summaryrefslogtreecommitdiffstats
path: root/g4f/Provider/Nexra.py
diff options
context:
space:
mode:
authorkqlio67 <kqlio67@users.noreply.github.com>2024-09-04 21:23:17 +0200
committerkqlio67 <kqlio67@users.noreply.github.com>2024-09-04 21:23:17 +0200
commit35b96f32d1671eb82f98cda751c46964954e9980 (patch)
tree833ad3e6fe43f5bc6e7b17a690dd4d527fc3b0ca /g4f/Provider/Nexra.py
parentfeat(image-processing): add support for data URI image handlin (diff)
downloadgpt4free-35b96f32d1671eb82f98cda751c46964954e9980.tar
gpt4free-35b96f32d1671eb82f98cda751c46964954e9980.tar.gz
gpt4free-35b96f32d1671eb82f98cda751c46964954e9980.tar.bz2
gpt4free-35b96f32d1671eb82f98cda751c46964954e9980.tar.lz
gpt4free-35b96f32d1671eb82f98cda751c46964954e9980.tar.xz
gpt4free-35b96f32d1671eb82f98cda751c46964954e9980.tar.zst
gpt4free-35b96f32d1671eb82f98cda751c46964954e9980.zip
Diffstat (limited to '')
-rw-r--r--g4f/Provider/Nexra.py130
1 files changed, 100 insertions, 30 deletions
diff --git a/g4f/Provider/Nexra.py b/g4f/Provider/Nexra.py
index 4914b930..e2c3e197 100644
--- a/g4f/Provider/Nexra.py
+++ b/g4f/Provider/Nexra.py
@@ -1,16 +1,19 @@
from __future__ import annotations
import json
+import base64
from aiohttp import ClientSession
+from typing import AsyncGenerator
from ..typing import AsyncResult, Messages
from .base_provider import AsyncGeneratorProvider, ProviderModelMixin
+from ..image import ImageResponse
from .helper import format_prompt
-
class Nexra(AsyncGeneratorProvider, ProviderModelMixin):
url = "https://nexra.aryahcr.cc"
- api_endpoint = "https://nexra.aryahcr.cc/api/chat/gpt"
+ api_endpoint_text = "https://nexra.aryahcr.cc/api/chat/gpt"
+ api_endpoint_image = "https://nexra.aryahcr.cc/api/image/complements"
working = True
supports_gpt_35_turbo = True
supports_gpt_4 = True
@@ -20,34 +23,19 @@ class Nexra(AsyncGeneratorProvider, ProviderModelMixin):
default_model = 'gpt-3.5-turbo'
models = [
- # Working with text
- 'gpt-4',
- 'gpt-4-0613',
- 'gpt-4-32k',
- 'gpt-4-0314',
- 'gpt-4-32k-0314',
-
- 'gpt-3.5-turbo',
- 'gpt-3.5-turbo-16k',
- 'gpt-3.5-turbo-0613',
- 'gpt-3.5-turbo-16k-0613',
- 'gpt-3.5-turbo-0301',
-
- 'gpt-3',
- 'text-davinci-003',
- 'text-davinci-002',
- 'code-davinci-002',
- 'text-curie-001',
- 'text-babbage-001',
- 'text-ada-001',
- 'davinci',
- 'curie',
- 'babbage',
- 'ada',
- 'babbage-002',
- 'davinci-002',
+ # Text models
+ 'gpt-4', 'gpt-4-0613', 'gpt-4-32k', 'gpt-4-0314', 'gpt-4-32k-0314',
+ 'gpt-3.5-turbo', 'gpt-3.5-turbo-16k', 'gpt-3.5-turbo-0613', 'gpt-3.5-turbo-16k-0613', 'gpt-3.5-turbo-0301',
+ 'gpt-3', 'text-davinci-003', 'text-davinci-002', 'code-davinci-002',
+ 'text-curie-001', 'text-babbage-001', 'text-ada-001',
+ 'davinci', 'curie', 'babbage', 'ada', 'babbage-002', 'davinci-002',
+ # Image models
+ 'dalle', 'dalle-mini', 'emi'
]
+ image_models = {"dalle", "dalle-mini", "emi"}
+ text_models = set(models) - image_models
+
model_aliases = {
"gpt-4": "gpt-4-0613",
"gpt-4": "gpt-4-32k",
@@ -90,9 +78,24 @@ class Nexra(AsyncGeneratorProvider, ProviderModelMixin):
messages: Messages,
proxy: str = None,
**kwargs
- ) -> AsyncResult:
+ ) -> AsyncGenerator[str | ImageResponse, None]:
model = cls.get_model(model)
+ if model in cls.image_models:
+ async for result in cls.create_image_async_generator(model, messages, proxy, **kwargs):
+ yield result
+ else:
+ async for result in cls.create_text_async_generator(model, messages, proxy, **kwargs):
+ yield result
+
+ @classmethod
+ async def create_text_async_generator(
+ cls,
+ model: str,
+ messages: Messages,
+ proxy: str = None,
+ **kwargs
+ ) -> AsyncGenerator[str, None]:
headers = {
"Content-Type": "application/json",
}
@@ -104,8 +107,75 @@ class Nexra(AsyncGeneratorProvider, ProviderModelMixin):
"markdown": False,
"stream": False,
}
- async with session.post(cls.api_endpoint, json=data, proxy=proxy) as response:
+ async with session.post(cls.api_endpoint_text, json=data, proxy=proxy) as response:
response.raise_for_status()
result = await response.text()
json_result = json.loads(result)
yield json_result["gpt"]
+
+ @classmethod
+ async def create_image_async_generator(
+ cls,
+ model: str,
+ messages: Messages,
+ proxy: str = None,
+ **kwargs
+ ) -> AsyncGenerator[ImageResponse | str, None]:
+ headers = {
+ "Content-Type": "application/json"
+ }
+
+ prompt = messages[-1]['content'] if messages else ""
+
+ data = {
+ "prompt": prompt,
+ "model": model
+ }
+
+ async def process_response(response_text: str) -> ImageResponse | None:
+ json_start = response_text.find('{')
+ if json_start != -1:
+ json_data = response_text[json_start:]
+ try:
+ response_data = json.loads(json_data)
+ image_data = response_data.get('images', [])[0]
+
+ if image_data.startswith('data:image/'):
+ return ImageResponse([image_data], "Generated image")
+
+ try:
+ base64.b64decode(image_data)
+ data_uri = f"data:image/jpeg;base64,{image_data}"
+ return ImageResponse([data_uri], "Generated image")
+ except:
+ print("Invalid base64 data")
+ return None
+ except json.JSONDecodeError:
+ print("Failed to parse JSON.")
+ else:
+ print("No JSON data found in the response.")
+ return None
+
+ async with ClientSession(headers=headers) as session:
+ async with session.post(cls.api_endpoint_image, json=data, proxy=proxy) as response:
+ response.raise_for_status()
+ response_text = await response.text()
+
+ image_response = await process_response(response_text)
+ if image_response:
+ yield image_response
+ else:
+ yield "Failed to process image data."
+
+ @classmethod
+ async def create_async(
+ cls,
+ model: str,
+ messages: Messages,
+ proxy: str = None,
+ **kwargs
+ ) -> str:
+ async for response in cls.create_async_generator(model, messages, proxy, **kwargs):
+ if isinstance(response, ImageResponse):
+ return response.images[0]
+ return response