Upload folder using huggingface_hub

env.txt

@@ -0,0 +1,8 @@
+conda create -n qwen_ft python=3.10 -y
+conda activate qwen_ft
+conda install nvidia/label/cuda-12.2.2::cuda-toolkit -y
+pip install torch==2.6.0 torchvision==0.21.0 deepspeed==0.17.1 triton==3.2.0 accelerate==1.7.0 torchcodec==0.2 peft==0.17.1
+pip install transformers==5.0.0.dev0
+pip install -e /root/Desktop/workspace/kwon/pinpoint/qwen_ft
+pip install qwen-vl-utils==0.0.14
+pip install matplotlib

qwen_ft.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e28d63bfee1a31bad10cc180e03b57cfe4401ff5f61eebd61cb83cfe373eb77
+size 1221066752

qwen_ft_env.txt

@@ -0,0 +1,249 @@
+# packages in environment at /opt/conda/envs/qwen_ft:
+#
+# Name                    Version                   Build  Channel
+_libgcc_mutex             0.1                        main    conda-forge
+_openmp_mutex             5.1                       1_gnu    defaults
+accelerate                1.7.0                    pypi_0    pypi
+alsa-lib                  1.2.14               hb9d3cd8_0    conda-forge
+anls                      0.0.2                    pypi_0    pypi
+annotated-types           0.7.0                    pypi_0    pypi
+anyio                     4.11.0                   pypi_0    pypi
+attr                      2.5.2                h39aace5_0    conda-forge
+av                        16.0.1                   pypi_0    pypi
+bzip2                     1.0.8                hda65f42_8    conda-forge
+ca-certificates           2025.10.5            hbd8a1cb_0    conda-forge
+certifi                   2025.10.5                pypi_0    pypi
+charset-normalizer        3.4.4                    pypi_0    pypi
+click                     8.3.0                    pypi_0    pypi
+contourpy                 1.3.2                    pypi_0    pypi
+cuda-cccl_linux-64        13.0.85              ha770c72_0    conda-forge
+cuda-command-line-tools   13.0.2               ha770c72_0    conda-forge
+cuda-compiler             12.2.2                        0    nvidia/label/cuda-12.2.2
+cuda-crt-dev_linux-64     13.0.88              ha770c72_0    conda-forge
+cuda-cudart               13.0.96              hecca717_0    conda-forge
+cuda-cudart-dev           13.0.96              hecca717_0    conda-forge
+cuda-cudart-dev_linux-64  13.0.96              h376f20c_0    conda-forge
+cuda-cudart-static        13.0.96              hecca717_0    conda-forge
+cuda-cudart-static_linux-64 13.0.96              h376f20c_0    conda-forge
+cuda-cudart_linux-64      13.0.96              h376f20c_0    conda-forge
+cuda-culibos-static       13.0.85              h676940d_0    conda-forge
+cuda-cuobjdump            13.0.85              hffce074_0    conda-forge
+cuda-cupti                13.0.85              h676940d_0    conda-forge
+cuda-cupti-dev            13.0.85              h676940d_0    conda-forge
+cuda-cuxxfilt             13.0.85              hffce074_0    conda-forge
+cuda-documentation        12.2.140                      0    nvidia/label/cuda-12.2.2
+cuda-driver-dev           13.0.96              hecca717_0    conda-forge
+cuda-driver-dev_linux-64  13.0.96              h376f20c_0    conda-forge
+cuda-gdb                  13.0.85              h1b59fc5_0    conda-forge
+cuda-libraries            13.0.2               ha770c72_0    conda-forge
+cuda-libraries-dev        13.0.2               ha770c72_0    conda-forge
+cuda-libraries-static     13.0.2               ha770c72_0    conda-forge
+cuda-nsight               13.0.85              h7938cbb_0    conda-forge
+cuda-nvcc                 12.2.140                      0    nvidia/label/cuda-12.2.2
+cuda-nvdisasm             13.0.85              hffce074_0    conda-forge
+cuda-nvml-dev             13.0.87              hffce074_0    conda-forge
+cuda-nvprune              13.0.85              hffce074_0    conda-forge
+cuda-nvrtc                13.0.88              hecca717_0    conda-forge
+cuda-nvrtc-dev            13.0.88              hecca717_0    conda-forge
+cuda-nvrtc-static         13.0.88              hecca717_0    conda-forge
+cuda-nvtx                 13.0.85              hecca717_0    conda-forge
+cuda-opencl               13.0.85              hecca717_0    conda-forge
+cuda-opencl-dev           13.0.85              hecca717_0    conda-forge
+cuda-profiler-api         13.0.85              h7938cbb_0    conda-forge
+cuda-sanitizer-api        13.0.85              h10ca0ad_0    conda-forge
+cuda-toolkit              12.2.2                        0    nvidia/label/cuda-12.2.2
+cuda-tools                13.0.2               ha770c72_0    conda-forge
+cuda-version              13.0                 hc7b4dd1_3    conda-forge
+cuda-visual-tools         13.0.2               ha770c72_0    conda-forge
+cycler                    0.12.1                   pypi_0    pypi
+dbus                      1.16.2               h3c4dab8_0    conda-forge
+deepspeed                 0.17.1                   pypi_0    pypi
+einops                    0.8.1                    pypi_0    pypi
+exceptiongroup            1.3.0                    pypi_0    pypi
+filelock                  3.20.0                   pypi_0    pypi
+flash-attn                2.7.4.post1              pypi_0    pypi
+font-ttf-dejavu-sans-mono 2.37                 hab24e00_0    conda-forge
+font-ttf-inconsolata      3.000                h77eed37_0    conda-forge
+font-ttf-source-code-pro  2.038                h77eed37_0    conda-forge
+font-ttf-ubuntu           0.83                 h77eed37_3    conda-forge
+fontconfig                2.15.0               h7e30c49_1    conda-forge
+fonts-conda-ecosystem     1                             0    conda-forge
+fonts-conda-forge         1                             0    conda-forge
+fonttools                 4.60.1                   pypi_0    pypi
+freetype                  2.14.1               ha770c72_0    conda-forge
+fsspec                    2025.10.0                pypi_0    pypi
+gds-tools                 1.15.1.6             hecca717_0    conda-forge
+gitdb                     4.0.12                   pypi_0    pypi
+gitpython                 3.1.45                   pypi_0    pypi
+gmp                       6.3.0                hac33072_2    conda-forge
+h11                       0.16.0                   pypi_0    pypi
+hf-xet                    1.2.0                    pypi_0    pypi
+hjson                     3.1.0                    pypi_0    pypi
+httpcore                  1.0.9                    pypi_0    pypi
+httpx                     0.28.1                   pypi_0    pypi
+huggingface-hub           1.0.0rc6                 pypi_0    pypi
+idna                      3.11                     pypi_0    pypi
+jinja2                    3.1.6                    pypi_0    pypi
+keyutils                  1.6.3                hb9d3cd8_0    conda-forge
+kiwisolver                1.4.9                    pypi_0    pypi
+krb5                      1.21.3               h659f571_0    conda-forge
+ld_impl_linux-64          2.44                 h1aa0949_4    conda-forge
+libcap                    2.76                 h0b2e76d_0    conda-forge
+libcublas                 13.1.0.3             h676940d_0    conda-forge
+libcublas-dev             13.1.0.3             h676940d_0    conda-forge
+libcublas-static          13.1.0.3             h676940d_0    conda-forge
+libcufft                  12.0.0.61            hecca717_0    conda-forge
+libcufft-dev              12.0.0.61            hecca717_0    conda-forge
+libcufft-static           12.0.0.61            hecca717_0    conda-forge
+libcufile                 1.15.1.6             hbc026e6_0    conda-forge
+libcufile-dev             1.15.1.6             hecca717_0    conda-forge
+libcufile-static          1.15.1.6             hecca717_0    conda-forge
+libcurand                 10.4.0.35            h676940d_1    conda-forge
+libcurand-dev             10.4.0.35            h676940d_1    conda-forge
+libcurand-static          10.4.0.35            h676940d_1    conda-forge
+libcusolver               12.0.4.66            h676940d_1    conda-forge
+libcusolver-dev           12.0.4.66            h676940d_1    conda-forge
+libcusolver-static        12.0.4.66            h676940d_1    conda-forge
+libcusparse               12.6.3.3             hecca717_0    conda-forge
+libcusparse-dev           12.6.3.3             hecca717_0    conda-forge
+libcusparse-static        12.6.3.3             hecca717_0    conda-forge
+libedit                   3.1.20250104    pl5321h7949ede_0    conda-forge
+libexpat                  2.7.1                hecca717_0    conda-forge
+libffi                    3.5.2                h9ec8514_0    conda-forge
+libfreetype               2.14.1               ha770c72_0    conda-forge
+libfreetype6              2.14.1               h73754d4_0    conda-forge
+libgcc                    15.2.0               h767d61c_7    conda-forge
+libgcc-ng                 15.2.0               h69a702a_7    conda-forge
+libgcrypt-lib             1.11.1               hb9d3cd8_0    conda-forge
+libglib                   2.86.1               h32235b2_1    conda-forge
+libglvnd                  1.7.0                ha4b6fd6_2    conda-forge
+libgomp                   15.2.0               h767d61c_7    conda-forge
+libgpg-error              1.55                 h3f2d84a_0    conda-forge
+libiconv                  1.18                 h3b78370_2    conda-forge
+liblzma                   5.8.1                hb9d3cd8_2    conda-forge
+libnl                     3.11.0               hb9d3cd8_0    conda-forge
+libnpp                    13.0.1.2             h676940d_0    conda-forge
+libnpp-dev                13.0.1.2             h676940d_0    conda-forge
+libnpp-static             13.0.1.2             h676940d_0    conda-forge
+libnsl                    2.0.1                hb9d3cd8_1    conda-forge
+libnuma                   2.0.19               hee96239_0    defaults
+libnvfatbin               13.0.85              hecca717_0    conda-forge
+libnvfatbin-dev           13.0.85              hecca717_0    conda-forge
+libnvfatbin-static        13.0.85              hecca717_0    conda-forge
+libnvjitlink              13.0.88              hecca717_0    conda-forge
+libnvjitlink-dev          13.0.88              hecca717_0    conda-forge
+libnvjitlink-static       13.0.88              hecca717_0    conda-forge
+libnvjpeg                 13.0.1.86            hecca717_0    conda-forge
+libnvjpeg-dev             13.0.1.86            ha770c72_0    conda-forge
+libnvjpeg-static          13.0.1.86            ha770c72_0    conda-forge
+libopengl                 1.7.0                ha4b6fd6_2    conda-forge
+libpng                    1.6.50               h421ea60_1    conda-forge
+libsqlite                 3.50.4               h0c1763c_0    conda-forge
+libstdcxx                 15.2.0               h8f9b012_7    conda-forge
+libstdcxx-ng              15.2.0               h4852527_7    conda-forge
+libsystemd0               257.9                h996ca69_0    conda-forge
+libudev1                  257.9                h085a93f_0    conda-forge
+libuuid                   2.41.2               he9a06e4_0    conda-forge
+libxcb                    1.17.0               h8a09558_0    conda-forge
+libxcrypt                 4.4.36               hd590300_1    conda-forge
+libxkbcommon              1.12.3               hca5e8e5_0    conda-forge
+libxkbfile                1.1.0                h166bdaf_1    conda-forge
+libxml2                   2.15.1               h031cc0b_0    conda-forge
+libxml2-16                2.15.1               hf2a90c1_0    conda-forge
+libzlib                   1.3.1                hb9d3cd8_2    conda-forge
+lz4-c                     1.10.0               h5888daf_1    conda-forge
+markupsafe                3.0.3                    pypi_0    pypi
+matplotlib                3.10.7                   pypi_0    pypi
+mpmath                    1.3.0                    pypi_0    pypi
+msgpack                   1.1.2                    pypi_0    pypi
+ncurses                   6.5                  h2d0b736_3    conda-forge
+networkx                  3.4.2                    pypi_0    pypi
+ninja                     1.13.0                   pypi_0    pypi
+nsight-compute            2025.3.1.4           h6a507f3_0    conda-forge
+nspr                      4.37                 h29cc59b_0    conda-forge
+nss                       3.117                h445c969_0    conda-forge
+numpy                     2.2.6                    pypi_0    pypi
+nvidia-cublas-cu12        12.4.5.8                 pypi_0    pypi
+nvidia-cuda-cupti-cu12    12.4.127                 pypi_0    pypi
+nvidia-cuda-nvrtc-cu12    12.4.127                 pypi_0    pypi
+nvidia-cuda-runtime-cu12  12.4.127                 pypi_0    pypi
+nvidia-cudnn-cu12         9.1.0.70                 pypi_0    pypi
+nvidia-cufft-cu12         11.2.1.3                 pypi_0    pypi
+nvidia-curand-cu12        10.3.5.147               pypi_0    pypi
+nvidia-cusolver-cu12      11.6.1.9                 pypi_0    pypi
+nvidia-cusparse-cu12      12.3.1.170               pypi_0    pypi
+nvidia-cusparselt-cu12    0.6.2                    pypi_0    pypi
+nvidia-nccl-cu12          2.21.5                   pypi_0    pypi
+nvidia-nvjitlink-cu12     12.4.127                 pypi_0    pypi
+nvidia-nvtx-cu12          12.4.127                 pypi_0    pypi
+ocl-icd                   2.3.3                hb9d3cd8_0    conda-forge
+opencl-headers            2025.06.13           h5888daf_0    conda-forge
+opencv-python             4.12.0.88                pypi_0    pypi
+openssl                   3.5.4                h26f9b46_0    conda-forge
+packaging                 25.0                     pypi_0    pypi
+parmap                    1.7.0                    pypi_0    pypi
+pcre2                     10.46                h1321c63_0    conda-forge
+peft                      0.17.1                   pypi_0    pypi
+pillow                    12.0.0                   pypi_0    pypi
+pip                       25.2               pyh8b19718_0    conda-forge
+platformdirs              4.5.0                    pypi_0    pypi
+protobuf                  6.33.0                   pypi_0    pypi
+psutil                    7.1.2                    pypi_0    pypi
+pthread-stubs             0.4               hb9d3cd8_1002    conda-forge
+py-cpuinfo                9.0.0                    pypi_0    pypi
+pydantic                  2.12.3                   pypi_0    pypi
+pydantic-core             2.41.4                   pypi_0    pypi
+pyparsing                 3.2.5                    pypi_0    pypi
+python                    3.10.19         h3c07f61_2_cpython    conda-forge
+python-dateutil           2.9.0.post0              pypi_0    pypi
+pyyaml                    6.0.3                    pypi_0    pypi
+qwen-vl-utils             0.0.14                   pypi_0    pypi
+rdma-core                 60.0                 hecca717_0    conda-forge
+readline                  8.3                  hc2a1206_0    defaults
+regex                     2025.10.23               pypi_0    pypi
+requests                  2.32.5                   pypi_0    pypi
+safetensors               0.6.2                    pypi_0    pypi
+sentry-sdk                2.43.0                   pypi_0    pypi
+setuptools                80.9.0             pyhff2d567_0    conda-forge
+six                       1.17.0                   pypi_0    pypi
+smmap                     5.0.2                    pypi_0    pypi
+sniffio                   1.3.1                    pypi_0    pypi
+sympy                     1.13.1                   pypi_0    pypi
+tk                        8.6.15               h54e0aa7_0    defaults
+tokenizers                0.22.1                   pypi_0    pypi
+torch                     2.6.0                    pypi_0    pypi
+torchcodec                0.2.0                    pypi_0    pypi
+torchvision               0.21.0                   pypi_0    pypi
+tqdm                      4.67.1                   pypi_0    pypi
+transformers              5.0.0.dev0               pypi_0    pypi
+triton                    3.2.0                    pypi_0    pypi
+typer-slim                0.20.0                   pypi_0    pypi
+typing-extensions         4.15.0                   pypi_0    pypi
+typing-inspection         0.4.2                    pypi_0    pypi
+tzdata                    2025b                h78e105d_0    conda-forge
+urllib3                   2.5.0                    pypi_0    pypi
+wandb                     0.22.3                   pypi_0    pypi
+wayland                   1.24.0               hd6090a7_1    conda-forge
+wheel                     0.45.1             pyhd8ed1ab_1    conda-forge
+xcb-util                  0.4.1                h4f16b4b_2    conda-forge
+xcb-util-cursor           0.1.5                hb9d3cd8_0    conda-forge
+xcb-util-image            0.4.0                hb711507_2    conda-forge
+xcb-util-keysyms          0.4.1                hb711507_0    conda-forge
+xcb-util-renderutil       0.3.10               hb711507_0    conda-forge
+xcb-util-wm               0.4.2                hb711507_0    conda-forge
+xkeyboard-config          2.46                 hb03c661_0    conda-forge
+xorg-libice               1.1.2                hb9d3cd8_0    conda-forge
+xorg-libsm                1.2.6                he73a12e_0    conda-forge
+xorg-libx11               1.8.12               h4f16b4b_0    conda-forge
+xorg-libxau               1.0.12               hb9d3cd8_0    conda-forge
+xorg-libxcomposite        0.4.6                hb9d3cd8_2    conda-forge
+xorg-libxdamage           1.1.6                hb9d3cd8_0    conda-forge
+xorg-libxdmcp             1.1.5                hb9d3cd8_0    conda-forge
+xorg-libxext              1.3.6                hb9d3cd8_0    conda-forge
+xorg-libxfixes            6.0.2                hb03c661_0    conda-forge
+xorg-libxi                1.8.2                hb9d3cd8_0    conda-forge
+xorg-libxrandr            1.5.4                hb9d3cd8_0    conda-forge
+xorg-libxrender           0.9.12               hb9d3cd8_0    conda-forge
+xorg-libxtst              1.2.5                hb9d3cd8_3    conda-forge
+zlib                      1.3.1                hb9d3cd8_2    conda-forge
+zstd                      1.5.7                hb8e6e7a_2    conda-forge

qwen_vanilla/gqa.py

@@ -0,0 +1,139 @@
+from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
+from qwen_vl_utils import process_vision_info
+
+from PIL import Image
+import copy
+import torch
+from tqdm import tqdm
+import json
+from anls import anls_score
+import torch.profiler
+import os
+
+# os.environ["CUDA_VISIBLE_DEVICES"] = '1'
+
+### Dataset Information ###
+data_path = "/root/Desktop/workspace/kwon/pinpoint/pinpoint_dataset/temp/image/images/"
+qa_path = "/root/Desktop/workspace/kwon/pinpoint/pinpoint_dataset/dataset_final/gqa/pinpoint_gqa_val.json"
+
+device_map = "auto"
+model_path = "Qwen/Qwen2-VL-7B-Instruct"
+# model_path = "/root/Desktop/workspace/kwon/pinpoint/qwen_pinpoint/ckpt/info_pinpoint02"
+
+# default: Load the model on the available device(s)
+model = Qwen2VLForConditionalGeneration.from_pretrained(
+    model_path, torch_dtype=torch.bfloat16, device_map=device_map
+)
+
+# We recommend enabling flash_attention_2 for better acceleration and memory saving, especially in multi-image and video scenarios.
+# model = Qwen2VLForConditionalGeneration.from_pretrained(
+#     "Qwen/Qwen2-VL-7B-Instruct",
+#     torch_dtype=torch.bfloat16,
+#     attn_implementation="flash_attention_2",
+#     device_map="auto",
+# )
+
+# default processer
+processor = AutoProcessor.from_pretrained(model_path)
+
+# The default range for the number of visual tokens per image in the model is 4-16384. You can set min_pixels and max_pixels according to your needs, such as a token count range of 256-1280, to balance speed and memory usage.
+# min_pixels = 12544 # original
+# min_pixels = 1204224
+min_pixels = 2408448
+# max_pixels = 1605632
+max_pixels = 3211264
+processor = AutoProcessor.from_pretrained(model_path, min_pixels=min_pixels, max_pixels=max_pixels)
+
+with open(qa_path, "r", encoding="utf-8") as file:
+    qa_data = json.load(file)
+
+total_ANLS = 0
+total_processed = 0
+total_len = 0
+total_time = 0.0 
+total_flops = 0.0
+
+pbar = tqdm(qa_data)
+start_event = torch.cuda.Event(enable_timing=True)
+end_event = torch.cuda.Event(enable_timing=True)
+
+for entry in pbar:
+    image_path = data_path + entry['image']
+    image = Image.open(image_path).convert("RGB")
+    ques = entry['question']
+
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {
+                    "type": "image",
+                    "image": image,
+                },
+                {"type": "text", "text": f"{ques} \n Give me just an answer."},
+            ],
+        }
+    ]
+
+    # Preparation for inference
+    start_event.record()
+    text = processor.apply_chat_template(
+        messages, tokenize=False, add_generation_prompt=True
+    )
+    image_inputs, video_inputs = process_vision_info(messages)
+    inputs = processor(
+        text=[text],
+        images=image_inputs,
+        videos=video_inputs,
+        padding=True,
+        return_tensors="pt",
+    )
+    inputs = inputs.to("cuda")
+
+    # Inference: Generation of the output
+
+    # with torch.no_grad():
+    #     with torch.profiler.profile(
+    #         activities=[torch.profiler.ProfilerActivity.CPU,torch.profiler.ProfilerActivity.CUDA],
+    #         with_flops=True,
+    #         profile_memory=False,
+    #         record_shapes=False
+    #     ) as prof:
+    #         generated_ids = model.generate(**inputs, max_new_tokens=128)
+    #     current_flops = sum([event.flops for event in prof.key_averages() if event.flops is not None])
+        
+    generated_ids = model.generate(**inputs, max_new_tokens=128)
+    current_flops = 0 
+
+    generated_ids_trimmed = [
+        out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
+    ]
+    output_text = processor.batch_decode(
+        generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
+    )
+    text_outputs = output_text[0]
+    end_event.record()
+    torch.cuda.synchronize()
+    elapsed_time = start_event.elapsed_time(end_event)
+    total_time += elapsed_time
+
+    ANLS_Score = anls_score(prediction=text_outputs, gold_labels=[entry['answer']])
+    print(entry['question'])
+    print(text_outputs)
+    print(ANLS_Score)
+    print("\n")
+    # Update counters
+    total_processed += 1
+    print(f"{total_time/ total_processed}ms")
+    total_ANLS += ANLS_Score
+    total_len += 0
+    total_flops += current_flops
+
+    # Calculate and update the accuracy in the progress bar description
+    if total_processed > 0:
+        pbar.set_description(f"Processing | ANLS: {total_ANLS / total_processed:.3f} | Token Length: {total_len / total_processed:.2f} | FLOPs: {(total_flops / (total_processed *1e12)):.2f} TFLOPs")
+
+print(f"\nFinal ANLS: {(total_ANLS / len(qa_data)):.4f}")
+print(f"Final Token Length: {total_len / len(qa_data):.2f}")
+print(f"Average FLOPs: {total_flops / (len(qa_data) *1e12):.2f} TFLOPs")
+print(f"Average Response Time : {total_time / len(qa_data):.4f}ms")

qwen_vanilla/info.py

@@ -0,0 +1,133 @@
+from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
+from qwen_vl_utils import process_vision_info
+
+from PIL import Image
+import copy
+import torch
+from tqdm import tqdm
+import json
+from anls import anls_score
+import torch.profiler
+import os
+
+### Dataset Information ###
+data_path = "/root/Desktop/workspace/kwon/pinpoint/pinpoint_dataset/infographic/images/"
+qa_path = "/root/Desktop/workspace/kwon/pinpoint/pinpoint_dataset/dataset_final/info/pinpoint_info_val.json"
+device_map = "auto"
+model_path = "Qwen/Qwen2-VL-7B-Instruct"
+# model_path = "/root/Desktop/workspace/kwon/pinpoint/qwen_pinpoint/ckpt/info_pinpoint02"
+
+# default: Load the model on the available device(s)
+model = Qwen2VLForConditionalGeneration.from_pretrained(
+    model_path, torch_dtype=torch.bfloat16, device_map=device_map
+)
+
+# We recommend enabling flash_attention_2 for better acceleration and memory saving, especially in multi-image and video scenarios.
+# model = Qwen2VLForConditionalGeneration.from_pretrained(
+#     "Qwen/Qwen2-VL-7B-Instruct",
+#     torch_dtype=torch.bfloat16,
+#     attn_implementation="flash_attention_2",
+#     device_map="auto",
+# )
+
+# default processer
+processor = AutoProcessor.from_pretrained(model_path)
+
+# The default range for the number of visual tokens per image in the model is 4-16384. You can set min_pixels and max_pixels according to your needs, such as a token count range of 256-1280, to balance speed and memory usage.
+min_pixels = 12544
+max_pixels = 3211264 # 4096 x 28 x 28
+processor = AutoProcessor.from_pretrained(model_path, min_pixels=min_pixels, max_pixels=max_pixels)
+
+with open(qa_path, "r", encoding="utf-8") as file:
+    qa_data = json.load(file)
+
+total_ANLS = 0
+total_processed = 0
+total_len = 0
+total_time = 0.0 
+total_flops = 0.0
+
+pbar = tqdm(qa_data)
+start_event = torch.cuda.Event(enable_timing=True)
+end_event = torch.cuda.Event(enable_timing=True)
+
+for entry in pbar:
+    image_path = data_path + entry['image']
+    image = Image.open(image_path).convert("RGB")
+    ques = entry['question']
+
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {
+                    "type": "image",
+                    "image": image,
+                },
+                {"type": "text", "text": f"{ques} \n Give me just an answer."},
+            ],
+        }
+    ]
+
+    # Preparation for inference
+    start_event.record()
+    text = processor.apply_chat_template(
+        messages, tokenize=False, add_generation_prompt=True
+    )
+    image_inputs, video_inputs = process_vision_info(messages)
+    inputs = processor(
+        text=[text],
+        images=image_inputs,
+        videos=video_inputs,
+        padding=True,
+        return_tensors="pt",
+    )
+    inputs = inputs.to("cuda")
+
+    # Inference: Generation of the output
+
+    # with torch.no_grad():
+    #     with torch.profiler.profile(
+    #         activities=[torch.profiler.ProfilerActivity.CPU,torch.profiler.ProfilerActivity.CUDA],
+    #         with_flops=True,
+    #         profile_memory=False,
+    #         record_shapes=False
+    #     ) as prof:
+    #         generated_ids = model.generate(**inputs, max_new_tokens=128)
+    #     current_flops = sum([event.flops for event in prof.key_averages() if event.flops is not None])
+        
+    generated_ids = model.generate(**inputs, max_new_tokens=128)
+    current_flops = 0 
+
+    generated_ids_trimmed = [
+        out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
+    ]
+    output_text = processor.batch_decode(
+        generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
+    )
+    text_outputs = output_text[0]
+    end_event.record()
+    torch.cuda.synchronize()
+    elapsed_time = start_event.elapsed_time(end_event)
+    total_time += elapsed_time
+
+    ANLS_Score = anls_score(prediction=text_outputs, gold_labels=entry['answers'])
+    print(entry['question'])
+    print(text_outputs)
+    print(ANLS_Score)
+    print("\n")
+    # Update counters
+    total_processed += 1
+    print(f"{total_time/ total_processed}ms")
+    total_ANLS += ANLS_Score
+    total_len += 0
+    total_flops += current_flops
+
+    # Calculate and update the accuracy in the progress bar description
+    if total_processed > 0:
+        pbar.set_description(f"Processing | ANLS: {total_ANLS / total_processed:.3f} | Token Length: {total_len / total_processed:.2f} | FLOPs: {(total_flops / (total_processed *1e12)):.2f} TFLOPs")
+
+print(f"\nFinal ANLS: {(total_ANLS / len(qa_data)):.4f}")
+print(f"Final Token Length: {total_len / len(qa_data):.2f}")
+print(f"Average FLOPs: {total_flops / (len(qa_data) *1e12):.2f} TFLOPs")
+print(f"Average Response Time : {total_time / len(qa_data):.4f}ms")

qwen_vanilla/mp.py

@@ -0,0 +1,136 @@
+from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
+from qwen_vl_utils import process_vision_info
+
+from PIL import Image
+import copy
+import torch
+from tqdm import tqdm
+import json
+from anls import anls_score
+import torch.profiler
+import os
+Image.MAX_IMAGE_PIXELS = None
+
+# os.environ["CUDA_VISIBLE_DEVICES"] = '1'
+
+### Dataset Information ###
+data_path = "/root/Desktop/workspace/kwon/pinpoint/pinpoint_dataset/dataset_final/mpdoc/combined_mpdoc/images/"
+qa_path = "/root/Desktop/workspace/kwon/pinpoint/pinpoint_dataset/dataset_final/mpdoc/combined_mpdoc/val.json"
+device_map = "auto"
+model_path = "Qwen/Qwen2-VL-7B-Instruct"
+# model_path = "/root/Desktop/workspace/kwon/pinpoint/qwen_pinpoint/ckpt/info_pinpoint02"
+
+# default: Load the model on the available device(s)
+model = Qwen2VLForConditionalGeneration.from_pretrained(
+    model_path, torch_dtype=torch.bfloat16, device_map=device_map
+)
+
+# We recommend enabling flash_attention_2 for better acceleration and memory saving, especially in multi-image and video scenarios.
+# model = Qwen2VLForConditionalGeneration.from_pretrained(
+#     "Qwen/Qwen2-VL-7B-Instruct",
+#     torch_dtype=torch.bfloat16,
+#     attn_implementation="flash_attention_2",
+#     device_map="auto",
+# )
+
+# default processer
+processor = AutoProcessor.from_pretrained(model_path)
+
+# The default range for the number of visual tokens per image in the model is 4-16384. You can set min_pixels and max_pixels according to your needs, such as a token count range of 256-1280, to balance speed and memory usage.
+min_pixels = 12544
+max_pixels = 3211264
+processor = AutoProcessor.from_pretrained(model_path, min_pixels=min_pixels, max_pixels=max_pixels)
+
+with open(qa_path, "r", encoding="utf-8") as file:
+    qa_data = json.load(file)
+
+total_ANLS = 0
+total_processed = 0
+total_len = 0
+total_time = 0.0 
+total_flops = 0.0
+
+pbar = tqdm(qa_data)
+start_event = torch.cuda.Event(enable_timing=True)
+end_event = torch.cuda.Event(enable_timing=True)
+
+for entry in pbar:
+    image_path = data_path + entry['image']
+    image = Image.open(image_path).convert("RGB")
+    ques = entry['question']
+
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {
+                    "type": "image",
+                    "image": image,
+                },
+                {"type": "text", "text": f"{ques} \n Give me just an answer."},
+            ],
+        }
+    ]
+
+    # Preparation for inference
+    start_event.record()
+    text = processor.apply_chat_template(
+        messages, tokenize=False, add_generation_prompt=True
+    )
+    image_inputs, video_inputs = process_vision_info(messages)
+    inputs = processor(
+        text=[text],
+        images=image_inputs,
+        videos=video_inputs,
+        padding=True,
+        return_tensors="pt",
+    )
+    inputs = inputs.to("cuda")
+
+    # Inference: Generation of the output
+
+    # with torch.no_grad():
+    #     with torch.profiler.profile(
+    #         activities=[torch.profiler.ProfilerActivity.CPU,torch.profiler.ProfilerActivity.CUDA],
+    #         with_flops=True,
+    #         profile_memory=False,
+    #         record_shapes=False
+    #     ) as prof:
+    #         generated_ids = model.generate(**inputs, max_new_tokens=128)
+    #     current_flops = sum([event.flops for event in prof.key_averages() if event.flops is not None])
+        
+    generated_ids = model.generate(**inputs, max_new_tokens=128)
+    current_flops = 0 
+
+    generated_ids_trimmed = [
+        out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
+    ]
+    output_text = processor.batch_decode(
+        generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
+    )
+    text_outputs = output_text[0]
+    end_event.record()
+    torch.cuda.synchronize()
+    elapsed_time = start_event.elapsed_time(end_event)
+    total_time += elapsed_time
+
+    ANLS_Score = anls_score(prediction=text_outputs, gold_labels=entry['answers'])
+    print(entry['question'])
+    print(text_outputs)
+    print(ANLS_Score)
+    print("\n")
+    # Update counters
+    total_processed += 1
+    print(f"{total_time/ total_processed}ms")
+    total_ANLS += ANLS_Score
+    total_len += 0
+    total_flops += current_flops
+
+    # Calculate and update the accuracy in the progress bar description
+    if total_processed > 0:
+        pbar.set_description(f"Processing | ANLS: {total_ANLS / total_processed:.3f} | Token Length: {total_len / total_processed:.2f} | FLOPs: {(total_flops / (total_processed *1e12)):.2f} TFLOPs")
+
+print(f"\nFinal ANLS: {(total_ANLS / len(qa_data)):.4f}")
+print(f"Final Token Length: {total_len / len(qa_data):.2f}")
+print(f"Average FLOPs: {total_flops / (len(qa_data) *1e12):.2f} TFLOPs")
+print(f"Average Response Time : {total_time / len(qa_data):.4f}ms")

qwen_vanilla/sp.py

@@ -0,0 +1,135 @@
+from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
+from qwen_vl_utils import process_vision_info
+
+from PIL import Image
+import copy
+import torch
+from tqdm import tqdm
+import json
+from anls import anls_score
+import torch.profiler
+import os
+
+# os.environ["CUDA_VISIBLE_DEVICES"] = '0'
+
+### Dataset Information ###
+data_path = "/root/Desktop/workspace/kwon/pinpoint/pinpoint_dataset/spdoc/"
+qa_path = "/root/Desktop/workspace/kwon/pinpoint/pinpoint_dataset/dataset_final/spdoc/pinpoint_spdoc_val.json"
+device_map = "auto"
+model_path = "Qwen/Qwen2-VL-7B-Instruct"
+# model_path = "/root/Desktop/workspace/kwon/pinpoint/qwen_pinpoint/ckpt/info_pinpoint02"
+
+# default: Load the model on the available device(s)
+model = Qwen2VLForConditionalGeneration.from_pretrained(
+    model_path, torch_dtype=torch.bfloat16, device_map=device_map
+)
+
+# We recommend enabling flash_attention_2 for better acceleration and memory saving, especially in multi-image and video scenarios.
+# model = Qwen2VLForConditionalGeneration.from_pretrained(
+#     "Qwen/Qwen2-VL-7B-Instruct",
+#     torch_dtype=torch.bfloat16,
+#     attn_implementation="flash_attention_2",
+#     device_map="auto",
+# )
+
+# default processer
+processor = AutoProcessor.from_pretrained(model_path)
+
+# The default range for the number of visual tokens per image in the model is 4-16384. You can set min_pixels and max_pixels according to your needs, such as a token count range of 256-1280, to balance speed and memory usage.
+min_pixels = 12544
+max_pixels = 3211264
+processor = AutoProcessor.from_pretrained(model_path, min_pixels=min_pixels, max_pixels=max_pixels)
+
+with open(qa_path, "r", encoding="utf-8") as file:
+    qa_data = json.load(file)
+
+total_ANLS = 0
+total_processed = 0
+total_len = 0
+total_time = 0.0 
+total_flops = 0.0
+
+pbar = tqdm(qa_data)
+start_event = torch.cuda.Event(enable_timing=True)
+end_event = torch.cuda.Event(enable_timing=True)
+
+for entry in pbar:
+    image_path = data_path + entry['image']
+    image = Image.open(image_path).convert("RGB")
+    ques = entry['question']
+
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {
+                    "type": "image",
+                    "image": image,
+                },
+                {"type": "text", "text": f"{ques} \n Give me just an answer."},
+            ],
+        }
+    ]
+
+    # Preparation for inference
+    start_event.record()
+    text = processor.apply_chat_template(
+        messages, tokenize=False, add_generation_prompt=True
+    )
+    image_inputs, video_inputs = process_vision_info(messages)
+    inputs = processor(
+        text=[text],
+        images=image_inputs,
+        videos=video_inputs,
+        padding=True,
+        return_tensors="pt",
+    )
+    inputs = inputs.to("cuda")
+
+    # Inference: Generation of the output
+
+    # with torch.no_grad():
+    #     with torch.profiler.profile(
+    #         activities=[torch.profiler.ProfilerActivity.CPU,torch.profiler.ProfilerActivity.CUDA],
+    #         with_flops=True,
+    #         profile_memory=False,
+    #         record_shapes=False
+    #     ) as prof:
+    #         generated_ids = model.generate(**inputs, max_new_tokens=128)
+    #     current_flops = sum([event.flops for event in prof.key_averages() if event.flops is not None])
+        
+    generated_ids = model.generate(**inputs, max_new_tokens=128)
+    current_flops = 0 
+
+    generated_ids_trimmed = [
+        out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
+    ]
+    output_text = processor.batch_decode(
+        generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
+    )
+    text_outputs = output_text[0]
+    end_event.record()
+    torch.cuda.synchronize()
+    elapsed_time = start_event.elapsed_time(end_event)
+    total_time += elapsed_time
+
+    ANLS_Score = anls_score(prediction=text_outputs, gold_labels=entry['answers'])
+    print(entry['question'])
+    print(text_outputs)
+    print(ANLS_Score)
+    print("\n")
+    # Update counters
+    total_processed += 1
+    print(f"{total_time/ total_processed}ms")
+    total_ANLS += ANLS_Score
+    total_len += 0
+    total_flops += current_flops
+
+    # Calculate and update the accuracy in the progress bar description
+    if total_processed > 0:
+        pbar.set_description(f"Processing | ANLS: {total_ANLS / total_processed:.3f} | Token Length: {total_len / total_processed:.2f} | FLOPs: {(total_flops / (total_processed *1e12)):.2f} TFLOPs")
+
+print(f"\nFinal ANLS: {(total_ANLS / len(qa_data)):.4f}")
+print(f"Final Token Length: {total_len / len(qa_data):.2f}")
+print(f"Average FLOPs: {total_flops / (len(qa_data) *1e12):.2f} TFLOPs")
+print(f"Average Response Time : {total_time / len(qa_data):.4f}ms")

textvqa.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:001a754102a81c95d33ea6d92b9b608f7292f8f2915a47e84b6ff62fda1a3eaa
+size 7078769179