VinaLlama2-14B Beta

GGUF Here: VinaLlama2-14B-GGUF

Top Features:

• Context Length: 32,768 tokens.
• VERY GOOD at reasoning, mathematics and creative writing.
• Works with Langchain Agent out-of-the-box.

Known Issues

• Still a bit struggling with Vietnamese fact (Hoang Sa & Truong Sa, Historical questions).
• Hallucination when reasoning.
• Can't do Vi-En/En-Vi translation (yet)!

Quick use:

VRAM Requirement: ~20GB

pip install transformers accelerate

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

model = AutoModelForCausalLM.from_pretrained(
    "vilm/VinaLlama2-14B",
    torch_dtype='auto',
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("vilm/VinaLlama2-14B")

prompt = "Một cộng một bằng mấy?"
messages = [
    {"role": "system", "content": "Bạn là trợ lí AI hữu ích."},
    {"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,
    max_new_tokens=1024,
    eos_token_id=tokenizer.eos_token_id,
    temperature=0.25,
)
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)[0]
print(response)