Update README.md

README.md

@@ -6,17 +6,95 @@ tags:
 - unsloth
 - mistral
 - trl
+- finance
+- crypto
+- lora
 license: apache-2.0
 language:
 - en
 ---
 
-# Uploaded  model
+# 🪙 CoinReason-7B (Proof of Concept)
 
-- **Developed by:** sarfras
-- **License:** apache-2.0
-- **Finetuned from model :** unsloth/mistral-7b-instruct-v0.3-bnb-4bit
+**⚠️ Prototype Warning:**
+This adapter was trained on a synthetic "Gold Standard" prototype dataset to demonstrate an **End-to-End MLOps Pipeline**. It is intended to showcase the *fine-tuning architecture* (Unsloth, QLoRA, Hugging Face integration) rather than to provide financial advice. It may overfit to specific training examples.
 
-This mistral model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth)
+## Model Overview
 
-[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
+**CoinReason-7B** is a specialized Low-Rank Adapter (LoRA) for the Mistral-7B Large Language Model. It is designed to analyze cryptocurrency social media text and output structured financial reasoning.
+
+Unlike standard sentiment models that output simple "Positive/Negative" labels, CoinReason attempts to generate:
+
+1. **Sentiment:** (Bullish/Bearish)
+2. **Explanation:** The logical reasoning behind the sentiment.
+3. **Market Implication:** A short-term predictive outlook for price action.
+
+## Technical Specifications
+
+* **Base Model:** [unsloth/mistral-7b-instruct-v0.3-bnb-4bit](https://huggingface.co/unsloth/mistral-7b-instruct-v0.3-bnb-4bit)
+* **Fine-Tuning Technique:** QLoRA (Quantized Low-Rank Adaptation)
+* **Quantization:** 4-bit (NF4) for efficient inference on edge hardware (T4 GPUs)
+* **Framework:** [Unsloth](https://github.com/unslothai/unsloth) (2x faster training) + Hugging Face Transformers
+
+## How to Use
+
+You can load this model using the `unsloth` library for fast inference.
+```python
+from unsloth import FastLanguageModel
+
+# 1. Load the model and adapters
+model, tokenizer = FastLanguageModel.from_pretrained(
+    "sarfras/coinreason-7b-lora",
+    max_seq_length = 2048,
+    dtype = None,
+    load_in_4bit = True,
+)
+FastLanguageModel.for_inference(model)
+
+# 2. Define the prompt format
+tweet = "Bitcoin volume is dying and we are stuck at resistance. I think we go down."
+
+prompt = f"""<s>[INST] Analyze the following Bitcoin market text for sentiment and short-horizon implication.
+
+Text: {tweet}
+
+Provide output in this exact format:
+Sentiment: [Bullish/Bearish]
+Explanation: [reasoning]
+Market Implication: [brief BTC price direction outlook][/INST]"""
+
+# 3. Generate
+inputs = tokenizer([prompt], return_tensors = "pt").to("cuda")
+outputs = model.generate(**inputs, max_new_tokens = 128, use_cache = True)
+print(tokenizer.batch_decode(outputs, skip_special_tokens=True)[0])
+```
+
+## Training Details
+
+* **Dataset:** Synthetic Financial Reasoning Dataset (Prototype v1)
+* **Objective:** Instruction Fine-Tuning (SFT)
+* **LoRA Rank (r):** 16
+* **LoRA Alpha:** 16
+* **Optimizer:** AdamW 8-bit
+
+## Example Output
+
+**Input:**
+*"Whales are dumping BTC heavily on Binance, price dropping fast below support."*
+
+**Model Prediction:**
+> **Sentiment:** Bearish
+>
+> **Explanation:** Large inflows of BTC to exchanges (Whale movement) typically signal an intent to sell, increasing sell-side pressure.
+>
+> **Market Implication:** Price likely to test the $60k support; a breakdown could trigger a flush to lower levels.
+
+---
+
+## Disclaimer
+
+This model is a **proof of concept** and should **NOT** be used for actual financial decision-making. Always conduct your own research and consult with qualified financial advisors before making investment decisions.
+
+---
+
+*Created by Sarfras as part of an End-to-End LLM Engineering Portfolio.*