internlm / Intern-S1-FP8

Intern-S1

Introduction

We introduce Intern-S1 , our most advanced open-source multimodal reasoning model to date. Intern-S1 combines strong general-task capabilities with state-of-the-art performance on a wide range of scientific tasks , rivaling leading closed-source commercial models. Built upon a 235B MoE language model (Qwen3) and a 6B Vision encoder (InternViT), Intern-S1 has been further pretrained on 5 trillion tokens of multimodal data, including over 2.5 trillion scientific-domain tokens . This enables the model to retain strong general capabilities while excelling in specialized scientific domains such as interpreting chemical structures, understanding protein sequences, and planning compound synthesis routes , making Intern-S1 to be a capable research assistant for real-world scientific applications. Features

• Strong performance across language and vision reasoning benchmarks, especially scientific tasks.

• Continuously pretrained on a massive 5T token dataset, with over 50% specialized scientific data, embedding deep domain expertise.

• Dynamic tokenizer enables native understanding of molecular formulas, protein sequences, and seismic signals.

Performance

We evaluate the Intern-S1 on various benchmarks including general datasets and scientifc datasets. We report the performance comparsion with the recent VLMs and LLMs below.

Benchmarks	Intern-S1	InternVL3-78B	Qwen2.5-VL-72B	DS-R1-0528	Qwen3-235B-A22B	Kimi-K2-Instruct	Gemini-2.5 Pro	o3	Grok-4
MMLU-Pro	83.5 ✅	73.0	72.1	83.4	82.2	82.7	86.0	85.0	85.9
MMMU	77.7 ✅	72.2	70.2	-	-	-	81.9	80.8	77.9
GPQA	77.3	49.9	49.0	80.6	71.1	77.8	83.8	83.3	87.5
MMStar	74.9 ✅	72.5	70.8	-	-	-	79.3	75.1	69.6
MathVista	81.5 👑	79.0	74.8	-	-	-	80.3	77.5	72.5
AIME2025	86.0	10.7	10.9	87.5	81.5	51.4	83.0	88.9	91.7
MathVision	62.5 ✅	43.1	38.1	-	-	-	73.0	67.7	67.3
IFEval	86.7	75.6	83.9	79.7	85.0	90.2	91.5	92.2	92.8
SFE	44.3 👑	36.2	30.5	-	-	-	43.0	37.7	31.2
Physics	44.0 ✅	23.1	15.7	-	-	-	40.0	47.9	42.8
SmolInstruct	51.0 👑	19.4	21.0	30.7	28.7	48.1	40.4	43.9	47.3
ChemBench	83.4 👑	61.3	61.6	75.6	75.8	75.3	82.8	81.6	83.3
MatBench	75.0 👑	49.3	51.5	57.7	52.1	61.7	61.7	61.6	67.9
MicroVQA	63.9 👑	59.1	53.0	-	-	-	63.1	58.3	59.5
ProteinLMBench	63.1	61.6	61.0	61.4	59.8	66.7	62.9	67.7	66.2
MSEarthMCQ	65.7 👑	57.2	37.6	-	-	-	59.9	61.0	58.0
XLRS-Bench	55.0 👑	49.3	50.9	-	-	-	45.2	43.6	45.4

Note : ✅ means the best performance among open-sourced models, 👑 indicates the best performance among all models.

We use the OpenCompass and VLMEvalkit to evaluate all models.

Quick Start

Sampling Parameters

We recommend using the following hyperparameters to ensure better results

top_p = 1.0
top_k = 50
min_p = 0.0
temperature = 0.7

Transformers

The following provides demo code illustrating how to generate based on text and multimodal inputs.

Please use transformers>=4.53.0 to ensure the model works normally.

Text input

from transformers import AutoProcessor, AutoModelForCausalLM
import torch

model_name = "internlm/Intern-S1"
processor = AutoProcessor.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", torch_dtype="auto", trust_remote_code=True)

messages = [
    {
        "role": "user",
        "content": [
            {"type": "text", "text": "tell me about an interesting physical phenomenon."},
        ],
    }
]

inputs = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt").to(model.device, dtype=torch.bfloat16)

generate_ids = model.generate(**inputs, max_new_tokens=32768)
decoded_output = processor.decode(generate_ids[0, inputs["input_ids"].shape[1] :], skip_special_tokens=True)
print(decoded_output)

Image input

from transformers import AutoProcessor, AutoModelForCausalLM
import torch

model_name = "internlm/Intern-S1"
processor = AutoProcessor.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", torch_dtype="auto", trust_remote_code=True)

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "http://images.cocodataset.org/val2017/000000039769.jpg"},
            {"type": "text", "text": "Please describe the image explicitly."},
        ],
    }
]

inputs = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt").to(model.device, dtype=torch.bfloat16)

generate_ids = model.generate(**inputs, max_new_tokens=32768)
decoded_output = processor.decode(generate_ids[0, inputs["input_ids"].shape[1] :], skip_special_tokens=True)
print(decoded_output)

Video input

Please ensure that the decord video decoding library is installed via pip install decord .

from transformers import AutoProcessor, AutoModelForCausalLM
import torch

model_name = "internlm/Intern-S1"
processor = AutoProcessor.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", torch_dtype="auto", trust_remote_code=True)

messages = [
        {
            "role": "user",
            "content": [
                {
                    "type": "video",
                    "url": "https://huggingface.co/datasets/hf-internal-testing/fixtures_videos/resolve/main/tennis.mp4",
                },
                {"type": "text", "text": "What type of shot is the man performing?"},
            ],
        }
    ]

inputs = processor.apply_chat_template(
        messages,
        return_tensors="pt",
        add_generation_prompt=True,
        video_load_backend="decord",
        tokenize=True,
        return_dict=True,
    ).to(model.device, dtype=torch.float16)

generate_ids = model.generate(**inputs, max_new_tokens=32768)
decoded_output = processor.decode(generate_ids[0, inputs["input_ids"].shape[1] :], skip_special_tokens=True)
print(decoded_output)

Serving

You can utilize one of the following LLM inference frameworks to create an OpenAI compatible server:

lmdeploy(>=0.9.2)

lmdeploy serve api_server internlm/Intern-S1-FP8 --reasoning-parser intern-s1 --tool-call-parser intern-s1 --tp 4

vllm

Coming soon.

sglang

Supporting Intern-S1 with SGLang is still in progress. Please refer to this PR .

CUDA_VISIBLE_DEVICES=0,1,2,3 \
    python3 -m sglang.launch_server \
    --model-path internlm/Intern-S1-FP8 \
    --trust-remote-code \
    --tp 4 \
    --port 8001 \
    --mem-fraction-static 0.85 \
    --enable-multimodal \
    --grammar-backend none

Advanced Usage

Tool Calling

Many Large Language Models (LLMs) now feature Tool Calling , a powerful capability that allows them to extend their functionality by interacting with external tools and APIs. This enables models to perform tasks like fetching up-to-the-minute information, running code, or calling functions within other applications.

A key advantage for developers is that a growing number of open-source LLMs are designed to be compatible with the OpenAI API. This means you can leverage the same familiar syntax and structure from the OpenAI library to implement tool calling with these open-source models. As a result, the code demonstrated in this tutorial is versatile—it works not just with OpenAI models, but with any model that follows the same interface standard.

To illustrate how this works, let's dive into a practical code example that uses tool calling to get the latest weather forecast (based on lmdeploy api server).

      
from openai import OpenAI
import json


def get_current_temperature(location: str, unit: str = "celsius"):
    """Get current temperature at a location.

    Args:
        location: The location to get the temperature for, in the format "City, State, Country".
        unit: The unit to return the temperature in. Defaults to "celsius". (choices: ["celsius", "fahrenheit"])

    Returns:
        the temperature, the location, and the unit in a dict
    """
    return {
        "temperature": 26.1,
        "location": location,
        "unit": unit,
    }


def get_temperature_date(location: str, date: str, unit: str = "celsius"):
    """Get temperature at a location and date.

    Args:
        location: The location to get the temperature for, in the format "City, State, Country".
        date: The date to get the temperature for, in the format "Year-Month-Day".
        unit: The unit to return the temperature in. Defaults to "celsius". (choices: ["celsius", "fahrenheit"])

    Returns:
        the temperature, the location, the date and the unit in a dict
    """
    return {
        "temperature": 25.9,
        "location": location,
        "date": date,
        "unit": unit,
    }

def get_function_by_name(name):
    if name == "get_current_temperature":
        return get_current_temperature
    if name == "get_temperature_date":
        return get_temperature_date

tools = [{
    'type': 'function',
    'function': {
        'name': 'get_current_temperature',
        'description': 'Get current temperature at a location.',
        'parameters': {
            'type': 'object',
            'properties': {
                'location': {
                    'type': 'string',
                    'description': 'The location to get the temperature for, in the format \'City, State, Country\'.'
                },
                'unit': {
                    'type': 'string',
                    'enum': [
                        'celsius',
                        'fahrenheit'
                    ],
                    'description': 'The unit to return the temperature in. Defaults to \'celsius\'.'
                }
            },
            'required': [
                'location'
            ]
        }
    }
}, {
    'type': 'function',
    'function': {
        'name': 'get_temperature_date',
        'description': 'Get temperature at a location and date.',
        'parameters': {
            'type': 'object',
            'properties': {
                'location': {
                    'type': 'string',
                    'description': 'The location to get the temperature for, in the format \'City, State, Country\'.'
                },
                'date': {
                    'type': 'string',
                    'description': 'The date to get the temperature for, in the format \'Year-Month-Day\'.'
                },
                'unit': {
                    'type': 'string',
                    'enum': [
                        'celsius',
                        'fahrenheit'
                    ],
                    'description': 'The unit to return the temperature in. Defaults to \'celsius\'.'
                }
            },
            'required': [
                'location',
                'date'
            ]
        }
    }
}]



messages = [
    {'role': 'user', 'content': 'Today is 2024-11-14, What\'s the temperature in San Francisco now? How about tomorrow?'}
]

openai_api_key = "EMPTY"
openai_api_base = "http://0.0.0.0:23333/v1"
client = OpenAI(
    api_key=openai_api_key,
    base_url=openai_api_base,
)
model_name = client.models.list().data[0].id
response = client.chat.completions.create(
    model=model_name,
    messages=messages,
    max_tokens=32768,
    temperature=0.8,
    top_p=0.8,
    stream=False,
    extra_body=dict(spaces_between_special_tokens=False, enable_thinking=False),
    tools=tools)
print(response.choices[0].message)
messages.append(response.choices[0].message)

for tool_call in response.choices[0].message.tool_calls:
    tool_call_args = json.loads(tool_call.function.arguments)
    tool_call_result = get_function_by_name(tool_call.function.name)(**tool_call_args)
    tool_call_result = json.dumps(tool_call_result, ensure_ascii=False)
    messages.append({
        'role': 'tool',
        'name': tool_call.function.name,
        'content': tool_call_result,
        'tool_call_id': tool_call.id
    })

response = client.chat.completions.create(
    model=model_name,
    messages=messages,
    temperature=0.8,
    top_p=0.8,
    stream=False,
    extra_body=dict(spaces_between_special_tokens=False, enable_thinking=False),
    tools=tools)
print(response.choices[0].message.content)

Switching Between Thinking and Non-Thinking Modes

Intern-S1 enables thinking mode by default, enhancing the model's reasoning capabilities to generate higher-quality responses. This feature can be disabled by setting enable_thinking=False in tokenizer.apply_chat_template

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=False  # think mode indicator
)

With LMDeploy serving Intern-S1 models, you can dynamically control the thinking mode by adjusting the enable_thinking parameter in your requests.

from openai import OpenAI
import json

messages = [
{
    'role': 'user',
    'content': 'who are you'
}, {
    'role': 'assistant',
    'content': 'I am an AI'
}, {
    'role': 'user',
    'content': 'AGI is?'
}]

openai_api_key = "EMPTY"
openai_api_base = "http://0.0.0.0:23333/v1"
client = OpenAI(
    api_key=openai_api_key,
    base_url=openai_api_base,
)
model_name = client.models.list().data[0].id

response = client.chat.completions.create(
    model=model_name,
    messages=messages,
    temperature=0.7,
    top_p=0.8,
    max_tokens=2048,
    extra_body={
        "enable_thinking": False,
    }
)
print(json.dumps(response.model_dump(), indent=2, ensure_ascii=False))

For vllm and sglang users, configure this through,

extra_body={
    "chat_template_kwargs": {"enable_thinking": False}
}