${reasoning.label ? reasoning.label :'Reasoning 🧠'}:
- ${reasoning.status ? escapeHtml(reasoning.status) : ' '}
+ ${reasoning.status ? escapeHtml(reasoning.status) : ''}
${inner_text}
`;
@@ -106,12 +117,18 @@ function render_reasoning_text(reasoning) {
}
function filter_message(text) {
+ if (Array.isArray(text)) {
+ return text;
+ }
return text.replaceAll(
/[\s\S]+/gm, ""
).replace(/ \[aborted\]$/g, "").replace(/ \[error\]$/g, "");
}
function filter_message_content(text) {
+ if (Array.isArray(text)) {
+ return text;
+ }
return text.replace(/ \[aborted\]$/g, "").replace(/ \[error\]$/g, "")
}
@@ -269,11 +286,12 @@ const register_message_buttons = async () => {
return
}
el.dataset.click = true;
- const provider_forms = document.querySelector(".provider_forms");
- const provider_form = provider_forms.querySelector(`#${el.dataset.provider}-form`);
const provider_link = el.querySelector("a");
provider_link?.addEventListener("click", async (event) => {
event.preventDefault();
+ await load_provider_parameters(el.dataset.provider);
+ const provider_forms = document.querySelector(".provider_forms");
+ const provider_form = provider_forms.querySelector(`#${el.dataset.provider}-form`);
if (provider_form) {
provider_form.classList.remove("hidden");
provider_forms.classList.remove("hidden");
@@ -281,11 +299,6 @@ const register_message_buttons = async () => {
}
return false;
});
- document.getElementById("close_provider_forms").addEventListener("click", async () => {
- provider_form.classList.add("hidden");
- provider_forms.classList.add("hidden");
- chat.classList.remove("hidden");
- });
});
chatBody.querySelectorAll(".message .fa-xmark").forEach(async (el) => {
@@ -479,23 +492,24 @@ const delete_conversations = async () => {
await new_conversation();
};
-const handle_ask = async (do_ask_gpt = true) => {
+const handle_ask = async (do_ask_gpt = true, message = null) => {
userInput.style.height = "82px";
userInput.focus();
await scroll_to_bottom();
- let message = userInput.value.trim();
- if (message.length <= 0) {
- return;
+ if (!message) {
+ message = userInput.value.trim();
+ if (!message) {
+ return;
+ }
+ userInput.value = "";
+ await count_input()
}
- userInput.value = "";
- await count_input()
- await add_conversation(window.conversation_id);
// Is message a url?
const expression = /^https?:\/\/(www\.)?[-a-zA-Z0-9@:%._\+~#=]{1,256}\.[a-zA-Z0-9()]{1,6}\b([-a-zA-Z0-9()@:%_\+.~#?&//=]*)$/gi;
const regex = new RegExp(expression);
- if (message.match(regex)) {
+ if (!Array.isArray(message) && message.match(regex)) {
paperclip.classList.add("blink");
const blob = new Blob([JSON.stringify([{url: message}])], { type: 'application/json' });
const file = new File([blob], 'downloads.json', { type: 'application/json' }); // Create File object
@@ -509,6 +523,8 @@ const handle_ask = async (do_ask_gpt = true) => {
connectToSSE(`/backend-api/v2/files/${bucket_id}`, false, bucket_id); //Retrieve and refine
return;
}
+
+ await add_conversation(window.conversation_id);
let message_index = await add_message(window.conversation_id, "user", message);
let message_id = get_message_id();
@@ -602,6 +618,12 @@ document.querySelector(".media-player .fa-x").addEventListener("click", ()=>{
media_player.removeChild(audio);
});
+document.getElementById("close_provider_forms").addEventListener("click", async () => {
+ const provider_forms = document.querySelector(".provider_forms");
+ provider_forms.classList.add("hidden");
+ chat.classList.remove("hidden");
+});
+
const prepare_messages = (messages, message_index = -1, do_continue = false, do_filter = true) => {
messages = [ ...messages ]
if (message_index != null) {
@@ -930,7 +952,6 @@ async function add_message_chunk(message, message_id, provider, scroll, finish_m
Object.entries(message.parameters).forEach(([key, value]) => {
parameters_storage[provider][key] = value;
});
- await load_provider_parameters(provider);
}
}
@@ -1332,6 +1353,9 @@ const new_conversation = async () => {
};
function merge_messages(message1, message2) {
+ if (Array.isArray(message2)) {
+ return message2;
+ }
let newContent = message2;
// Remove start tokens
if (newContent.startsWith("```")) {
@@ -1530,20 +1554,20 @@ const load_conversation = async (conversation, scroll=true) => {
});
if (countTokensEnabled && window.GPTTokenizer_cl100k_base) {
- const filtered = prepare_messages(messages, null, true, false);
- if (filtered.length > 0) {
- last_model = last_model?.startsWith("gpt-3") ? "gpt-3.5-turbo" : "gpt-4"
- let count_total = GPTTokenizer_cl100k_base?.encodeChat(filtered, last_model).length
- if (count_total > 0) {
- elements.push(`
-
-
- ${conversation.updated ? toLocaleDateString(conversation.updated) : ""}
- ${shareIcon} ${escapeHtml(conversation.new_title ? conversation.new_title : conversation.title)}
-
-
-
-
-
-
+ let convo = document.createElement("div");
+ convo.classList.add("convo");
+ convo.id = `convo-${conversation.id}`;
+ convo.innerHTML = `
+
+
+ ${conversation.updated ? toLocaleDateString(conversation.updated) : ""}
+ ${shareIcon} ${escapeHtml(conversation.new_title ? conversation.new_title : conversation.title)}
- `);
+
+
+
+
+
+ `;
+ box_conversations.appendChild(convo);
});
- await clear_conversations();
- box_conversations.innerHTML += html.join("");
};
const hide_input = document.querySelector(".chat-toolbar .hide-input");
@@ -1800,6 +1825,13 @@ const uuid = () => {
);
};
+function generateSecureRandomString(length = 128) {
+ const chars = 'abcdefghijklmnopqrstuvwxyz0123456789';
+ const array = new Uint8Array(length);
+ crypto.getRandomValues(array);
+ return Array.from(array, byte => chars[byte % chars.length]).join('');
+}
+
function get_message_id() {
random_bytes = (Math.floor(Math.random() * 1338377565) + 2956589730).toString(
2
@@ -2003,6 +2035,9 @@ function count_chars(text) {
}
function count_words_and_tokens(text, model, completion_tokens, prompt_tokens) {
+ if (Array.isArray(text)) {
+ return "";
+ }
text = filter_message(text);
return `(${count_words(text)} words, ${count_chars(text)} chars, ${completion_tokens ? completion_tokens : count_tokens(model, text, prompt_tokens)} tokens)`;
}
@@ -2626,12 +2661,12 @@ async function upload_files(fileInput) {
fileInput.value = "";
}
if (result.media) {
+ const media = [];
result.media.forEach((filename)=> {
const url = `/files/${bucket_id}/media/${filename}`;
- image_storage[url] = {bucket_id: bucket_id, name: filename};
+ media.push({bucket_id: bucket_id, name: filename, url: url});
});
- mediaSelect.classList.remove("hidden");
- renderMediaSelect();
+ await handle_ask(false, media);
}
}
diff --git a/g4f/gui/server/backend_api.py b/g4f/gui/server/backend_api.py
index 0627b1f8..d6b3f60a 100644
--- a/g4f/gui/server/backend_api.py
+++ b/g4f/gui/server/backend_api.py
@@ -340,8 +340,7 @@ class Backend_Api(Api):
@app.route('/files/
/media/', methods=['GET'])
def get_media(bucket_id, filename, dirname: str = None):
- bucket_dir = get_bucket_dir(secure_filename(bucket_id), secure_filename(dirname))
- media_dir = os.path.join(bucket_dir, "media")
+ media_dir = get_bucket_dir(dirname, bucket_id, "media")
try:
return send_from_directory(os.path.abspath(media_dir), filename)
except NotFound:
@@ -391,15 +390,14 @@ class Backend_Api(Api):
@self.app.route('/backend-api/v2/chat/', methods=['GET'])
def get_chat(share_id: str) -> str:
share_id = secure_filename(share_id)
- if self.chat_cache.get(share_id, 0) == request.headers.get("if-none-match", 0):
+ if self.chat_cache.get(share_id, 0) == int(request.headers.get("if-none-match", 0)):
return jsonify({"error": {"message": "Not modified"}}), 304
- bucket_dir = get_bucket_dir(share_id)
- file = os.path.join(bucket_dir, "chat.json")
+ file = get_bucket_dir(share_id, "chat.json")
if not os.path.isfile(file):
return jsonify({"error": {"message": "Not found"}}), 404
with open(file, 'r') as f:
chat_data = json.load(f)
- if chat_data.get("updated", 0) == request.headers.get("if-none-match", 0):
+ if chat_data.get("updated", 0) == int(request.headers.get("if-none-match", 0)):
return jsonify({"error": {"message": "Not modified"}}), 304
self.chat_cache[share_id] = chat_data.get("updated", 0)
return jsonify(chat_data), 200
diff --git a/g4f/image/__init__.py b/g4f/image/__init__.py
index bfc85bb6..435d599a 100644
--- a/g4f/image/__init__.py
+++ b/g4f/image/__init__.py
@@ -103,7 +103,7 @@ def is_data_an_media(data, filename: str = None) -> str:
return is_accepted_format(data)
return is_data_uri_an_image(data)
-def is_data_an_audio(data_uri: str, filename: str = None) -> str:
+def is_data_an_audio(data_uri: str = None, filename: str = None) -> str:
if filename:
if filename.endswith(".wav"):
return "audio/wav"
diff --git a/g4f/image/copy_images.py b/g4f/image/copy_images.py
index 01a55c75..df4b9ae0 100644
--- a/g4f/image/copy_images.py
+++ b/g4f/image/copy_images.py
@@ -2,17 +2,18 @@ from __future__ import annotations
import os
import time
-import uuid
import asyncio
import hashlib
import re
from typing import AsyncIterator
from urllib.parse import quote, unquote
from aiohttp import ClientSession, ClientError
+from urllib.parse import urlparse
from ..typing import Optional, Cookies
from ..requests.aiohttp import get_connector, StreamResponse
from ..image import MEDIA_TYPE_MAP, EXTENSIONS_MAP
+from ..tools.files import secure_filename
from ..providers.response import ImageResponse, AudioResponse, VideoResponse
from ..Provider.template import BackendApi
from . import is_accepted_format, extract_data_uri
@@ -23,13 +24,15 @@ images_dir = "./generated_images"
def get_media_extension(media: str) -> str:
"""Extract media file extension from URL or filename"""
- match = re.search(r"\.(j?[a-z]{3})(?:\?|$)", media, re.IGNORECASE)
- extension = match.group(1).lower() if match else ""
+ path = urlparse(media).path
+ extension = os.path.splitext(path)[1]
+ if not extension:
+ extension = os.path.splitext(media)[1]
if not extension:
return ""
- if extension not in EXTENSIONS_MAP:
+ if extension[1:] not in EXTENSIONS_MAP:
raise ValueError(f"Unsupported media extension: {extension} in: {media}")
- return f".{extension}"
+ return extension
def ensure_images_dir():
"""Create images directory if it doesn't exist"""
@@ -43,19 +46,6 @@ def get_source_url(image: str, default: str = None) -> str:
return decoded_url
return default
-def secure_filename(filename: str) -> str:
- if filename is None:
- return None
- # Keep letters, numbers, basic punctuation and all Unicode chars
- filename = re.sub(
- r'[^\w.,_-]+',
- '_',
- unquote(filename).strip(),
- flags=re.UNICODE
- )
- filename = filename[:100].strip(".,_-")
- return filename
-
def is_valid_media_type(content_type: str) -> bool:
return content_type in MEDIA_TYPE_MAP or content_type.startswith("audio/") or content_type.startswith("video/")
diff --git a/g4f/providers/asyncio.py b/g4f/providers/asyncio.py
index 5bcdb212..9dc59a13 100644
--- a/g4f/providers/asyncio.py
+++ b/g4f/providers/asyncio.py
@@ -72,9 +72,8 @@ async def to_async_iterator(iterator) -> AsyncIterator:
if hasattr(iterator, '__aiter__'):
async for item in iterator:
yield item
- return
- try:
+ elif asyncio.iscoroutine(iterator):
+ yield await iterator
+ else:
for item in iterator:
- yield item
- except TypeError:
- yield await iterator
\ No newline at end of file
+ yield item
\ No newline at end of file
diff --git a/g4f/providers/helper.py b/g4f/providers/helper.py
index 6c93d703..df7add0d 100644
--- a/g4f/providers/helper.py
+++ b/g4f/providers/helper.py
@@ -6,6 +6,15 @@ import string
from ..typing import Messages, Cookies, AsyncIterator, Iterator
from .. import debug
+def to_string(value) -> str:
+ if isinstance(value, str):
+ return value
+ elif isinstance(value, dict):
+ return value.get("text")
+ elif isinstance(value, list):
+ return "".join([to_string(v) for v in value if v.get("type") == "text"])
+ return str(value)
+
def format_prompt(messages: Messages, add_special_tokens: bool = False, do_continue: bool = False, include_system: bool = True) -> str:
"""
Format a series of messages into a single string, optionally adding special tokens.
@@ -18,11 +27,16 @@ def format_prompt(messages: Messages, add_special_tokens: bool = False, do_conti
str: A formatted string containing all messages.
"""
if not add_special_tokens and len(messages) <= 1:
- return messages[0]["content"]
- formatted = "\n".join([
- f'{message["role"].capitalize()}: {message["content"]}'
+ return to_string(messages[0]["content"])
+ messages = [
+ (message["role"], to_string(message["content"]))
for message in messages
- if include_system or message["role"] != "system"
+ if include_system or message.get("role") != "system"
+ ]
+ formatted = "\n".join([
+ f'{role.capitalize()}: {content}'
+ for role, content in messages
+ if content.strip()
])
if do_continue:
return formatted
@@ -34,11 +48,13 @@ def get_system_prompt(messages: Messages) -> str:
def get_last_user_message(messages: Messages) -> str:
user_messages = []
last_message = None if len(messages) == 0 else messages[-1]
+ messages = messages.copy()
while last_message is not None and messages:
last_message = messages.pop()
if last_message["role"] == "user":
- if isinstance(last_message["content"], str):
- user_messages.append(last_message["content"].strip())
+ content = to_string(last_message["content"]).strip()
+ if content:
+ user_messages.append(content)
else:
return "\n".join(user_messages[::-1])
return "\n".join(user_messages[::-1])
diff --git a/g4f/tools/files.py b/g4f/tools/files.py
index 36491ea8..368564d9 100644
--- a/g4f/tools/files.py
+++ b/g4f/tools/files.py
@@ -1,22 +1,19 @@
from __future__ import annotations
+import re
import os
import json
from pathlib import Path
from typing import Iterator, Optional, AsyncIterator
from aiohttp import ClientSession, ClientError, ClientResponse, ClientTimeout
import urllib.parse
+from urllib.parse import unquote
import time
import zipfile
import asyncio
import hashlib
import base64
-try:
- from werkzeug.utils import secure_filename
-except ImportError:
- secure_filename = os.path.basename
-
try:
import PyPDF2
from PyPDF2.errors import PdfReadError
@@ -83,6 +80,19 @@ PLAIN_CACHE = "plain.cache"
DOWNLOADS_FILE = "downloads.json"
FILE_LIST = "files.txt"
+def secure_filename(filename: str) -> str:
+ if filename is None:
+ return None
+ # Keep letters, numbers, basic punctuation and all Unicode chars
+ filename = re.sub(
+ r'[^\w.,_-]+',
+ '_',
+ unquote(filename).strip(),
+ flags=re.UNICODE
+ )
+ filename = filename[:100].strip(".,_-")
+ return filename
+
def supports_filename(filename: str):
if filename.endswith(".pdf"):
if has_pypdf2:
@@ -118,10 +128,8 @@ def supports_filename(filename: str):
return True
return False
-def get_bucket_dir(bucket_id: str, dirname: str = None):
- if dirname is None:
- return os.path.join(get_cookies_dir(), "buckets", bucket_id)
- return os.path.join(get_cookies_dir(), "buckets", dirname, bucket_id)
+def get_bucket_dir(*parts):
+ return os.path.join(get_cookies_dir(), "buckets", *[secure_filename(part) for part in parts if part])
def get_buckets():
buckets_dir = os.path.join(get_cookies_dir(), "buckets")
diff --git a/g4f/tools/media.py b/g4f/tools/media.py
new file mode 100644
index 00000000..20819f3c
--- /dev/null
+++ b/g4f/tools/media.py
@@ -0,0 +1,82 @@
+from __future__ import annotations
+
+import os
+import base64
+from typing import Iterator, Union
+from pathlib import Path
+
+from ..typing import Messages
+from ..image import is_data_an_media, is_data_an_audio, to_input_audio, to_data_uri
+from .files import get_bucket_dir
+
+def render_media(bucket_id: str, name: str, url: str, as_path: bool = False, as_base64: bool = False) -> Union[str, Path]:
+ if (not as_base64 or url.startswith("/")):
+ file = Path(get_bucket_dir(bucket_id, "media", name))
+ if as_path:
+ return file
+ data = file.read_bytes()
+ data_base64 = base64.b64encode(data).decode()
+ if as_base64:
+ return data_base64
+ return f"data:{is_data_an_media(data, name)};base64,{data_base64}"
+
+def render_part(part: dict) -> dict:
+ if "type" in part:
+ return part
+ filename = part.get("name")
+ if filename.endswith(".wav") or filename.endswith(".mp3"):
+ return {
+ "type": "input_audio",
+ "input_audio": {
+ "data": render_media(**part, as_base64=True),
+ "format": "wav" if filename.endswith(".wav") else "mp3"
+ }
+ }
+ return {
+ "type": "image_url",
+ "image_url": {"url": render_media(**part)}
+ }
+
+def merge_media(media: list, messages: list) -> Iterator:
+ buffer = []
+ for message in messages:
+ if message.get("role") == "user":
+ content = message.get("content")
+ if isinstance(content, list):
+ for part in content:
+ if "type" not in part:
+ path = render_media(**part, as_path=True)
+ buffer.append((path, os.path.basename(path)))
+ elif part.get("type") == "image_url":
+ buffer.append((part.get("image_url"), None))
+ else:
+ buffer = []
+ yield from buffer
+ if media is not None:
+ yield from media
+
+def render_messages(messages: Messages, media: list = None) -> Iterator:
+ for idx, message in enumerate(messages):
+ if isinstance(message["content"], list):
+ yield {
+ **message,
+ "content": [render_part(part) for part in message["content"] if part]
+ }
+ else:
+ if media is not None and idx == len(messages) - 1:
+ yield {
+ **message,
+ "content": [
+ {
+ "type": "input_audio",
+ "input_audio": to_input_audio(media_data, filename)
+ }
+ if is_data_an_audio(media_data, filename) else {
+ "type": "image_url",
+ "image_url": {"url": to_data_uri(media_data)}
+ }
+ for media_data, filename in media
+ ] + ([{"type": "text", "text": message["content"]}] if isinstance(message["content"], str) else [])
+ }
+ else:
+ yield message
\ No newline at end of file