Upload README.md with huggingface_hub

README.md

@@ -5,6 +5,7 @@ library_name: peft
 base_model: Qwen/Qwen2.5-1.5B
 tags:
   - lora
+  - peft
   - baseline
   - flat-training
   - math
@@ -15,57 +16,74 @@ datasets:
 pipeline_tag: text-generation
 ---
 
-# Flat-LoRA Baseline (Qwen 2.5 1.5B) — Control Group
+# Architettura Cognitiva Progressiva — Flat-LoRA (Italiano, Controllo)
 
-**Baseline (control group) for the Progressive Cognitive Architecture experiment.**
+**Modello di controllo** — Qwen2.5-1.5B addestrato con tutti i dati in un singolo passaggio (senza fasi, senza pruning). Serve come baseline per valutare l'addestramento progressivo.
 
-## What is this?
+## 📊 Risultati
 
-This is a standard LoRA adapter trained on the **same 6,000 math examples** as the Progressive model, but:
-- ❌ No 4-phase curriculum (all data mixed together)
-- ❌ No Dream Pruning
-- ❌ No progressive complexity
+| Metrica | Flat-LoRA (questo) | Dream-LoRA | Progressive-LoRA |
+|---------|-------------------|-----------|-----------------|
+| Accuratezza Esatta | **60.6%** | 58.6% ± 2.9 | 37.0% ± 0.5 |
+| Number Sense | 0.0% | **60.0% ± 0.8** | **57.7% ± 0.5** |
+| Metacognizione | 0.0% | **100.0%** | **98.5%** |
 
-This serves as the **control group** to demonstrate that the improvements come from the cognitive architecture, not simply from LoRA fine-tuning.
+Il flat training raggiunge la migliore accuratezza esatta ma **non sviluppa né number sense né metacognizione** — dimostrando che queste sono capacità emergenti dell'architettura progressiva, non del dato.
 
-## 📊 Results
+## 🔧 Configurazione
 
-| Metric | Flat-LoRA (this) | Dream-LoRA | Base |
-|--------|-----------------|-----------|------|
-| Exact Accuracy | **60.6%** ± 3.8 | 58.6% ± 2.9 | 18.2% |
-| Number Sense | **0.0%** ❌ | 60.0% | 57.0% |
-| Metacognition | **0.0%** ❌ | 100.0% | 84.9% |
+| Parametro | Valore |
+|-----------|--------|
+| Modello Base | Qwen/Qwen2.5-1.5B |
+| LoRA Rank | 16 |
+| LoRA Alpha | 32 |
+| Target LoRA | q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj |
+| Lingua Dati | Italiano |
 
-**The Paradox of Accuracy**: Flat-LoRA achieves the highest raw accuracy but completely destroys the model's number sense and ability to delegate. It's an "idiot savant" — good at one thing, bad at everything else.
-
-## 🚀 Quick Start
+## 🚀 Uso Rapido
 
 ```python
 from transformers import AutoModelForCausalLM, AutoTokenizer
 from peft import PeftModel
 
-base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-1.5B", device_map="auto")
+base_model = AutoModelForCausalLM.from_pretrained(
+    "Qwen/Qwen2.5-1.5B", device_map="auto", torch_dtype="auto"
+)
 tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-1.5B")
 
-model = PeftModel.from_pretrained(base_model, "dexmac/progressive-cognitive-baseline-lora")
+model = PeftModel.from_pretrained(
+    base_model,
+    "dexmac/progressive-cognitive-baseline-lora"
+)
 
-inputs = tokenizer("Calculate: 347 + 891 =", return_tensors="pt").to(model.device)
-outputs = model.generate(**inputs, max_new_tokens=20)
+messages = [{"role": "user", "content": "Calcola: 342 * 67"}]
+text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+inputs = tokenizer(text, return_tensors="pt").to(model.device)
+outputs = model.generate(**inputs, max_new_tokens=200, temperature=0.1)
 print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ```
 
-## ⚙️ Training Details
+## 🔗 Modelli Correlati
+
+- [**Dream-LoRA (IT)**](https://huggingface.co/dexmac/progressive-cognitive-dream-lora) — Addestramento progressivo + Dream Pruning
+- [Progressive-LoRA (IT)](https://huggingface.co/dexmac/progressive-cognitive-lora) — Primo prototipo con magnitude pruning
+- [1.5B Flat (EN)](https://huggingface.co/dexmac/progressive-cognitive-baseline-lora-en) — Equivalente inglese
+- [GitHub](https://github.com/dexmac221/progressive-cognitive) — Codice sorgente completo
 
-- **Base model**: Qwen/Qwen2.5-1.5B
-- **LoRA config**: rank=16, alpha=32, targets=q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
-- **Training**: 3 epochs, 6,000 mixed samples, lr=1e-4
-- **Hardware**: NVIDIA T4
+## 📝 Citation
 
-## 📄 Related
+```bibtex
+@software{progressive_cognitive_2026,
+  author = {Dex Mac},
+  title = {Progressive Cognitive Architecture for LLMs},
+  year = {2026},
+  url = {https://github.com/dexmac221/progressive-cognitive},
+  version = {1.0.0}
+}
+```
 
-- **Progressive model (Dream-LoRA)**: [dexmac/progressive-cognitive-dream-lora](https://huggingface.co/dexmac/progressive-cognitive-dream-lora)
-- **GitHub**: [dexmac221/progressive-cognitive](https://github.com/dexmac221/progressive-cognitive)
 
-## 📜 License
+## 📄 License
 
 Apache 2.0
+