Maknee / minigpt4.cpp

Port of MiniGPT4 in C++ (4bit, 5bit, 6bit, 8bit, 16bit CPU inference with GGML)
MIT License
549 stars 26 forks source link

[error] translate text from english to chinese #3

Open subineru opened 1 year ago

subineru commented 1 year ago

utf8_error

felladrin commented 11 months ago

Hi, @Maknee. Could you recheck this?

I found a reliable way to reproduce this issue:

from huggingface_hub import hf_hub_download
import minigpt4_library

model_path = hf_hub_download(
    repo_id="maknee/minigpt4-13b-ggml",
    filename="minigpt4-13B-f16.bin",
    repo_type="dataset",
)

llm_model_path = hf_hub_download(
    repo_id="maknee/ggml-vicuna-v0-quantized",
    filename="ggml-vicuna-13B-v0-q5_k.bin",
    repo_type="dataset",
)

minigpt4_chatbot = minigpt4_library.MiniGPT4ChatBot(model_path, llm_model_path)

for output in minigpt4_chatbot.generate(
    "tell me about yourself",
    limit=1024,
    temp=0,
    top_k=0,
    top_p=0,
    repeat_penalty=1,
):
    print(output)

Using an emoticon as prompt makes it fail even earlier (it fails before printing anything on the terminal):

for output in minigpt4_chatbot.generate(
     "😀",
    limit=1024,
    temp=0,
    top_k=0,
    top_p=0,
    repeat_penalty=1,
):
    print(output)

I think minigpt4.cpp would require a similar solution like the one applied here:

Maknee commented 11 months ago

Thanks @felladrin. Interesting, I'll have to look into this with that PR in mind. Unfortunately, I haven't tested any other input besides english characters/words.