haotian-liu / LLaVA

[NeurIPS'23 Oral] Visual Instruction Tuning (LLaVA) built towards GPT-4V level capabilities and beyond.
https://llava.hliu.cc
Apache License 2.0
20.4k stars 2.26k forks source link

[Question] When training with the llm of vicuna_v1_1, the conversation_lib.default_conversation is set to be conversation_lib.conv_templates["vicuna_v1_1"], while template_name = "llava_v1" in the 'gradio_web_server.py' in this case, does this have any effect? #129

Open xuhzyy opened 1 year ago

xuhzyy commented 1 year ago

Question

No response

haotian-liu commented 1 year ago

Hi @xuhzyy, thank you for the question. This is intentional as we do not want to include the model names like "LLaVA" in training. Thanks.

YanqiDai commented 11 months ago

I have a similar confusion: the conversation template used in the fine-tuning stage of llava-v1.5 is conv_vicuna_v1, while conv_llava_v1 is used in the inference files. The only difference between the two is "user" and "human", why is this? I'm sorry that I didn't understand your previous answer. I would appreciate a more detailed elucidation. Thanks.

Zeqiang-Lai commented 11 months ago

I am very confused about the conversation templates of llava-v1.5 as well(pretraining, fine-tuning, and inference use three different templates), @haotian-liu would appreciate for the response.

I have a similar confusion: the conversation template used in the fine-tuning stage of llava-v1.5 is conv_vicuna_v1, while conv_llava_v1 is used in the inference files. The only difference between the two is "user" and "human", why is this? I'm sorry that I didn't understand your previous answer. I would appreciate a more detailed elucidation. Thanks.

linhaojia13 commented 7 months ago

I have a similar confusion: the conversation template used in the fine-tuning stage of llava-v1.5 is conv_vicuna_v1, while conv_llava_v1 is used in the inference files. The only difference between the two is "user" and "human", why is this? I'm sorry that I didn't understand your previous answer. I would appreciate a more detailed elucidation. Thanks.

I am confused with this, too. @haotian-liu

conv_vicuna_v1 = Conversation(
    system="A chat between a curious **user** and an artificial intelligence assistant. "
    "The assistant gives helpful, detailed, and polite answers to the user's questions.",
    roles=("USER", "ASSISTANT"),
    version="v1",
    messages=(),
    offset=0,
    sep_style=SeparatorStyle.TWO,
    sep=" ",
    sep2="</s>",
)
conv_llava_v1 = Conversation(
    system="A chat between a curious **human** and an artificial intelligence assistant. "
           "The assistant gives helpful, detailed, and polite answers to the human's questions.",
    roles=("USER", "ASSISTANT"),
    version="v1",
    messages=(),
    offset=0,
    sep_style=SeparatorStyle.TWO,
    sep=" ",
    sep2="</s>",
)

Is it a typo?

darkpromise98 commented 4 months ago

some question~