Spaces:
Sleeping
Sleeping
File size: 3,862 Bytes
b1c0d9b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 |
import transformers
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig, pipeline
from huggingface_hub import notebook_login
from datasets import Dataset
import pandas as pd
from IPython.display import display
from google.colab import userdata
!rm -rf ~/.cache/huggingface/
notebook_login()
import pandas as pd
df = pd.read_csv("/content/sample.csv")
print("Shape of dataset:", df.shape)
display(df.head(5))
inbound_df = df[df["inbound"] == True]
outbound_df = df[df["inbound"] == False]
merged_df = pd.merge(
inbound_df,
outbound_df,
left_on="tweet_id",
right_on="in_response_to_tweet_id",
suffixes=("_customer", "_brand")
)
merged_df = merged_df[["tweet_id_customer", "text_customer",
"tweet_id_brand", "text_brand"]]
display(merged_df.head())
def build_chat_example(row):
return {
"prompt": f"User: {row['text_customer']}\nAssistant:",
"response": row["text_brand"]
}
paired_data = merged_df.apply(build_chat_example, axis=1).to_list()
from datasets import Dataset
dataset = Dataset.from_list(paired_data)
dataset = dataset.train_test_split(test_size=0.1, seed=42)
train_dataset = dataset["train"]
eval_dataset = dataset["test"]
print(train_dataset[0])
model_id = "meta-llama/Meta-Llama-3-8B"
import bitsandbytes as bnb
from transformers import BitsAndBytesConfig
import torch # Import torch here
bnb_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_use_double_quant = True,
bnb_4bit_quant_type = "nf4",
bnb_4bit_compute_dtype = torch.bfloat16
)
tokenizer = AutoTokenizer.from_pretrained(model_id)
tokenizer.pad_token = tokenizer.eos_token
model = AutoModelForCausalLM.from_pretrained(
model_id,
device_map = "auto",
quantization_config = bnb_config,
)
llama_pipeline = pipeline(
"text-generation",
model=model,
tokenizer=tokenizer,
torch_dtype=torch.bfloat16,
device_map="auto"
)
from peft import LoraConfig, get_peft_model, TaskType
lora_config = LoraConfig(
r=16,
lora_alpha=32,
lora_dropout=0.05,
bias="none",
task_type=TaskType.CAUSAL_LM,
target_modules=["q_proj", "v_proj"]
)
model = get_peft_model(model, lora_config)
print("LoRA layers added to the model!")
def tokenize_function(examples):
full_texts = [
f"{p}\n{r}" for p, r in zip(examples["prompt"], examples["response"])
]
return tokenizer(full_texts, truncation=True, max_length=512)
train_tokenized = train_dataset.map(tokenize_function, batched=True)
eval_tokenized = eval_dataset.map(tokenize_function, batched=True)
from transformers import TrainingArguments, Trainer
training_args = TrainingArguments(
output_dir="./results",
overwrite_output_dir=True,
num_train_epochs=3,
per_device_train_batch_size=2,
per_device_eval_batch_size=2,
eval_strategy="epoch",
save_strategy="epoch",
logging_steps=50,
fp16=True,
report_to="none"
)
def tokenize_function(examples):
full_texts = [
f"{p}\n{r}" for p, r in zip(examples["prompt"], examples["response"])
]
tokenized_inputs = tokenizer(
full_texts,
truncation=True,
max_length=512,
padding="max_length",
return_tensors="pt"
)
return tokenized_inputs
train_tokenized = train_dataset.map(tokenize_function, batched=True)
eval_tokenized = eval_dataset.map(tokenize_function, batched=True)
model.eval()
test_prompt = "User: What is an IPhone?\nAssistant:"
inputs = tokenizer(test_prompt, return_tensors="pt").to("cuda")
with torch.no_grad():
outputs = model.generate(
**inputs,
max_new_tokens=100,
do_sample=True,
top_k=50,
top_p=0.9
)
print("=== Model Reply ===")
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
|