datatab/Yugo55A-4bit

Yugo55A-GPT 4bit

• Developed by: datatab
• License: mit

🏆 Results

Results obtained through the Serbian LLM evaluation, released by Aleksa Gordić: serbian-llm-eval

• Evaluation was conducted on a 4-bit version of the model due to hardware resource constraints.

MODEL	ARC-E	ARC-C	Hellaswag	BoolQ	Winogrande	OpenbookQA	PiQA
*Yugo55-GPT-v4-4bit	51.41	36.00	57.51	80.92	65.75	34.70	70.54
Yugo55A-GPT	51.52	37.78	57.52	84.40	65.43	35.60	69.43

🔗 Merge Details

Merge Method

This is a merge of pre-trained language models created using mergekit. This model was merged using the linear merge method.

Models Merged

The following models were included in the merge:

• datatab/Yugo55-GPT-v4
• datatab/Yugo55-GPT-DPO-v1-chkp-300
• mlabonne/AlphaMonarch-7B
• NousResearch/Nous-Hermes-2-Mistral-7B-DPO

🧩 Configuration

The following YAML configuration was used to produce this model:

models:
  - model: datatab/Yugo55-GPT-v4
    parameters:
      weight: 1.0
  - model: datatab/Yugo55-GPT-DPO-v1-chkp-300
    parameters:
      weight: 1.0
  - model: mlabonne/AlphaMonarch-7B
    parameters:
      weight: 0.5
  - model: NousResearch/Nous-Hermes-2-Mistral-7B-DPO
    parameters:
      weight: 0.5
merge_method: linear
dtype: float16

💻 Usage

!pip -q install git+https://github.com/huggingface/transformers # need to install from github
!pip install -q datasets loralib sentencepiece
!pip -q install bitsandbytes accelerate

from IPython.display import HTML, display

def set_css():
  display(HTML('''
  <style>
    pre {
        white-space: pre-wrap;
    }
  </style>
  '''))
get_ipython().events.register('pre_run_cell', set_css)

import torch
import transformers
from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
    "datatab/Yugo55A-GPT", torch_dtype="auto"
)

tokenizer = AutoTokenizer.from_pretrained(
    "datatab/Yugo55A-GPT", torch_dtype="auto"
)

from typing import Optional
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer


def generate(
    user_content: str, system_content: Optional[str] = ""
) -> str:
    system_content = "Ispod je uputstvo koje opisuje zadatak, upareno sa unosom koji pruža dodatni kontekst. Napišite odgovor koji na odgovarajući način kompletira zahtev."

    messages = [
        {
            "role": "system",
            "content": system_content,
        },
        {"role": "user", "content": user_content},
    ]

    tokenized_chat = tokenizer.apply_chat_template(
        messages, tokenize=True, add_generation_prompt=True, return_tensors="pt"
    ).to("cuda")

    text_streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
    output = model.generate(
        tokenized_chat,
        streamer=text_streamer,
        max_new_tokens=2048,
        temperature=0.1,
        repetition_penalty=1.11,
        top_p=0.92,
        top_k=1000,
        pad_token_id=tokenizer.pad_token_id,
        eos_token_id=tokenizer.eos_token_id,
        do_sample=True,
    )

    generated_text = tokenizer.decode(output[0], skip_special_tokens=True)

generate("Nabroj mi sve planete suncevog sistemai reci mi koja je najveca planeta")

generate("Koja je razlika između lame, vikune i alpake?")

generate("Napišite kratku e-poruku Semu Altmanu dajući razloge za GPT-4 otvorenog koda")