---
license: apache-2.0
base_model: bn22/Mistral-7B-Instruct-v0.1-sharded
tags:
- generated_from_trainer
model-index:
- name: outputs
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# UrduNewsMistral7B

This model is a fine-tuned version of [bn22/Mistral-7B-Instruct-v0.1-sharded](https://huggingface.co/bn22/Mistral-7B-Instruct-v0.1-sharded) on [Urdu-Instruct-News-Article-Generation](https://huggingface.co/datasets/AhmadMustafa/Urdu-Instruct-News-Article-Generation) dataset.
It achieves the following results on the evaluation set:
- Loss: 1.1925

## Usage
Here is an example of how you would load:
```python
import torch
from peft import PeftModel    
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

model_name = "bn22/Mistral-7B-Instruct-v0.1-sharded"
adapters_name = 'mwz/UrduNewsMistral7B'

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    load_in_4bit=True,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    max_memory= {i: '24000MB' for i in range(torch.cuda.device_count())},
    quantization_config=BitsAndBytesConfig(
        load_in_4bit=True,
        bnb_4bit_compute_dtype=torch.bfloat16,
        bnb_4bit_use_double_quant=True,
        bnb_4bit_quant_type='nf4'
    ),
)
model = PeftModel.from_pretrained(model, adapters_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

```
Inference can then be performed as usual with HF models as follows:
```python
prompt = "پیٹرول کی قیمت میں 2روپے 50 پیسے اضافہ"
formatted_prompt = (
    f"اس دی گی ایک خبر سے متعلق ایک مضمون لکھیں"
    f"### Human: {prompt} ### Assistant:"
)
inputs = tokenizer(formatted_prompt, return_tensors="pt").to("cuda:0")
outputs = model.generate(inputs=inputs.input_ids, max_new_tokens=20)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```
Expected output similar to the following:
```
اس دی گی ایک خبر سے متعلق ایک مضمون لکھیں۔ خبر:### Human:  پیٹرول کی قیمت میں 2روپے 50 پیسے اضافہ ### Assistant: بہترین پیٹرول کے ان
```


## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 0.0002
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.05
- num_epochs: 50
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:----:|:---------------:|
| 1.1387        | 5.71  | 10   | 1.2070          |
| 0.9337        | 11.43 | 20   | 1.1634          |
| 0.8676        | 17.14 | 30   | 1.1697          |
| 0.8065        | 22.86 | 40   | 1.1868          |
| 0.7759        | 28.57 | 50   | 1.1925          |


### Framework versions

- Transformers 4.36.0.dev0
- Pytorch 2.1.0+cu121
- Datasets 2.15.0
- Tokenizers 0.15.0