|
--- |
|
license: llama2 |
|
datasets: |
|
- uonlp/CulturaX |
|
language: |
|
- tr |
|
- en |
|
pipeline_tag: text-generation |
|
metrics: |
|
- accuracy |
|
- bleu |
|
--- |
|
|
|
|
|
|
|
# Commencis-LLM |
|
|
|
<!-- Provide a quick summary of what the model is/does. --> |
|
Commencis LLM is a generative model based on the Mistral 7B model. The base model adapts Mistral 7B to Turkish Banking specifically by training on a diverse dataset obtained through various methods, encompassing general Turkish and banking data. |
|
## Model Description |
|
<!-- Provide a longer summary of what this model is. --> |
|
|
|
- **Developed by:** [Commencis](https://www.commencis.com) |
|
- **Language(s):** Turkish |
|
- **Finetuned from model:** [Mistral 7B](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2) |
|
- **Input:** Model input text only |
|
- **Output:** Model generates text only |
|
- **Blog Post**: |
|
|
|
## Training Details |
|
Alignment phase consists of two stages: supervised fine-tuning (SFT) and Reward Modeling with Reinforcement learning from human feedback (RLHF). |
|
|
|
The SFT phase was done on the a mixture of synthetic datasets generated from comprehensive banking dictionary data, synthetic datasets generated from banking-based domain and sub-domain headings, and derived from the CulturaX Turkish dataset by filtering. It was trained with three epochs. We used a learning rate 2e-5, lora rank 64 and maximum sequence length 1024 tokens. |
|
|
|
### Usage |
|
|
|
### Suggested Inference Parameters |
|
- Temperature: 0.5 |
|
- Repetition penalty: 1.0 |
|
- Top-p: 0.9 |
|
|
|
```python |
|
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline |
|
|
|
class TextGenerationAssistant: |
|
def __init__(self, model_id:str): |
|
self.tokenizer = AutoTokenizer.from_pretrained(model_id) |
|
self.model = AutoModelForCausalLM.from_pretrained(model_id, device_map='auto', load_in_8bit=True) |
|
self.pipe = pipeline("text-generation", |
|
model=self.model, |
|
tokenizer=self.tokenizer, |
|
device_map="auto", |
|
max_new_tokens=1024, |
|
return_full_text=True, |
|
repetition_penalty=1.0 |
|
) |
|
|
|
self.sampling_params = dict(do_sample=True, temperature=0.5, top_k=50, top_p=0.9) |
|
self.SYSTEM_PROMPT = "Sen yardımcı bir asistansın. Sana verilen talimat ve girdilere en uygun cevapları üreteceksin. \n\n\n" |
|
|
|
def format_prompt(self, user_input): |
|
return "[INST] " + self.SYSTEM_PROMPT + user_input + " [/INST]" |
|
|
|
def generate_response(self, user_query): |
|
prompt = self.format_prompt(user_query) |
|
outputs = self.pipe(prompt, **self.sampling_params) |
|
return outputs[0]["generated_text"].split("[/INST]")[-1] |
|
|
|
|
|
assistant = TextGenerationAssistant(model_id="Commencis/Commencis-LLM") |
|
|
|
# Enter your query here. |
|
user_query = "Faiz oranları yükseldiğinde kredilerim nasıl etkilenir?" |
|
response = assistant.generate_response(user_query) |
|
print(response) |
|
|
|
``` |
|
|
|
### Chat Template |
|
|
|
```python |
|
from transformers import AutoTokenizer |
|
import transformers |
|
import torch |
|
|
|
model = "Commencis/Commencis-LLM" |
|
messages = [{"role": "user", "content": "Faiz oranları yükseldiğinde kredilerim nasıl etkilenir?"}] |
|
|
|
tokenizer = AutoTokenizer.from_pretrained(model) |
|
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) |
|
pipeline = transformers.pipeline( |
|
"text-generation", |
|
model=model, |
|
torch_dtype=torch.float16, |
|
device_map="auto", |
|
) |
|
|
|
outputs = pipeline(prompt, max_new_tokens=1024, do_sample=True, temperature=0.5, top_k=50, top_p=0.9) |
|
print(outputs[0]["generated_text"]) |
|
``` |
|
|
|
## Bias, Risks, and Limitations |
|
|
|
<!-- This section is meant to convey both technical and sociotechnical limitations. --> |
|
|
|
Like all LLMs, Commencis-LLM has certain limitations: |
|
- Hallucination: Model may sometimes generate responses that contain plausible-sounding but factually incorrect or irrelevant information. |
|
- Code Switching: The model might unintentionally switch between languages or dialects within a single response, affecting the coherence and understandability of the output. |
|
- Repetition: The Model may produce repetitive phrases or sentences, leading to less engaging and informative responses. |
|
- Coding and Math: The model's performance in generating accurate code or solving complex mathematical problems may be limited. |
|
- Toxicity: The model could inadvertently generate responses containing inappropriate or harmful content. |
