---
library_name: peft

license: apache-2.0
base_model: meta-llama/Llama-2-7b-hf
datasets:
- cfilt/iitb-english-hindi
language:
- en
- hi
metrics:
- bleu
---


# Finetuning

This model is a fine-tuned version of [meta-llama/Llama-2-7b-hf](https://huggingface.co/meta-llama/Llama-2-7b-hf) on the IITB English to Hindi dataset.
source group: English
target group: Hindi


## Model description

meta-llama/Llama-2-7b-hf finetuned for translation task in Hindi language

## Training and evaluation data

cfilt/iitb-english-hindi

### Training hyperparameters

The following hyperparameters were used during training:
- num_train_epochs=1
- per_device_train_batch_size=4
- per_device_eval_batch_size = 4
- gradient_accumulation_steps=1
- optim="paged_adamw_32bit"
- learning_rate=2e-4
- weight_decay=0.001
- fp16=True
- max_grad_norm=0.3
- max_steps=-1
- warmup_ratio=0.03
- group_by_length=True
- lr_scheduler_type="constant"
### Benchamark Evaluation
- BLEU score on Tatoeba: 12.605968092174914
- BLUE score on IN-22: 25.893729634826876

## Training procedure


The following `bitsandbytes` quantization config was used during training:
- load_in_8bit: False
- load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: False
- bnb_4bit_compute_dtype: float16


### Framework versions
- PEFT 0.4.0
- Transformers 4.42.3
- Pytorch 2.1.2
- Datasets 2.20.0
- Tokenizers 0.19.1