6.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.6.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.7.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.7.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.7.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.7.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.7.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.7.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.8.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.8.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.8.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.8.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.8.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.8.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.9.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.9.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.9.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.9.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.9.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.9.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.10.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.10.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.10.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.10.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.10.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.10.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.11.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.11.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.11.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.11.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.11.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.11.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.12.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.12.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.12.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.12.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.12.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.12.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.13.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.13.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.13.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.13.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.13.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.13.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.14.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.14.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.14.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.14.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.14.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.14.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.15.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.15.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.15.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.15.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.15.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.15.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.16.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.16.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.16.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.16.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.16.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.16.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.17.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.17.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.17.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.17.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.17.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.17.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.18.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.18.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.18.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.18.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.18.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.18.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.19.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.19.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.19.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.19.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.19.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.19.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.20.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.20.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.20.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.20.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.20.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.20.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.21.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.21.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.21.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.21.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.21.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.21.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.22.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.22.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.22.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.22.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.22.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.22.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.23.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.23.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.23.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.23.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.23.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.23.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.24.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.24.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.24.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.24.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.24.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.24.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.25.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.25.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.25.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.25.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.25.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.25.mlp.fc2",
	"vision_tower.vision_model.encoder.layers.26.self_attn.k_proj",
	"vision_tower.vision_model.encoder.layers.26.self_attn.v_proj",
	"vision_tower.vision_model.encoder.layers.26.self_attn.q_proj",
	"vision_tower.vision_model.encoder.layers.26.self_attn.out_proj",
	"vision_tower.vision_model.encoder.layers.26.mlp.fc1",
	"vision_tower.vision_model.encoder.layers.26.mlp.fc2",
	"multi_modal_projector.cross_attn.q_proj",
	"multi_modal_projector.cross_attn.k_proj",
	"multi_modal_projector.cross_attn.v_proj",
	"multi_modal_projector.cross_attn.linear",
	"multi_modal_projector.ffn.linear_in",
	"multi_modal_projector.ffn.linear_out",
	"language_model.lm_head"
	],
	"kv_cache_scheme": null,
	"quant_method": "compressed-tensors",
	"quantization_status": "compressed"
	},
	"text_config": {
	"hidden_size": 2560,
	"intermediate_size": 13568,
	"max_position_embeddings": 65536,
	"model_type": "aria_moe_lm",
	"moe_intermediate_size": 1664,
	"moe_num_experts": 64,
	"moe_topk": 6,
	"num_attention_heads": 20,
	"num_experts_per_tok": 6,
	"num_hidden_layers": 28,
	"num_key_value_heads": 20,
	"rope_theta": 5000000,
	"vocab_size": 100352
	},
	"torch_dtype": "bfloat16",
	"transformers_version": "4.45.0",
	"vision_config": {
	"_flash_attn_2_enabled": true,
	"architectures": [
	"AriaVisionModel"
	],
	"hidden_size": 1152,
	"image_size": 980,
	"intermediate_size": 4304,
	"model_type": "aria_vision_model",
	"num_attention_heads": 16,
	"num_hidden_layers": 27,
	"patch_size": 14,
	"torch_dtype": "bfloat16"
	}
	}