File size: 1,066 Bytes
72003ee
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
---
language: es
license: apache-2.0
tags:
- text-generation
- transformer
- pytorch
---

# MTP Mini - Modelo Mejorado 20x

Modelo transformer con arquitectura avanzada entrenado en GPU T4.

## Arquitectura
- **Parámetros**: ~310.7M (310,708,225)
- **Vocabulario**: 8000 tokens
- **Capas**: 24
- **Dimensión**: 1024
- **Contexto**: 2048 tokens

## Mejoras
- ✅ RoPE, RMSNorm, SwiGLU
- ✅ Flash Attention
- ✅ Gradient Checkpointing
- ✅ Mixed Precision FP16
- ✅ Anti-alucinación
- ✅ Confidence Scoring

## Uso
```python
import torch, pickle
from tokenizer import MTPTokenizer
from model import MTPMiniModel

with open('mtp_mini.pkl', 'rb') as f:
    data = pickle.load(f)

tokenizer = MTPTokenizer('mtp_tokenizer.model')
model = MTPMiniModel(**data['config']['model'])
model.load_state_dict(data['model_state_dict'])
model.eval()

prompt = "¿Qué es la IA?"
ids = torch.tensor([tokenizer.encode(prompt)]).unsqueeze(0)
output = model.generate(ids, max_new_tokens=150)
print(tokenizer.decode(output[0].tolist()))
```

Entrenado en Google Colab con GPU T4.