LLM EXPLORER
 LLM Explorer: A Curated Large Language Model Directory and Analytics
Dark Theme
  Agents   Benchmarks   LLM  List    LLM Hosting    LLM Leaderboards    Blog    Newsfeed    Advertise

K2S3 V0.1 by Changgil

 »  All LLMs  »  Changgil  »  K2S3 V0.1   URL Share it on

  6   

K2S3 V0.1 is an open-source language model by Changgil. Features: 14.4b LLM, VRAM: 28.7GB, Context: 32K, License: cc-by-nc-4.0, HF Score: 64.2, LLM Explorer Score: 0.19, Arc: 60.3, HellaSwag: 79.7, MMLU: 59.3, TruthfulQA: 47.8, WinoGrande: 76.1, GSM8K: 61.7.

  En   Endpoints compatible   Ko   Mistral   Region:us   Safetensors   Sharded   Tensorflow
Model Card on HF 🤗: https://huggingface.co/Changgil/K2S3-v0.1 

K2S3 V0.1 Benchmarks

MMLU Pro: 17.36
GPQA: 1.9
MUSR: 7.84
BBH: 24.56
IFEval: 32.77 vs 88 (so35)-62.8%
ARC: 60.32 vs 96.7 (so35)-37.6%
HellaSwag: 79.65 vs 95.3 (gpt4)-16.4%
MMLU: 59.3 vs 88.3 (so35)-32.8%
TruthfulQA: 47.81 vs 59 (gpt4)-19%
WinoGrande: 76.09 vs 87.5 (gpt4)-13%
GSM8K: 61.71 vs 96.4 (so35)-36%
MATH Lvl 5: 4.61
LLME Score: 0.18884
nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").
K2S3 V0.1 (Changgil/K2S3-v0.1)
🌟 Advertise your project 🚀

K2S3 V0.1 Parameters and Internals

Additional Notes 
The model has been enhanced with Korean vocabulary and merges to the tokenizer.
Supported Languages 
en (English), ko (Korean)
Training Details 
Data Sources:
alpaca-gpt4-data, The OpenOrca Dataset
Methodology:
Depth up scaling and full parameter tuning with SFT (Supervised Fine-Tuning)
Hardware Used:
A100 (80G*2EA) GPUs
Model Architecture:
Enhanced base model with depth up scaling
LLM NameK2S3 V0.1
Repository 🤗https://huggingface.co/Changgil/K2S3-v0.1 
Model Size14.4b
Required VRAM28.7 GB
Updated2026-04-09
MaintainerChanggil
Model Typemistral
Model Files  5.0 GB: 1-of-6   4.9 GB: 2-of-6   5.0 GB: 3-of-6   4.9 GB: 4-of-6   4.9 GB: 5-of-6   4.0 GB: 6-of-6
Supported Languagesen ko
Model ArchitectureMistralForCausalLM
Licensecc-by-nc-4.0
Context Length32768
Model Max Length32768
Transformers Version4.38.0.dev0
Tokenizer ClassLlamaTokenizer
Padding Token<unk>
Vocabulary Size48000
Torch Data Typefloat16

Best Alternatives to K2S3 V0.1

Best Alternatives
Context / RAM
Downloads
Likes
Krutrim 2 Instruct1000K / 49.3 GB14736
Ft V1 Violet1000K / 24.5 GB50
Mistral Large Instruct 2407128K / 226.7 GB7491859
Tiny Random MistralForCausalLM128K / 0 GB32521
Winterreise M732K / 14.4 GB00
Frostwind V2.1 M732K / 14.4 GB00
MistralLite32K / 14.4 GB11345435
MistralLite32K / 14.4 GB61777430
...ydaz Web AI Reasoner BaseModel32K / 14.4 GB01
Tess XS V1.3 Yarn 128K32K / 14.5 GB332013
Note: green Score (e.g. "73.2") means that the model is better than Changgil/K2S3-v0.1.

Rank the K2S3 V0.1 Capabilities

🆘 Have you tried this model? Rate its performance. This feedback would greatly assist ML community in identifying the most suitable model for their needs. Your contribution really does make a difference! 🌟

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  

What open-source LLMs or SLMs are you in search of? 52721 in total.

Email us: info@extractum.io. Our Privacy Policy  | Terms and Conditions  | Suggest an improvement.
Our Social Media →  
Original data from HuggingFace, OpenCompass and various public git repos.
Check out Ag3ntum — our secure, self-hosted AI agent for server management.
Release v20260328a
  Support LLM Explorer