What are the hardware requirements for SmolLM2 135M?

SmolLM2 135M requires approximately 0.3 GB of VRAM and supports a context window of 8K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed SmolLM2 135M and how large is it?

SmolLM2 135M is developed by HuggingFaceTB, a model with 135m parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

Where can I download or evaluate SmolLM2 135M?

SmolLM2 135M is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

SmolLM2 135M by HuggingFaceTB — VRAM 0.3GB, 8K context

Name: SmolLM2 135M
Author: HuggingFaceTB

SmolLM2 135M is an open-source language model by HuggingFaceTB. Features: 135m LLM, VRAM: 0.3GB, Context: 8K, License: apache-2.0, LLM Explorer Score: 0.28.

Arxiv:2502.02737 Deploy:azure En Endpoints compatible Eval-results Llama Region:us Safetensors

Model Card on HF 🤗: https://huggingface.co/HuggingFaceTB/SmolLM2-135M

SmolLM2 135M Parameters and Internals

Model Type

text generation

Use Cases

Areas:

On-device computing, Instruction following

Applications:

Summarization, Text rewriting, Function calling

Primary Use Cases:

English content generation, Instruction following

Limitations:

Models primarily understand and generate English., Content may not be factually accurate or logically consistent., Presence of biases inherent to training data.

Considerations:

Models are assistive tools and should not be used as definitive information sources.

Additional Notes

Memory footprint of the 135M model is 723.56 MB when loaded.

Supported Languages

en (High proficiency)

Training Details

Data Sources:

FineWeb-Edu, DCLM, The Stack, UltraFeedback

Data Volume:

2 trillion tokens

Methodology:

Direct Preference Optimization (DPO), Supervised Fine-tuning (SFT)

Context Length:

8000

Hardware Used:

64 H100 GPUs

Model Architecture:

Transformer decoder

Input Output

Input Format:

Token sequences encoded with a tokenizer

Accepted Modalities:

text

Output Format:

Generated token sequences

Performance Tips:

Use multiple GPUs and specific precision settings (e.g., bfloat16) for optimal performance

LLM Name	SmolLM2 135M
Repository 🤗	https://huggingface.co/HuggingFaceTB/SmolLM2-135M
Model Size	135m
Required VRAM	0.3 GB
Updated	2026-07-22
Maintainer	HuggingFaceTB
Model Type	llama
Model Files	0.3 GB
Supported Languages	en
Model Architecture	LlamaForCausalLM
License	apache-2.0
Context Length	8192
Model Max Length	8192
Transformers Version	4.40.1
Tokenizer Class	GPT2Tokenizer
Vocabulary Size	49152
Torch Data Type	bfloat16

Rank the SmolLM2 135M capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Quantized Models of the SmolLM2 135M

Model	Likes	Downloads	VRAM
SmolLM2 135M Bnb 4bit	4	233	0 GB

Best Alternatives to SmolLM2 135M

Best Alternatives	Context / RAM	Downloads	Likes
Digest Sft2	8K / 0.5 GB	178	0
SmolLM2 135M Instruct	8K / 0.3 GB	1857297	388
SmolLM2 135M Reasoning 5K	8K / 0.3 GB	276	3
... Subtitle Postprocessor Merged	8K / 0.5 GB	16	1
Bori 2 135M Base	8K / 0 GB	10	0
Nuevamol 135M Base	8K / 0.3 GB	5	0
SmolLM2 135M Instruction Tuned	8K / 0.3 GB	10	0
Smollm2 Helpbot 135M	8K / 0.3 GB	5	1
BokantLM0.1 135M Deepseek	8K / 0.5 GB	14	0
SmolLM2 135M Mofbandgap	8K / 0.3 GB	7	0

Note: green Score (e.g. "73.2") means that the model is better than HuggingFaceTB/SmolLM2-135M.