Views
No views yet
| Metrica | Progressive-LoRA (questo) | Dream-LoRA | Flat-LoRA |
|---|---|---|---|
| Accuratezza Esatta | 37.0% ± 0.5 | 58.6% ± 2.9 | 60.6% |
| Number Sense | 57.7% ± 0.5 | 60.0% ± 0.8 | 0.0% |
| Metacognizione | 98.5% | 100.0% | 0.0% |
| 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 |
| Tipo Pruning | Magnitude (azzeramento pesi piccoli) |
| Lingua Dati | Italiano |
1from transformers import AutoModelForCausalLM, AutoTokenizer
2from peft import PeftModel
3
4base_model = AutoModelForCausalLM.from_pretrained(
5 "Qwen/Qwen2.5-1.5B", device_map="auto", torch_dtype="auto"
6)
7tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-1.5B")
8
9model = PeftModel.from_pretrained(
10 base_model,
11 "dexmac/progressive-cognitive-lora",
12 subfolder="lora_adapters"
13)
14
15messages = [{"role": "user", "content": "Calcola: 342 * 67"}]
16text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
17inputs = tokenizer(text, return_tensors="pt").to(model.device)
18outputs = model.generate(**inputs, max_new_tokens=200, temperature=0.1)
19print(tokenizer.decode(outputs[0], skip_special_tokens=True))1@software{progressive_cognitive_2026,
2 author = {Dex Mac},
3 title = {Progressive Cognitive Architecture for LLMs},
4 year = {2026},
5 url = {https://github.com/dexmac221/progressive-cognitive},
6 version = {1.0.0}
7}