thelamapi/next-14b

1---2language:3- tr4- en5- de6- es7- fr8- ru9- zh10- ja11- ko12license: mit13tags:14- turkish15- türkiye16- reasoning17- ai18- lamapi19- gemma320- next21- next-x122- text-generation23- open-source24- 14b25- large-language-model26- llm27- transformer28- artificial-intelligence29- machine-learning30- nlp31- multilingual32- instruction-tuned33- chat34- generative-ai35- optimized36- trl37- sft38- cognitive39- analytical40- enterprise41pipeline_tag: text-generation42datasets:43- mlabonne/FineTome-100k44- CognitiveKernel/CognitiveKernel-Pro-SFT45- OpenSPG/KAG-Thinker-training-dataset46- Gryphe/ChatGPT-4o-Writing-Prompts47- QuixiAI/dolphin-r148- uclanlp/Brief-Pro49library_name: transformers50---51 52<img src='assets/banner.png'>53 54# 🧠 Next 14B (l310)55 56### *Türkiye’s First Reasoning-Capable AI Model — Logical, Analytical, and Enterprise-Ready*57 58[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://opensource.org/licenses/MIT)59[![Language: Multilingual](https://img.shields.io/badge/Language-Multilingual-red.svg)]()60[![HuggingFace](https://img.shields.io/badge/🤗-Lamapi/Next--14B-orange.svg)](https://huggingface.co/Lamapi/next-14b)61[![Discord](https://cdn.modrinth.com/data/cached_images/e84c69448cbf878a167f996d63e1a253437fcea2.png)](https://discord.gg/XgH4EpyPD2)62 63---64 65## 📖 Overview66 67**Next 14B** is a **14-billion parameter large language model (LLM)** built upon **Qwen 3 architecture**, trained to achieve **superior reasoning and analytical capabilities**.  68It is **Türkiye’s first reasoning-capable AI model**, designed to think, infer, and make decisions — **not just respond**.69 70Unlike vision-based models, **Next 14B focuses on pure cognitive performance**, mastering complex problem solving, abstract logic, and human-level understanding in both **Turkish and English**.71 72---73 74## ⚡ Highlights75 76- 🇹🇷 **Türkiye’s first reasoning-capable AI model**77- 🧠 **Advanced logical, analytical, and inferential reasoning**78- 🌍 **High multilingual understanding (Turkish, English, and beyond)**79- 🏢 **Enterprise-grade stability and consistency**80- 💬 **Instruction-tuned for dialogue, problem solving, and analysis**81 82---83 84## 📊 Benchmark Performance85 86<table>87  <thead>88    <tr>89      <th>Model</th>90      <th>MMLU (5-shot) %</th>91      <th>MMLU-Pro %</th>92      <th>GSM8K %</th>93      <th>MATH %</th>94    </tr>95  </thead>96  <tbody>97    <tr>98      <td><strong>Next 14B (Thinking)</strong></td>99      <td><strong>94.6</strong></td>100      <td><strong>93.2</strong></td>101      <td><strong>98.8</strong></td>102      <td>92.7</td>103    </tr>104    <tr>105      <td>Next 12B</td>106      <td>92.7</td>107      <td>84.4</td>108      <td>95.3</td>109      <td>87.2</td>110    </tr>111    <tr class="next">112      <td>Next 8B (Thinking)</td>113      <td>91.0</td>114      <td>88.5</td>115      <td>96.2</td>116      <td>88.0</td>117    </tr>118    <tr>119      <td>GPT-5</td>120      <td>92.5</td>121      <td>87.0</td>122      <td>98.4</td>123      <td><strong>96.0</strong></td>124    </tr>125    <tr>126      <td>Claude Opus 4.1 (Thinking)</td>127      <td>~92.0</td>128      <td>87.8</td>129      <td>84.7</td>130      <td>95.4</td>131    </tr>132  </tbody>133</table>134 135---136 137## 🚀 Installation & Usage138 139```python140from transformers import AutoTokenizer, AutoModelForCausalLM141import torch142 143model_id = "Lamapi/next-14b"144 145tokenizer = AutoTokenizer.from_pretrained(model_id)146model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")147 148messages = [149    {"role": "system", "content": "You are Next-X1, a reasoning-capable AI assistant created by Lamapi. You think deeply, reason logically, and always answer concisely. Proudly made in Turkey."},150    {"role": "user", "content": "Explain why the sky appears blue using logical reasoning."}151]152 153prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)154inputs = tokenizer(prompt, return_tensors="pt").to(model.device)155 156outputs = model.generate(**inputs, max_new_tokens=150)157print(tokenizer.decode(outputs[0], skip_special_tokens=True))158````159 160---161 162## 🧩 Key Features163 164| Feature                                       | Description                                                                    |165| --------------------------------------------- | ------------------------------------------------------------------------------ |166| 🧠 **Advanced Reasoning**                     | Excels in abstract logic, critical thinking, and long-form analysis.           |167| 🇹🇷 **Cultural & Multilingual Intelligence** | Deep Turkish understanding, alongside fluent English and 30+ languages.        |168| ⚙️ **Optimized for Efficiency**               | Available in quantized formats (Q8_0, Q4_K_M, FP16).                           |169| 🧮 **Mathematical & Analytical Skill**        | Performs exceptionally in structured problem solving and scientific reasoning. |170| 🧩 **Non-Vision Architecture**                | Focused purely on cognitive and linguistic understanding.                      |171| 🏢 **Enterprise Reliability**                 | Consistent, interpretable outputs for professional use cases.                  |172 173---174 175## 📐 Model Specifications176 177| Specification     | Details                                                            |178| ----------------- | ------------------------------------------------------------------ |179| **Base Model**    | Qwen 3                                                             |180| **Parameters**    | 14 Billion                                                         |181| **Architecture**  | Transformer (Causal LLM)                                           |182| **Modalities**    | Text-only                                                          |183| **Fine-Tuning**   | Instruction-tuned and reinforced with cognitive reasoning datasets |184| **Optimizations** | Quantization-ready, FP16 support                                   |185| **Primary Focus** | Reasoning, logic, decision-making, and language understanding      |186 187---188 189## 🎯 Ideal Use Cases190 191* **Analytical Chatbots** for business and enterprise logic192* **Research Assistance** — scientific, legal, or data-heavy reasoning193* **Education & Tutoring** — explain concepts step-by-step194* **Creative Writing** — coherent story logic and worldbuilding195* **Code & Algorithm Design** — reasoning-based code generation196* **Decision Support Systems** — scenario evaluation and inference197 198---199 200## 💡 Performance Highlights201 202* **Superior Reasoning:** Outperforms previous-generation 12B models in logic-based benchmarks.203* **Robust Mathematical Understanding:** Handles symbolic reasoning and complex equations.204* **Consistent Long-Context Memory:** Capable of tracking context across multi-turn conversations.205* **Professional Reliability:** Built for critical enterprise and research applications.206 207---208 209## 📄 License210 211Licensed under the **MIT License** — free for commercial and non-commercial use. Attribution is appreciated.212 213---214 215## 📞 Contact & Support216 217* 📧 **Email:** [lamapicontact@gmail.com](mailto:lamapicontact@gmail.com)218* 🤗 **HuggingFace:** [Lamapi](https://huggingface.co/Lamapi)219 220---221 222> **Next 14B** — Türkiye’s first *reasoning-capable* large language model, combining **logical depth**, **analytical intelligence**, and **enterprise reliability**.223 224[![Follow on HuggingFace](https://img.shields.io/badge/Follow-HuggingFace-yellow?logo=huggingface)](https://huggingface.co/Lamapi)225 226```