ritika-kumar/finetuned-llama2-7b-en-hi

Finetuning

This model is a fine-tuned version of 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