Does LLaVA-Next support in-context(few-shot) inference?

[waltonfuture]

Thanks for your work! Can I input multi-images and multi-instructions for few-shot inference?

[carlos-havier]

Also very interested in this for few-shot image classification. So far, I haven't been able to get good results. Is it possible to do it with LLaVA-NeXT out of the box, or would it need fine tune for this use?

[YepJin]

same here, hope the authors can have some feedback

[ChunyuanLI]

The recently released LLaVA-NeXT (Interleave) supports the a variety of daily-life multi-image scenarios, but it is NOT specifically trained for in-context-learning.

[waltonfuture]

# from .demo_modelpart import InferenceDemo
import gradio as gr
import os
# import time
import cv2
import torch
# import random
import numpy as np

from llava import conversation as conversation_lib
from llava.constants import DEFAULT_IMAGE_TOKEN

from llava.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN
from llava.conversation import conv_templates, SeparatorStyle
from llava.model.builder import load_pretrained_model
from llava.utils import disable_torch_init
from llava.mm_utils import tokenizer_image_token, get_model_name_from_path, KeywordsStoppingCriteria

from PIL import Image

import requests
from PIL import Image
from io import BytesIO
from transformers import TextStreamer

torch.manual_seed(42)

class InferenceDemo(object):
    def __init__(self,args,model_path) -> None:
        disable_torch_init()

        model_name = get_model_name_from_path(args.model_path)
        self.tokenizer, self.model, self.image_processor, self.context_len = load_pretrained_model(args.model_path, args.model_base, model_name)

        # if "llama-2" in model_name.lower():
        #     conv_mode = "llava_llama_2"
        # elif "v1" in model_name.lower():
        #     conv_mode = "llava_v1"
        # elif "mpt" in model_name.lower():
        #     conv_mode = "mpt"
        # elif 'qwen' in model_name.lower():
        #     conv_mode = "qwen_1_5"
        # else:
        #     conv_mode = "llava_v0"
        conv_mode = "qwen_1_5"
        if args.conv_mode is not None and conv_mode != args.conv_mode:
            print("[WARNING] the auto inferred conversation mode is {}, while `--conv-mode` is {}, using {}".format(conv_mode, args.conv_mode, args.conv_mode))
        else:
            args.conv_mode = conv_mode
            pass
        self.conv_mode=conv_mode
        self.conversation = conv_templates[args.conv_mode].copy()
        self.num_frames = args.num_frames
        pass

def load_image(image_file):
    if image_file.startswith("http") or image_file.startswith("https"):
        response = requests.get(image_file)
        if response.status_code == 200:
            image = Image.open(BytesIO(response.content)).convert("RGB")
        else:
            print('failed to load the image')
    else:
        print('Load image from local file')
        print(image_file)
        image = Image.open(image_file).convert("RGB")

    return image

if __name__ == "__main__":
    import argparse
    argparser = argparse.ArgumentParser()
    argparser.add_argument("--model_path", default="/data/weilai/weilai_code/model_weights/llava-next-interleave-7b", type=str)
    # argparser.add_argument("--model-path", type=str, default="facebook/opt-350m")
    argparser.add_argument("--model-base", type=str, default=None)
    argparser.add_argument("--num-gpus", type=int, default=1)
    argparser.add_argument("--conv-mode", type=str, default=None)
    argparser.add_argument("--temperature", type=float, default=0.2)
    argparser.add_argument("--max-new-tokens", type=int, default=512)
    argparser.add_argument("--num_frames", type=int, default=16)
    argparser.add_argument("--load-8bit", action="store_true")
    argparser.add_argument("--load-4bit", action="store_true")
    argparser.add_argument("--debug", action="store_true")

    args = argparser.parse_args()
    model_path = args.model_path
    filt_invalid="cut"
    our_chatbot = InferenceDemo(args,model_path)
    images_this_term = ["/data/weilai/weilai_code/datasets/CodeLLaVA/images/matplotlib_images/matplotlib_Aligning_Labels_0029.jpg","/data/weilai/weilai_code/datasets/CodeLLaVA/images/matplotlib_images/matplotlib_Aligning_Labels_0031.jpg"]
    image_list=[]
    for f in images_this_term:
        image_list.append(load_image(f))
    image_tensor = [our_chatbot.image_processor.preprocess(f, return_tensors="pt")["pixel_values"][0].half().to(our_chatbot.model.device) for f in image_list]

    image_tensor = torch.stack(image_tensor)
    image_token = DEFAULT_IMAGE_TOKEN*len(image_list)
    # if our_chatbot.model.config.mm_use_im_start_end:
    #     inp = DEFAULT_IM_START_TOKEN + image_token + DEFAULT_IM_END_TOKEN + "\n" + inp
    # else:
    inp='How to edit image1 to make it look like image2?'
    inp = image_token + "\n" + inp
    our_chatbot.conversation.append_message(our_chatbot.conversation.roles[0], inp)
    our_chatbot.conversation.append_message(our_chatbot.conversation.roles[1], None)
    prompt = our_chatbot.conversation.get_prompt()
    print(prompt)
    input_ids = tokenizer_image_token(prompt, our_chatbot.tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt").unsqueeze(0).to(our_chatbot.model.device)
    stop_str = our_chatbot.conversation.sep if our_chatbot.conversation.sep_style != SeparatorStyle.TWO else our_chatbot.conversation.sep2
    keywords = [stop_str]
    stopping_criteria = KeywordsStoppingCriteria(keywords, our_chatbot.tokenizer, input_ids)
    # import pdb;pdb.set_trace()
    with torch.inference_mode():
        output_ids = our_chatbot.model.generate(input_ids, images=image_tensor, do_sample=True, temperature=0.2, max_new_tokens=1024,use_cache=True)

    outputs = our_chatbot.tokenizer.decode(output_ids[0]).strip()
    # if outputs.endswith(stop_str):
    #     outputs = outputs[:-len(stop_str)]
    print(outputs)

I use this script for inference, but the model can only outputs "<|im_end|>". What's wrong with my script? Thanks a lot for help! @ChunyuanLI