File size: 1,806 Bytes
98985f3
 
 
b5dfd52
bbc0512
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60a0592
bbc0512
98985f3
2789d18
 
b084026
e363f01
 
 
 
 
 
 
2789d18
e363f01
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
from transformers import AutoTokenizer
import gradio as gr

def tokenize(input_text):
    llama_tokens = len(llama_tokenizer(input_text, add_special_tokens=True)["input_ids"])
    mistral_tokens = len(mistral_tokenizer(input_text, add_special_tokens=True)["input_ids"])
    gpt2_tokens = len(gpt2_tokenizer(input_text, add_special_tokens=True)["input_ids"])
    gpt_neox_tokens = len(gpt_neox_tokenizer(input_text, add_special_tokens=True)["input_ids"])
    falcon_tokens = len(falcon_tokenizer(input_text, add_special_tokens=True)["input_ids"])
    phi2_tokens = len(phi2_tokenizer(input_text, add_special_tokens=True)["input_ids"])
    t5_tokens = len(t5_tokenizer(input_text, add_special_tokens=True)["input_ids"])

    token_lengths = {
        "LLaMa": llama_tokens,
        "Mistral": mistral_tokens,
        "GPT-2/GPT-J": gpt2_tokens,
        "GPT-NeoX": gpt_neox_tokens,
        "Falcon": falcon_tokens,
        "Phi-2": phi2_tokens,
        "T5": t5_tokens
    }

    sorted_tokens = sorted(token_lengths.items(), key=lambda x: x[1], reverse=True)
    result = "\n".join([f"{name}: {length}" for name, length in sorted_tokens])

    return result


if __name__ == "__main__":
    llama_tokenizer = AutoTokenizer.from_pretrained("TheBloke/Llama-2-7B-fp16")
    mistral_tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")
    gpt2_tokenizer = AutoTokenizer.from_pretrained("gpt2")
    gpt_neox_tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neox-20b")
    falcon_tokenizer = AutoTokenizer.from_pretrained("tiiuae/falcon-7b")
    phi2_tokenizer = AutoTokenizer.from_pretrained("microsoft/phi-2")
    t5_tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-xxl")

    iface = gr.Interface(fn=tokenize, inputs=gr.Textbox(lines=7), outputs="text")
    iface.launch()