---
language:
- en
license: apache-2.0
tags:
- text-generation-inference
- transformers
- unsloth
- mistral
- trl
- sft
base_model: unsloth/mistral-7b-bnb-4bit
---

# Nepali GPT
Nepali GPT is a large Nepali language fine-tuned model based on Mixtral_7B.The fine-tuning process uses Unsloth, expediting the training process for optimal efficiency.



## Model Description
* Model type: A 7B fine-tuned model
* Primary Language(s): Nepali
* License: Mistral


### Installation
```
#Install Unsloth
%%capture
import torch
major_version, minor_version = torch.cuda.get_device_capability()
# Must install separately since Colab has torch 2.2.1, which breaks packages
!pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
if major_version >= 8:
    # Use this for new GPUs like Ampere, Hopper GPUs (RTX 30xx, RTX 40xx, A100, H100, L40)
    !pip install --no-deps packaging ninja einops flash-attn xformers trl peft accelerate bitsandbytes
else:
    # Use this for older GPUs (V100, Tesla T4, RTX 20xx)
    !pip install --no-deps xformers trl peft accelerate bitsandbytes
pass
```

### Model loading
```
from unsloth import FastLanguageModel
import torch
max_seq_length = 2048
dtype = None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+
load_in_4bit = True # Use 4bit quantization to reduce memory usage. Can be False.

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "Heem2/NEPALIGPT-1.0",
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
)

prompt = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
{}

### Input:
{}

### Response:
{}"""

```
### Inference
```
FastLanguageModel.for_inference(model)
inputs = tokenizer(
[
    prompt.format(
        "नेपालको बारेमा व्याख्या गर्नुहोस्?", # instruction
        "संस्कृति, भाषा, भूगोल, राजनीति, जलवायु", # input
        "", # output - leave this blank for generation!
    )
], return_tensors = "pt").to("cuda")

outputs = model.generate(**inputs, max_new_tokens = 1000, use_cache = True)
tokenizer.batch_decode(outputs)


```

### Citation Information

If you find this model useful, please consider giving 👏 and citing:

```
@heem2
}
```

### Contributions

- This is developed by Hem Bahadur Gurung.Feel free to DM  if you have any questions.