Upload folder using huggingface_hub

.ipynb_checkpoints/config-checkpoint.json

@@ -0,0 +1,26 @@
+{
+  "_name_or_path": "/notebooks/qwenv",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "bos_token_id": 151643,
+  "eos_token_id": 151643,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 6144,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 32,
+  "pad_token_id": 0,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.32.0.dev0",
+  "use_cache": false,
+  "vocab_size": 151936
+}

.ipynb_checkpoints/generation_config-checkpoint.json

@@ -0,0 +1,20 @@
+{
+  "chat_format": "chatml",
+  "decay_bound": 0.0,
+  "decay_factor": 1.0,
+  "do_sample": true,
+  "eos_token_id": 151643,
+  "factual_nucleus_sampling": false,
+  "max_context_size": 1024,
+  "max_generate_size": 512,
+  "max_new_tokens": 512,
+  "pad_token_id": 151643,
+  "stop_words_ids": [
+    [
+      151643
+    ]
+  ],
+  "top_k": 0,
+  "top_p": 0.8,
+  "transformers_version": "4.32.0.dev0"
+}

.ipynb_checkpoints/special_tokens_map-checkpoint.json

@@ -0,0 +1,5 @@
+{
+  "bos_token": "<|endoftext|>",
+  "eos_token": "<|endoftext|>",
+  "unk_token": "<|endoftext|>"
+}

.ipynb_checkpoints/tokenization_qwen-checkpoint.py

@@ -0,0 +1,566 @@
+# Copyright (c) Alibaba Cloud.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+"""Tokenization classes for QWen."""
+
+import base64
+import logging
+import os
+import requests
+import unicodedata
+from typing import Collection, Dict, List, Set, Tuple, Union, Any, Callable, Optional
+
+import tiktoken
+import numpy as np
+from PIL import Image
+from PIL import ImageFont
+from PIL import ImageDraw
+from transformers import PreTrainedTokenizer, AddedToken
+from transformers.utils import try_to_load_from_cache
+
+import matplotlib.pyplot as plt
+import matplotlib.colors as mcolors
+from matplotlib.font_manager import FontProperties
+
+logger = logging.getLogger(__name__)
+
+
+VOCAB_FILES_NAMES = {"vocab_file": "qwen.tiktoken", "ttf": "SimSun.ttf"}
+
+PAT_STR = r"""(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+"""
+ENDOFTEXT = "<|endoftext|>"
+IMSTART = "<|im_start|>"
+IMEND = "<|im_end|>"
+# as the default behavior is changed to allow special tokens in
+# regular texts, the surface forms of special tokens need to be
+# as different as possible to minimize the impact
+EXTRAS = tuple((f"<|extra_{i}|>" for i in range(205)))
+SPECIAL_TOKENS = (
+    ENDOFTEXT,
+    IMSTART,
+    IMEND,
+) + EXTRAS
+IMG_TOKEN_SPAN = 256
+
+
+def _load_tiktoken_bpe(tiktoken_bpe_file: str) -> Dict[bytes, int]:
+    with open(tiktoken_bpe_file, "rb") as f:
+        contents = f.read()
+    return {
+        base64.b64decode(token): int(rank)
+        for token, rank in (line.split() for line in contents.splitlines() if line)
+    }
+
+def _list_find(
+    input_list: List[Any],
+    candidates: Tuple[Any],
+    start: int = 0,
+):
+    for i in range(start, len(input_list)):
+        if input_list[i] in candidates:
+            return i
+    return -1
+
+def _replace_closed_tag(
+    input_tokens: List[Any],
+    start_tags: Union[Any, Tuple[Any]],
+    end_tags: Union[Any, Tuple[Any]],
+    inclusive_replace_func: Callable,
+    exclusive_replace_func: Callable = lambda x: x,
+):
+    if isinstance(start_tags, (str, int)):
+        start_tags = (start_tags,)
+    if isinstance(end_tags, (str, int)):
+        end_tags = (end_tags,)
+    assert len(start_tags) == len(end_tags)
+
+    output_tokens = []
+    end = 0
+    while True:
+        start = _list_find(input_tokens, start_tags, end)
+        if start == -1:
+            break
+        output_tokens.extend(exclusive_replace_func(input_tokens[end : start]))
+        tag_idx = start_tags.index(input_tokens[start])
+        end = _list_find(input_tokens, (end_tags[tag_idx],), start)
+        if end == -1:
+            raise ValueError("Unclosed image token")
+        output_tokens.extend(inclusive_replace_func(input_tokens[start : end + 1]))
+        end += 1
+    output_tokens.extend(exclusive_replace_func(input_tokens[end : ]))
+    return output_tokens
+
+class QWenTokenizer(PreTrainedTokenizer):
+    """QWen tokenizer."""
+
+    vocab_files_names = VOCAB_FILES_NAMES
+
+    def __init__(
+        self,
+        vocab_file,
+        errors="replace",
+        image_start_tag='<img>',
+        image_end_tag='</img>',
+        image_pad_tag='<imgpad>',
+        ref_start_tag='<ref>',
+        ref_end_tag='</ref>',
+        box_start_tag='<box>',
+        box_end_tag='</box>',
+        quad_start_tag='<quad>',
+        quad_end_tag='</quad>',
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.image_start_tag = image_start_tag
+        self.image_end_tag = image_end_tag
+        self.image_pad_tag = image_pad_tag
+        self.ref_start_tag = ref_start_tag
+        self.ref_end_tag = ref_end_tag
+        self.box_start_tag = box_start_tag
+        self.box_end_tag = box_end_tag
+        self.quad_start_tag = quad_start_tag
+        self.quad_end_tag = quad_end_tag
+        self.IMAGE_ST = (
+            ref_start_tag, ref_end_tag,
+            box_start_tag, box_end_tag,
+            quad_start_tag, quad_end_tag,
+            image_start_tag, image_end_tag,
+            image_pad_tag
+        )
+
+        self.errors = errors  # how to handle errors in decoding
+
+        self.mergeable_ranks = _load_tiktoken_bpe(vocab_file)  # type: dict[bytes, int]
+        self.special_tokens = {
+            token: index
+            for index, token in enumerate(
+                SPECIAL_TOKENS + self.IMAGE_ST, start=len(self.mergeable_ranks)
+            )
+        }
+        self.img_start_id = self.special_tokens[self.image_start_tag]
+        self.img_end_id = self.special_tokens[self.image_end_tag]
+        self.img_pad_id = self.special_tokens[self.image_pad_tag]
+        self.ref_start_id = self.special_tokens[self.ref_start_tag]
+        self.ref_end_id = self.special_tokens[self.ref_end_tag]
+        self.box_start_id = self.special_tokens[self.box_start_tag]
+        self.box_end_id = self.special_tokens[self.box_end_tag]
+        self.quad_start_id = self.special_tokens[self.quad_start_tag]
+        self.quad_end_id = self.special_tokens[self.quad_end_tag]
+
+        enc = tiktoken.Encoding(
+            "Qwen",
+            pat_str=PAT_STR,
+            mergeable_ranks=self.mergeable_ranks,
+            special_tokens=self.special_tokens,
+        )
+        assert (
+            len(self.mergeable_ranks) + len(self.special_tokens) == enc.n_vocab
+        ), f"{len(self.mergeable_ranks) + len(self.special_tokens)} != {enc.n_vocab} in encoding"
+
+        self.decoder = {
+            v: k for k, v in self.mergeable_ranks.items()
+        }  # type: dict[int, bytes|str]
+        self.decoder.update({v: k for k, v in self.special_tokens.items()})
+
+        self.tokenizer = enc  # type: tiktoken.Encoding
+
+        self.eod_id = self.tokenizer.eot_token
+        self.im_start_id = self.special_tokens[IMSTART]
+        self.im_end_id = self.special_tokens[IMEND]
+
+    def __len__(self) -> int:
+        return self.tokenizer.n_vocab
+
+    def get_vocab(self) -> Dict[bytes, int]:
+        return self.mergeable_ranks
+
+    def convert_tokens_to_ids(
+        self, tokens: Union[bytes, str, List[Union[bytes, str]]]
+    ) -> List[int]:
+        ids = []
+        if isinstance(tokens, (str, bytes)):
+            if tokens in self.special_tokens:
+                return self.special_tokens[tokens]
+            else:
+                return self.mergeable_ranks.get(tokens)
+        for token in tokens:
+            if token in self.special_tokens:
+                ids.append(self.special_tokens[token])
+            else:
+                ids.append(self.mergeable_ranks.get(token))
+        return ids
+
+    def _add_tokens(self, new_tokens: Union[List[str], List[AddedToken]], special_tokens: bool = False) -> int:
+        if not special_tokens and new_tokens:
+            raise ValueError('Adding regular tokens is not supported')
+        for token in new_tokens:
+            surface_form = token.content if isinstance(token, AddedToken) else token
+            if surface_form not in SPECIAL_TOKENS + self.IMAGE_ST:
+                raise ValueError('Adding unknown special tokens is not supported')
+        return 0
+
+    def save_vocabulary(self, save_directory: str, **kwargs) -> Tuple[str]:
+        """
+        Save only the vocabulary of the tokenizer (vocabulary).
+
+        Returns:
+            `Tuple(str)`: Paths to the files saved.
+        """
+        file_path = os.path.join(save_directory, "qwen.tiktoken")
+        with open(file_path, "w", encoding="utf8") as w:
+            for k, v in self.mergeable_ranks.items():
+                line = base64.b64encode(k).decode("utf8") + " " + str(v) + "\n"
+                w.write(line)
+        return (file_path,)
+
+    def tokenize(
+        self,
+        text: str,
+        allowed_special: Union[Set, str] = "all",
+        disallowed_special: Union[Collection, str] = (),
+        **kwargs,
+    ) -> List[Union[bytes, str]]:
+        """
+        Converts a string in a sequence of tokens.
+
+        Args:
+            text (`str`):
+                The sequence to be encoded.
+            allowed_special (`Literal["all"]` or `set`):
+                The surface forms of the tokens to be encoded as special tokens in regular texts.
+                Default to "all".
+            disallowed_special (`Literal["all"]` or `Collection`):
+                The surface forms of the tokens that should not be in regular texts and trigger errors.
+                Default to an empty tuple.
+
+            kwargs (additional keyword arguments, *optional*):
+                Will be passed to the underlying model specific encode method.
+
+        Returns:
+            `List[bytes|str]`: The list of tokens.
+        """
+        tokens = []
+        text = unicodedata.normalize("NFC", text)
+
+        # this implementation takes a detour: text -> token id -> token surface forms
+        for t in self.tokenizer.encode(
+            text, allowed_special=allowed_special, disallowed_special=disallowed_special
+        ):
+            tokens.append(self.decoder[t])
+
+        def _encode_imgurl(img_tokens):
+            assert img_tokens[0] == self.image_start_tag and img_tokens[-1] == self.image_end_tag
+            img_tokens = img_tokens[1:-1]
+            img_url = b''.join(img_tokens)
+            out_img_tokens = list(map(self.decoder.get, img_url))
+            if len(out_img_tokens) > IMG_TOKEN_SPAN:
+                raise ValueError("The content in {}..{} is too long".format(
+                    self.image_start_tag, self.image_end_tag))
+            out_img_tokens.extend([self.image_pad_tag] * (IMG_TOKEN_SPAN - len(out_img_tokens)))
+            out_img_tokens = [self.image_start_tag] + out_img_tokens + [self.image_end_tag]
+            return out_img_tokens
+
+        return _replace_closed_tag(tokens, self.image_start_tag, self.image_end_tag, _encode_imgurl)
+
+    def convert_tokens_to_string(self, tokens: List[Union[bytes, str]]) -> str:
+        """
+        Converts a sequence of tokens in a single string.
+        """
+        text = ""
+        temp = b""
+        for t in tokens:
+            if isinstance(t, str):
+                if temp:
+                    text += temp.decode("utf-8", errors=self.errors)
+                    temp = b""
+                text += t
+            elif isinstance(t, bytes):
+                temp += t
+            else:
+                raise TypeError("token should only be of type types or str")
+        if temp:
+            text += temp.decode("utf-8", errors=self.errors)
+        return text
+
+    @property
+    def vocab_size(self):
+        return self.tokenizer.n_vocab
+
+    def _convert_id_to_token(self, index: int) -> Union[bytes, str]:
+        """Converts an id to a token, special tokens included"""
+        if index in self.decoder:
+            return self.decoder[index]
+        raise ValueError("unknown ids")
+
+    def _convert_token_to_id(self, token: Union[bytes, str]) -> int:
+        """Converts a token to an id using the vocab, special tokens included"""
+        if token in self.special_tokens:
+            return self.special_tokens[token]
+        if token in self.mergeable_ranks:
+            return self.mergeable_ranks[token]
+        raise ValueError("unknown token")
+
+    def _tokenize(self, text: str, **kwargs):
+        """
+        Converts a string in a sequence of tokens (string), using the tokenizer. Split in words for word-based
+        vocabulary or sub-words for sub-word-based vocabularies (BPE/SentencePieces/WordPieces).
+
+        Do NOT take care of added tokens.
+        """
+        raise NotImplementedError
+
+    def _decode(
+        self,
+        token_ids: Union[int, List[int]],
+        skip_special_tokens: bool = False,
+        errors: str = None,
+        **kwargs,
+    ) -> str:
+        if isinstance(token_ids, int):
+            token_ids = [token_ids]
+
+        def _decode_imgurl(img_token_ids):
+            assert img_token_ids[0] == self.img_start_id and img_token_ids[-1] == self.img_end_id
+            img_token_ids = img_token_ids[1:-1]
+            img_token_ids = img_token_ids[ : img_token_ids.index(self.img_pad_id)]
+            img_url = bytes(img_token_ids).decode('utf-8')
+            return [self.img_start_id] + self.tokenizer.encode(img_url) + [self.img_end_id]
+
+        token_ids = _replace_closed_tag(token_ids, self.img_start_id, self.img_end_id, _decode_imgurl)
+
+        if skip_special_tokens:
+            token_ids = [i for i in token_ids if i < self.eod_id]
+        return self.tokenizer.decode(token_ids, errors=errors or self.errors)
+
+    def to_list_format(self, text: str):
+        text = unicodedata.normalize("NFC", text)
+        token_ids = self.tokenizer.encode(
+            text, allowed_special=set(self.IMAGE_ST + (ENDOFTEXT,)))
+
+        def _encode_vl_info(tokens):
+            if len(tokens) == 0:
+                return []
+            if tokens[0] == self.img_start_id and tokens[-1] == self.img_end_id:
+                key = 'image'
+            elif tokens[0] == self.ref_start_id and tokens[-1] == self.ref_end_id:
+                key = 'ref'
+            elif tokens[0] == self.box_start_id and tokens[-1] == self.box_end_id:
+                key = 'box'
+            elif tokens[0] == self.quad_start_id and tokens[-1] == self.quad_end_id:
+                key = 'quad'
+            else:
+                _tobytes = lambda x: x.encode('utf-8') if isinstance(x, str) else x
+                return [{'text': b''.join(map(_tobytes, map(self.decoder.get, tokens))).decode('utf-8')}]
+            val = b''.join(map(self.decoder.get, tokens[1:-1])).decode('utf-8')
+            return [{key: val}]
+
+        return _replace_closed_tag(
+            token_ids,
+            (self.img_start_id, self.ref_start_id, self.box_start_id, self.quad_start_id),
+            (self.img_end_id, self.ref_end_id, self.box_end_id, self.quad_end_id),
+            _encode_vl_info,
+            _encode_vl_info,
+        )
+
+    def from_list_format(self, list_format: List[Dict]):
+        text = ''
+        num_images = 0
+        for ele in list_format:
+            if 'image' in ele:
+                num_images += 1
+                text += f'Picture {num_images}:'
+                text += self.image_start_tag + ele['image'] + self.image_end_tag
+                text += '\n'
+            elif 'text' in ele:
+                text += ele['text']
+            elif 'box' in ele:
+                if 'ref' in ele:
+                    text += self.ref_start_tag + ele['ref'] + self.ref_end_tag
+                for box in ele['box']:
+                    text += self.box_start_tag + '(%d,%d),(%d,%d)' % (box[0], box[1], box[2], box[3]) + self.box_end_tag
+            else:
+                raise ValueError("Unsupport element: " + str(ele))
+        return text
+
+    def _fetch_latest_picture(self, response, history):
+        if history is None:
+            history = []
+        _history = history + [(response, None)]
+        for q, r in _history[::-1]:
+            for ele in self.to_list_format(q)[::-1]:
+                if 'image' in ele:
+                    return ele['image']
+        return None
+
+    def _fetch_all_box_with_ref(self, text):
+        list_format = self.to_list_format(text)
+        output = []
+        for i, ele in enumerate(list_format):
+            if 'box' in ele:
+                bbox = tuple(map(int, ele['box'].replace('(', '').replace(')', '').split(',')))
+                assert len(bbox) == 4
+                output.append({'box': bbox})
+                if i > 0 and 'ref' in list_format[i-1]:
+                    output[-1]['ref'] = list_format[i-1]['ref'].strip()
+        return output
+
+    def draw_bbox_on_latest_picture(
+        self,
+        response,
+        history=None,
+    ) -> Optional[Image.Image]:
+        image = self._fetch_latest_picture(response, history)
+        if image is None:
+            return None
+        if image.startswith("http://") or image.startswith("https://"):
+            image = Image.open(requests.get(image, stream=True).raw).convert("RGB")
+            h, w = image.height, image.width
+        else:
+            image = plt.imread(image)
+            h, w = image.shape[0], image.shape[1]
+        visualizer = Visualizer(image)
+
+        boxes = self._fetch_all_box_with_ref(response)
+        if not boxes:
+            return None
+        color = random.choice([_ for _ in mcolors.TABLEAU_COLORS.keys()]) # init color
+        for box in boxes:
+            if 'ref' in box: # random new color for new refexps
+                color = random.choice([_ for _ in mcolors.TABLEAU_COLORS.keys()])
+            x1, y1, x2, y2 = box['box']
+            x1, y1, x2, y2 = (int(x1 / 1000 * w), int(y1 / 1000 * h), int(x2 / 1000 * w), int(y2 / 1000 * h))
+            visualizer.draw_box((x1, y1, x2, y2), alpha=1, edge_color=color)
+            if 'ref' in box:
+                visualizer.draw_text(box['ref'], (x1, y1), color=color, horizontal_alignment="left")
+        return visualizer.output
+
+
+import colorsys
+import logging
+import math
+import numpy as np
+import matplotlib as mpl
+import matplotlib.colors as mplc
+import matplotlib.figure as mplfigure
+import torch
+from matplotlib.backends.backend_agg import FigureCanvasAgg
+from PIL import Image
+import random
+
+logger = logging.getLogger(__name__)
+
+
+class VisImage:
+    def __init__(self, img, scale=1.0):
+        self.img = img
+        self.scale = scale
+        self.width, self.height = img.shape[1], img.shape[0]
+        self._setup_figure(img)
+
+    def _setup_figure(self, img):
+        fig = mplfigure.Figure(frameon=False)
+        self.dpi = fig.get_dpi()
+        # add a small 1e-2 to avoid precision lost due to matplotlib's truncation
+        # (https://github.com/matplotlib/matplotlib/issues/15363)
+        fig.set_size_inches(
+            (self.width * self.scale + 1e-2) / self.dpi,
+            (self.height * self.scale + 1e-2) / self.dpi,
+        )
+        self.canvas = FigureCanvasAgg(fig)
+        # self.canvas = mpl.backends.backend_cairo.FigureCanvasCairo(fig)
+        ax = fig.add_axes([0.0, 0.0, 1.0, 1.0])
+        ax.axis("off")
+        self.fig = fig
+        self.ax = ax
+        self.reset_image(img)
+
+    def reset_image(self, img):
+        img = img.astype("uint8")
+        self.ax.imshow(img, extent=(0, self.width, self.height, 0), interpolation="nearest")
+
+    def save(self, filepath):
+        self.fig.savefig(filepath)
+
+    def get_image(self):
+        canvas = self.canvas
+        s, (width, height) = canvas.print_to_buffer()
+
+        buffer = np.frombuffer(s, dtype="uint8")
+
+        img_rgba = buffer.reshape(height, width, 4)
+        rgb, alpha = np.split(img_rgba, [3], axis=2)
+        return rgb.astype("uint8")
+
+
+class Visualizer:
+    def __init__(self, img_rgb, metadata=None, scale=1.0):
+        self.img = np.asarray(img_rgb).clip(0, 255).astype(np.uint8)
+        self.font_path = try_to_load_from_cache("Qwen/Qwen-VL-Chat", "SimSun.ttf")
+        self.output = VisImage(self.img, scale=scale)
+        self.cpu_device = torch.device("cpu")
+
+        # too small texts are useless, therefore clamp to 14
+        self._default_font_size = max(
+            np.sqrt(self.output.height * self.output.width) // 30, 15 // scale
+        )
+
+    def draw_text(
+        self,
+        text,
+        position,
+        *,
+        font_size=None,
+        color="g",
+        horizontal_alignment="center",
+        rotation=0,
+    ):
+        if not font_size:
+            font_size = self._default_font_size
+
+        # since the text background is dark, we don't want the text to be dark
+        color = np.maximum(list(mplc.to_rgb(color)), 0.2)
+        color[np.argmax(color)] = max(0.8, np.max(color))
+
+        x, y = position
+        self.output.ax.text(
+            x,
+            y,
+            text,
+            size=font_size * self.output.scale,
+            fontproperties=FontProperties(fname=self.font_path),
+            bbox={"facecolor": "black", "alpha": 0.8, "pad": 0.7, "edgecolor": "none"},
+            verticalalignment="top",
+            horizontalalignment=horizontal_alignment,
+            color=color,
+            zorder=10,
+            rotation=rotation,
+        )
+        return self.output
+
+    def draw_box(self, box_coord, alpha=0.5, edge_color="g", line_style="-"):
+        
+        x0, y0, x1, y1 = box_coord
+        width = x1 - x0
+        height = y1 - y0
+
+        linewidth = max(self._default_font_size / 4, 1)
+
+        self.output.ax.add_patch(
+            mpl.patches.Rectangle(
+                (x0, y0),
+                width,
+                height,
+                fill=False,
+                edgecolor=edge_color,
+                linewidth=linewidth * self.output.scale,
+                alpha=alpha,
+                linestyle=line_style,
+            )
+        )
+        return self.output
+
+    def get_output(self):
+        
+        return self.output

.ipynb_checkpoints/tokenizer_config-checkpoint.json

@@ -0,0 +1,10 @@
+{
+  "model_max_length": 999999999999999999,
+  "tokenizer_class": "QWenTokenizer",
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_qwen.QWenTokenizer",
+      null
+      ]
+  }
+}

config.json

@@ -0,0 +1,26 @@
+{
+  "_name_or_path": "/notebooks/pub1",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "bos_token_id": 151643,
+  "eos_token_id": 151643,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 6144,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 32,
+  "pad_token_id": 0,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.32.0.dev0",
+  "use_cache": false,
+  "vocab_size": 151936
+}

generation_config.json

@@ -0,0 +1,20 @@
+{
+  "chat_format": "chatml",
+  "decay_bound": 0.0,
+  "decay_factor": 1.0,
+  "do_sample": true,
+  "eos_token_id": 151643,
+  "factual_nucleus_sampling": false,
+  "max_context_size": 1024,
+  "max_generate_size": 512,
+  "max_new_tokens": 512,
+  "pad_token_id": 151643,
+  "stop_words_ids": [
+    [
+      151643
+    ]
+  ],
+  "top_k": 0,
+  "top_p": 0.8,
+  "transformers_version": "4.32.0.dev0"
+}

pytorch_model-00001-of-00008.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:530544f45e358bdef6e89de90b1080ae00539242bee59bd5b620c867842a068f
+size 1964005971

pytorch_model-00002-of-00008.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0788b211ea06307fc12af22c67f40b4da7413dfe3d3b3bfba67f1562e3697b7
+size 1933673729

pytorch_model-00003-of-00008.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c11031bababa6d8628fba295ffbe78206e25b86ba746460e0fb3eda9fe98060d
+size 1933673729

pytorch_model-00004-of-00008.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a7f9275ddb70230b630f2fdd5ea902cc651a117737bbb960c3d57c201bb9f08
+size 1990296303

pytorch_model-00005-of-00008.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:648cca7986994ccdb372dd88208f6670f12dfc6d596f2011e5b1560e48481654
+size 1990296897

pytorch_model-00006-of-00008.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd7e821703872b76a9e4055362a449e030e6b5c1301be5464df9b37cebfc2657
+size 1990296897

pytorch_model-00007-of-00008.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2876ed1f592aabb3c02dde1441e3cf486029d02287280146c69f60dad81c46d9
+size 1990296897

pytorch_model-00008-of-00008.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d4cd3230172a536631e7e29e72508a7fb517cef05d94f3bac8f59fa67da6ca3
+size 1649439207

pytorch_model.bin.index.json

@@ -0,0 +1,330 @@
+{
+  "metadata": {
+    "total_size": 15441870848
+  },
+  "weight_map": {
+    "lm_head.weight": "pytorch_model-00008-of-00008.bin",
+    "model.embed_tokens.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.0.input_layernorm.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.0.mlp.down_proj.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.0.mlp.gate_proj.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.0.mlp.up_proj.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.0.post_attention_layernorm.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.0.self_attn.k_proj.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.0.self_attn.o_proj.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.0.self_attn.q_proj.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.0.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00008.bin",
+    "model.layers.0.self_attn.v_proj.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.1.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.1.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.1.mlp.gate_proj.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.1.mlp.up_proj.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.1.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.1.self_attn.k_proj.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.1.self_attn.o_proj.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.1.self_attn.q_proj.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.1.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00008.bin",
+    "model.layers.1.self_attn.v_proj.weight": "pytorch_model-00001-of-00008.bin",
+    "model.layers.10.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.10.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.10.mlp.gate_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.10.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.10.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.10.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.10.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.10.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.10.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
+    "model.layers.10.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.11.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.11.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.11.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.11.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.11.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.11.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.11.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.11.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.11.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
+    "model.layers.11.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.12.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.12.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.12.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.12.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.12.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.12.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.12.self_attn.o_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.12.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.12.self_attn.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
+    "model.layers.12.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.13.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.13.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.13.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.13.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.13.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.13.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.13.self_attn.o_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.13.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.13.self_attn.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
+    "model.layers.13.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.14.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.14.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.14.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.14.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.14.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.14.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.14.self_attn.o_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.14.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.14.self_attn.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
+    "model.layers.14.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.15.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.15.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.15.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.15.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.15.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.15.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.15.self_attn.o_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.15.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.15.self_attn.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
+    "model.layers.15.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.16.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.16.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.16.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.16.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.16.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.16.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.16.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.16.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.16.self_attn.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
+    "model.layers.16.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
+    "model.layers.17.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.17.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.17.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.17.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.17.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.17.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.17.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.17.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.17.self_attn.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
+    "model.layers.17.self_attn.v_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.18.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.18.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.18.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.18.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.18.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.18.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.18.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.18.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.18.self_attn.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
+    "model.layers.18.self_attn.v_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.19.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.19.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.19.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.19.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.19.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.19.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.19.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.19.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.19.self_attn.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
+    "model.layers.19.self_attn.v_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.2.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.2.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.2.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.2.mlp.up_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.2.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.2.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.2.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.2.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.2.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
+    "model.layers.2.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.20.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.20.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.20.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.20.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.20.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.20.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.20.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.20.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.20.self_attn.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
+    "model.layers.20.self_attn.v_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.21.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.21.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.21.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.21.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.21.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.21.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.21.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.21.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
+    "model.layers.21.self_attn.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
+    "model.layers.21.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.22.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.22.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.22.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.22.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.22.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.22.self_attn.k_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.22.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.22.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.22.self_attn.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
+    "model.layers.22.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.23.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.23.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.23.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.23.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.23.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.23.self_attn.k_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.23.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.23.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.23.self_attn.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
+    "model.layers.23.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.24.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.24.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.24.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.24.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.24.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.24.self_attn.k_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.24.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.24.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.24.self_attn.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
+    "model.layers.24.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.25.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.25.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.25.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.25.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.25.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.25.self_attn.k_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.25.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.25.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.25.self_attn.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
+    "model.layers.25.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.26.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.26.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.26.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.26.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.26.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.26.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.26.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.26.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
+    "model.layers.26.self_attn.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
+    "model.layers.26.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.27.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.27.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.27.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.27.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.27.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.27.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.27.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.27.self_attn.q_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.27.self_attn.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
+    "model.layers.27.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.28.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.28.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.28.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.28.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.28.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.28.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.28.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.28.self_attn.q_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.28.self_attn.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
+    "model.layers.28.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.29.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.29.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.29.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.29.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.29.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.29.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.29.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.29.self_attn.q_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.29.self_attn.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
+    "model.layers.29.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.3.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.3.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.3.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.3.mlp.up_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.3.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.3.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.3.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.3.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.3.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
+    "model.layers.3.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.30.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.30.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.30.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.30.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.30.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.30.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.30.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.30.self_attn.q_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.30.self_attn.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
+    "model.layers.30.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
+    "model.layers.31.input_layernorm.weight": "pytorch_model-00008-of-00008.bin",
+    "model.layers.31.mlp.down_proj.weight": "pytorch_model-00008-of-00008.bin",
+    "model.layers.31.mlp.gate_proj.weight": "pytorch_model-00008-of-00008.bin",
+    "model.layers.31.mlp.up_proj.weight": "pytorch_model-00008-of-00008.bin",
+    "model.layers.31.post_attention_layernorm.weight": "pytorch_model-00008-of-00008.bin",
+    "model.layers.31.self_attn.k_proj.weight": "pytorch_model-00008-of-00008.bin",
+    "model.layers.31.self_attn.o_proj.weight": "pytorch_model-00008-of-00008.bin",
+    "model.layers.31.self_attn.q_proj.weight": "pytorch_model-00008-of-00008.bin",
+    "model.layers.31.self_attn.rotary_emb.inv_freq": "pytorch_model-00008-of-00008.bin",
+    "model.layers.31.self_attn.v_proj.weight": "pytorch_model-00008-of-00008.bin",
+    "model.layers.4.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.4.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.4.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.4.mlp.up_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.4.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.4.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.4.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.4.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.4.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
+    "model.layers.4.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.5.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.5.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.5.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.5.mlp.up_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.5.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.5.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.5.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.5.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.5.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
+    "model.layers.5.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.6.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.6.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.6.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.6.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.6.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.6.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.6.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.6.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.6.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
+    "model.layers.6.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
+    "model.layers.7.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.7.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.7.mlp.gate_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.7.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.7.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.7.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.7.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.7.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.7.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
+    "model.layers.7.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.8.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.8.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.8.mlp.gate_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.8.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.8.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.8.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.8.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.8.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.8.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
+    "model.layers.8.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.9.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.9.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.9.mlp.gate_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.9.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.9.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.9.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.9.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.9.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.layers.9.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
+    "model.layers.9.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
+    "model.norm.weight": "pytorch_model-00008-of-00008.bin"
+  }
+}

special_tokens_map.json

@@ -0,0 +1,5 @@
+{
+  "bos_token": "<|endoftext|>",
+  "eos_token": "<|endoftext|>",
+  "unk_token": "<|endoftext|>"
+}

tokenization_qwen.py

@@ -0,0 +1,566 @@
+# Copyright (c) Alibaba Cloud.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+"""Tokenization classes for QWen."""
+
+import base64
+import logging
+import os
+import requests
+import unicodedata
+from typing import Collection, Dict, List, Set, Tuple, Union, Any, Callable, Optional
+
+import tiktoken
+import numpy as np
+from PIL import Image
+from PIL import ImageFont
+from PIL import ImageDraw
+from transformers import PreTrainedTokenizer, AddedToken
+from transformers.utils import try_to_load_from_cache
+
+import matplotlib.pyplot as plt
+import matplotlib.colors as mcolors
+from matplotlib.font_manager import FontProperties
+
+logger = logging.getLogger(__name__)
+
+
+VOCAB_FILES_NAMES = {"vocab_file": "qwen.tiktoken", "ttf": "SimSun.ttf"}
+
+PAT_STR = r"""(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+"""
+ENDOFTEXT = "<|endoftext|>"
+IMSTART = "<|im_start|>"
+IMEND = "<|im_end|>"
+# as the default behavior is changed to allow special tokens in
+# regular texts, the surface forms of special tokens need to be
+# as different as possible to minimize the impact
+EXTRAS = tuple((f"<|extra_{i}|>" for i in range(205)))
+SPECIAL_TOKENS = (
+    ENDOFTEXT,
+    IMSTART,
+    IMEND,
+) + EXTRAS
+IMG_TOKEN_SPAN = 256
+
+
+def _load_tiktoken_bpe(tiktoken_bpe_file: str) -> Dict[bytes, int]:
+    with open(tiktoken_bpe_file, "rb") as f:
+        contents = f.read()
+    return {
+        base64.b64decode(token): int(rank)
+        for token, rank in (line.split() for line in contents.splitlines() if line)
+    }
+
+def _list_find(
+    input_list: List[Any],
+    candidates: Tuple[Any],
+    start: int = 0,
+):
+    for i in range(start, len(input_list)):
+        if input_list[i] in candidates:
+            return i
+    return -1
+
+def _replace_closed_tag(
+    input_tokens: List[Any],
+    start_tags: Union[Any, Tuple[Any]],
+    end_tags: Union[Any, Tuple[Any]],
+    inclusive_replace_func: Callable,
+    exclusive_replace_func: Callable = lambda x: x,
+):
+    if isinstance(start_tags, (str, int)):
+        start_tags = (start_tags,)
+    if isinstance(end_tags, (str, int)):
+        end_tags = (end_tags,)
+    assert len(start_tags) == len(end_tags)
+
+    output_tokens = []
+    end = 0
+    while True:
+        start = _list_find(input_tokens, start_tags, end)
+        if start == -1:
+            break
+        output_tokens.extend(exclusive_replace_func(input_tokens[end : start]))
+        tag_idx = start_tags.index(input_tokens[start])
+        end = _list_find(input_tokens, (end_tags[tag_idx],), start)
+        if end == -1:
+            raise ValueError("Unclosed image token")
+        output_tokens.extend(inclusive_replace_func(input_tokens[start : end + 1]))
+        end += 1
+    output_tokens.extend(exclusive_replace_func(input_tokens[end : ]))
+    return output_tokens
+
+class QWenTokenizer(PreTrainedTokenizer):
+    """QWen tokenizer."""
+
+    vocab_files_names = VOCAB_FILES_NAMES
+
+    def __init__(
+        self,
+        vocab_file,
+        errors="replace",
+        image_start_tag='<img>',
+        image_end_tag='</img>',
+        image_pad_tag='<imgpad>',
+        ref_start_tag='<ref>',
+        ref_end_tag='</ref>',
+        box_start_tag='<box>',
+        box_end_tag='</box>',
+        quad_start_tag='<quad>',
+        quad_end_tag='</quad>',
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.image_start_tag = image_start_tag
+        self.image_end_tag = image_end_tag
+        self.image_pad_tag = image_pad_tag
+        self.ref_start_tag = ref_start_tag
+        self.ref_end_tag = ref_end_tag
+        self.box_start_tag = box_start_tag
+        self.box_end_tag = box_end_tag
+        self.quad_start_tag = quad_start_tag
+        self.quad_end_tag = quad_end_tag
+        self.IMAGE_ST = (
+            ref_start_tag, ref_end_tag,
+            box_start_tag, box_end_tag,
+            quad_start_tag, quad_end_tag,
+            image_start_tag, image_end_tag,
+            image_pad_tag
+        )
+
+        self.errors = errors  # how to handle errors in decoding
+
+        self.mergeable_ranks = _load_tiktoken_bpe(vocab_file)  # type: dict[bytes, int]
+        self.special_tokens = {
+            token: index
+            for index, token in enumerate(
+                SPECIAL_TOKENS + self.IMAGE_ST, start=len(self.mergeable_ranks)
+            )
+        }
+        self.img_start_id = self.special_tokens[self.image_start_tag]
+        self.img_end_id = self.special_tokens[self.image_end_tag]
+        self.img_pad_id = self.special_tokens[self.image_pad_tag]
+        self.ref_start_id = self.special_tokens[self.ref_start_tag]
+        self.ref_end_id = self.special_tokens[self.ref_end_tag]
+        self.box_start_id = self.special_tokens[self.box_start_tag]
+        self.box_end_id = self.special_tokens[self.box_end_tag]
+        self.quad_start_id = self.special_tokens[self.quad_start_tag]
+        self.quad_end_id = self.special_tokens[self.quad_end_tag]
+
+        enc = tiktoken.Encoding(
+            "Qwen",
+            pat_str=PAT_STR,
+            mergeable_ranks=self.mergeable_ranks,
+            special_tokens=self.special_tokens,
+        )
+        assert (
+            len(self.mergeable_ranks) + len(self.special_tokens) == enc.n_vocab
+        ), f"{len(self.mergeable_ranks) + len(self.special_tokens)} != {enc.n_vocab} in encoding"
+
+        self.decoder = {
+            v: k for k, v in self.mergeable_ranks.items()
+        }  # type: dict[int, bytes|str]
+        self.decoder.update({v: k for k, v in self.special_tokens.items()})
+
+        self.tokenizer = enc  # type: tiktoken.Encoding
+
+        self.eod_id = self.tokenizer.eot_token
+        self.im_start_id = self.special_tokens[IMSTART]
+        self.im_end_id = self.special_tokens[IMEND]
+
+    def __len__(self) -> int:
+        return self.tokenizer.n_vocab
+
+    def get_vocab(self) -> Dict[bytes, int]:
+        return self.mergeable_ranks
+
+    def convert_tokens_to_ids(
+        self, tokens: Union[bytes, str, List[Union[bytes, str]]]
+    ) -> List[int]:
+        ids = []
+        if isinstance(tokens, (str, bytes)):
+            if tokens in self.special_tokens:
+                return self.special_tokens[tokens]
+            else:
+                return self.mergeable_ranks.get(tokens)
+        for token in tokens:
+            if token in self.special_tokens:
+                ids.append(self.special_tokens[token])
+            else:
+                ids.append(self.mergeable_ranks.get(token))
+        return ids
+
+    def _add_tokens(self, new_tokens: Union[List[str], List[AddedToken]], special_tokens: bool = False) -> int:
+        if not special_tokens and new_tokens:
+            raise ValueError('Adding regular tokens is not supported')
+        for token in new_tokens:
+            surface_form = token.content if isinstance(token, AddedToken) else token
+            if surface_form not in SPECIAL_TOKENS + self.IMAGE_ST:
+                raise ValueError('Adding unknown special tokens is not supported')
+        return 0
+
+    def save_vocabulary(self, save_directory: str, **kwargs) -> Tuple[str]:
+        """
+        Save only the vocabulary of the tokenizer (vocabulary).
+
+        Returns:
+            `Tuple(str)`: Paths to the files saved.
+        """
+        file_path = os.path.join(save_directory, "qwen.tiktoken")
+        with open(file_path, "w", encoding="utf8") as w:
+            for k, v in self.mergeable_ranks.items():
+                line = base64.b64encode(k).decode("utf8") + " " + str(v) + "\n"
+                w.write(line)
+        return (file_path,)
+
+    def tokenize(
+        self,
+        text: str,
+        allowed_special: Union[Set, str] = "all",
+        disallowed_special: Union[Collection, str] = (),
+        **kwargs,
+    ) -> List[Union[bytes, str]]:
+        """
+        Converts a string in a sequence of tokens.
+
+        Args:
+            text (`str`):
+                The sequence to be encoded.
+            allowed_special (`Literal["all"]` or `set`):
+                The surface forms of the tokens to be encoded as special tokens in regular texts.
+                Default to "all".
+            disallowed_special (`Literal["all"]` or `Collection`):
+                The surface forms of the tokens that should not be in regular texts and trigger errors.
+                Default to an empty tuple.
+
+            kwargs (additional keyword arguments, *optional*):
+                Will be passed to the underlying model specific encode method.
+
+        Returns:
+            `List[bytes|str]`: The list of tokens.
+        """
+        tokens = []
+        text = unicodedata.normalize("NFC", text)
+
+        # this implementation takes a detour: text -> token id -> token surface forms
+        for t in self.tokenizer.encode(
+            text, allowed_special=allowed_special, disallowed_special=disallowed_special
+        ):
+            tokens.append(self.decoder[t])
+
+        def _encode_imgurl(img_tokens):
+            assert img_tokens[0] == self.image_start_tag and img_tokens[-1] == self.image_end_tag
+            img_tokens = img_tokens[1:-1]
+            img_url = b''.join(img_tokens)
+            out_img_tokens = list(map(self.decoder.get, img_url))
+            if len(out_img_tokens) > IMG_TOKEN_SPAN:
+                raise ValueError("The content in {}..{} is too long".format(
+                    self.image_start_tag, self.image_end_tag))
+            out_img_tokens.extend([self.image_pad_tag] * (IMG_TOKEN_SPAN - len(out_img_tokens)))
+            out_img_tokens = [self.image_start_tag] + out_img_tokens + [self.image_end_tag]
+            return out_img_tokens
+
+        return _replace_closed_tag(tokens, self.image_start_tag, self.image_end_tag, _encode_imgurl)
+
+    def convert_tokens_to_string(self, tokens: List[Union[bytes, str]]) -> str:
+        """
+        Converts a sequence of tokens in a single string.
+        """
+        text = ""
+        temp = b""
+        for t in tokens:
+            if isinstance(t, str):
+                if temp:
+                    text += temp.decode("utf-8", errors=self.errors)
+                    temp = b""
+                text += t
+            elif isinstance(t, bytes):
+                temp += t
+            else:
+                raise TypeError("token should only be of type types or str")
+        if temp:
+            text += temp.decode("utf-8", errors=self.errors)
+        return text
+
+    @property
+    def vocab_size(self):
+        return self.tokenizer.n_vocab
+
+    def _convert_id_to_token(self, index: int) -> Union[bytes, str]:
+        """Converts an id to a token, special tokens included"""
+        if index in self.decoder:
+            return self.decoder[index]
+        raise ValueError("unknown ids")
+
+    def _convert_token_to_id(self, token: Union[bytes, str]) -> int:
+        """Converts a token to an id using the vocab, special tokens included"""
+        if token in self.special_tokens:
+            return self.special_tokens[token]
+        if token in self.mergeable_ranks:
+            return self.mergeable_ranks[token]
+        raise ValueError("unknown token")
+
+    def _tokenize(self, text: str, **kwargs):
+        """
+        Converts a string in a sequence of tokens (string), using the tokenizer. Split in words for word-based
+        vocabulary or sub-words for sub-word-based vocabularies (BPE/SentencePieces/WordPieces).
+
+        Do NOT take care of added tokens.
+        """
+        raise NotImplementedError
+
+    def _decode(
+        self,
+        token_ids: Union[int, List[int]],
+        skip_special_tokens: bool = False,
+        errors: str = None,
+        **kwargs,
+    ) -> str:
+        if isinstance(token_ids, int):
+            token_ids = [token_ids]
+
+        def _decode_imgurl(img_token_ids):
+            assert img_token_ids[0] == self.img_start_id and img_token_ids[-1] == self.img_end_id
+            img_token_ids = img_token_ids[1:-1]
+            img_token_ids = img_token_ids[ : img_token_ids.index(self.img_pad_id)]
+            img_url = bytes(img_token_ids).decode('utf-8')
+            return [self.img_start_id] + self.tokenizer.encode(img_url) + [self.img_end_id]
+
+        token_ids = _replace_closed_tag(token_ids, self.img_start_id, self.img_end_id, _decode_imgurl)
+
+        if skip_special_tokens:
+            token_ids = [i for i in token_ids if i < self.eod_id]
+        return self.tokenizer.decode(token_ids, errors=errors or self.errors)
+
+    def to_list_format(self, text: str):
+        text = unicodedata.normalize("NFC", text)
+        token_ids = self.tokenizer.encode(
+            text, allowed_special=set(self.IMAGE_ST + (ENDOFTEXT,)))
+
+        def _encode_vl_info(tokens):
+            if len(tokens) == 0:
+                return []
+            if tokens[0] == self.img_start_id and tokens[-1] == self.img_end_id:
+                key = 'image'
+            elif tokens[0] == self.ref_start_id and tokens[-1] == self.ref_end_id:
+                key = 'ref'
+            elif tokens[0] == self.box_start_id and tokens[-1] == self.box_end_id:
+                key = 'box'
+            elif tokens[0] == self.quad_start_id and tokens[-1] == self.quad_end_id:
+                key = 'quad'
+            else:
+                _tobytes = lambda x: x.encode('utf-8') if isinstance(x, str) else x
+                return [{'text': b''.join(map(_tobytes, map(self.decoder.get, tokens))).decode('utf-8')}]
+            val = b''.join(map(self.decoder.get, tokens[1:-1])).decode('utf-8')
+            return [{key: val}]
+
+        return _replace_closed_tag(
+            token_ids,
+            (self.img_start_id, self.ref_start_id, self.box_start_id, self.quad_start_id),
+            (self.img_end_id, self.ref_end_id, self.box_end_id, self.quad_end_id),
+            _encode_vl_info,
+            _encode_vl_info,
+        )
+
+    def from_list_format(self, list_format: List[Dict]):
+        text = ''
+        num_images = 0
+        for ele in list_format:
+            if 'image' in ele:
+                num_images += 1
+                text += f'Picture {num_images}:'
+                text += self.image_start_tag + ele['image'] + self.image_end_tag
+                text += '\n'
+            elif 'text' in ele:
+                text += ele['text']
+            elif 'box' in ele:
+                if 'ref' in ele:
+                    text += self.ref_start_tag + ele['ref'] + self.ref_end_tag
+                for box in ele['box']:
+                    text += self.box_start_tag + '(%d,%d),(%d,%d)' % (box[0], box[1], box[2], box[3]) + self.box_end_tag
+            else:
+                raise ValueError("Unsupport element: " + str(ele))
+        return text
+
+    def _fetch_latest_picture(self, response, history):
+        if history is None:
+            history = []
+        _history = history + [(response, None)]
+        for q, r in _history[::-1]:
+            for ele in self.to_list_format(q)[::-1]:
+                if 'image' in ele:
+                    return ele['image']
+        return None
+
+    def _fetch_all_box_with_ref(self, text):
+        list_format = self.to_list_format(text)
+        output = []
+        for i, ele in enumerate(list_format):
+            if 'box' in ele:
+                bbox = tuple(map(int, ele['box'].replace('(', '').replace(')', '').split(',')))
+                assert len(bbox) == 4
+                output.append({'box': bbox})
+                if i > 0 and 'ref' in list_format[i-1]:
+                    output[-1]['ref'] = list_format[i-1]['ref'].strip()
+        return output
+
+    def draw_bbox_on_latest_picture(
+        self,
+        response,
+        history=None,
+    ) -> Optional[Image.Image]:
+        image = self._fetch_latest_picture(response, history)
+        if image is None:
+            return None
+        if image.startswith("http://") or image.startswith("https://"):
+            image = Image.open(requests.get(image, stream=True).raw).convert("RGB")
+            h, w = image.height, image.width
+        else:
+            image = plt.imread(image)
+            h, w = image.shape[0], image.shape[1]
+        visualizer = Visualizer(image)
+
+        boxes = self._fetch_all_box_with_ref(response)
+        if not boxes:
+            return None
+        color = random.choice([_ for _ in mcolors.TABLEAU_COLORS.keys()]) # init color
+        for box in boxes:
+            if 'ref' in box: # random new color for new refexps
+                color = random.choice([_ for _ in mcolors.TABLEAU_COLORS.keys()])
+            x1, y1, x2, y2 = box['box']
+            x1, y1, x2, y2 = (int(x1 / 1000 * w), int(y1 / 1000 * h), int(x2 / 1000 * w), int(y2 / 1000 * h))
+            visualizer.draw_box((x1, y1, x2, y2), alpha=1, edge_color=color)
+            if 'ref' in box:
+                visualizer.draw_text(box['ref'], (x1, y1), color=color, horizontal_alignment="left")
+        return visualizer.output
+
+
+import colorsys
+import logging
+import math
+import numpy as np
+import matplotlib as mpl
+import matplotlib.colors as mplc
+import matplotlib.figure as mplfigure
+import torch
+from matplotlib.backends.backend_agg import FigureCanvasAgg
+from PIL import Image
+import random
+
+logger = logging.getLogger(__name__)
+
+
+class VisImage:
+    def __init__(self, img, scale=1.0):
+        self.img = img
+        self.scale = scale
+        self.width, self.height = img.shape[1], img.shape[0]
+        self._setup_figure(img)
+
+    def _setup_figure(self, img):
+        fig = mplfigure.Figure(frameon=False)
+        self.dpi = fig.get_dpi()
+        # add a small 1e-2 to avoid precision lost due to matplotlib's truncation
+        # (https://github.com/matplotlib/matplotlib/issues/15363)
+        fig.set_size_inches(
+            (self.width * self.scale + 1e-2) / self.dpi,
+            (self.height * self.scale + 1e-2) / self.dpi,
+        )
+        self.canvas = FigureCanvasAgg(fig)
+        # self.canvas = mpl.backends.backend_cairo.FigureCanvasCairo(fig)
+        ax = fig.add_axes([0.0, 0.0, 1.0, 1.0])
+        ax.axis("off")
+        self.fig = fig
+        self.ax = ax
+        self.reset_image(img)
+
+    def reset_image(self, img):
+        img = img.astype("uint8")
+        self.ax.imshow(img, extent=(0, self.width, self.height, 0), interpolation="nearest")
+
+    def save(self, filepath):
+        self.fig.savefig(filepath)
+
+    def get_image(self):
+        canvas = self.canvas
+        s, (width, height) = canvas.print_to_buffer()
+
+        buffer = np.frombuffer(s, dtype="uint8")
+
+        img_rgba = buffer.reshape(height, width, 4)
+        rgb, alpha = np.split(img_rgba, [3], axis=2)
+        return rgb.astype("uint8")
+
+
+class Visualizer:
+    def __init__(self, img_rgb, metadata=None, scale=1.0):
+        self.img = np.asarray(img_rgb).clip(0, 255).astype(np.uint8)
+        self.font_path = try_to_load_from_cache("Qwen/Qwen-VL-Chat", "SimSun.ttf")
+        self.output = VisImage(self.img, scale=scale)
+        self.cpu_device = torch.device("cpu")
+
+        # too small texts are useless, therefore clamp to 14
+        self._default_font_size = max(
+            np.sqrt(self.output.height * self.output.width) // 30, 15 // scale
+        )
+
+    def draw_text(
+        self,
+        text,
+        position,
+        *,
+        font_size=None,
+        color="g",
+        horizontal_alignment="center",
+        rotation=0,
+    ):
+        if not font_size:
+            font_size = self._default_font_size
+
+        # since the text background is dark, we don't want the text to be dark
+        color = np.maximum(list(mplc.to_rgb(color)), 0.2)
+        color[np.argmax(color)] = max(0.8, np.max(color))
+
+        x, y = position
+        self.output.ax.text(
+            x,
+            y,
+            text,
+            size=font_size * self.output.scale,
+            fontproperties=FontProperties(fname=self.font_path),
+            bbox={"facecolor": "black", "alpha": 0.8, "pad": 0.7, "edgecolor": "none"},
+            verticalalignment="top",
+            horizontalalignment=horizontal_alignment,
+            color=color,
+            zorder=10,
+            rotation=rotation,
+        )
+        return self.output
+
+    def draw_box(self, box_coord, alpha=0.5, edge_color="g", line_style="-"):
+        
+        x0, y0, x1, y1 = box_coord
+        width = x1 - x0
+        height = y1 - y0
+
+        linewidth = max(self._default_font_size / 4, 1)
+
+        self.output.ax.add_patch(
+            mpl.patches.Rectangle(
+                (x0, y0),
+                width,
+                height,
+                fill=False,
+                edgecolor=edge_color,
+                linewidth=linewidth * self.output.scale,
+                alpha=alpha,
+                linestyle=line_style,
+            )
+        )
+        return self.output
+
+    def get_output(self):
+        
+        return self.output

tokenizer_config.json

@@ -0,0 +1,10 @@
+{
+  "model_max_length": 999999999999999999,
+  "tokenizer_class": "QWenTokenizer",
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_qwen.QWenTokenizer",
+      null
+      ]
+  }
+}

vision.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f293b74085184879a81aa6b67301fd7156ffff8eaee4fa4ea65a1f75a4b44f5d
+size 3871409417

visual.py

@@ -0,0 +1,428 @@
+# Copyright (c) Alibaba Cloud.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+from collections import OrderedDict
+import math
+import requests
+from io import BytesIO
+from functools import partial
+from PIL import Image
+from typing import Callable, Optional, Sequence, Tuple, List
+import numpy as np
+
+import torch
+from torch import nn
+from torch.nn import functional as F
+from torch.nn.init import trunc_normal_
+from torchvision import transforms
+from torchvision.transforms import InterpolationMode
+
+
+def get_abs_pos(abs_pos, tgt_size):
+    # abs_pos: L, C
+    # tgt_size: M
+    # return: M, C
+    src_size = int(math.sqrt(abs_pos.size(0)))
+    tgt_size = int(math.sqrt(tgt_size))
+    dtype = abs_pos.dtype
+
+    if src_size != tgt_size:
+        return F.interpolate(
+            abs_pos.float().reshape(1, src_size, src_size, -1).permute(0, 3, 1, 2),
+            size=(tgt_size, tgt_size),
+            mode="bicubic",
+            align_corners=False,
+        ).permute(0, 2, 3, 1).flatten(0, 2).to(dtype=dtype)
+    else:
+        return abs_pos
+
+# https://github.com/facebookresearch/mae/blob/efb2a8062c206524e35e47d04501ed4f544c0ae8/util/pos_embed.py#L20
+def get_2d_sincos_pos_embed(embed_dim, grid_size, cls_token=False):
+    """
+    grid_size: int of the grid height and width
+    return:
+    pos_embed: [grid_size*grid_size, embed_dim] or [1+grid_size*grid_size, embed_dim] (w/ or w/o cls_token)
+    """
+    grid_h = np.arange(grid_size, dtype=np.float32)
+    grid_w = np.arange(grid_size, dtype=np.float32)
+    grid = np.meshgrid(grid_w, grid_h)  # here w goes first
+    grid = np.stack(grid, axis=0)
+
+    grid = grid.reshape([2, 1, grid_size, grid_size])
+    pos_embed = get_2d_sincos_pos_embed_from_grid(embed_dim, grid)
+    if cls_token:
+        pos_embed = np.concatenate([np.zeros([1, embed_dim]), pos_embed], axis=0)
+    return pos_embed
+
+
+def get_2d_sincos_pos_embed_from_grid(embed_dim, grid):
+    assert embed_dim % 2 == 0
+
+    # use half of dimensions to encode grid_h
+    emb_h = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[0])  # (H*W, D/2)
+    emb_w = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[1])  # (H*W, D/2)
+
+    emb = np.concatenate([emb_h, emb_w], axis=1) # (H*W, D)
+    return emb
+
+
+def get_1d_sincos_pos_embed_from_grid(embed_dim, pos):
+    """
+    embed_dim: output dimension for each position
+    pos: a list of positions to be encoded: size (M,)
+    out: (M, D)
+    """
+    assert embed_dim % 2 == 0
+    omega = np.arange(embed_dim // 2, dtype=np.float32)
+    omega /= embed_dim / 2.
+    omega = 1. / 10000**omega  # (D/2,)
+
+    pos = pos.reshape(-1)  # (M,)
+    out = np.einsum('m,d->md', pos, omega)  # (M, D/2), outer product
+
+    emb_sin = np.sin(out) # (M, D/2)
+    emb_cos = np.cos(out) # (M, D/2)
+
+    emb = np.concatenate([emb_sin, emb_cos], axis=1)  # (M, D)
+    return emb
+
+
+class Resampler(nn.Module):
+    """
+    A 2D perceiver-resampler network with one cross attention layers by
+        (grid_size**2) learnable queries and 2d sincos pos_emb
+    Outputs:
+        A tensor with the shape of (grid_size**2, embed_dim)
+    """
+    def __init__(
+            self,
+            grid_size,
+            embed_dim,
+            num_heads,
+            kv_dim=None,
+            norm_layer=nn.LayerNorm
+    ):
+        super().__init__()
+        self.num_queries = grid_size ** 2
+        self.embed_dim = embed_dim
+        self.num_heads = num_heads
+
+        self.pos_embed = nn.Parameter(
+            torch.from_numpy(get_2d_sincos_pos_embed(embed_dim, grid_size)).float()
+        ).requires_grad_(False)
+
+        self.query = nn.Parameter(torch.zeros(self.num_queries, embed_dim))
+        trunc_normal_(self.query, std=.02)
+
+        if kv_dim is not None and kv_dim != embed_dim:
+            self.kv_proj = nn.Linear(kv_dim, embed_dim, bias=False)
+        else:
+            self.kv_proj = nn.Identity()
+
+        self.attn = nn.MultiheadAttention(embed_dim, num_heads)
+        self.ln_q = norm_layer(embed_dim)
+        self.ln_kv = norm_layer(embed_dim)
+        
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight, std=.02)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                nn.init.constant_(m.bias, 0)
+        elif isinstance(m, nn.LayerNorm):
+            nn.init.constant_(m.bias, 0)
+            nn.init.constant_(m.weight, 1.0)
+
+    def forward(self, x, attn_mask=None):
+
+        pos_embed = get_abs_pos(self.pos_embed, x.size(1))
+
+        x = self.kv_proj(x)
+        x = self.ln_kv(x).permute(1, 0, 2)
+
+        N = x.shape[1]
+        q = self.ln_q(self.query)
+        out = self.attn(
+            self._repeat(q, N) + self.pos_embed.unsqueeze(1),
+            x + pos_embed.unsqueeze(1),
+            x,
+            attn_mask=attn_mask)[0]
+        return out.permute(1, 0, 2)
+
+    def _repeat(self, query, N: int):
+        return query.unsqueeze(1).repeat(1, N, 1)
+
+
+class VisualAttention(nn.Module):
+    """self-attention layer class.
+
+    Self-attention layer takes input with size [s, b, h]
+    and returns output of the same size.
+    """
+
+    def __init__(self, embed_dim, num_heads,
+                 bias=True, kdim=None, vdim=None):
+        super(VisualAttention, self).__init__()
+        self.embed_dim = embed_dim
+        self.kdim = kdim if kdim is not None else embed_dim
+        self.vdim = vdim if vdim is not None else embed_dim
+        self._qkv_same_embed_dim = self.kdim == embed_dim and self.vdim == embed_dim
+
+        self.num_heads = num_heads
+
+        # Per attention head and per partition values.
+        assert embed_dim % num_heads == 0
+        self.hidden_size_per_attention_head = embed_dim // num_heads
+        self.num_attention_heads_per_partition = num_heads
+        self.hidden_size_per_partition = embed_dim
+
+        # Strided linear layer.
+        assert self._qkv_same_embed_dim, 'Only Support SelfAttention Currently'
+        self.in_proj = nn.Linear(embed_dim, 3 * embed_dim)
+        self.out_proj = nn.Linear(embed_dim, embed_dim)
+        self.norm_factor = math.sqrt(self.hidden_size_per_attention_head)
+
+    def forward(self, query, key, value, attn_mask = None):
+        # query/key/value: [sq, b, h]
+        sq, b, _ = query.size()
+
+        assert query is key, 'Only Support Self-Attention Currently'
+        sk = sq
+        mixed_x_layer = self.in_proj(query)
+
+        # [sq, b, (np * 3 * hn)] --> [sq, b, np, 3 * hn]
+        new_tensor_shape = mixed_x_layer.size()[:-1] + \
+            (self.num_attention_heads_per_partition,
+             3 * self.hidden_size_per_attention_head)
+        mixed_x_layer = mixed_x_layer.view(*new_tensor_shape)
+
+        # [sq, b, np, 3 * hn] --> 3 [sq, b, np, hn]
+        query_layer, key_layer, value_layer = mixed_x_layer.split(
+            self.hidden_size_per_attention_head, dim=-1)
+
+        # [sq, b, np, hn] -> [sq, b * np, hn]
+        query_layer = query_layer.view(sq,
+            b * self.num_attention_heads_per_partition,
+            self.hidden_size_per_attention_head).transpose(0, 1)
+        # [sk, b, np, hn] -> [sk, b * np, hn]
+        key_layer = key_layer.view(sk,
+            b * self.num_attention_heads_per_partition,
+            self.hidden_size_per_attention_head).transpose(0, 1)
+
+        q_scaled = query_layer / self.norm_factor
+        if attn_mask is not None:
+            attention_probs = torch.baddbmm(attn_mask, q_scaled, key_layer.transpose(-2, -1))
+        else:
+            attention_probs = torch.bmm(q_scaled, key_layer.transpose(-2, -1))
+        attention_probs = attention_probs.softmax(dim=-1)
+
+        value_layer = value_layer.view(sk,
+            b * self.num_attention_heads_per_partition,
+            self.hidden_size_per_attention_head).transpose(0, 1)
+
+        # matmul: [b * np, sq, hn]
+        context_layer = torch.bmm(attention_probs, value_layer)
+
+        # change view [b, np, sq, hn]
+        context_layer = context_layer.view(b,
+            self.num_attention_heads_per_partition,
+            sq, self.hidden_size_per_attention_head)
+
+        # [b, np, sq, hn] --> [sq, b, np, hn]
+        context_layer = context_layer.permute(2, 0, 1, 3).contiguous()
+
+        # [sq, b, np, hn] --> [sq, b, hp]
+        new_context_layer_shape = context_layer.size()[:-2] + \
+            (self.hidden_size_per_partition,)
+        context_layer = context_layer.view(*new_context_layer_shape)
+
+        output = self.out_proj(context_layer)
+
+        return output
+
+
+class VisualAttentionBlock(nn.Module):
+    def __init__(
+            self,
+            d_model: int,
+            n_head: int,
+            mlp_ratio: float = 4.0,
+            act_layer: Callable = nn.GELU,
+            norm_layer: Callable = nn.LayerNorm,
+            is_cross_attention: bool = False,
+    ):
+        super().__init__()
+
+        self.ln_1 = norm_layer(d_model)
+        if is_cross_attention:
+            self.ln_1_kv = norm_layer(d_model)
+
+        self.ln_2 = norm_layer(d_model)
+        mlp_width = int(d_model * mlp_ratio)
+        self.attn = VisualAttention(d_model, n_head)
+        self.mlp = nn.Sequential(OrderedDict([
+            ("c_fc", nn.Linear(d_model, mlp_width)),
+            ("gelu", act_layer()),
+            ("c_proj", nn.Linear(mlp_width, d_model))
+        ]))
+
+    def attention(
+            self,
+            q_x: torch.Tensor,
+            k_x: Optional[torch.Tensor] = None,
+            v_x: Optional[torch.Tensor] = None,
+            attn_mask: Optional[torch.Tensor] = None,
+    ):
+        k_x = k_x if k_x is not None else q_x
+        v_x = v_x if v_x is not None else q_x
+
+        attn_mask = attn_mask.to(q_x.dtype) if attn_mask is not None else None
+        return self.attn(q_x, k_x, v_x, attn_mask=attn_mask)
+
+    def forward(
+            self,
+            q_x: torch.Tensor,
+            k_x: Optional[torch.Tensor] = None,
+            v_x: Optional[torch.Tensor] = None,
+            attn_mask: Optional[torch.Tensor] = None,
+    ):
+        k_x = self.ln_1_kv(k_x) if hasattr(self, "ln_1_kv") and k_x is not None else None
+        v_x = self.ln_1_kv(v_x) if hasattr(self, "ln_1_kv") and v_x is not None else None
+
+        x = q_x + self.attention(q_x=self.ln_1(q_x), k_x=k_x, v_x=v_x, attn_mask=attn_mask)
+        x = x + self.mlp(self.ln_2(x))
+        return x
+
+
+class TransformerBlock(nn.Module):
+    def __init__(
+            self,
+            width: int,
+            layers: int,
+            heads: int,
+            mlp_ratio: float = 4.0,
+            act_layer: Callable = nn.GELU,
+            norm_layer: Callable = nn.LayerNorm,
+    ):
+        super().__init__()
+        self.width = width
+        self.layers = layers
+
+        self.resblocks = nn.ModuleList([
+            VisualAttentionBlock(
+                width, heads, mlp_ratio, act_layer=act_layer, norm_layer=norm_layer)
+            for _ in range(layers)
+        ])
+
+    def get_cast_dtype(self) -> torch.dtype:
+        return self.resblocks[0].mlp.c_fc.weight.dtype
+
+    def get_cast_device(self) -> torch.device:
+        return self.resblocks[0].mlp.c_fc.weight.device
+
+    def forward(self, x: torch.Tensor, attn_mask: Optional[torch.Tensor] = None):
+        for r in self.resblocks:
+            x = r(x, attn_mask=attn_mask)
+        return x
+
+
+class VisionTransformer(nn.Module):
+
+    def __init__(
+            self,
+            image_size: int,
+            patch_size: int,
+            width: int,
+            layers: int,
+            heads: int,
+            mlp_ratio: float,
+            n_queries: int = 256,
+            output_dim: int = 512,
+            **kwargs
+    ):
+        super().__init__()
+        image_height, image_width = self.image_size = (image_size, image_size)
+        patch_height, patch_width = self.patch_size = (patch_size, patch_size)
+        self.grid_size = (image_height // patch_height, image_width // patch_width)
+        self.output_dim = output_dim
+
+        mean = (0.48145466, 0.4578275, 0.40821073)
+        std = (0.26862954, 0.26130258, 0.27577711)
+        self.image_transform = transforms.Compose([
+            transforms.Resize(
+                (image_size, image_size),
+                interpolation=InterpolationMode.BICUBIC
+            ),
+            transforms.ToTensor(),
+            transforms.Normalize(mean=mean, std=std),
+        ])
+
+        self.conv1 = nn.Conv2d(in_channels=3, out_channels=width, kernel_size=patch_size, stride=patch_size, bias=False)
+
+        # class embeddings and positional embeddings
+        scale = width ** -0.5
+        self.positional_embedding = nn.Parameter(scale * torch.randn(256, width))
+
+        norm_layer = partial(nn.LayerNorm, eps=1e-6)
+        act_layer = nn.GELU
+
+        self.ln_pre = norm_layer(width)
+        self.transformer = TransformerBlock(
+            width,
+            layers,
+            heads,
+            mlp_ratio,
+            act_layer=act_layer,
+            norm_layer=norm_layer,
+        )
+
+        self.attn_pool = Resampler(
+            grid_size=int(math.sqrt(n_queries)),
+            embed_dim=output_dim,
+            num_heads=output_dim // 128,
+            kv_dim=width,
+            norm_layer=norm_layer,
+        )
+        self.ln_post = norm_layer(output_dim)
+        self.proj = nn.Parameter((output_dim** -0.5) * torch.randn(output_dim, output_dim))
+
+    def forward(self, x: torch.Tensor):
+        x = x.to(
+            dtype=self.transformer.get_cast_dtype(),
+            device=self.transformer.get_cast_device(),
+        )
+        # to patches
+        x = self.conv1(x)  # shape = [*, width, grid, grid]
+        x = x.reshape(x.shape[0], x.shape[1], -1)  # shape = [*, width, grid ** 2]
+        x = x.permute(0, 2, 1)  # shape = [*, grid ** 2, width]
+
+        x = x + get_abs_pos(self.positional_embedding, x.size(1))
+
+        x = self.ln_pre(x)
+
+        x = x.permute(1, 0, 2)  # NLD -> LND
+        x = self.transformer(x)
+        x = x.permute(1, 0, 2)  # LND -> NLD
+
+        x = self.attn_pool(x)
+        x = self.ln_post(x)
+        x = x @ self.proj
+
+        return x
+
+    def encode(self, image_paths):
+        images = []
+        for image_path in image_paths:
+            if isinstance(image_path, Image.Image):
+                image = image_path
+            elif image_path.startswith("http://") or image_path.startswith("https://"):
+                image = Image.open(requests.get(image_path, stream=True).raw)
+            else:
+                image = Image.open(image_path)
+            image = image.convert("RGB")
+            images.append(self.image_transform(image))
+        images = torch.stack(images, dim=0)
+        return self(images)