WebTokenizer / app.py
xzuyn's picture
Add more models
e9cb382 verified
raw
history blame
4.25 kB
from transformers import AutoTokenizer
import gradio as gr
def formatarr(input):
return "["+",".join(str(x) for x in input)+"]"
def tokenize(input_text):
llama_tokens = llama_tokenizer(input_text, add_special_tokens=True)["input_ids"]
llama3_tokens = llama3_tokenizer(input_text, add_special_tokens=True)["input_ids"]
mistral_tokens = mistral_tokenizer(input_text, add_special_tokens=True)["input_ids"]
gpt2_tokens = gpt2_tokenizer(input_text, add_special_tokens=True)["input_ids"]
gpt_neox_tokens = gpt_neox_tokenizer(input_text, add_special_tokens=True)["input_ids"]
falcon_tokens = falcon_tokenizer(input_text, add_special_tokens=True)["input_ids"]
phi2_tokens = phi2_tokenizer(input_text, add_special_tokens=True)["input_ids"]
phi3_tokens = phi3_tokenizer(input_text, add_special_tokens=True)["input_ids"]
t5_tokens = t5_tokenizer(input_text, add_special_tokens=True)["input_ids"]
gemma_tokens = gemma_tokenizer(input_text, add_special_tokens=True)["input_ids"]
command_r_tokens = command_r_tokenizer(input_text, add_special_tokens=True)["input_ids"]
qwen_tokens = qwen_tokenizer(input_text, add_special_tokens=True)["input_ids"]
codeqwen_tokens = codeqwen_tokenizer(input_text, add_special_tokens=True)["input_ids"]
rwkv4_tokens = rwkv4_tokenizer(input_text, add_special_tokens=True)["input_ids"]
rwkv5_tokens = rwkv5_tokenizer(input_text, add_special_tokens=True)["input_ids"]
deepseek_tokens = deepseek_tokenizer(input_text, add_special_tokens=True)["input_ids"]
internlm_tokens = internlm_tokenizer(input_text, add_special_tokens=True)["input_ids"]
internlm2_tokens = internlm2_tokenizer(input_text, add_special_tokens=True)["input_ids"]
results = {
"LLaMa-1/LLaMa-2": llama_tokens,
"LLaMa-3": llama3_tokens,
"Mistral": mistral_tokens,
"GPT-2/GPT-J": gpt2_tokens,
"GPT-NeoX": gpt_neox_tokens,
"Falcon": falcon_tokens,
"Phi-1/Phi-2": phi2_tokens,
"Phi-3": phi3_tokens,
"T5": t5_tokens,
"Gemma": gemma_tokens,
"Command-R": command_r_tokens,
"Qwen/Qwen1.5": qwen_tokens,
"CodeQwen": codeqwen_tokens,
"RWKV-v4": rwkv4_tokens,
"RWKV-v5/RWKV-v6": rwkv5_tokens,
"DeepSeek": deepseek_tokens,
"InternLM": internlm_tokens,
"InternLM2": internlm2_tokens
}
toks = ""
for model, tokens in results.items():
toks += f"\n{model} gets {len(tokens)} tokens: {formatarr(tokens)}"
return toks
if __name__ == "__main__":
llama_tokenizer = AutoTokenizer.from_pretrained("TheBloke/Llama-2-7B-fp16")
llama3_tokenizer = AutoTokenizer.from_pretrained("unsloth/llama-3-8b")
mistral_tokenizer = AutoTokenizer.from_pretrained("mistral-community/Mistral-7B-v0.2")
gpt2_tokenizer = AutoTokenizer.from_pretrained("openai-community/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")
phi3_tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-4k-instruct")
t5_tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-xxl")
gemma_tokenizer = AutoTokenizer.from_pretrained("alpindale/gemma-2b")
command_r_tokenizer = AutoTokenizer.from_pretrained("CohereForAI/c4ai-command-r-plus")
qwen_tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen1.5-7B")
codeqwen_tokenizer = AutoTokenizer.from_pretrained("Qwen/CodeQwen1.5-7B")
rwkv4_tokenizer = AutoTokenizer.from_pretrained("RWKV/rwkv-4-14b-pile", trust_remote_code=True)
rwkv5_tokenizer = AutoTokenizer.from_pretrained("RWKV/v5-EagleX-v2-7B-HF", trust_remote_code=True)
deepseek_tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/DeepSeek-V2", trust_remote_code=True)
internlm_tokenizer = AutoTokenizer.from_pretrained("internlm/internlm-20b", trust_remote_code=True)
internlm2_tokenizer = AutoTokenizer.from_pretrained("internlm/internlm2-20b", trust_remote_code=True)
iface = gr.Interface(
fn=tokenize, inputs=gr.Textbox(label="Input Text", lines=19), outputs="text"
)
iface.launch()