Update README.md

README.md

@@ -11,12 +11,20 @@ tags:
 base_model: unsloth/llama-3-8b-bnb-4bit
 ---
 
-# Uploaded  model
+#  LLama 3 for router module in RAG
+
+While developing complex RAG applications, I found a common need for router functionality to map user queries to different system workflows (and APIs).  The router acts as a dispatcher that can enhance responsiveness and accuracy by choosing the best workflow or API based on the query context. This implies that we need to produce structured output from unstructured input text.     
+
+To this end, and to avoid external dependencies for this part of the system, I will undertake a simple exercise to fine-tune the new Llama 3 model to process text input and generate JSON-like output. My hope is that we can seamlessly integrate various models to reinforce complex applications in production settings, building a robust critical infraestructures for the semantical modules.    
+
+For training, we will use structured data from [azizshaw](https://huggingface.co/azizshaw/text_to_json). The dataset has 485 rows and contains 'input', 'output' and 'instruction' columns.    
 
+For a quick evaluation, let's use another dataset for text-to-JSON, the **Diverse Restricted JSON Data Extraction**, curated by: The paraloq analytics team ([here](https://huggingface.co/datasets/paraloq/json_data_extraction))
+
+
+# Uploaded  model
 - **Developed by:** sccastillo
 - **License:** apache-2.0
 - **Finetuned from model :** unsloth/llama-3-8b-bnb-4bit
-
 This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
 
-[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)