# SmolLM2-135M-Instruct by Hugging Face Smol Models Research Source: https://savrn.com/models/smollm2-135m-instruct Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- ## Runs On What it takes to serve SmolLM2-135M-Instruct (135M parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them. | Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits | | --- | --- | --- | --- | --- | --- | | 16-bit | 0.3 GB | 0.3 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 8-bit | 0.1 GB | 0.2 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 4-bit | 0.1 GB | 0.1 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the [SAVRN Index](https://savrn.com/ai-index/pricing/gpus), read Oct 25, 2026. [SmolLM2-135M-Instruct on every accelerator the SAVRN Index prices, at every precision](https://savrn.com/models/smollm2-135m-instruct/gpus) ## SAVRN's Notes on SmolLM2-135M-Instruct We keep a slot in every facility for very small text models, and SmolLM2-135M-Instruct from Hugging Face Smol Models Research is the kind that fills it. At 135 million parameters and 0.3 GB of memory at 16-bit, it fits on any device, and the cheapest Index host, one MI300X with 192 GB at $1.85 an hour on-demand, would hold hundreds of instances. Its work is on-device text generation: short instructions, classification, routing, the traffic you do not send to a large model. Context is 8,192 tokens, enough for a page, not a contract. Apache 2.0 allows commercial use, and our record shows it derived from and quantized from the SmolLM2-135M base, described in arXiv:2502.02737. Before deploying, confirm the base model's terms travel with it, match the September 22, 2025 update to the version you pulled, and note the onnx export for CPU serving. SAVRN Research, 2026-09-18 ## Model Card By Hugging Face Smol Models Research, published under apache-2.0, revision 12fd25f77366. ### SmolLM2 ### Table of Contents 1. Model Summary 2. Limitations 3. Training 4. License 5. Citation ### Model Summary SmolLM2 is a family of compact language models available in three size: 135M, 360M, and 1.7B parameters. They are capable of solving a wide range of tasks while being lightweight enough to run on-device. More details in our paper https://arxiv.org/abs/2502.02737 SmolLM2 demonstrates significant advances over its predecessor SmolLM1, particularly in instruction following, knowledge, reasoning. The 135M model was trained on 2 trillion tokens using a diverse dataset combination: FineWeb-Edu, DCLM, The Stack, along with new filtered datasets we curated and will release soon. We developed the instruct version through supervised fine-tuning (SFT) using a combination of public datasets and our own curated datasets. We then applied Direct Preference Optimization (DPO) using [UltraFeedback](https://huggingface.co/datasets/HuggingFaceH4/ultrafeedback_binarized). [Read the full model card (610 words)](https://savrn.com/models/smollm2-135m-instruct/card) ## Configuration Architecture LlamaForCausalLM Context length (tokens) 8,192 Layers 30 Hidden size 576 Feed-forward size 1,536 Attention heads 9 Key/value heads 3 Vocabulary size 49,152 RoPE base 100,000 Stored precision bfloat16 Model type llama ## Identity and Version Repository HuggingFaceTB/SmolLM2-135M-Instruct Publisher Hugging Face Smol Models Research Task Text generation Modality Text Library transformers Parameters 135M parameters Languages en Revision 12fd25f77366fa6b3b4b768ec3050bf629380bac First published 2024-10-31 Last updated 2025-09-22 ## Files and Weights 25 files, 2.0 GB in total. The weights are 10 files totalling 2.0 GB in bin, onnx, safetensors. Weights10 files · 2.0 GB Configuration7 files · 60.5 KB Tokenizer4 files · 3.4 MB Documentation1 file · 6.8 KB Other2 files · 79.0 KB Repository1 file · 1.5 KB Every file | File | Type | Size | SHA-256 | | --- | --- | --- | --- | | model.safetensors | Weights | 269.1 MB | 5af571cbf074 | | onnx/model.onnx | Weights | 540.3 MB | 1e790f1d6fac | | onnx/model_bnb4.onnx | Weights | 175.4 MB | bc32b446540f | | onnx/model_fp16.onnx | Weights | 270.3 MB | e85e706ce14a | | onnx/model_int8.onnx | Weights | 137.1 MB | a7c33f9ef85d | | onnx/model_q4.onnx | Weights | 182.1 MB | 933577110303 | | onnx/model_q4f16.onnx | Weights | 117.7 MB | 9358cd4ce037 | | onnx/model_quantized.onnx | Weights | 137.1 MB | ecc1a19eece6 | | onnx/model_uint8.onnx | Weights | 137.1 MB | ecc1a19eece6 | | training_args.bin | Weights | 6.5 KB | 90f508e5769b | | all_results.json | Configuration | 783 B | — | | config.json | Configuration | 861 B | — | | eval_results.json | Configuration | 586 B | — | | generation_config.json | Configuration | 132 B | — | | special_tokens_map.json | Configuration | 655 B | — | | train_results.json | Configuration | 232 B | — | | trainer_state.json | Configuration | 57.2 KB | — | | README.md | Documentation | 6.8 KB | — | | runs/Oct31_10-14-22_ip-26-0-174-36/events.out.tfevents.1730370128.ip-26-0-174-36.3239327.0 | Other | 78.2 KB | 774faae98f36 | | runs/Oct31_10-14-22_ip-26-0-174-36/events.out.tfevents.1730376056.ip-26-0-174-36.3239327.1 | Other | 828 B | 043a35029ecb | | .gitattributes | Repository | 1.5 KB | — | | merges.txt | Tokenizer | 466.4 KB | — | | tokenizer.json | Tokenizer | 2.1 MB | — | | tokenizer_config.json | Tokenizer | 3.8 KB | — | | vocab.json | Tokenizer | 800.7 KB | — | ## License and Download License apache-2.0 Access Open weights, no gate Download size 2.0 GB [Download from Hugging Face Smol Models Research](https://huggingface.co/HuggingFaceTB/SmolLM2-135M-Instruct) Released by Hugging Face Smol Models Research through its official repository on Hugging Face. [Read the license](https://www.apache.org/licenses/LICENSE-2.0). ## Built From - Derived from [HuggingFaceTB/SmolLM2-135M](https://savrn.com/models/smollm2-135m) - Described by [arXiv:2502.02737](https://savrn.com/papers/smollm2-when-smol-goes-big-data-centric-training-of-a-small-language-model) - Quantized from [HuggingFaceTB/SmolLM2-135M](https://savrn.com/models/smollm2-135m) ## Memory Requirements | Precision | Weights in memory | | --- | --- | | As published | 2.0 GB | | 16-bit | 0.3 GB | | 8-bit | 0.1 GB | | 4-bit | 0.1 GB | Weights only, from the published parameter count; the key-value cache and runtime add to this. ## Built on This Model - Derived from[tuned-SmolLM2-135M-Instruct-test-s](https://savrn.com/models/tuned-smollm2-135m-instruct-test-s) - Derived from[nanoVLM-230M-packing-ab](https://savrn.com/models/nanovlm-230m-packing-ab) - Quantized from[SmolLM2-135M-Instruct-Q8_0-GGUF](https://savrn.com/models/smollm2-135m-instruct-q8-0-gguf) - Derived from[SmolLM2-135M-Instruct-Q8_0-GGUF](https://savrn.com/models/smollm2-135m-instruct-q8-0-gguf) ## Compare SmolLM2-135M-Instruct - [gpt2 vs SmolLM2-135M-Instruct](https://savrn.com/models/compare/gpt2-vs-smollm2-135m-instruct) - [pythia-160m vs SmolLM2-135M-Instruct](https://savrn.com/models/compare/pythia-160m-vs-smollm2-135m-instruct) ## Questions About SmolLM2-135M-Instruct ### How much GPU memory does SmolLM2-135M-Instruct need? About 0.3 GB at 16-bit and 0.1 GB at 4-bit: the weights (135M parameters) plus a working margin. A long context needs more. ### What is the cheapest GPU to run SmolLM2-135M-Instruct on? At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists. ### Can I use SmolLM2-135M-Instruct commercially? Yes. SmolLM2-135M-Instruct is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors. ### What is SmolLM2-135M-Instruct's context length? 8,192 tokens, from the maximum position embeddings in its published configuration. ## Similar Models Model · Text generation ### [SmolLM2-135M](https://savrn.com/models/smollm2-135m) [Hugging Face Smol Models Research](https://savrn.com/model-publishers/huggingfacetb) SmolLM2 is a family of compact language models available in three size: 135M, 360M, and 1.7B parameters. They are capable of solving a wide range of tasks while being lightweight enough to run on-device. More details in our paper: https://arxiv.org/abs/2502.02737 SmolLM2 demonstrates significant advances over its predecessor SmolLM1, particularly in instruction following, knowledge, reasoning. The 135M model was trained on 2 trillion tokens using a diverse dataset combination: FineWeb-Edu, DCLM, The Stack, along with new filtered datasets we curated and will release soon. We developed the instruct version through supervised fine-tuning (SFT) using a combination of public datasets and our… Open weights apache-2.0 135M parameters 8,192 tokens transformers [View model](https://savrn.com/models/smollm2-135m) Model · Text generation ### [tuned-SmolLM2-135M-Instruct-test-s](https://savrn.com/models/tuned-smollm2-135m-instruct-test-s) [Zied Ben hadj Amor](https://savrn.com/model-publishers/mrbhazied) This model is a fine-tuned version of HuggingFaceTB/SmolLM2-135M-Instruct on an unknown dataset. The following hyperparameters were used during training: - learningrate: 2e-05 - trainbatchsize: 1 - evalbatchsize: 1 - gradientaccumulationsteps: 4 - totaltrainbatchsize: 4 - lrschedulertype: linear - numepochs: 1 - Transformers 5.17.0 - Pytorch 2.11.0+cu130 - Datasets 4.8.5 - Tokenizers 0.23.2 Open weights apache-2.0 135M parameters 8,192 tokens transformers [View model](https://savrn.com/models/tuned-smollm2-135m-instruct-test-s) Model · Text generation ### [SmolLM2-FT-MyDataset](https://savrn.com/models/smollm2-ft-mydataset) [Roy Mazumder](https://savrn.com/model-publishers/aritroroy0) This model is a fine-tuned version of HuggingFaceTB/SmolLM2-135M on an unknown dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 4 - evalbatchsize: 8 - lrschedulertype: linear - trainingsteps: 1000 - Transformers 4.46.3 - Pytorch 2.11.0+cu130 - Datasets 4.8.5 - Tokenizers 0.20.3 Open weights apache-2.0 135M parameters 8,192 tokens transformers [View model](https://savrn.com/models/smollm2-ft-mydataset) Model · Text generation ### [SmolLM2-FT-MyDataset](https://savrn.com/models/smollm2-ft-mydataset-2) [Andrew Day](https://savrn.com/model-publishers/ad565700) Open weights 135M parameters transformers [View model](https://savrn.com/models/smollm2-ft-mydataset-2) Model · Text generation ### [Smol-GPT-Fp8](https://savrn.com/models/smol-gpt-fp8) [Gggff123](https://savrn.com/model-publishers/gggff123) This llama model was trained 2x faster with Unsloth and Huggingface's TRL library. Open weights apache-2.0 135M parameters 8,192 tokens transformers [View model](https://savrn.com/models/smol-gpt-fp8) Model · Text generation ### [Smol-GPT-RAW](https://savrn.com/models/smol-gpt-raw) [Gggff123](https://savrn.com/model-publishers/gggff123) This llama model was trained 2x faster with Unsloth and Huggingface's TRL library. Open weights apache-2.0 135M parameters 8,192 tokens transformers [View model](https://savrn.com/models/smol-gpt-raw) ## Hugging Face Smol Models Research [All models and datasets](https://savrn.com/model-publishers/huggingfacetb) ## Versions - [12fd25f77366](https://savrn.com/models/smollm2-135m-instruct/versions/12fd25f77366) · current 2026-10-25 ## Explore More - [All text generation models](https://savrn.com/models/tasks/text-generation) - [All models under apache-2.0](https://savrn.com/models/licenses/apache-2-0) - [Model comparisons](https://savrn.com/models/comparisons) - [The model directory](https://savrn.com/models) - [Open model prices by host](https://savrn.com/ai-index/pricing/open-models) ## Source - Repository metadata, read 2026-10-19. - [Hugging Face record](https://huggingface.co/HuggingFaceTB/SmolLM2-135M-Instruct) - [How the hub is built](https://savrn.com/model-hub/methodology)