metadata
datasets:
- HuggingFaceH4/CodeAlpaca_20K
language:
- en
library_name: transformers
pipeline_tag: text-generation
tags:
- code
- LLaMa2
LLaMaCoder
Model Description
LLaMaCoder is based on LLaMa2 7B language model, finetuned using LoRA adaptors.
Usage
Generate code with LLaMaCoder in 4bit model according to the following python snippet:
from transformers import AutoModelForCausalLM, BitsAndBytesConfig, AutoTokenizer
import torch
MODEL_NAME = "Sakuna/LLaMaCoderAll"
device = "cuda:0"
bnb_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_compute_dtype=torch.float16,
)
model = AutoModelForCausalLM.from_pretrained(
MODEL_NAME,
quantization_config=bnb_config,
trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token
model = model.to(device)
model.eval()
prompt = "Write a Java program to calculate the factorial of a given number k"
input = f"{prompt}\n### Solution:\n"
device = "cuda:0"
inputs = tokenizer(input, return_tensors="pt").to(device)
outputs = model.generate(**inputs, max_length=256, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))