---
license: gpl-3.0
datasets:
- tatsu-lab/alpaca
- yizhongw/self_instruct
- anon8231489123/ShareGPT_Vicuna_unfiltered
- NeelNanda/pile-10k
language:
- en
- es
- ar
- fr
- fa
metrics:
- accuracy
- bleu
pipeline_tag: text-generation
---


this model uses Task classification and the conversation is between USER and Answer or AI 

# NOTE ⚠️


This model is a finetuned version of Kolla with LGeM data With Respect to them and changes some data and optimizers
The model includes pre-trained Weights so it is GNU v3.0 licensed as same as Original Llama Model 

# Using Model in Huggingface Transformers

## Examples 🚀 

```text
CONVERSATION: USER: how can I start to work out more \n
```
```text
Q&A: USER: how can I start to work out more \n
```
```text
INFO: USER: how can I start to work out more \n
```

```python
from transformers import LlamaTokenizer, LlamaForCausalLM, pipeline
import torch
import textwrap
tokenizer = LlamaTokenizer.from_pretrained("erfanzar/LGeM-7B-MT")

model = LlamaForCausalLM.from_pretrained(
    'erfanzar/LGeM-7B-MT',
    load_in_8bit=True, 
    device_map='auto',
    torch_dtype=torch.float16,
)

pipe_line = pipeline(
    "text-generation",
    model=model, 
    tokenizer=tokenizer, 
    max_length=256, # Your max length here
    temperature=1, # Temperature (use 1 for good performance)
    top_p=0.95,
)

verify_text = lambda txt : '\n'.join([textwrap.fill(txt, width=90) for txt in txt.split('\n')])
with torch.no_grad():
  output = pipe_line('CONVERSATION: USER: code a program for me to check internet connection in python ? ')
  print(verify_text(output[0]['generated_text']))

```
### Generate Method to get res Text by Text

```python

def generate(model_,input_ids_,tokeinzer_,max_length:int=256,temperature :float= 1,eos_token_id:int=2):
  with torch.no_grad():
    before_start = len(input_ids_[0])+1
    for _ in range(max_length):
      out = model_(
          input_ids=input_ids_,
          return_dict=True,
      )
      opa = torch.nn.functional.softmax(out.logits[:,-1,:]/temperature)
      Camila = torch.multinomial(opa,1)
      input_ids_ = torch.cat([input_ids_,Camila],-1)
      clear_output(wait=True)
      print(f"\r{tokeinzer_.decode(input_ids_[0],skip_special_tokens=True)[before_start:]}",end='')
      if Camila[0].item() == eos_token_id:
        break
      yield tokeinzer_.decode(Camila[0],skip_special_tokens=True)
  return f"{tokeinzer_.decode(input_ids_[0],skip_special_tokens=True)[before_start:]}"
```


### Result 

```python
import socket
import time

def check_internet_connection():
    try:
        s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
        s.connect(("www.google.com", 80))
        print("Internet connection is active.")
    except:
        print("Internet connection is not active.")

if __name__ == "__main__":

  check_internet_connection()
```


# Using Model in OST


### LGeM 🚀

- what is LGeM, LGeM is a CausalLM Model that is trained on self instruct data (Alpaca data) and for initialization of the first train of the main model (weights are available) I used pre weights from Alpaca LoRA (open source) 

- it's Decoder Only
- built-in Pytorch
- you can simply import models like

```python
from modules import LGeMForCausalLM
```

- and Training code is available at LGeM-Train.py (check source)
- training parameters 
- - learning rate 1e-4
- - AdamW (weight decay 1e-2)
- - batch 2
- - A 100 80GB used for training (4 X)
- - Train Time 800 hours
- - budget 760 $
``` shell
python3 LGeM-train.py
```