jossoft/my-amature-llm

my-turkish-llm

Bu model sıfırdan oluşturulmuş özel bir Transformer tabanlı dil modelidir.

Model Detayları

• Model Tipi: Custom Transformer-based Language Model
• Framework: PyTorch
• Dil: Türkçe

Kullanım

🎯 Kolay Yöntem: AutoTokenizer (Önerilen)

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Model ve tokenizer yükle (trust_remote_code gerekli)
model = AutoModelForCausalLM.from_pretrained("jossoft/my-turkish-llm", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("jossoft/my-turkish-llm", trust_remote_code=True)

# Text generation
prompt = "merhaba"
input_ids = tokenizer.encode(prompt, return_tensors="pt")

with torch.no_grad():
    output = model.generate(input_ids, max_length=50)
    
generated_text = tokenizer.decode(output[0])
print(generated_text)

⚠️ Not: trust_remote_code=True parametresi gereklidir çünkü özel tokenizer kullanılıyor.

🔧 Alternatif: Manuel Yükleme

Tıklayın (özel kod gerektirir)

from transformers import AutoModelForCausalLM
import torch
import importlib.util
import sys
from huggingface_hub import hf_hub_download

# Model yükle
model = AutoModelForCausalLM.from_pretrained("jossoft/my-turkish-llm")

# Özel tokenizer'ı yükle
model_path = "jossoft/my-turkish-llm"
tokenizer_file = hf_hub_download(repo_id=model_path, filename="my_tokenizer.py")
vocab_file = hf_hub_download(repo_id=model_path, filename="vocab.json")

# Tokenizer modülünü dinamik yükle
spec = importlib.util.spec_from_file_location("my_tokenizer", tokenizer_file)
tokenizer_module = importlib.util.module_from_spec(spec)
sys.modules["my_tokenizer"] = tokenizer_module
spec.loader.exec_module(tokenizer_module)

# Tokenizer oluştur
LongestMatchTokenizer = tokenizer_module.LongestMatchTokenizer
tokenizer = LongestMatchTokenizer(vocab_file)

# Text generation
prompt = "merhaba"
input_ids = tokenizer.encode(prompt)

with torch.no_grad():
    output = model.generate(input_ids, max_length=50)
    
generated_text = tokenizer.decode(output[0].tolist())
print(generated_text)

💻 Lokal Kullanım

import torch
from llm_factory.model import BasicModel
from llm_factory.my_tokenizer import LongestMatchTokenizer
from save_model import load_model

# Model yükle
model, tokenizer, config = load_model("./saved_model")

# Text generation
prompt = "merhaba"
input_ids = tokenizer.encode(prompt)
output = model.generate(input_ids)
generated_text = tokenizer.decode(output.squeeze().tolist())
print(generated_text)

Model Mimarisi

• Multi-Head Attention with Rotary Position Encoding (RoPE)
• Gated MLP (Feed-Forward)
• RMS Normalization
• Causal Self-Attention

Lisans

MIT