Update README.md

README.md

@@ -1,23 +1,71 @@
 ---
+license: apache-2.0
+base_model: unsloth/Qwen2.5-3B-Instruct-bnb-4bit
+language:
+- en
+- de
 tags:
-- gguf
-- llama.cpp
+- creative
+- flash
+- roleplay
 - unsloth
-
+- lazyloopstudio
+- nomi
+- creative_writing
+- nomi flash
+- nomi 1
+model_name: Nomi-1-Flash
+datasets:
+- databricks/databricks-dolly-15k
+metrics:
+- character
+pipeline_tag: text-generation
+library_name: transformers
 ---
 
-# Nomi-1-Flash : GGUF
+<div align="center">
+  
+![Znomi_nbg](https://cdn-uploads.huggingface.co/production/uploads/6921fa6332f7fb129563d495/9oOMYWNoxBDmF854MciTn.png)
+
+</div>
+
+# Nomi-1-Flash ⚡
+
+**Nomi-1-Flash** is a high-speed, creative-focused AI companion based on the Qwen 2.5 3B architecture. Developed by **LazyLoopStudio**, this model is fine-tuned to prioritize vibrant personality, creative vocabulary, and rapid responses over rigid technical instruction following.
+
+
+
+## 📊 Evaluation & Benchmarks
+
+We believe in transparency. Nomi-1-Flash was tested using a custom evaluation suite to measure its "Creative First" approach.
+
+| Metric | Score | Interpretation |
+| :--- | :---: | :--- |
+| **Creativity Index** | **100.0%** | Exceptional vocabulary diversity and imaginative flair. |
+| **General Knowledge (MMLU)** | **48.0%** | Solid factual foundation, comparable to mid-sized models. |
+| **Instruction Following (IFEval)** | **33.3%** | Low. Nomi tends to prioritize style over strict formatting rules. |
+
+### Summary
+Nomi-1-Flash is **not** a coding or logic expert. She is a storyteller and a conversationalist. While her instruction following is lower than the base model, her creative output is significantly more engaging and human-like.
+
+## 🚀 Quick Start (Inference)
+
+To use Nomi-1-Flash in Python (requires `unsloth` or `transformers`):
+
+```python
+from unsloth import FastLanguageModel
+import torch
 
-This model was finetuned and converted to GGUF format using [Unsloth](https://github.com/unslothai/unsloth).
+model, tokenizer = FastLanguageModel.from_pretrained(
+    model_name = "LazyLoopStudio/Nomi-1-Flash",
+    max_seq_length = 2048,
+    load_in_4bit = True,
+)
+FastLanguageModel.for_inference(model)
 
-**Example usage**:
-- For text only LLMs:    `./llama.cpp/llama-cli -hf LazyLoopStudio/Nomi-1-Flash --jinja`
-- For multimodal models: `./llama.cpp/llama-mtmd-cli -hf LazyLoopStudio/Nomi-1-Flash --jinja`
+prompt = "Write a creative opening for a story about a neon-lit cloud city."
+inputs = tokenizer([f"<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n"], return_tensors="pt").to("cuda")
 
-## Available Model files:
-- `Qwen2.5-3B-Instruct.Q4_K_M.gguf`
+outputs = model.generate(**inputs, max_new_tokens=256)
+print(tokenizer.batch_decode(outputs)[0])
 
-## Ollama
-An Ollama Modelfile is included for easy deployment.
-This was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth)
-[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)