metadata

license: llama3
language:
  - tr

CEREBRUM LLM

CERE-LLMA-3-8b-TR

This model is an fine-tuned version of a Llama3 8b Large Language Model (LLM) for Turkish. It was trained on a high quality Turkish instruction sets created from various open-source and internal resources. Turkish Instruction dataset carefully annotated to carry out Turkish instructions in an accurate and organized manner.

Model Details

Base Model: LLMA 3 8B based LLM
Tokenizer Extension: Specifically extended for Turkish
Training Dataset: Cleaned Turkish raw data with 5 billion tokens, custom Turkish instruction sets
Training Method: Initially with DORA, followed by fine-tuning with LORA

Benchmark Results

Winogrande_tr: 56.16
TruthfulQA_tr_v0.2: 47.46
Mmlu_tr_v0.2: 46.46
HellaSwag_tr_v0.2: 48.87
GSM8k_tr_v0.2: 25.43
Arc_tr_v0.2: 41.97

Usage Examples


from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda" # the device to load the model onto

model = AutoModelForCausalLM.from_pretrained(
    "Cerebrum/cere-llama-3-8b-tr",
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("Cerebrum/cere-llama-3-8b-tr")

prompt = "Python'da ekrana 'Merhaba Dünya' nasıl yazılır?"
messages = [
    {"role": "system", "content": "Sen, Cerebrum Tech tarafından üretilen ve verilen talimatları takip ederek en iyi cevabı üretmeye çalışan yardımcı bir yapay zekasın."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(device)

generated_ids = model.generate(
    model_inputs.input_ids,
    temperature=0.3,
    top_k=50,
    top_p=0.9,
    max_new_tokens=512,
    repetition_penalty=1,
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]