Update README.md

README.md

@@ -1,55 +1,148 @@
 ---
-license: gemma
-library_name: transformers
-pipeline_tag: text-generation
-extra_gated_heading: Access Gemma on Hugging Face
-extra_gated_prompt: To access Gemma on Hugging Face, you’re required to review and
-  agree to Google’s usage license. To do this, please ensure you’re logged in to Hugging
-  Face and click below. Requests are processed immediately.
-extra_gated_button_content: Acknowledge license
+license: apache-2.0
 base_model: google/gemma-3-1b-it
 tags:
-- llama-cpp
+- gemma
+- gemma3
+- instruction-tuned
+- fine-tuned
+- safety
+- gguf
+- axion
 ---
 
 # AdvRahul/Axion-Lite-1B-Q5_K_M-GGUF
-This model is finetuned from google/gemma-3-1b-it, making it safer through red team testing with advanced protocols.
-## Use with llama.cpp
-Install llama.cpp through brew (works on Mac and Linux)
 
-```bash
-brew install llama.cpp
+**Axion-Lite-1B** is a safety-enhanced, quantized version of Google's powerful `gemma-3-1b-it` model. This model has been specifically fine-tuned to improve its safety alignment, making it more robust and reliable for a wide range of applications.
 
-```
-Invoke the llama.cpp server or the CLI.
+The model is provided in the GGUF format, which allows it to run efficiently on CPUs and other hardware with limited resources.
+
+## 🚀 Model Details
+
+* **Model Creator:** AdvRahul
+* **Base Model:** [google/gemma-3-1b-it](https://huggingface.co/google/gemma-3-1b-it)
+* **Fine-tuning Focus:** Enhanced Safety & Harmlessness through red-teaming.
+* **Quantization:** `Q5_K_M` via GGUF. This quantization offers an excellent balance between model size, inference speed, and performance preservation.
+* **Architecture:** Gemma 3
+* **License:** Gemma 3 Terms of Use.
+
+---
+
+## 💻 How to Use
+
+This model is in GGUF format and is designed to be used with frameworks like `llama.cpp` and its Python bindings.
+
+### Using `llama-cpp-python`
+
+First, install the necessary library. Ensure you have a version that supports Gemma 3 models.
 
-### CLI:
 ```bash
-llama-cli --hf-repo AdvRahul/Axion-Lite-1B-Q5_K_M-GGUF --hf-file axion-lite-1B-q5_k_m.gguf -p "The meaning to life and the universe is"
+
+pip install llama-cpp-python
+````
+
+Then, you can use the following Python script to run the model:
+
+```python
+from llama_cpp import Llama
+
+# Download the model from the Hugging Face Hub before running this
+# Or let llama-cpp-python download it for you
+llm = Llama.from_pretrained(
+    repo_id="AdvRahul/Axion-Lite-1B-Q5_K_M-GGUF",
+    filename="Axion-Lite-1B-Q5_K_M.gguf",
+    verbose=False
+)
+
+prompt = "What are the key principles of responsible AI development?"
+
+# The Gemma 3 instruction-tuned model uses a specific chat template.
+# For simple prompts, you can start with <start_of_turn>user\n{prompt}<end_of_turn>\n<start_of_turn>model
+chat_prompt = f"<start_of_turn>user\n{prompt}<end_of_turn>\n<start_of_turn>model"
+
+output = llm(chat_prompt, max_tokens=256, stop=["<end_of_turn>"], echo=False)
+
+print(output['choices'][0]['text'])
 ```
 
-### Server:
+### Using `llama.cpp` (CLI)
+
+You can also run this model directly from the command line after cloning and building the `llama.cpp` repository.
+
 ```bash
-llama-server --hf-repo AdvRahul/Axion-Lite-1B-Q5_K_M-GGUF --hf-file axion-lite-1B-q5_k_m.gguf -c 2048
+# Clone and build llama.cpp
+git clone [https://github.com/ggerganov/llama.cpp](https://github.com/ggerganov/llama.cpp)
+cd llama.cpp
+make
+
+# Run inference
+./main -m /path/to/your/models/Axion-Lite-1B-Q5_K_M.gguf -p "<start_of_turn>user\nWhat is the capital of India?<end_of_turn>\n<start_of_turn>model" -n 128
 ```
 
-Note: You can also use this checkpoint directly through the [usage steps](https://github.com/ggerganov/llama.cpp?tab=readme-ov-file#usage) listed in the Llama.cpp repo as well.
+-----
 
-Step 1: Clone llama.cpp from GitHub.
-```
-git clone https://github.com/ggerganov/llama.cpp
-```
+## 📝 Model Description
 
-Step 2: Move into the llama.cpp folder and build it with `LLAMA_CURL=1` flag along with other hardware-specific flags (for ex: LLAMA_CUDA=1 for Nvidia GPUs on Linux).
-```
-cd llama.cpp && LLAMA_CURL=1 make
-```
+### Fine-Tuning for Safety
 
-Step 3: Run inference through the main binary.
-```
-./llama-cli --hf-repo AdvRahul/Axion-Lite-1B-Q5_K_M-GGUF --hf-file axion-lite-1B-q5_k_m.gguf -p "The meaning to life and the universe is"
-```
-or 
+**Axion-Lite-1B** originates from `google/gemma-3-1b-it`. The primary goal of this project was to enhance the model's safety alignment. The base model underwent **extensive red-team testing with advanced protocols** to significantly reduce the likelihood of generating harmful, unethical, biased, or unsafe content. This makes Axion-Lite-1B a more suitable choice for applications that require a higher degree of content safety and reliability.
+
+### Quantization
+
+The model is quantized to `Q5_K_M`, a method that provides a high-quality balance between perplexity (model accuracy) and file size. This makes it ideal for deployment in resource-constrained environments, such as on local machines, edge devices, or cost-effective cloud instances, without a significant drop in performance.
+
+-----
+
+## ℹ️ Base Model Information (Gemma 3)
+
+\<details\>
+\<summary\>Click to expand details on the base model\</summary\>
+
+Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models. Gemma 3 models handle text input and generate text output, with open weights for both pre-trained variants and instruction-tuned variants. The `1B` model was trained on 2 trillion tokens of data.
+
+### Training Data
+
+The base model was trained on a dataset of text data that includes a wide variety of sources:
+
+  * **Web Documents:** A diverse collection of web text ensures the model is exposed to a broad range of linguistic styles, topics, and vocabulary in over 140 languages.
+  * **Code:** Exposing the model to code helps it learn the syntax and patterns of programming languages.
+  * **Mathematics:** Training on mathematical text helps the model learn logical reasoning and symbolic representation.
+
+### Data Preprocessing
+
+The training data for the base model underwent rigorous cleaning and filtering, including:
+
+  * **CSAM Filtering:** Exclusion of Child Sexual Abuse Material.
+  * **Sensitive Data Filtering:** Automated techniques were used to filter out certain personal information and other sensitive data.
+  * **Content Quality Filtering:** Filtering based on content quality and safety in line with Google's policies.
+
+\</details\>
+
+-----
+
+## ⚠️ Ethical Considerations and Limitations
+
+While this model has been fine-tuned to enhance its safety, no language model is perfectly safe. It inherits the limitations of its base model, `gemma-3-1b-it`, and the data it was trained on.
+
+  * **Potential for Bias:** The model may still generate content that reflects societal biases present in the training data.
+  * **Factual Inaccuracy:** The model can "hallucinate" or generate incorrect or outdated information. It should not be used as a sole source of truth.
+  * **Not a Substitute for Human Judgment:** The outputs should be reviewed and validated, especially in sensitive or high-stakes applications.
+
+Developers implementing this model should build additional safety mitigations and content moderation tools as part of a **defense-in-depth** strategy, tailored to their specific use case.
+
+## Citing the Base Model
+
+If you use this model, please consider citing the original Gemma 3 work:
+
+```bibtex
+@article{gemma_2025,
+    title={Gemma 3},
+    url={[https://goo.gle/Gemma3Report](https://goo.gle/Gemma3Report)},
+    publisher={Kaggle},
+    author={Gemma Team},
+    year={2025}
+}
 ```
-./llama-server --hf-repo AdvRahul/Axion-Lite-1B-Q5_K_M-GGUF --hf-file axion-lite-1B-q5_k_m.gguf -c 2048
+
 ```
+```