Update README.md

README.md

@@ -1,3 +1,147 @@
----
-license: mit
----
+---
+license: mit
+language:
+- pt
+pipeline_tag: text-generation
+---
+
+## 🧠 MiniBot-0.9M-Instruct
+
+Instruction-tuned GPT-2 style language model (~900K parameters) optimized for Portuguese conversational tasks.
+
+## 📌 Model Overview
+
+MiniBot-0.9M-Instruct is an instruction-tuned version of MiniBot-0.9M-Base, designed to better follow prompts, respond to user inputs, and generate more coherent conversational outputs in Portuguese.
+
+Built on a GPT-2 architecture (~0.9M parameters), this model was fine-tuned on conversational and instruction-style data to improve usability in real-world interactions.
+
+🎯 Key Characteristics
+🇧🇷 Language: Portuguese (primary)
+🧠 Architecture: GPT-2 style (decoder-only Transformer)
+🔤 Embeddings: GPT-2 compatible
+📉 Parameters: ~900K
+⚙️ Base Model: MiniBot-0.9M-Base
+🎯 Fine-tuning: Instruction tuning (supervised)
+✅ Alignment: Basic prompt-following behavior
+🧠 What Changed from Base?
+
+Compared to the base model:
+
+Feature	Base	Instruct
+Prompt understanding	❌	✅
+Conversational flow	⚠️	✅
+Instruction following	❌	✅
+Coherence	Baixa	Melhorada
+Usability	Experimental	Practical
+
+👉 The model is now significantly more usable in chat scenarios.
+
+🏗️ Architecture
+
+Same core as base:
+
+Decoder-only Transformer (GPT-2 style)
+Token + positional embeddings
+Self-attention + MLP blocks
+Autoregressive generation
+
+No architectural changes — only behavioral improvement via fine-tuning.
+
+📚 Fine-Tuning
+Dataset
+
+The model was fine-tuned on a Portuguese instruction-style conversational dataset, including:
+
+Perguntas e respostas
+Instruções simples
+Chat estilo assistente
+Roleplay básico
+Conversas naturais
+Format
+User: Me explique o que é gravidade
+Bot: A gravidade é a força que atrai objetos com massa...
+Strategy
+Supervised fine-tuning (SFT)
+Pattern learning for instruction-following
+No RLHF or preference optimization
+💡 Capabilities
+
+✅ Strengths:
+
+Seguir instruções simples
+Responder perguntas básicas
+Conversar de forma mais natural
+Melhor coerência em respostas curtas
+Estrutura de diálogo mais consistente
+
+❌ Limitations:
+
+Raciocínio ainda limitado
+Pode errar fatos
+Não mantém contexto longo
+Sensível a prompts mal estruturados
+
+👉 Mesmo com instruct tuning, ainda é um modelo extremamente pequeno.
+
+🚀 Usage
+Hugging Face Transformers
+```Python
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+model_name = "AxionLab-official/MiniBot-0.9M-Instruct"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(model_name)
+
+prompt = "User: Me diga uma curiosidade sobre o espaço\nBot:"
+inputs = tokenizer(prompt, return_tensors="pt")
+
+outputs = model.generate(
+    **inputs,
+    max_new_tokens=80,
+    temperature=0.7,
+    top_p=0.9,
+    do_sample=True
+)
+
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))
+```
+
+⚙️ Recommended Settings
+
+Para melhor qualidade:
+
+temperature: 0.6 – 0.8
+top_p: 0.85 – 0.95
+do_sample: True
+max_new_tokens: 40 – 100
+
+👉 Instruct models tendem a performar melhor com menos aleatoriedade.
+
+🧪 Intended Use
+💬 Chatbots leves em português
+🎮 NPCs e jogos
+🧠 Testes de fine-tuning
+📚 Educação em NLP
+⚡ Aplicações locais (CPU-only)
+⚠️ Limitations
+
+Modelo extremamente pequeno
+Sem alinhamento robusto
+Pode gerar respostas incorretas
+Não adequado para produção crítica
+
+🔮 Future Work
+🧠 Reasoning-tuned version (MiniBot-Reason)
+📈 Scaling para 1M–10M parâmetros
+📚 Dataset mais diverso
+🤖 Melhor alinhamento de respostas
+🧩 Tool-use experiments
+
+📜 License
+
+MIT
+
+👤 Author
+
+Developed by AxionLab