Instructions to use hsmin92/internvl35-fp8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use hsmin92/internvl35-fp8 with Transformers:

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

pipe = pipeline("image-text-to-text", model="hsmin92/internvl35-fp8")
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("hsmin92/internvl35-fp8")
model = AutoModelForMultimodalLM.from_pretrained("hsmin92/internvl35-fp8", 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 hsmin92/internvl35-fp8 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "hsmin92/internvl35-fp8"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "hsmin92/internvl35-fp8",
		"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/hsmin92/internvl35-fp8

SGLang

How to use hsmin92/internvl35-fp8 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 "hsmin92/internvl35-fp8" \
    --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": "hsmin92/internvl35-fp8",
		"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 "hsmin92/internvl35-fp8" \
        --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": "hsmin92/internvl35-fp8",
		"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 hsmin92/internvl35-fp8 with Docker Model Runner:
```
docker model run hf.co/hsmin92/internvl35-fp8
```

internvl35-fp8 / config.json

hsmin92

Upload InternVL3.5-4B FP8 Dynamic checkpoint and vLLM guide

e8805c1 verified 9 days ago

Raw

History Blame Contribute Delete

13.2 kB

	{
	"architectures": [
	"InternVLForConditionalGeneration"
	],
	"downsample_ratio": 0.5,
	"dtype": "bfloat16",
	"image_seq_length": 256,
	"image_token_id": 151671,
	"model_type": "internvl",
	"projector_hidden_act": "gelu",
	"quantization_config": {
	"config_groups": {
	"group_0": {
	"format": "float-quantized",
	"input_activations": {
	"actorder": null,
	"block_structure": null,
	"dynamic": true,
	"group_size": null,
	"num_bits": 8,
	"observer": null,
	"observer_kwargs": {},
	"scale_dtype": null,
	"strategy": "token",
	"symmetric": true,
	"type": "float",
	"zp_dtype": null
	},
	"output_activations": null,
	"targets": [
	"Linear"
	],
	"weights": {
	"actorder": null,
	"block_structure": null,
	"dynamic": false,
	"group_size": null,
	"num_bits": 8,
	"observer": "memoryless_minmax",
	"observer_kwargs": {},
	"scale_dtype": null,
	"strategy": "channel",
	"symmetric": true,
	"type": "float",
	"zp_dtype": null
	}
	}
	},
	"format": "float-quantized",
	"global_compression_ratio": null,
	"ignore": [
	"model.vision_tower.encoder.layer.0.attention.q_proj",
	"model.vision_tower.encoder.layer.0.attention.k_proj",
	"model.vision_tower.encoder.layer.0.attention.v_proj",
	"model.vision_tower.encoder.layer.0.attention.projection_layer",
	"model.vision_tower.encoder.layer.0.mlp.fc1",
	"model.vision_tower.encoder.layer.0.mlp.fc2",
	"model.vision_tower.encoder.layer.1.attention.q_proj",
	"model.vision_tower.encoder.layer.1.attention.k_proj",
	"model.vision_tower.encoder.layer.1.attention.v_proj",
	"model.vision_tower.encoder.layer.1.attention.projection_layer",
	"model.vision_tower.encoder.layer.1.mlp.fc1",
	"model.vision_tower.encoder.layer.1.mlp.fc2",
	"model.vision_tower.encoder.layer.2.attention.q_proj",
	"model.vision_tower.encoder.layer.2.attention.k_proj",
	"model.vision_tower.encoder.layer.2.attention.v_proj",
	"model.vision_tower.encoder.layer.2.attention.projection_layer",
	"model.vision_tower.encoder.layer.2.mlp.fc1",
	"model.vision_tower.encoder.layer.2.mlp.fc2",
	"model.vision_tower.encoder.layer.3.attention.q_proj",
	"model.vision_tower.encoder.layer.3.attention.k_proj",
	"model.vision_tower.encoder.layer.3.attention.v_proj",
	"model.vision_tower.encoder.layer.3.attention.projection_layer",
	"model.vision_tower.encoder.layer.3.mlp.fc1",
	"model.vision_tower.encoder.layer.3.mlp.fc2",
	"model.vision_tower.encoder.layer.4.attention.q_proj",
	"model.vision_tower.encoder.layer.4.attention.k_proj",
	"model.vision_tower.encoder.layer.4.attention.v_proj",
	"model.vision_tower.encoder.layer.4.attention.projection_layer",
	"model.vision_tower.encoder.layer.4.mlp.fc1",
	"model.vision_tower.encoder.layer.4.mlp.fc2",
	"model.vision_tower.encoder.layer.5.attention.q_proj",
	"model.vision_tower.encoder.layer.5.attention.k_proj",
	"model.vision_tower.encoder.layer.5.attention.v_proj",
	"model.vision_tower.encoder.layer.5.attention.projection_layer",
	"model.vision_tower.encoder.layer.5.mlp.fc1",
	"model.vision_tower.encoder.layer.5.mlp.fc2",
	"model.vision_tower.encoder.layer.6.attention.q_proj",
	"model.vision_tower.encoder.layer.6.attention.k_proj",
	"model.vision_tower.encoder.layer.6.attention.v_proj",
	"model.vision_tower.encoder.layer.6.attention.projection_layer",
	"model.vision_tower.encoder.layer.6.mlp.fc1",
	"model.vision_tower.encoder.layer.6.mlp.fc2",
	"model.vision_tower.encoder.layer.7.attention.q_proj",
	"model.vision_tower.encoder.layer.7.attention.k_proj",
	"model.vision_tower.encoder.layer.7.attention.v_proj",
	"model.vision_tower.encoder.layer.7.attention.projection_layer",
	"model.vision_tower.encoder.layer.7.mlp.fc1",
	"model.vision_tower.encoder.layer.7.mlp.fc2",
	"model.vision_tower.encoder.layer.8.attention.q_proj",
	"model.vision_tower.encoder.layer.8.attention.k_proj",
	"model.vision_tower.encoder.layer.8.attention.v_proj",
	"model.vision_tower.encoder.layer.8.attention.projection_layer",
	"model.vision_tower.encoder.layer.8.mlp.fc1",
	"model.vision_tower.encoder.layer.8.mlp.fc2",
	"model.vision_tower.encoder.layer.9.attention.q_proj",
	"model.vision_tower.encoder.layer.9.attention.k_proj",
	"model.vision_tower.encoder.layer.9.attention.v_proj",
	"model.vision_tower.encoder.layer.9.attention.projection_layer",
	"model.vision_tower.encoder.layer.9.mlp.fc1",
	"model.vision_tower.encoder.layer.9.mlp.fc2",
	"model.vision_tower.encoder.layer.10.attention.q_proj",
	"model.vision_tower.encoder.layer.10.attention.k_proj",
	"model.vision_tower.encoder.layer.10.attention.v_proj",
	"model.vision_tower.encoder.layer.10.attention.projection_layer",
	"model.vision_tower.encoder.layer.10.mlp.fc1",
	"model.vision_tower.encoder.layer.10.mlp.fc2",
	"model.vision_tower.encoder.layer.11.attention.q_proj",
	"model.vision_tower.encoder.layer.11.attention.k_proj",
	"model.vision_tower.encoder.layer.11.attention.v_proj",
	"model.vision_tower.encoder.layer.11.attention.projection_layer",
	"model.vision_tower.encoder.layer.11.mlp.fc1",
	"model.vision_tower.encoder.layer.11.mlp.fc2",
	"model.vision_tower.encoder.layer.12.attention.q_proj",
	"model.vision_tower.encoder.layer.12.attention.k_proj",
	"model.vision_tower.encoder.layer.12.attention.v_proj",
	"model.vision_tower.encoder.layer.12.attention.projection_layer",
	"model.vision_tower.encoder.layer.12.mlp.fc1",
	"model.vision_tower.encoder.layer.12.mlp.fc2",
	"model.vision_tower.encoder.layer.13.attention.q_proj",
	"model.vision_tower.encoder.layer.13.attention.k_proj",
	"model.vision_tower.encoder.layer.13.attention.v_proj",
	"model.vision_tower.encoder.layer.13.attention.projection_layer",
	"model.vision_tower.encoder.layer.13.mlp.fc1",
	"model.vision_tower.encoder.layer.13.mlp.fc2",
	"model.vision_tower.encoder.layer.14.attention.q_proj",
	"model.vision_tower.encoder.layer.14.attention.k_proj",
	"model.vision_tower.encoder.layer.14.attention.v_proj",
	"model.vision_tower.encoder.layer.14.attention.projection_layer",
	"model.vision_tower.encoder.layer.14.mlp.fc1",
	"model.vision_tower.encoder.layer.14.mlp.fc2",
	"model.vision_tower.encoder.layer.15.attention.q_proj",
	"model.vision_tower.encoder.layer.15.attention.k_proj",
	"model.vision_tower.encoder.layer.15.attention.v_proj",
	"model.vision_tower.encoder.layer.15.attention.projection_layer",
	"model.vision_tower.encoder.layer.15.mlp.fc1",
	"model.vision_tower.encoder.layer.15.mlp.fc2",
	"model.vision_tower.encoder.layer.16.attention.q_proj",
	"model.vision_tower.encoder.layer.16.attention.k_proj",
	"model.vision_tower.encoder.layer.16.attention.v_proj",
	"model.vision_tower.encoder.layer.16.attention.projection_layer",
	"model.vision_tower.encoder.layer.16.mlp.fc1",
	"model.vision_tower.encoder.layer.16.mlp.fc2",
	"model.vision_tower.encoder.layer.17.attention.q_proj",
	"model.vision_tower.encoder.layer.17.attention.k_proj",
	"model.vision_tower.encoder.layer.17.attention.v_proj",
	"model.vision_tower.encoder.layer.17.attention.projection_layer",
	"model.vision_tower.encoder.layer.17.mlp.fc1",
	"model.vision_tower.encoder.layer.17.mlp.fc2",
	"model.vision_tower.encoder.layer.18.attention.q_proj",
	"model.vision_tower.encoder.layer.18.attention.k_proj",
	"model.vision_tower.encoder.layer.18.attention.v_proj",
	"model.vision_tower.encoder.layer.18.attention.projection_layer",
	"model.vision_tower.encoder.layer.18.mlp.fc1",
	"model.vision_tower.encoder.layer.18.mlp.fc2",
	"model.vision_tower.encoder.layer.19.attention.q_proj",
	"model.vision_tower.encoder.layer.19.attention.k_proj",
	"model.vision_tower.encoder.layer.19.attention.v_proj",
	"model.vision_tower.encoder.layer.19.attention.projection_layer",
	"model.vision_tower.encoder.layer.19.mlp.fc1",
	"model.vision_tower.encoder.layer.19.mlp.fc2",
	"model.vision_tower.encoder.layer.20.attention.q_proj",
	"model.vision_tower.encoder.layer.20.attention.k_proj",
	"model.vision_tower.encoder.layer.20.attention.v_proj",
	"model.vision_tower.encoder.layer.20.attention.projection_layer",
	"model.vision_tower.encoder.layer.20.mlp.fc1",
	"model.vision_tower.encoder.layer.20.mlp.fc2",
	"model.vision_tower.encoder.layer.21.attention.q_proj",
	"model.vision_tower.encoder.layer.21.attention.k_proj",
	"model.vision_tower.encoder.layer.21.attention.v_proj",
	"model.vision_tower.encoder.layer.21.attention.projection_layer",
	"model.vision_tower.encoder.layer.21.mlp.fc1",
	"model.vision_tower.encoder.layer.21.mlp.fc2",
	"model.vision_tower.encoder.layer.22.attention.q_proj",
	"model.vision_tower.encoder.layer.22.attention.k_proj",
	"model.vision_tower.encoder.layer.22.attention.v_proj",
	"model.vision_tower.encoder.layer.22.attention.projection_layer",
	"model.vision_tower.encoder.layer.22.mlp.fc1",
	"model.vision_tower.encoder.layer.22.mlp.fc2",
	"model.vision_tower.encoder.layer.23.attention.q_proj",
	"model.vision_tower.encoder.layer.23.attention.k_proj",
	"model.vision_tower.encoder.layer.23.attention.v_proj",
	"model.vision_tower.encoder.layer.23.attention.projection_layer",
	"model.vision_tower.encoder.layer.23.mlp.fc1",
	"model.vision_tower.encoder.layer.23.mlp.fc2",
	"model.multi_modal_projector.linear_1",
	"model.multi_modal_projector.linear_2",
	"lm_head"
	],
	"kv_cache_scheme": null,
	"quant_method": "compressed-tensors",
	"quantization_status": "compressed",
	"sparsity_config": {},
	"transform_config": {},
	"version": "0.17.1"
	},
	"text_config": {
	"_name_or_path": "/root/codespace/checkpoints/Qwen3-4B",
	"architectures": [
	"Qwen3ForCausalLM"
	],
	"attention_bias": false,
	"attention_dropout": 0.0,
	"bos_token_id": 151643,
	"debug": false,
	"dtype": "bfloat16",
	"eos_token_id": 151645,
	"ep_size": 1,
	"head_dim": 128,
	"hidden_act": "silu",
	"hidden_size": 2560,
	"initializer_range": 0.02,
	"intermediate_size": 9728,
	"layer_types": [
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention",
	"full_attention"
	],
	"max_position_embeddings": 40960,
	"max_window_layers": 36,
	"micro_forward": false,
	"model_type": "qwen3",
	"num_attention_heads": 32,
	"num_hidden_layers": 36,
	"num_key_value_heads": 8,
	"pad_token_id": null,
	"rms_norm_eps": 1e-06,
	"rope_parameters": {
	"rope_theta": 1000000,
	"rope_type": "default"
	},
	"skip_checkpoint": false,
	"sliding_window": null,
	"tie_word_embeddings": false,
	"use_cache": true,
	"use_deepep": false,
	"use_sliding_window": false,
	"vocab_size": 151936
	},
	"tie_word_embeddings": false,
	"transformers_version": "5.10.1",
	"vision_config": {
	"architectures": [
	"InternVisionModel"
	],
	"attention_bias": true,
	"attention_dropout": 0.0,
	"dropout": 0.0,
	"dtype": "bfloat16",
	"hidden_act": "gelu",
	"hidden_dropout_prob": 0.0,
	"hidden_size": 1024,
	"image_size": [
	448,
	448
	],
	"initializer_factor": 0.1,
	"initializer_range": 1e-10,
	"intermediate_size": 4096,
	"layer_norm_eps": 1e-06,
	"layer_scale_init_value": 0.1,
	"model_type": "internvl_vision",
	"norm_type": "layer_norm",
	"num_attention_heads": 16,
	"num_channels": 3,
	"num_hidden_layers": 24,
	"patch_size": [
	14,
	14
	],
	"projection_dropout": 0.0,
	"use_absolute_position_embeddings": true,
	"use_mask_token": false,
	"use_mean_pooling": true,
	"use_qk_norm": false
	},
	"vision_feature_layer": -1,
	"vision_feature_select_strategy": "default"
	}