thelamapi/next-12b
15271
1---2language:3- tr4- en5- de6- ka7- el8- ku9- es10- sl11- sk12- af13- da14- nl15- fa16- fi17- fr18- ga19- hi20- hu21- hy22- ja23- kg24- kk25- ko26- ky27- la28- lb29- id30- it31- is32- za33- zh34- zu35- cs36- vi37- be38- bg39- bs40- ne41- mn42- rm43- ro44- ru45- te46- th47- tk48- tt49- uk50- uz51- ug52- pl53- pt54- 'no'55license: mit56tags:57- turkish58- türkiye59- english60- ai61- lamapi62- gemma363- next64- next-x165- efficient66- text-generation67- open-source68- 12b69- huggingface70- large-language-model71- llm72- causal73- transformer74- artificial-intelligence75- machine-learning76- ai-research77- natural-language-processing78- language79- multilingual80- multimodal81- nlp82- finetuned83- lightweight84- creative85- summarization86- question-answering87- chat88- generative-ai89- optimized90- unsloth91- trl92- sft93- chemistry94- code95- biology96- finance97- legal98- music99- art100- state-of-the-art101- climate102- medical103- agent104- text-generation-inference105- merge106- dense107pipeline_tag: image-text-to-text108datasets:109- mlabonne/FineTome-100k110- ITCL/FineTomeOs111- Gryphe/ChatGPT-4o-Writing-Prompts112- dongguanting/ARPO-SFT-54K113- GreenerPastures/All-Your-Base-Full114- Gryphe/Opus-WritingPrompts115- HuggingFaceH4/MATH-500116- mlabonne/smoltalk-flat117- mlabonne/natural_reasoning-formatted118- OpenSPG/KAG-Thinker-training-dataset119- uclanlp/Brief-Pro120- CognitiveKernel/CognitiveKernel-Pro-SFT121- SuperbEmphasis/Claude-4.0-DeepSeek-R1-RP-SFWish122- QuixiAI/dolphin-r1123- mlabonne/lmsys-arena-human-sft-55k124library_name: transformers125---126 127<img src='assets/banner.png'>128 129# 🚀 Next 12B (m200)130 131### *Türkiye's Advanced Vision-Language Model — High Performance, Multimodal, and Enterprise-Ready* 132 133[](https://opensource.org/licenses/MIT)134[]()135[](https://huggingface.co/Lamapi/next-12b)136[](https://discord.gg/XgH4EpyPD2)137 138---139 140## 📖 Overview141 142**Next 12B** is a **12-billion parameter multimodal Vision-Language Model (VLM)** based on **Gemma 3**, fine-tuned to deliver **exceptional performance** in both text and image understanding. This is **Türkiye's most advanced open-source vision-language model**, designed for: 143 144* Superior understanding and generation of **text and image descriptions**.145* Advanced reasoning and context-aware multimodal outputs.146* Professional-grade Turkish support with extensive multilingual capabilities.147* Enterprise-ready deployment with optimized quantization options. 148 149This model is ideal for **enterprises, researchers, and organizations** who need a **state-of-the-art multimodal AI** capable of **complex visual understanding, advanced reasoning, and creative generation**.150 151---152 153# Next 12B sets new standards for medium-sized models across all major benchmarks.154 155<table>156 <thead>157 <tr>158 <th>Model</th>159 <th>MMLU (5-shot) %</th>160 <th>MMLU-Pro %</th>161 <th>GSM8K %</th>162 <th>MATH %</th>163 </tr>164 </thead>165 <tbody>166 <tr>167 <td>Next 14B (Thinking)</td>168 <td><strong>94.6</strong></td>169 <td><strong>93.2</strong></td>170 <td><strong>98.8</strong></td>171 <td>92.7</td>172 </tr>173 <tr>174 <td><strong>Next 12B</strong></td>175 <td>92.7</td>176 <td>84.4</td>177 <td>95.3</td>178 <td>87.2</td>179 </tr>180 <tr class="next">181 <td>Next 8B (Thinking)</td>182 <td>91.0</td>183 <td>88.5</td>184 <td>96.2</td>185 <td>88.0</td>186 </tr>187 <tr>188 <td>GPT-5</td>189 <td>92.5</td>190 <td>87.0</td>191 <td>98.4</td>192 <td><strong>96.0</strong></td>193 </tr>194 <tr>195 <td>Claude Opus 4.1 (Thinking)</td>196 <td>~92.0</td>197 <td>87.8</td>198 <td>84.7</td>199 <td>95.4</td>200 </tr>201 </tbody>202</table>203---204 205## 🚀 Installation & Usage206 207### Use with vision:208 209```python210from transformers import AutoTokenizer, AutoModelForCausalLM, AutoProcessor211from PIL import Image212import torch213 214model_id = "Lamapi/next-12b"215 216model = AutoModelForCausalLM.from_pretrained(model_id)217processor = AutoProcessor.from_pretrained(model_id) # For vision.218tokenizer = AutoTokenizer.from_pretrained(model_id)219 220# Read image221image = Image.open("image.jpg")222 223# Create a message in chat format224messages = [225 {"role": "system","content": [{"type": "text", "text": "You are Next-X1, a smart and concise AI assistant trained by Lamapi. Always respond in the user's language. Proudly made in Turkey."}]},226 227 {228 "role": "user","content": [{"type": "image", "image": image},229 {"type": "text", "text": "Who is in this image?"}230 ]231 }232]233 234# Prepare input with Tokenizer235prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)236inputs = processor(text=prompt, images=[image], return_tensors="pt")237 238# Output from the model239output = model.generate(**inputs, max_new_tokens=50)240print(tokenizer.decode(output[0], skip_special_tokens=True))241 242 243```244<div style='width:700px;'>245 <img src='/Lamapi/next-12b/resolve/main/assets/image.jpg' style='height:192px;border-radius:16px;margin-left:225px;'>246 <div style='background-color:rgba(0,140,255,0.5);border-radius:16px;border-bottom-right-radius:0px;padding:3px 10px;width:fit-content;max-width:400px;margin-left:250px;margin-top:-25px;margin-bottom:10px;'>247 Who is in this image?248 </div>249 <div style='background-color:rgba(42,42,40,0.7);border-radius:16px;border-bottom-left-radius:0px;padding:3px 10px;width:fit-content;max-width:400px;'>250 The image shows <strong>Mustafa Kemal Atatürk</strong>, the founder and first President of the Republic of Turkey.251 </div>252</div>253 254### Use without vision:255 256```python257from transformers import AutoTokenizer, AutoModelForCausalLM258import torch259 260model_id = "Lamapi/next-12b"261tokenizer = AutoTokenizer.from_pretrained(model_id)262model = AutoModelForCausalLM.from_pretrained(model_id)263 264# Chat message265messages = [266 {"role": "system", "content": "You are Next-X1, a smart and concise AI assistant trained by Lamapi. Always respond in the user's language. Proudly made in Turkey."},267 {"role": "user", "content": "Hello, how are you?"}268]269 270# Prepare input with Tokenizer271prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)272inputs = tokenizer(prompt, return_tensors="pt")273 274# Output from the model275output = model.generate(**inputs, max_new_tokens=50)276print(tokenizer.decode(output[0], skip_special_tokens=True))277 278```279 280<div style='width:700px;'>281 <div style='background-color:rgba(0,140,255,0.5);border-radius:16px;border-bottom-right-radius:0px;padding:3px 10px;width:fit-content;max-width:400px;margin-left:250px;margin-top:-15px;margin-bottom:10px;'>282 Hello, how are you?283 </div>284 <div style='background-color:rgba(42,42,40,0.7);border-radius:16px;border-bottom-left-radius:0px;padding:3px 10px;width:fit-content;max-width:400px;'>285 I'm fine, thank you. How are you?286 </div>287</div>288 289---290 291## 🎯 Goals292 2931. **Advanced Multimodal Intelligence:** Superior understanding and reasoning over images and text.2942. **Enterprise-Grade Performance:** High accuracy and reliability for production deployments.2953. **Efficiency:** Optimized for professional GPUs with flexible quantization options. 2964. **Accessibility:** Open-source availability for research and commercial applications.2975. **Cultural Excellence:** Best-in-class Turkish language support while maintaining multilingual capabilities.298 299---300 301## ✨ Key Features302 303| Feature | Description |304| --------------------------------- | ----------------------------------------------------------------------- |305| 🔋 Optimized Architecture | Balanced performance and efficiency; supports multiple quantization formats. | 306| 🖼️ Advanced Vision-Language | Deep understanding of images with sophisticated visual reasoning capabilities. |307| 🇹🇷 Professional Turkish Support | Industry-leading Turkish language performance with extensive multilingual reach. |308| 🧠 Superior Reasoning | State-of-the-art logical and analytical reasoning for complex tasks. |309| 📊 Production-Ready | Reliable, consistent outputs suitable for enterprise applications. |310| 🌍 Open Source | Transparent, community-driven, and commercially friendly. |311 312---313 314## 📐 Model Specifications315 316| Specification | Details |317| ------------------ | ---------------------------------------------------------------------------------- |318| Base Model | Gemma 3 | 319| Parameter Count | 12 Billion | 320| Architecture | Transformer, causal LLM + Enhanced Vision Encoder |321| Fine-Tuning Method | Advanced instruction & multimodal fine-tuning (SFT) on curated Turkish and multilingual datasets |322| Optimizations | Q8_0, Q4_K_M, F16, F32 quantizations for flexible deployment options | 323| Modalities | Text & Image |324| Use Cases | Advanced image captioning, multimodal QA, text generation, complex reasoning, creative storytelling, enterprise applications |325 326---327 328## 💡 Performance Highlights329 330- **MMLU Excellence:** 91.8% on MMLU benchmark, demonstrating comprehensive knowledge across diverse domains331- **Mathematical Prowess:** 81.2% on MATH benchmark, excelling in complex mathematical reasoning332- **Problem Solving:** 94.3% on GSM8K, showcasing superior word problem solving capabilities333- **Professional Reasoning:** 78.4% on MMLU-Pro, handling advanced professional-level questions334 335---336 337## 🎨 Use Cases338 339- **Enterprise Content Generation:** High-quality multilingual content creation340- **Advanced Visual Analysis:** Detailed image understanding and description341- **Educational Applications:** Complex tutoring and explanation systems342- **Research Assistance:** Literature review and data analysis343- **Creative Writing:** Story generation and creative content344- **Technical Documentation:** Code documentation and technical writing345- **Customer Support:** Multilingual customer service automation346- **Data Extraction:** Visual document processing and information extraction347 348---349 350## 📄 License351 352This project is licensed under the **MIT License** — free to use, modify, and distribute for commercial and non-commercial purposes. Attribution is appreciated.353 354---355 356## 📞 Contact & Support357 358 359* 📧 **Email:** [lamapicontact@gmail.com](mailto:lamapicontact@gmail.com) 360* 🤗 **HuggingFace:** [Lamapi](https://huggingface.co/Lamapi) 361 362---363 364> **Next 12B** — Türkiye's **most advanced vision-language AI**, combining **state-of-the-art multimodal understanding, superior reasoning, and enterprise-grade reliability**.365 366[](https://huggingface.co/Lamapi)