In [ ]:

import pandas as pd
from openai import OpenAI
import re
import time
import pickle
import re

import pandas as pd from openai import OpenAI import re import time import pickle import re

Set up¶

In [ ]:

client = OpenAI(base_url="http://localhost:1234/v1", api_key="lm-studio")

client = OpenAI(base_url="http://localhost:1234/v1", api_key="lm-studio")

In [ ]:

def get_text_between_quotes(input_string):
    pattern = r'"([^"]*)"'
    matches = re.findall(pattern, input_string)
    try:
        return matches[0]
    except Exception as e:
        return input_string

def get_text_between_quotes(input_string): pattern = r'"([^"]*)"' matches = re.findall(pattern, input_string) try: return matches[0] except Exception as e: return input_string

In [ ]:

with open('./dataset/full_formal_script_temp0.8.pkl', 'rb') as fp:
    data_formal = pickle.load(fp)

with open('./dataset/full_informal_script_temp0.8.pkl', 'rb') as fp:
    data_informal = pickle.load(fp)

with open('./dataset/full_novel_script_temp0.8.pkl', 'rb') as fp:
    data_novel = pickle.load(fp)

df = pd.read_csv('./dataset/raw.csv', index_col='uid')

with open('./dataset/full_formal_script_temp0.8.pkl', 'rb') as fp: data_formal = pickle.load(fp) with open('./dataset/full_informal_script_temp0.8.pkl', 'rb') as fp: data_informal = pickle.load(fp) with open('./dataset/full_novel_script_temp0.8.pkl', 'rb') as fp: data_novel = pickle.load(fp) df = pd.read_csv('./dataset/raw.csv', index_col='uid')

In [ ]:

data_novel

data_novel

Load model¶

Load finetuned model¶

In [ ]:

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel, PeftConfig

import torch from transformers import AutoTokenizer, AutoModelForCausalLM from peft import PeftModel, PeftConfig

In [ ]:

model_path = "./Tuning/checkpoint/Phi-3-mini-4k-instruct"
token_path = f"{model_path}/tokenize"

model_path = "./Tuning/checkpoint/Phi-3-mini-4k-instruct" token_path = f"{model_path}/tokenize"

In [ ]:

tokenizer = AutoTokenizer.from_pretrained(token_path)

tokenizer = AutoTokenizer.from_pretrained(token_path)

In [ ]:

peft_config = PeftConfig.from_pretrained(model_path)

peft_config = PeftConfig.from_pretrained(model_path)

In [ ]:

base_model = AutoModelForCausalLM.from_pretrained(
    peft_config.base_model_name_or_path,
    torch_dtype=torch.float16,
    device_map="auto",
)

base_model = AutoModelForCausalLM.from_pretrained( peft_config.base_model_name_or_path, torch_dtype=torch.float16, device_map="auto", )

In [ ]:

model = PeftModel.from_pretrained(base_model, model_path)

model = PeftModel.from_pretrained(base_model, model_path)

In [ ]:

model = model.merge_and_unload()

model = model.merge_and_unload()

In [ ]:

model.eval()

model.eval()

In [ ]:

def generate_text(prompt, max_length=400, temperature=0.7):
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_length=max_length,
            num_return_sequences=1,
            temperature=temperature,
            do_sample=True,
        )
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

def generate_text(prompt, max_length=400, temperature=0.7): inputs = tokenizer(prompt, return_tensors="pt").to(model.device) with torch.no_grad(): outputs = model.generate( **inputs, max_length=max_length, num_return_sequences=1, temperature=temperature, do_sample=True, ) return tokenizer.decode(outputs[0], skip_special_tokens=True)

In [ ]:

def get_dict(text):
    res = None
    try:
        res = re.findall('{.*}', text)[0]
        res = eval(res)
    except:
        print(f"Error: {text}")
    return res

def get_dict(text): res = None try: res = re.findall('{.*}', text)[0] res = eval(res) except: print(f"Error: {text}") return res

In [ ]:

prompt = """<|system|>
You are a helpful, smart, kind, and efficient AI assistant. You always fulfill the user's requests to the best of your ability.<|end|>
<|user|>
Extract name, age, job from the sentence to json format. if the information doesn't exist fill null.

Sentence: 'John Doe is a 35-year-old software engineer with a passion for artificial intelligence.'<|end|>
<|assistant|>
"""

generated_text_low_temp = generate_text(prompt, temperature=0.1)
get_dict(generated_text_low_temp)

prompt = """<|system|> You are a helpful, smart, kind, and efficient AI assistant. You always fulfill the user's requests to the best of your ability.<|end|> <|user|> Extract name, age, job from the sentence to json format. if the information doesn't exist fill null. Sentence: 'John Doe is a 35-year-old software engineer with a passion for artificial intelligence.'<|end|> <|assistant|> """ generated_text_low_temp = generate_text(prompt, temperature=0.1) get_dict(generated_text_low_temp)

Predict¶

In [ ]:

def call_llm(data):
    promtp = """Extract name, age, job from the sentence to json format. 
    if the information doesn't exsits fill null.
    
    sentence: '{}'
    """
    
    start_time = time.time()
    completion = client.chat.completions.create(
      model="microsoft/Phi-3-mini-4k-instruct-gguf",
      messages=[
        {"role": "system", "content": "You are a helpful, smart, kind, and efficient AI assistant. You always fulfill the user's requests to the best of your ability."},
        {"role": "user", "content": promtp.format(data)}
      ],
      temperature=0,
    )
    end_time = time.time()
    content = completion.choices[0].message.content
    return end_time-start_time, content.strip().replace('null', '\"\"')

def call_llm_model(data):
    prompt = f"""<|system|>
You are a helpful, smart, kind, and efficient AI assistant. You always fulfill the user's requests to the best of your ability.<|end|>
<|user|>
Extract name, age, job from the sentence to json format. if the information doesn't exist fill null.

Sentence: '{data}'<|end|>
<|assistant|>"""
    start_time = time.time()
    generated_text_low_temp = generate_text(prompt, temperature=0.1)
    end_time = time.time()
    return end_time-start_time, get_dict(generated_text_low_temp)

def call_llm(data): promtp = """Extract name, age, job from the sentence to json format. if the information doesn't exsits fill null. sentence: '{}' """ start_time = time.time() completion = client.chat.completions.create( model="microsoft/Phi-3-mini-4k-instruct-gguf", messages=[ {"role": "system", "content": "You are a helpful, smart, kind, and efficient AI assistant. You always fulfill the user's requests to the best of your ability."}, {"role": "user", "content": promtp.format(data)} ], temperature=0, ) end_time = time.time() content = completion.choices[0].message.content return end_time-start_time, content.strip().replace('null', '\"\"') def call_llm_model(data): prompt = f"""<|system|> You are a helpful, smart, kind, and efficient AI assistant. You always fulfill the user's requests to the best of your ability.<|end|> <|user|> Extract name, age, job from the sentence to json format. if the information doesn't exist fill null. Sentence: '{data}'<|end|> <|assistant|>""" start_time = time.time() generated_text_low_temp = generate_text(prompt, temperature=0.1) end_time = time.time() return end_time-start_time, get_dict(generated_text_low_temp)

In [ ]:

# Predict formal data
predict_data = {}
for temp in data_formal:
    for id in data_formal[temp]:
        # _, result = call_llm(data_formal[temp][id]) # Before finetune
        _, result = call_llm_model(data_formal[temp][id]) # After finetune
        try:
            if not isinstance(result, dict):
                predict_data[id] = eval(result)
            else:
                predict_data[id] = result
        except:
            print(result)

# with open('./dataset/predict_formal_script.pkl', 'wb') as fp:
#     pickle.dump(predict_data, fp)
with open('./dataset/finetune_predict_formal_script.pkl', 'wb') as fp:
    pickle.dump(predict_data, fp)

# Predict formal data predict_data = {} for temp in data_formal: for id in data_formal[temp]: # _, result = call_llm(data_formal[temp][id]) # Before finetune _, result = call_llm_model(data_formal[temp][id]) # After finetune try: if not isinstance(result, dict): predict_data[id] = eval(result) else: predict_data[id] = result except: print(result) # with open('./dataset/predict_formal_script.pkl', 'wb') as fp: # pickle.dump(predict_data, fp) with open('./dataset/finetune_predict_formal_script.pkl', 'wb') as fp: pickle.dump(predict_data, fp)

In [ ]:

# Predict informal data
predict_data = {}
for temp in data_informal:
    for id in data_informal[temp]:
        # _, result = call_llm(data_informal[temp][id]) # Before finetune
        _, result = call_llm_model(data_informal[temp][id]) # After finetune
        try:
            if not isinstance(result, dict):
                predict_data[id] = eval(result)
            else:
                predict_data[id] = result
        except:
            print(result)

# with open('./dataset/predict_informal_script.pkl', 'wb') as fp:
#     pickle.dump(predict_data, fp)
with open('./dataset/finetune_predict_informal_script.pkl', 'wb') as fp:
    pickle.dump(predict_data, fp)

# Predict informal data predict_data = {} for temp in data_informal: for id in data_informal[temp]: # _, result = call_llm(data_informal[temp][id]) # Before finetune _, result = call_llm_model(data_informal[temp][id]) # After finetune try: if not isinstance(result, dict): predict_data[id] = eval(result) else: predict_data[id] = result except: print(result) # with open('./dataset/predict_informal_script.pkl', 'wb') as fp: # pickle.dump(predict_data, fp) with open('./dataset/finetune_predict_informal_script.pkl', 'wb') as fp: pickle.dump(predict_data, fp)

In [ ]:

# Predict novel data
predict_data = {}
for temp in data_novel:
    for id in data_novel[temp]:
        # _, result = call_llm(data_novel[temp][id]) # Before finetune
        _, result = call_llm_model(data_novel[temp][id]) # After finetune
        try:
            if not isinstance(result, dict):
                predict_data[id] = eval(result)
            else:
                predict_data[id] = result
        except:
            print(result)

# with open('./dataset/predict_novel_script.pkl', 'wb') as fp:
#     pickle.dump(predict_data, fp)
with open('./dataset/finetune_predict_novel_script.pkl', 'wb') as fp:
    pickle.dump(predict_data, fp)

# Predict novel data predict_data = {} for temp in data_novel: for id in data_novel[temp]: # _, result = call_llm(data_novel[temp][id]) # Before finetune _, result = call_llm_model(data_novel[temp][id]) # After finetune try: if not isinstance(result, dict): predict_data[id] = eval(result) else: predict_data[id] = result except: print(result) # with open('./dataset/predict_novel_script.pkl', 'wb') as fp: # pickle.dump(predict_data, fp) with open('./dataset/finetune_predict_novel_script.pkl', 'wb') as fp: pickle.dump(predict_data, fp)

Evaluate¶

Calculate score¶

In [ ]:

with open('./dataset/predict_formal_script.pkl', 'rb') as fp:
    predict_formal = pickle.load(fp)

with open('./dataset/predict_informal_script.pkl', 'rb') as fp:
    predict_informal = pickle.load(fp)

with open('./dataset/predict_novel_script.pkl', 'rb') as fp:
    predict_novel = pickle.load(fp)

with open('./dataset/finetune_predict_formal_script.pkl', 'rb') as fp:
    finetune_predict_formal = pickle.load(fp)

with open('./dataset/finetune_predict_informal_script.pkl', 'rb') as fp:
    finetune_predict_informal = pickle.load(fp)

with open('./dataset/finetune_predict_novel_script.pkl', 'rb') as fp:
    finetune_predict_novel = pickle.load(fp)

with open('./dataset/full_formal_script_temp0.8.pkl', 'rb') as fp:
    data_formal = pickle.load(fp)

with open('./dataset/full_informal_script_temp0.8.pkl', 'rb') as fp:
    data_informal = pickle.load(fp)

with open('./dataset/full_novel_script_temp0.8.pkl', 'rb') as fp:
    data_novel = pickle.load(fp)

df = pd.read_csv('./dataset/raw.csv', index_col='uid')

with open('./dataset/predict_formal_script.pkl', 'rb') as fp: predict_formal = pickle.load(fp) with open('./dataset/predict_informal_script.pkl', 'rb') as fp: predict_informal = pickle.load(fp) with open('./dataset/predict_novel_script.pkl', 'rb') as fp: predict_novel = pickle.load(fp) with open('./dataset/finetune_predict_formal_script.pkl', 'rb') as fp: finetune_predict_formal = pickle.load(fp) with open('./dataset/finetune_predict_informal_script.pkl', 'rb') as fp: finetune_predict_informal = pickle.load(fp) with open('./dataset/finetune_predict_novel_script.pkl', 'rb') as fp: finetune_predict_novel = pickle.load(fp) with open('./dataset/full_formal_script_temp0.8.pkl', 'rb') as fp: data_formal = pickle.load(fp) with open('./dataset/full_informal_script_temp0.8.pkl', 'rb') as fp: data_informal = pickle.load(fp) with open('./dataset/full_novel_script_temp0.8.pkl', 'rb') as fp: data_novel = pickle.load(fp) df = pd.read_csv('./dataset/raw.csv', index_col='uid')

In [ ]:

datasets = ['formal', 'informal', 'novel']
features = ['name', 'age', 'job']
for dataset in datasets:
    for feature in features:
        key = f"pred_{dataset}_{feature}"
        key = f"finetune_pred_{dataset}_{feature}"
        df[key] = None

datasets = ['formal', 'informal', 'novel'] features = ['name', 'age', 'job'] for dataset in datasets: for feature in features: key = f"pred_{dataset}_{feature}" key = f"finetune_pred_{dataset}_{feature}" df[key] = None

Evaluate finetune model¶

In [ ]:

for i in finetune_predict_formal:
    dataset = 'formal'
    for feature in features:
        key = f"finetune_pred_{dataset}_{feature}"
        try:
            if isinstance(finetune_predict_formal[i][feature], list):
                df.loc[i, key] =  " ".join(finetune_predict_formal[i][feature])
            else:
                df.loc[i, key] =  finetune_predict_formal[i][feature]
        except Exception as e:
            print(f"{e}")

for i in finetune_predict_informal:
    dataset = 'informal'
    for feature in features:
        key = f"finetune_pred_{dataset}_{feature}"
        try:
            if isinstance(finetune_predict_informal[i][feature], list):
                df.loc[i, key] =  " ".join(finetune_predict_informal[i][feature])
            else:
                df.loc[i, key] =  finetune_predict_informal[i][feature]
        except Exception as e:
            print(f"{e}")

for i in finetune_predict_novel:
    dataset = 'novel'
    for feature in features:
        key = f"finetune_pred_{dataset}_{feature}"
        try:
            if isinstance(finetune_predict_novel[i][feature], list):
                df.loc[i, key] =  " ".join(finetune_predict_novel[i][feature])
            else:
                df.loc[i, key] =  finetune_predict_novel[i][feature]
        except Exception as e:
            print(f"{e}")

for i in finetune_predict_formal: dataset = 'formal' for feature in features: key = f"finetune_pred_{dataset}_{feature}" try: if isinstance(finetune_predict_formal[i][feature], list): df.loc[i, key] = " ".join(finetune_predict_formal[i][feature]) else: df.loc[i, key] = finetune_predict_formal[i][feature] except Exception as e: print(f"{e}") for i in finetune_predict_informal: dataset = 'informal' for feature in features: key = f"finetune_pred_{dataset}_{feature}" try: if isinstance(finetune_predict_informal[i][feature], list): df.loc[i, key] = " ".join(finetune_predict_informal[i][feature]) else: df.loc[i, key] = finetune_predict_informal[i][feature] except Exception as e: print(f"{e}") for i in finetune_predict_novel: dataset = 'novel' for feature in features: key = f"finetune_pred_{dataset}_{feature}" try: if isinstance(finetune_predict_novel[i][feature], list): df.loc[i, key] = " ".join(finetune_predict_novel[i][feature]) else: df.loc[i, key] = finetune_predict_novel[i][feature] except Exception as e: print(f"{e}")

Evaluate base model¶

In [ ]:

for i in predict_formal:
    dataset = 'formal'
    for feature in features:
        key = f"pred_{dataset}_{feature}"
        try:
            if isinstance(predict_formal[i][feature], list):
                df.loc[i, key] =  " ".join(predict_formal[i][feature])
            else:
                df.loc[i, key] =  predict_formal[i][feature]
        except Exception as e:
            print(f"{e}")

for i in predict_informal:
    dataset = 'informal'
    for feature in features:
        key = f"pred_{dataset}_{feature}"
        try:
            if isinstance(predict_informal[i][feature], list):
                df.loc[i, key] =  " ".join(predict_informal[i][feature])
            else:
                df.loc[i, key] =  predict_informal[i][feature]
        except Exception as e:
            print(f"{e}")

for i in predict_novel:
    dataset = 'novel'
    for feature in features:
        key = f"pred_{dataset}_{feature}"
        try:
            if isinstance(predict_novel[i][feature], list):
                df.loc[i, key] =  " ".join(predict_novel[i][feature])
            else:
                df.loc[i, key] =  predict_novel[i][feature]
        except Exception as e:
            print(f"{e}")

for i in predict_formal: dataset = 'formal' for feature in features: key = f"pred_{dataset}_{feature}" try: if isinstance(predict_formal[i][feature], list): df.loc[i, key] = " ".join(predict_formal[i][feature]) else: df.loc[i, key] = predict_formal[i][feature] except Exception as e: print(f"{e}") for i in predict_informal: dataset = 'informal' for feature in features: key = f"pred_{dataset}_{feature}" try: if isinstance(predict_informal[i][feature], list): df.loc[i, key] = " ".join(predict_informal[i][feature]) else: df.loc[i, key] = predict_informal[i][feature] except Exception as e: print(f"{e}") for i in predict_novel: dataset = 'novel' for feature in features: key = f"pred_{dataset}_{feature}" try: if isinstance(predict_novel[i][feature], list): df.loc[i, key] = " ".join(predict_novel[i][feature]) else: df.loc[i, key] = predict_novel[i][feature] except Exception as e: print(f"{e}")

In [ ]:

for dataset in datasets:
    for feature in features:
        key = f"pred_{dataset}_{feature}"
        score_key = f"pred_{dataset}_{feature}_score"
        df[score_key] = df[feature] == df[key]
        
        key = f"finetune_pred_{dataset}_{feature}"
        score_key = f"finetune_pred_{dataset}_{feature}_score"
        df[score_key] = df[feature] == df[key]

for dataset in datasets: for feature in features: key = f"pred_{dataset}_{feature}" score_key = f"pred_{dataset}_{feature}_score" df[score_key] = df[feature] == df[key] key = f"finetune_pred_{dataset}_{feature}" score_key = f"finetune_pred_{dataset}_{feature}_score" df[score_key] = df[feature] == df[key]

Calculate all dataset¶

In [ ]:

lenght = len(df)
for dataset in datasets:
    print(f"-------- {dataset.capitalize()} --------")
    for feature in features:
        score_key = f"pred_{dataset}_{feature}_score"
        print(f"{feature.capitalize()} accuracy: {sum(df[score_key]) / len(df)}")
    print(f"========================================\n")


for dataset in datasets:
    print(f"-------- Finetune - {dataset.capitalize()} --------")
    for feature in features:
        score_key = f"finetune_pred_{dataset}_{feature}_score"
        print(f"{feature.capitalize()} accuracy: {sum(df[score_key]) / len(df)}")
    print(f"========================================\n")

lenght = len(df) for dataset in datasets: print(f"-------- {dataset.capitalize()} --------") for feature in features: score_key = f"pred_{dataset}_{feature}_score" print(f"{feature.capitalize()} accuracy: {sum(df[score_key]) / len(df)}") print(f"========================================\n") for dataset in datasets: print(f"-------- Finetune - {dataset.capitalize()} --------") for feature in features: score_key = f"finetune_pred_{dataset}_{feature}_score" print(f"{feature.capitalize()} accuracy: {sum(df[score_key]) / len(df)}") print(f"========================================\n")

In [ ]:

# -------- Formal --------
# Name accuracy: 0.8715
# Age accuracy: 0.959
# Job accuracy: 0.612
# ========================================

# -------- Informal --------
# Name accuracy: 0.8985
# Age accuracy: 0.989
# Job accuracy: 0.686
# ========================================

# -------- Novel --------
# Name accuracy: 0.882
# Age accuracy: 0.95
# Job accuracy: 0.5505
# ========================================

# -------- Finetune - Formal --------
# Name accuracy: 0.984
# Age accuracy: 0.984
# Job accuracy: 0.9995
# ========================================

# -------- Finetune - Informal --------
# Name accuracy: 0.9775
# Age accuracy: 0.9985
# Job accuracy: 1.0
# ========================================

# -------- Finetune - Novel --------
# Name accuracy: 0.949
# Age accuracy: 0.9595
# Job accuracy: 0.9885
# ========================================

# -------- Formal -------- # Name accuracy: 0.8715 # Age accuracy: 0.959 # Job accuracy: 0.612 # ======================================== # -------- Informal -------- # Name accuracy: 0.8985 # Age accuracy: 0.989 # Job accuracy: 0.686 # ======================================== # -------- Novel -------- # Name accuracy: 0.882 # Age accuracy: 0.95 # Job accuracy: 0.5505 # ======================================== # -------- Finetune - Formal -------- # Name accuracy: 0.984 # Age accuracy: 0.984 # Job accuracy: 0.9995 # ======================================== # -------- Finetune - Informal -------- # Name accuracy: 0.9775 # Age accuracy: 0.9985 # Job accuracy: 1.0 # ======================================== # -------- Finetune - Novel -------- # Name accuracy: 0.949 # Age accuracy: 0.9595 # Job accuracy: 0.9885 # ========================================

Calculate only test dataset¶

In [ ]:

lenght = len(df)
test_df = df[1800:]
for dataset in datasets:
    print(f"-------- Test dataset - {dataset.capitalize()} --------")
    for feature in features:
        score_key = f"pred_{dataset}_{feature}_score"
        print(f"{feature.capitalize()} accuracy: {sum(test_df[score_key]) / len(test_df)}")
    print(f"========================================\n")


for dataset in datasets:
    print(f"-------- Finetune test dataset - {dataset.capitalize()} --------")
    for feature in features:
        score_key = f"finetune_pred_{dataset}_{feature}_score"
        print(f"{feature.capitalize()} accuracy: {sum(test_df[score_key]) / len(test_df)}")
    print(f"========================================\n")

lenght = len(df) test_df = df[1800:] for dataset in datasets: print(f"-------- Test dataset - {dataset.capitalize()} --------") for feature in features: score_key = f"pred_{dataset}_{feature}_score" print(f"{feature.capitalize()} accuracy: {sum(test_df[score_key]) / len(test_df)}") print(f"========================================\n") for dataset in datasets: print(f"-------- Finetune test dataset - {dataset.capitalize()} --------") for feature in features: score_key = f"finetune_pred_{dataset}_{feature}_score" print(f"{feature.capitalize()} accuracy: {sum(test_df[score_key]) / len(test_df)}") print(f"========================================\n")

In [ ]:

# -------- Test dataset - Formal --------
# Name accuracy: 0.87
# Age accuracy: 0.965
# Job accuracy: 0.685
# ========================================

# -------- Test dataset - Informal --------
# Name accuracy: 0.895
# Age accuracy: 0.98
# Job accuracy: 0.75
# ========================================

# -------- Test dataset - Novel --------
# Name accuracy: 0.875
# Age accuracy: 0.935
# Job accuracy: 0.55
# ========================================

# -------- Finetune test dataset - Formal --------
# Name accuracy: 0.965
# Age accuracy: 0.975
# Job accuracy: 1.0
# ========================================

# -------- Finetune test dataset - Informal --------
# Name accuracy: 0.96
# Age accuracy: 0.995
# Job accuracy: 1.0
# ========================================

# -------- Finetune test dataset - Novel --------
# Name accuracy: 0.925
# Age accuracy: 0.93
# Job accuracy: 0.955
# ========================================

# -------- Test dataset - Formal -------- # Name accuracy: 0.87 # Age accuracy: 0.965 # Job accuracy: 0.685 # ======================================== # -------- Test dataset - Informal -------- # Name accuracy: 0.895 # Age accuracy: 0.98 # Job accuracy: 0.75 # ======================================== # -------- Test dataset - Novel -------- # Name accuracy: 0.875 # Age accuracy: 0.935 # Job accuracy: 0.55 # ======================================== # -------- Finetune test dataset - Formal -------- # Name accuracy: 0.965 # Age accuracy: 0.975 # Job accuracy: 1.0 # ======================================== # -------- Finetune test dataset - Informal -------- # Name accuracy: 0.96 # Age accuracy: 0.995 # Job accuracy: 1.0 # ======================================== # -------- Finetune test dataset - Novel -------- # Name accuracy: 0.925 # Age accuracy: 0.93 # Job accuracy: 0.955 # ========================================