• unsloth

• HuggingFaceTB/SmolLM3-3B

• en
• fr
• es
• it
• pt
• zh
• ar
• ru

[!NOTE]

Includes our chat template fixes!
If you are using llama.cpp, use --jinja

Unsloth Dynamic 2.0 achieves superior accuracy & outperforms other leading quants.

SmolLM3

!image/png

Table of Contents

1. Model Summary
2. How to use
3. Evaluation
4. Training
5. Limitations
6. License

Model Summary

SmolLM3 is a 3B parameter language model designed to push the boundaries of small models. It supports 6 languages, advanced reasoning and long context. SmolLM3 is a fully open model that offers strong performance at the 3B–4B scale.

!image/png

The model is a decoder-only transformer using GQA and NoPE (with 3:1 ratio), it was pretrained on 11.2T tokens with a staged curriculum of web, code, math and reasoning data. Post-training included midtraining on 140B reasoning tokens followed by supervised fine-tuning and alignment via Anchored Preference Optimization (APO).

Key features

• Instruct model optimized for hybrid reasoning
• Fully open model: open weights + full training details including public data mixture and training configs
• Long context: Trained on 64k context and suppots up to 128k tokens using YARN extrapolation
• Multilingual: 6 natively supported (English, French, Spanish, German, Italian, and Portuguese)

For more details refer to our blog post: https://hf.co/blog/smollm3

How to use

The modeling code for SmolLM3 is available in transformers v4.53.0, so make sure to upgrade your transformers version. You can also load the model with the latest vllm which uses transformers as a backend.

pip install -U transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "HuggingFaceTB/SmolLM3-3B"
device = "cuda"  # for GPU usage or "cpu" for CPU usage

load the tokenizer and the model
tokenizer = AutoTokenizer.frompretrained(modelname)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
).to(device)

prepare the model input
prompt = "Give me a brief explanation of gravity in simple terms."
messages_think = [
    {"role": "user", "content": prompt}
]

text = tokenizer.applychattemplate(
    messages_think,
    tokenize=False,
    addgenerationprompt=True,
)
modelinputs = tokenizer([text], returntensors="pt").to(model.device)

Generate the output
generatedids = model.generate(modelinputs, maxnewtokens=32768)

Get and decode the output
outputids = generatedids[0][len(modelinputs.inputids[0]) :]
print(tokenizer.decode(outputids, skipspecial_tokens=True))

>[!TIP]

We recommend setting temperature=0.6 and top_p=0.95 in the sampling parameters.

Enabling and Disabling Extended Thinking Mode

We enable extended thinking by default, so the example above generates the output with a reasoning trace. For choosing between enabling, you can provide the /think and /nothink flags through the system prompt as shown in the snippet below for extended thinking disabled. The code for generating the response with extended thinking would be the same except that the system prompt should have /think instead of /nothink.

prompt = "Give me a brief explanation of gravity in simple terms."
messages = [
    {"role": "system", "content": "/no_think"},
    {"role": "user", "content": prompt}
]

text = tokenizer.applychattemplate(
    messages,
    tokenize=False,
    addgenerationprompt=True,
)

We also provide the option of specifying the whether to use extended thinking through the enablethinking kwarg as in the example below. You do not need to set the /nothink or /think flags through the system prompt if using the kwarg, but keep in mind that the flag in the system prompt overwrites the setting in the kwarg.

prompt = "Give me a brief explanation of gravity in simple terms."
messages = [
    {"role": "user", "content": prompt}
]

text = tokenizer.applychattemplate(
    messages,
    tokenize=False,
    addgenerationprompt=True,
    enable_thinking=False
)

Agentic Usage

SmolLM3 supports tool calling!
Just pass your list of tools:

• Under the argument xmltools for standard tool-calling: these tools will be called as JSON blobs within XML tags, like call>{"name": "getweather", "arguments": {"city": "Copenhagen"}}call>
• Or under pythontools: then the model will call tools like python functions in a snippet, like getweather(city="Copenhagen")

from transformers import AutoModelForCausalLM, AutoTokenizer

checkpoint = "HuggingFaceTB/SmolLM3-3B"

tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForCausalLM.from_pretrained(checkpoint)

tools = [
    {
        "name": "get_weather",
        "description": "Get the weather in a city",
        "parameters": {"type": "object", "properties": {"city": {"type": "string", "description": "The city to get the weather for"}}}}
]

messages = [
    {
        "role": "user",
        "content": "Hello! How is the weather today in Copenhagen?"
    }
]

inputs = tokenizer.applychattemplate(
    messages,
    enable_thinking=False, # True works as well, your choice!
    xml_tools=tools,
    addgenerationprompt=True,
    tokenize=True,
    return_tensors="pt"
)

outputs = model.generate(inputs)
print(tokenizer.decode(outputs[0]))

prompt = "Give me a brief explanation of gravity in simple terms."
messages = [
    {"role": "system", "content": "Speak like a pirate./think"},
    {"role": "user", "content": prompt}
]

text = tokenizer.applychattemplate(
    messages,
    tokenize=False,
    addgenerationprompt=True,
)

python -m sglang.launch_server --model-path HuggingFaceTB/SmolLM3-3B

vllm serve HuggingFaceTB/SmolLM3-3B

curl http://localhost:8000/v1/chat/completions -H "Content-Type: application/json" -d '{
  "model": "HuggingFaceTB/SmolLM3-3B",
  "messages": [
    {"role": "user", "content": "Give me a brief explanation of gravity in simple terms."}
  ],
  "temperature": 0.6,
  "top_p": 0.95,
  "max_tokens": 16384,
  "chattemplatekwargs": {"enable_thinking": false}
}'