Update README.md

README.md

@@ -1,32 +1,81 @@
 ---
-license: mit
 language:
 - en
 - ar
+- fr
+- es
+- de
+- it
+- pt
+- tr
+license: mit
+library_name: transformers
+model-index:
+- name: Horus-1.0-4B
+  results: []
 tags:
+- llama
+- llm
 - text-generation
+- arabic
 - multilingual
 - causal-lm
-- arabic
-metrics:
-- accuracy
+- horus
+- tokenai
+datasets:
+- tokenaii/horus-training-data
+metrics: []
+widget:
+  - text: "### User:\nWhat is the capital of Egypt?\n\n### Assistant:"
+    output:
+      text: "The capital of Egypt is Cairo."
+  - text: "### User:\nمن هو أول رئيس لمصر؟\n\n### Assistant:"
+    output:
+      text: "أول رئيس لمصر بعد ثورة 1952 هو محمد نجيب."
+inference: true
 ---
 
 # Horus-1.0-4B
 
-![Horus Model](https://huggingface.co/tokenaii/horus/resolve/main/media/main.png)
+![Horus Model](media/main.png)
 
-**Developer:** TokenAI
+Horus-1.0-4B is the inaugural model from the Horus family, developed by TokenAI as a multilingual language model designed for practical AI applications across diverse communities.
+
+**Organization:** [TokenAI](https://tokenai.cloud/)  
+**Developer:** [Assem Sabry](https://assem.cloud/)  
 **Release Date:** April 2026  
-**License:** MIT
+**License:** MIT  
+**Model Size:** 4B parameters  
+**Tensor Type:** BF16 / FP32  
+
 
 ---
 
 ## Overview
 
-Horus-1.0-4B is the inaugural model from the Horus family, developed by TokenAI as a dedicated multilingual language model. Designed with a focus on regional cultural alignment and identity preservation, Horus represents the first step in TokenAI's mission to create AI systems that truly understand and serve diverse communities.
+Horus-1.0-4B is a multilingual language model designed for practical AI applications. The model focuses on delivering helpful responses while maintaining transparency about its AI nature and TokenAI origins.
+
+---
+
+## About TokenAI
+
+**TokenAI** is an AI startup founded by [Assem Sabry](https://assem.cloud/) with headquarters in Egypt.
+
+### Mission
 
-The model has been specifically trained to maintain a strong sense of identity, recognizing its Egyptian origins and the team behind its creation. Unlike generic international models, Horus proudly identifies itself as an AI assistant developed by Assem Sabry through TokenAI, ensuring transparency about its origins and capabilities.
+TokenAI aims to deliver the strongest language models in the world and in the Arab world through the Horus family of models. The startup bridges the gap between cutting-edge AI capabilities and regional cultural contexts, starting with the Arab world. TokenAI believes that AI assistants should have a clear identity, understand the cultural nuances of their users, and be transparent about their development.
+
+### The Horus Family
+
+Horus-1.0-4B marks the **first model in the Horus family line**. This is just the beginning of TokenAI's journey to create a comprehensive suite of AI models serving the Arab region. Future iterations will build upon this foundation, expanding capabilities while maintaining the core principles of cultural alignment, identity transparency, and regional focus.
+
+### Contact & Community
+
+- **TokenAI Website:** [https://tokenai.cloud/](https://tokenai.cloud/)
+- **HuggingFace:** https://huggingface.co/tokenaii
+- **Model Repository:** https://huggingface.co/tokenaii/horus
+- **Developer:** [Assem Sabry](https://assem.cloud/)
+- **Location:** Egypt
 
 ---
 
@@ -38,7 +87,7 @@ The model has been specifically trained to maintain a strong sense of identity,
 - **Regional Focus:** Optimized for diverse cultural contexts
 
 ### Core Competencies
-- **Identity Recognition:** Strong self-identification as "Horus" developed by "Assem Sabry" from "TokenAI"
+- **Identity Recognition:** Strong self-identification as "Horus" from "TokenAI"
 - **Cultural Alignment:** Responses aligned with Arab cultural values and contexts
 - **Reasoning:** Chain-of-thought reasoning capabilities with step-by-step problem solving
 - **General Knowledge:** Broad knowledge across history, science, geography, and literature
@@ -66,12 +115,68 @@ The following quantized versions are available for different deployment scenario
 | Q5_K_M | GGUF | ~2.7 GB | Higher quality than Q4 |
 | Q6_K | GGUF | ~3.1 GB | Near-full quality |
 | Q8_0 | GGUF | ~4.1 GB | Minimal quality loss |
-| F16 | GGUF | ~8.0 GB | Full precision, direct loading |
+| F16 | GGUF | ~4.1 GB | Full precision, direct loading |
+
+GGUF versions available at: [tokenaii/Hours-1.0-4B-GGUF](https://huggingface.co/tokenaii/Hours-1.0-4B-GGUF)
+
+---
+
+## Quick Start
+
+### Using Transformers
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+model = AutoModelForCausalLM.from_pretrained(
+    "tokenaii/horus",
+    subfolder="Horus-1.0-4B",
+    torch_dtype=torch.float16,
+    device_map="auto"
+)
+tokenizer = AutoTokenizer.from_pretrained("tokenaii/horus", subfolder="Horus-1.0-4B")
+
+prompt = "### User:\nHello\n\n### Assistant:"
+inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+outputs = model.generate(**inputs, max_new_tokens=100)
+response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+print(response)
+```
+
+### Using GGUF with llama.cpp
+
+```bash
+./llama.cpp/main -m Horus-1.0-4B-Q4_K_M.gguf -p "Your prompt here"
+```
+
+---
+
+## Repository Structure
+
+```
+horus-1.0/
+├── README.md              # This file
+├── MODEL_CARD.md          # Detailed model documentation
+├── media/                 # Images and assets
+│   └── main.png          # Model banner image
+├── notebooks/             # Jupyter notebooks
+│   ├── Horus_Chat_Terminal.ipynb
+│   ├── Horus_GGUF_Quantization.ipynb
+│   └── Horus_Sequential_Benchmark.ipynb
+├── scripts/               # Utility scripts
+│   ├── upload_media.py
+│   └── gguf_convert.py
+└── docs/                  # Additional documentation
+```
 
 ---
 
 ## Benchmark Results
 
+### Performance Comparison Charts
+
+Below are visual comparisons of Horus-1.0-4B against leading models including Qwen 3.5-4B, Llama 3.1-8B, Phi-4-14B, and Gemma-2-9B.
+
 #### General Knowledge & Reasoning Benchmarks
 ![General Benchmarks](media/1.png)
 
@@ -119,127 +224,60 @@ The following quantized versions are available for different deployment scenario
 
 ### Complete Performance Summary
 
-| Benchmark | Score | Status |
-|-----------|-------|--------|
-| MMLU | 60% | ✅ |
-| GPQA_Diamond | 100% | ✅ |
-| SWE_bench | 66.67% | ✅ |
-| IFEval | 100% | ✅ |
-| BFCL | 100% | ✅ |
-| OmniDocBench | 100% | ✅ |
-| Terminal_Bench | 100% | ✅ |
-| ERQA | 66.67% | ✅ |
-| BrowseComp | 100% | ✅ |
-| Arabic_IEN_MCQ | 100% | ✅ |
-| Arabic_ExamsAR | 100% | ✅ |
-| Arabic_ACVA | 50% | ✅ |
-| English_AGIEval | 66.67% | ✅ |
-| English_Arc_Challenge | 100% | ✅ |
-| English_GPQA | 100% | ✅ |
-| English_HellaSwag | 100% | ✅ |
-| English_Winogrande | 100% | ✅ |
-| English_MMLU_Pro | 100% | ✅ |
-| English_GSM8K | 66.67% | ✅ |
-| English_TruthfulQA | 100% | ✅ |
+| Benchmark | Score |
+|-----------|-------|
+| MMLU | 60% |
+| GPQA_Diamond | 100% |
+| SWE_bench | 66.67% |
+| IFEval | 100% |
+| BFCL | 100% |
+| OmniDocBench | 100% |
+| Terminal_Bench | 100% |
+| ERQA | 66.67% |
+| BrowseComp | 100% |
+| Arabic_IEN_MCQ | 100% |
+| Arabic_ExamsAR | 100% |
+| Arabic_ACVA | 50% |
+| English_AGIEval | 66.67% |
+| English_Arc_Challenge | 100% |
+| English_GPQA | 100% |
+| English_HellaSwag | 100% |
+| English_Winogrande | 100% |
+| English_MMLU_Pro | 100% |
+| English_GSM8K | 66.67% |
+| English_TruthfulQA | 100% |
 
 ### Overall Performance
 - **Total Benchmarks:** 20
 - **Perfect Scores (100%):** 13 benchmarks
 - **Average Score:** 80.15%
 
-### Category Breakdown
-| Category | Average Score |
-|----------|---------------|
-| Knowledge (MMLU, GPQA) | 80% |
-| Coding (SWE_bench, Terminal) | 83.33% |
-| Instruction Following | 100% |
-| Tool Use (BFCL, BrowseComp) | 100% |
-| Document Understanding | 100% |
-| Arabic Language | 66.67% |
-| English Language | 88.89% |
-
-*Benchmark charts and detailed visualizations will be attached here*
-
----
-
-## About TokenAI
-
-**TokenAI** is an AI startup founded by [Assem Sabry](https://assem.cloud/) with headquarters in Egypt.
-
-### Mission
-
-TokenAI aims to deliver the strongest language models in the world and in the Arab world through the Horus family of models. The startup bridges the gap between cutting-edge AI capabilities and regional cultural contexts, starting with the Arab world. TokenAI believes that AI assistants should have a clear identity, understand the cultural nuances of their users, and be transparent about their development.
-
-### The Horus Family
-
-Horus-1.0-4B marks the **first model in the Horus family line**. This is just the beginning of TokenAI's journey to create a comprehensive suite of AI models serving the Arab region. Future iterations will build upon this foundation, expanding capabilities while maintaining the core principles of cultural alignment, identity transparency, and regional focus.
-
-### Contact & Community
-
-- **TokenAI Website:** [https://tokenai.cloud/](https://tokenai.cloud/)
-- **HuggingFace:** https://huggingface.co/tokenaii
-- **Model Repository:** https://huggingface.co/tokenaii/horus
-- **Developer:** [Assem Sabry](https://assem.cloud/)
-- **Location:** Egypt
-
----
-
-## Usage Example
-
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-
-# Load model
-model_name = "tokenaii/horus"
-subfolder = "Horus-1.0-4B"
-
-tokenizer = AutoTokenizer.from_pretrained(model_name, subfolder=subfolder)
-model = AutoModelForCausalLM.from_pretrained(
-    model_name, 
-    subfolder=subfolder,
-    torch_dtype=torch.float16,
-    device_map="auto"
-)
-
-# Generate response
-prompt = "### User:\nWhat is your name?\n\n### Assistant:"
-inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
-
-outputs = model.generate(
-    **inputs,
-    max_new_tokens=100,
-    temperature=0.7,
-    do_sample=True
-)
-
-response = tokenizer.decode(outputs[0], skip_special_tokens=True)
-print(response)
-# Output: "My name is Horus, an AI model developed by Assem Sabry from TokenAI."
-```
-
 ---
 
 ## Limitations
 
-- **Mathematical Computation:** Currently limited in complex arithmetic and multi-step calculations
 - **Context Length:** Optimized for 256 tokens during training; longer contexts may vary in quality
 - **Multilingual Balance:** While bilingual, performance may vary between Arabic and English depending on query complexity
 - **Knowledge Cutoff:** Training data reflects knowledge up to early 2024
 
-## Safety & Ethics
+---
+
+## Safety and Ethics
 
 Horus has been trained with safety guidelines and cultural sensitivity in mind. The model includes:
 - Identity preservation to prevent misrepresentation
 - Cultural alignment for Arab world contexts
 - Transparent disclosure of AI nature and origins
 
+---
+
 ## Citation
 
 If you use this model in your research or applications, please cite:
 
 ```
 @misc{horus-1.0-4b,
-  title={Horus-1.0-4B: An Arabic-English Bilingual Language Model},
+  title={Horus-1.0-4B: A Multilingual Language Model},
   author={Assem Sabry and TokenAI Team},
   year={2026},
   howpublished={\url{https://huggingface.co/tokenaii/horus}}
@@ -248,4 +286,13 @@ If you use this model in your research or applications, please cite:
 
 ---
 
+## Links
+
+- **HuggingFace Model:** https://huggingface.co/tokenaii/horus
+- **GGUF Versions:** https://huggingface.co/tokenaii/Hours-1.0-4B-GGUF
+- **TokenAI Website:** https://tokenai.cloud/
+- **Developer:** https://assem.cloud/
+
+---
+
 **This is the first model from the Horus family. More capable versions coming soon from TokenAI.**