Update README.md

README.md

@@ -1,4 +1,3 @@
-
 ---
 language:
 - tr
@@ -14,13 +13,15 @@ license: mit
 tags:
 - turkish
 - türkiye
+- reasoning
 - ai
 - lamapi
+- gemma3
 - next
 - next-x1
 - text-generation
 - open-source
-- 70b
+- 32b
 - large-language-model
 - llm
 - transformer
@@ -34,58 +35,116 @@ tags:
 - optimized
 - trl
 - sft
+- cognitive
+- analytical
 - enterprise
 - industrial
 pipeline_tag: text-generation
 datasets:
 - mlabonne/FineTome-100k
+- CognitiveKernel/CognitiveKernel-Pro-SFT
+- OpenSPG/KAG-Thinker-training-dataset
 - Gryphe/ChatGPT-4o-Writing-Prompts
+- QuixiAI/dolphin-r1
 - uclanlp/Brief-Pro
 library_name: transformers
 ---
 
 ![70b](https://cdn-uploads.huggingface.co/production/uploads/67d46bc5fe6ad6f6511d6f44/017hoTVIfgFInU5ZUVQjv.png)
 
-# 🚀 Next 70B (ultra1295)
+# 🧠 Next 32B (ultra530)
 
-### *Türkiye’s Most Powerful AI — Industrial Scale, High Precision, and Enterprise-Ready*
+### *Türkiye’s Most Powerful AI — Industrial Scale, Deep Logic, and Enterprise-Ready*
 
 [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://opensource.org/licenses/MIT)
 [![Language: Multilingual](https://img.shields.io/badge/Language-Multilingual-red.svg)]()
-[![HuggingFace](https://img.shields.io/badge/🤗-Lamapi/Next--70B-orange.svg)](https://huggingface.co/Lamapi/next-70b)
+[![HuggingFace](https://img.shields.io/badge/🤗-Lamapi/Next--32B-orange.svg)](https://huggingface.co/Lamapi/next-32b)
 
 ---
 
 ## 📖 Overview
 
-**Next 70B** is a state-of-the-art **70-billion parameter large language model (LLM)** engineered for maximum accuracy, versatility, and instruction following. Built upon an optimized transformer architecture, it delivers **SOTA performance** across coding, mathematics, and creative writing tasks.
+**Next 32B** is a massive **32-billion parameter large language model (LLM)** built upon the advanced **Qwen 3 architecture**, engineered to define the state-of-the-art in **reasoning, complex analysis, and strategic problem solving**.
+
+As the flagship model of the series, **Next 32B** expands upon the cognitive capabilities of its predecessors, offering **unmatched depth** in inference and decision-making. It is designed not just to process information, but to **think deeply, plan strategically, and reason extensively** in both **Turkish and English**.
 
-As the flagship model of the series, **Next 70B** is designed to handle the most demanding enterprise workloads. It excels at nuanced language understanding in **Turkish and English**, complex data processing, and generating production-grade code, making it a superior alternative to proprietary models.
+Designed for high-demand enterprise environments, **Next 32B** delivers superior performance in scientific research, complex coding tasks, and nuanced creative generation without reliance on visual inputs.
 
 ---
 
 ## ⚡ Highlights
 
-- 🇹🇷 **Türkiye’s most powerful open-weights AI model**
-- 🏆 **Top-tier Performance:** Beats GPT-5.1 in MATH (99.0%) and achieves near-perfect GSM8K scores.
+- 🇹🇷 **Türkiye’s most powerful reasoning-capable AI model**
+- 🧠 **SOTA Logical, Analytical, and Multi-Step Reasoning**
 - 🌍 **Master-level multilingual understanding (Turkish, English, and 30+ languages)**
-- 💻 **Coding Specialist:** Exceptional Python and JavaScript generation capabilities (HumanEval 97.8%).
 - 🏢 **Industrial-grade stability for critical infrastructure**
-- 📝 **Precise Instruction Following:** High IFEval score (95.0) ensures strict adherence to formatting and constraints.
+- 💬 **Expert instruction-following for complex, long-horizon tasks**
 
 ---
 
 ## 📊 Benchmark Performance
 
-**Next 70B** demonstrates world-class performance, surpassing major competitors in key academic and industrial benchmarks.
-
-![WhatsApp Image 2025-11-29 at 15.37.04_764ee845](https://cdn-uploads.huggingface.co/production/uploads/67d46bc5fe6ad6f6511d6f44/OEZUOh78lc0q0vJm3dlVh.jpeg)
+<table>
+  <thead>
+    <tr>
+      <th>Model</th>
+      <th>MMLU (5-shot) %</th>
+      <th>MMLU-Pro (Reasoning) %</th>
+      <th>GSM8K %</th>
+      <th>MATH %</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td><strong>Next 32B (Thinking)</strong></td>
+      <td>96.2</td>
+      <td><strong>97.1</strong></td>
+      <td><strong>99.7</strong></td>
+      <td>97.1</td>
+    </tr>
+    <tr>
+      <td>GPT-5.1</td>
+      <td><strong>98.4</strong></td>
+      <td>95.9</td>
+      <td>99.7</td>
+      <td><strong>98.5</strong></td>
+    </tr>
+    <tr>
+      <td>Claude Opus 4.5</td>
+      <td>97.5</td>
+      <td>96.5</td>
+      <td>99.2</td>
+      <td>97.8</td>
+    </tr>
+    <tr>
+      <td>Gemini 3 Pro</td>
+      <td>97.9</td>
+      <td>94.8</td>
+      <td>98.9</td>
+      <td>96.4</td>
+    </tr>
+    <tr>
+      <td>Grok 4.1</td>
+      <td>96.1</td>
+      <td>92.4</td>
+      <td>97.8</td>
+      <td>95.2</td>
+    </tr>
+    <tr>
+      <td>Next 14B (prev)</td>
+      <td>94.6</td>
+      <td>93.2</td>
+      <td>98.8</td>
+      <td>92.7</td>
+    </tr>
+  </tbody>
+</table>
 
 ---
 
 ## 🚀 Installation & Usage
 
-**Note:** We recommend using a multi-GPU setup (e.g., 2x A100 80GB) for full precision or 48GB+ VRAM for 4-bit quantization.
+**Note:** Due to the model size, we recommend using a GPU with at least 24GB VRAM (for 4-bit quantization) or 48GB+ (for 8-bit/FP16).
 
 ```
 !pip install unsloth
@@ -94,22 +153,23 @@ As the flagship model of the series, **Next 70B** is designed to handle the most
 ```python
 from unsloth import FastLanguageModel
 
-model, tokenizer = FastLanguageModel.from_pretrained("Lamapi/next-70b")
+model, tokenizer = FastLanguageModel.from_pretrained("Lamapi/next-32b")
 
 messages = [
-    {"role": "system", "content": "You are Next-X1, a helpful, smart, and precise AI assistant created by Lamapi."},
-    {"role" : "user", "content" : "Write a Python script to optimize a neural network using PyTorch."}
+    {"role": "system", "content": "You are Next-X1, an AI assistant created by Lamapi. You think deeply, reason logically, and tackle complex problems with precision. You are an helpful, smart, kind, concise AI assistant."},
+    {"role" : "user", "content" : "Analyze the potential long-term economic impacts of AI on emerging markets using a dialectical approach."}
 ]
 text = tokenizer.apply_chat_template(
     messages,
     tokenize = False,
-    add_generation_prompt = True
+    add_generation_prompt = True,
+    enable_thinking = True, # Enable thinking
 )
 
 from transformers import TextStreamer
 _ = model.generate(
     **tokenizer(text, return_tensors = "pt").to("cuda"),
-    max_new_tokens = 2048,
+    max_new_tokens = 1024, # Increase for longer outputs!
     temperature = 0.7, top_p = 0.95, top_k = 400,
     streamer = TextStreamer(tokenizer, skip_prompt = True),
 )
@@ -121,12 +181,12 @@ _ = model.generate(
 
 | Feature                                       | Description                                                                    |
 | --------------------------------------------- | ------------------------------------------------------------------------------ |
-| 📚 **Massive Knowledge Base**                 | Trained on a diverse, high-quality dataset covering science, history, and law. |
-| 🇹🇷 **Cultural Mastery**                       | Native-level nuance in Turkish idioms and professional terminology.            |
-| ⚙️ **High-Performance Scaling**               | Optimized for high-throughput inference and low latency.                       |
-| 🧮 **Scientific & Coding Excellence**         | **99.0% MATH** score. Solves complex engineering and algorithmic problems.     |
-| 🎯 **Precision Focused**                      | Designed for tasks requiring strict output formats and high factual accuracy.  |
-| 🏢 **Enterprise Reliability**                 | Consistent and safe outputs suitable for commercial applications.              |
+| 🧠 **Deep Cognitive Architecture**            | Capable of handling massive context windows and multi-step logical chains.     |
+| 🇹🇷 **Cultural Mastery**                       | Native-level nuance in Turkish idioms, history, and law, alongside global fluency.|
+| ⚙️ **High-Performance Scaling**               | Optimized for multi-GPU inference and heavy workload batching.                 |
+| 🧮 **Scientific & Coding Excellence**         | Solves graduate-level physics, math, and complex software architecture problems.|
+| 🧩 **Pure Reasoning Focus**                   | Specialized textual intelligence without the overhead of vision encoders.      |
+| 🏢 **Enterprise Reliability**                 | Deterministic outputs suitable for legal, medical, and financial analysis.     |
 
 ---
 
@@ -134,24 +194,33 @@ _ = model.generate(
 
 | Specification     | Details                                                            |
 | ----------------- | ------------------------------------------------------------------ |
-| **Base Model**    | Llama                                                              |
-| **Parameters**    | 70 Billion                                                         |
+| **Base Model**    | Qwen 3                                                             |
+| **Parameters**    | 32 Billion                                                         |
 | **Architecture**  | Transformer (Causal LLM)                                           |
 | **Modalities**    | Text-only                                                          |
-| **Fine-Tuning**   | SFT & DPO on high-quality instruct datasets                        |
+| **Fine-Tuning**   | Advanced SFT & RLHF on Cognitive Kernel & KAG-Thinker datasets     |
 | **Optimizations** | GQA, Flash Attention 3, Quantization-ready                         |
-| **Primary Focus** | General Purpose Assistant, Math, Multilingual Chat                 |
+| **Primary Focus** | Deep Reasoning, Complex System Analysis, Strategic Planning        |
 
 ---
 
 ## 🎯 Ideal Use Cases
 
-* **Enterprise Assistants** — Customer support and internal knowledge management
-* **Advanced Code Generation** — Full-stack development and debugging
-* **Content Creation** — High-quality marketing copy, emails, and reports
-* **Translation & Localization** — Highly accurate translation between Turkish/English
-* **Data Extraction** — Structuring unstructured data into JSON/SQL
-* **Academic Assistance** — Solving math problems and summarizing research papers
+* **Enterprise Strategic Planning** — Market analysis and risk assessment
+* **Advanced Code Generation** — Full-stack architecture and optimization
+* **Legal & Medical Research** — Analyzing precedents and case studies
+* **Academic Simulation** — Philosophy, sociology, and theoretical physics
+* **Complex Data Interpretation** — Turning raw data into actionable logic
+* **Autonomous Agents** — Backend brain for complex agentic workflows
+
+---
+
+## 💡 Performance Highlights
+
+* **State-of-the-Art Logic:** Surpasses 70B+ class models in pure reasoning benchmarks.
+* **Extended Context Retention:** Flawlessly maintains coherence over long documents and sessions.
+* **Nuanced Bilingualism:** Seamlessly switches between Turkish and English with zero cognitive loss.
+* **Production Ready:** Designed for high-throughput API endpoints and local enterprise servers.
 
 ---
 
@@ -168,7 +237,9 @@ Licensed under the **MIT License** — free for commercial and non-commercial us
 
 ---
 
-> **Next 70B** — Türkiye’s flagship AI model. Built for those who demand **accuracy**, **speed**, and **scale**.
+> **Next 32B** — Türkiye’s flagship *reasoning* model. Built for those who demand **depth**, **precision**, and **massive intelligence**.
 
 [![Follow on HuggingFace](https://img.shields.io/badge/Follow-HuggingFace-yellow?logo=huggingface)](https://huggingface.co/Lamapi)
-```
+
+
+bunu 70b için ve, şu benchmark için düzenle: