SmolLM2-360M-Instruct

SmolLM2-360M-Instruct

Answers chat-style questions and handles rewriting and summaries with a 362M-parameter instruction-tuned model that runs on modest hardware. For developers who want a small chat model with a bit more headroom than 135M.

Made by HuggingFaceTB on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: SmolLM2-360M-Instruct is the chat-tuned version of the 360M model in the SmolLM2 family. It starts from the 360M base model, which saw 4 trillion training tokens. The team then added supervised fine-tuning on public and curated datasets and a Direct Preference Optimization pass. It reads chat messages and replies in English.

Who it suits: It suits developers who want a small assistant model that fits on a laptop or a modest GPU and who can live with short, simple answers.

Industry | Who uses it | How they use it | Example
Software | App developers | Embed a small assistant that runs locally | A writing helper inside a desktop tool
Content tools | Product teams | Reword or shorten text without a cloud call | Condensing a note into two lines
Education | Teachers and students | Explore how chat templates and sampling work | A demo comparing temperature settings
Research | ML researchers | Fine-tune a compact instruction model | Adapting it to a company's tone

Skip it if you need dependable facts or arithmetic, since small models of this size make many mistakes.

Where to use it:

• Python scripts with the transformers library, on CPU or GPU
• The TRL command line, to chat from a terminal
• Windows, macOS or Linux with Python

Use case | Who | What they get back
Question answering | Developers | A short reply to a chat message
Rewriting and summaries | Content teams | Reworded or condensed text, which the card lists as supported
Terminal chat | Hobbyists | A session you start with one command

How to set it up: You need these first.

• Python with pip
• The transformers library, plus accelerate if you use several GPUs
• About 724 MB for the BF16 weights, or about 181 MB at 4-bit by estimate

1. Install the transformers library with pip.
2. Load the model by its Hugging Face name.
3. Put your message in a list, apply the chat template and generate.

pip install transformers

pip install -U huggingface_hub
hf download HuggingFaceTB/SmolLM2-360M-Instruct

How to use it: Pass your messages through the chat template, then generate with the low-temperature settings the card uses. For a terminal chat, install TRL and run its chat command pointed at this model.

from transformers import AutoModelForCausalLM, AutoTokenizer
checkpoint = "HuggingFaceTB/SmolLM2-360M-Instruct"

device = "cuda" # for GPU usage or "cpu" for CPU usage
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
# for multiple GPUs install accelerate and do `model = AutoModelForCausalLM.from_pretrained(checkpoint, device_map="auto")`
model = AutoModelForCausalLM.from_pretrained(checkpoint).to(device)

messages = [{"role": "user", "content": "What is the capital of France."}]
input_text=tokenizer.apply_chat_template(messages, tokenize=False)
print(input_text)
inputs = tokenizer.encode(input_text, return_tensors="pt").to(device)
outputs = model.generate(inputs, max_new_tokens=50, temperature=0.2, top_p=0.9, do_sample=True)
print(tokenizer.decode(outputs[0]))

pip install trl
trl chat --model_name_or_path HuggingFaceTB/SmolLM2-360M-Instruct --device cpu

Tool | What it does
apply_chat_template | Formats your messages the way the model expects
model.generate | Produces the reply, here capped at 50 new tokens
trl chat | Opens an interactive chat in the terminal

Example: Send the single message asking for the capital of France. With a temperature of 0.2 the model gives a short, steady reply and the script prints the prompt followed by that reply. The terminal chat command lets you keep asking follow-ups.

For newcomers and experts: If you are new, this is the SmolLM2 size you can chat with. Start with the TRL command on CPU, ask one plain question and judge the reply length. Keep expectations modest for a model this small.

If you know the field, the card scores the base model above Qwen2.5-0.5B on HellaSwag, ARC, PIQA and TriviaQA, while Qwen2.5-0.5B is far ahead on GSM8K. Function calling is listed only for the 1.7B sibling. The card's Transformers.js example loads the 135M model by name, so swap in this model's name if you use it.

SmolLM2-360M base model
Supervised fine-tuning on chat data
DPO on preference data
Chat template formats your messages
Model writes the reply

Good to know:

• Language: English only
• Size: 362M parameters in Safetensors
• Family: 135M and 1.7B siblings exist
• Community: other people have made 139 quantized and 200 fine-tuned versions

• License: Apache 2.0
• Cost to run: none named, the source lists no API key, account or paid service
• Data it reaches: the messages you give your own script, and the source names no web or account access
• Not for: function calling or arithmetic, since the card ties function calling to the 1.7B size and the base model scores 3.2 on GSM8K

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 362M
• Weights: about 724 MB in BF16; about 181 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 221,273
• Likes: 244
• License: apache-2.0
• Languages: en
• Based on: HuggingFaceTB/SmolLM2-360M
• Quantized versions: 138
• Fine-tunes: 200
• Tags: transformers, tensorboard, onnx, safetensors, llama, text-generation, transformers.js, conversational

Links

• Model: https://huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct
• Files: https://huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download HuggingFaceTB/SmolLM2-360M-Instruct

Then follow the usage in its model card. Nothing in this kit runs by itself.