Instructions to use InternScience/Agents-A1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use InternScience/Agents-A1 with Transformers:

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

pipe = pipeline("text-generation", model="InternScience/Agents-A1")
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("InternScience/Agents-A1")
model = AutoModelForMultimodalLM.from_pretrained("InternScience/Agents-A1", 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 InternScience/Agents-A1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "InternScience/Agents-A1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "InternScience/Agents-A1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/InternScience/Agents-A1

SGLang

How to use InternScience/Agents-A1 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 "InternScience/Agents-A1" \
    --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": "InternScience/Agents-A1",
		"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 "InternScience/Agents-A1" \
        --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": "InternScience/Agents-A1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use InternScience/Agents-A1 with Docker Model Runner:
```
docker model run hf.co/InternScience/Agents-A1
```

NVFP4 quantization

#19

by Duonglv - opened 29 days ago

Discussion

Duonglv

29 days ago

Hello,
You did a amazing job. This model is very strong.
Could you guys release a NVFP4 quantized model with some evaluations between this one and FP16 model?
Thank so much.

QBANIN

27 days ago

https://huggingface.co/s-batman/Agents-A1-NVFP4-MTP-GGUF/tree/main

Duonglv

27 days ago

@QBANIN That is GGUF version. I meant a NVFP4 version for running in vLLM. You can see the nvdia org, they have quantized many nvfp4 models with evaluations.

sHEL1562

14 days ago

In an effort to make this all accessible for consumer cards. I can't speak for much but I work heavily with VLLM, and have only a single 5090 at my disposal, so here are two non-GGUF serve commands.

This one has limitation of 21gb to weights. Adjust cache or sequences to lower, as the below commands fill up to around 30gb maximum fully loaded, fitting perfectly.

INT4 (non blackwell)

(Click to Expand)

First run, use: vllm serve compute1/Agents-A1-GPTQ-INT4-Sym

CUDA_VISIBLE_DEVICES=0
vllm serve /path/to/.cache/huggingface/hub/models--compute1--Agents-A1-GPTQ-INT4-Sym/snapshots/somestring/
--served-model-name compute1/Agents-A1-GPTQ-INT4-Sym
--host 0.0.0.0
--port 8000
--tensor-parallel-size 1
--max-model-len 196608
--gpu-memory-utilization 0.87
--kv-cache-dtype fp8_e5m2
--max-num-seqs 2
--tool-call-parser qwen3_coder
--reasoning-parser qwen3
--enable-auto-tool-choice
--trust-remote-code

^^^ Great INT4 , performs as well as ~Q5/Q6 gguf, if not better.

NVFP4 (heavier, blackwell)

(Click to Expand)

First run, use: vllm serve cyankiwi/Agents-A1-AWQ-NVFP4

CUDA_VISIBLE_DEVICES=0
VLLM_USE_FLASHINFER_SAMPLER=0
vllm serve /path/to/.cache/huggingface/hub/models--cyankiwi--Agents-A1-AWQ-NVFP4/snapshots/somestring/
--served-model-name cyankiwi/Agents-A1-AWQ-NVFP4
--host 0.0.0.0
--port 8000
--moe-backend marlin
--tensor-parallel-size 1
--max-model-len 172032
--gpu-memory-utilization 0.88
--kv-cache-dtype fp8
--max-num-seqs 2
--tool-call-parser qwen3_coder
--reasoning-parser qwen3
--enable-auto-tool-choice
--trust-remote-code

^^^ Great NVFP4, performs as well as ~Q6 gguf, if not better.

For sm120 - i.e. 5090, consumer blackwell
Non-negotiable moe-backend and flashinfer sampler env var - discovered from (this model - link to highlight) - where they broke the vision heads 🤷

YMMV with system ram for profiling/warmup - Specs: 5090 + 64GB — vLLM runtime on CUDA host

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment