summaryrefslogtreecommitdiffstats
path: root/g4f
diff options
context:
space:
mode:
Diffstat (limited to 'g4f')
-rw-r--r--g4f/Provider/LiteIcoding.py97
-rw-r--r--g4f/models.py102
2 files changed, 143 insertions, 56 deletions
diff --git a/g4f/Provider/LiteIcoding.py b/g4f/Provider/LiteIcoding.py
new file mode 100644
index 00000000..6aa407ca
--- /dev/null
+++ b/g4f/Provider/LiteIcoding.py
@@ -0,0 +1,97 @@
+from __future__ import annotations
+
+from aiohttp import ClientSession, ClientResponseError
+from ..typing import AsyncResult, Messages
+from .base_provider import AsyncGeneratorProvider, ProviderModelMixin
+from .helper import format_prompt
+
+
+class LiteIcoding(AsyncGeneratorProvider, ProviderModelMixin):
+ url = "https://lite.icoding.ink"
+ api_endpoint = "/api/v1/gpt/message"
+ working = True
+ supports_gpt_4 = True
+ default_model = "gpt-4o"
+ models = [
+ 'gpt-4o',
+ 'gpt-4-turbo',
+ 'claude-3',
+ 'claude-3.5',
+ 'gemini-1.5',
+ ]
+
+ @classmethod
+ async def create_async_generator(
+ cls,
+ model: str,
+ messages: Messages,
+ proxy: str = None,
+ **kwargs
+ ) -> AsyncResult:
+ headers = {
+ "Accept": "*/*",
+ "Accept-Language": "en-US,en;q=0.9",
+ "Authorization": "Bearer null",
+ "Connection": "keep-alive",
+ "Content-Type": "application/json;charset=utf-8",
+ "DNT": "1",
+ "Origin": cls.url,
+ "Referer": f"{cls.url}/",
+ "Sec-Fetch-Dest": "empty",
+ "Sec-Fetch-Mode": "cors",
+ "Sec-Fetch-Site": "same-origin",
+ "User-Agent": (
+ "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) "
+ "Chrome/126.0.0.0 Safari/537.36"
+ ),
+ "sec-ch-ua": '"Not/A)Brand";v="8", "Chromium";v="126"',
+ "sec-ch-ua-mobile": "?0",
+ "sec-ch-ua-platform": '"Linux"',
+ }
+
+ data = {
+ "model": model,
+ "chatId": "-1",
+ "messages": [
+ {
+ "role": msg["role"],
+ "content": msg["content"],
+ "time": msg.get("time", ""),
+ "attachments": msg.get("attachments", []),
+ }
+ for msg in messages
+ ],
+ "plugins": [],
+ "systemPrompt": "",
+ "temperature": 0.5,
+ }
+
+ async with ClientSession(headers=headers) as session:
+ try:
+ async with session.post(
+ f"{cls.url}{cls.api_endpoint}", json=data, proxy=proxy
+ ) as response:
+ response.raise_for_status()
+ buffer = ""
+ full_response = ""
+ async for chunk in response.content.iter_any():
+ if chunk:
+ buffer += chunk.decode()
+ while "\n\n" in buffer:
+ part, buffer = buffer.split("\n\n", 1)
+ if part.startswith("data: "):
+ content = part[6:].strip()
+ if content and content != "[DONE]":
+ content = content.strip('"')
+ full_response += content
+
+ full_response = full_response.replace('" "', ' ')
+ yield full_response.strip()
+
+ except ClientResponseError as e:
+ raise RuntimeError(
+ f"ClientResponseError {e.status}: {e.message}, url={e.request_info.url}, data={data}"
+ ) from e
+
+ except Exception as e:
+ raise RuntimeError(f"Unexpected error: {str(e)}") from e
diff --git a/g4f/models.py b/g4f/models.py
index bcc50b9f..c963115a 100644
--- a/g4f/models.py
+++ b/g4f/models.py
@@ -25,6 +25,7 @@ from .Provider import (
HuggingFace,
Koala,
Liaobots,
+ LiteIcoding,
Marsyoo,
MetaAI,
OpenaiChat,
@@ -114,24 +115,6 @@ gpt_35_turbo = Model(
])
)
-gpt_35_turbo_16k = Model(
- name = 'gpt-3.5-turbo-16k',
- base_provider = 'openai',
- best_provider = gpt_35_long.best_provider
-)
-
-gpt_35_turbo_16k_0613 = Model(
- name = 'gpt-3.5-turbo-16k-0613',
- base_provider = 'openai',
- best_provider = gpt_35_long.best_provider
-)
-
-gpt_35_turbo_0613 = Model(
- name = 'gpt-3.5-turbo-0613',
- base_provider = 'openai',
- best_provider = gpt_35_turbo.best_provider
-)
-
# gpt-4
gpt_4 = Model(
name = 'gpt-4',
@@ -141,36 +124,18 @@ gpt_4 = Model(
])
)
-gpt_4_0613 = Model(
- name = 'gpt-4-0613',
- base_provider = 'openai',
- best_provider = gpt_4.best_provider
-)
-
-gpt_4_32k = Model(
- name = 'gpt-4-32k',
- base_provider = 'openai',
- best_provider = gpt_4.best_provider
-)
-
-gpt_4_32k_0613 = Model(
- name = 'gpt-4-32k-0613',
- base_provider = 'openai',
- best_provider = gpt_4.best_provider
-)
-
gpt_4_turbo = Model(
name = 'gpt-4-turbo',
base_provider = 'openai',
best_provider = IterListProvider([
- Bing, Liaobots
+ Bing, Liaobots, LiteIcoding
])
)
gpt_4o = Model(
name = 'gpt-4o',
base_provider = 'openai',
best_provider = IterListProvider([
- You, Liaobots, Chatgpt4o, AI365VIP, OpenaiChat, Marsyoo
+ You, Liaobots, Chatgpt4o, AI365VIP, OpenaiChat, Marsyoo, LiteIcoding
])
)
@@ -210,7 +175,7 @@ llama_3_70b_instruct = Model(
best_provider = IterListProvider([DeepInfra, PerplexityLabs, Replicate])
)
-llama3_70b_instruct = Model(
+llama_3_70b_instruct = Model(
name = "meta/meta-llama-3-70b-instruct",
base_provider = "meta",
best_provider = IterListProvider([ReplicateHome])
@@ -222,13 +187,13 @@ llama_3_70b_chat_hf = Model(
best_provider = IterListProvider([DDG])
)
-llama_3_1_70b_Instruct = Model(
+llama_3_1_70b_instruct = Model(
name = "meta-llama/Meta-Llama-3.1-70B-Instruct",
base_provider = "meta",
best_provider = IterListProvider([HuggingChat, HuggingFace])
)
-llama_3_1_405b_Instruct_FP8 = Model(
+llama_3_1_405b_instruct_FP8 = Model(
name = "meta-llama/Meta-Llama-3.1-405B-Instruct-FP8",
base_provider = "meta",
best_provider = IterListProvider([HuggingChat, HuggingFace])
@@ -258,7 +223,7 @@ Nous_Hermes_2_Mixtral_8x7B_DPO = Model(
### 01-ai ###
-Yi_1_5_34B_Chat = Model(
+Yi_1_5_34B_chat = Model(
name = "01-ai/Yi-1.5-34B-Chat",
base_provider = "01-ai",
best_provider = IterListProvider([HuggingFace, HuggingChat])
@@ -284,7 +249,7 @@ gemini = Model(
gemini_pro = Model(
name = 'gemini-pro',
base_provider = 'Google',
- best_provider = IterListProvider([GeminiPro, You, ChatGot, GeminiProChat, Liaobots])
+ best_provider = IterListProvider([GeminiPro, You, ChatGot, GeminiProChat, Liaobots, LiteIcoding])
)
gemini_flash = Model(
@@ -293,6 +258,12 @@ gemini_flash = Model(
best_provider = IterListProvider([Liaobots])
)
+gemini_1_5 = Model(
+ name = 'gemini-1.5',
+ base_provider = 'Google',
+ best_provider = IterListProvider([LiteIcoding])
+)
+
# gemma
gemma_2b_it = Model(
name = 'gemma-2b-it',
@@ -356,6 +327,18 @@ claude_3_haiku = Model(
best_provider = IterListProvider([DDG, AI365VIP, Liaobots])
)
+claude_3 = Model(
+ name = 'claude-3',
+ base_provider = 'Anthropic',
+ best_provider = IterListProvider([LiteIcoding])
+)
+
+claude_3_5 = Model(
+ name = 'claude-3.5',
+ base_provider = 'Anthropic',
+ best_provider = IterListProvider([LiteIcoding])
+)
+
### Reka AI ###
reka_core = Model(
@@ -420,7 +403,7 @@ deepseek_chat = Model(
### Qwen ###
-Qwen2_7B_Instruct = Model(
+Qwen2_7B_instruct = Model(
name = 'Qwen2-7B-Instruct',
base_provider = 'Qwen',
best_provider = IterListProvider([FreeChatgpt])
@@ -442,7 +425,7 @@ chatglm3_6B = Model(
### 01-ai ###
-Yi_1_5_9B_Chat = Model(
+Yi_1_5_9B_chat = Model(
name = 'Yi-1.5-9B-Chat',
base_provider = '01-ai',
best_provider = IterListProvider([FreeChatgpt])
@@ -518,12 +501,12 @@ class ModelUtils:
### Meta ###
"meta-ai": meta,
- 'llama-3-8b': llama_3_8b_instruct,
- 'llama-3-70b': llama_3_70b_instruct,
+ 'llama-3-8b-instruct': llama_3_8b_instruct,
+ 'llama-3-70b-instruct': llama_3_70b_instruct,
'llama-3-70b-chat': llama_3_70b_chat_hf,
- 'llama-3-70b-instruct': llama3_70b_instruct,
- 'llama-3.1-70b': llama_3_1_70b_Instruct,
- 'llama-3.1-405b': llama_3_1_405b_Instruct_FP8,
+ 'llama-3-70b-instruct': llama_3_70b_instruct,
+ 'llama-3.1-70b-instruct': llama_3_1_70b_instruct,
+ 'llama-3.1-405b-instruct': llama_3_1_405b_instruct_FP8,
### Mistral (Opensource) ###
'mixtral-8x7b': mixtral_8x7b,
@@ -533,15 +516,16 @@ class ModelUtils:
'Nous-Hermes-2-Mixtral-8x7B-DPO': Nous_Hermes_2_Mixtral_8x7B_DPO,
### 01-ai ###
- 'Yi-1.5-34b': Yi_1_5_34B_Chat,
+ 'Yi-1.5-34b-chat': Yi_1_5_34B_chat,
### Microsoft ###
- 'Phi-3-mini-4k': Phi_3_mini_4k_instruct,
+ 'Phi-3-mini-4k-instruct': Phi_3_mini_4k_instruct,
### Google ###
# gemini
'gemini': gemini,
'gemini-pro': gemini_pro,
+ 'gemini-pro': gemini_1_5,
'gemini-flash': gemini_flash,
# gemma
@@ -553,16 +537,22 @@ class ModelUtils:
'claude-2': claude_2,
'claude-2.0': claude_2_0,
'claude-2.1': claude_2_1,
+
'claude-3-opus': claude_3_opus,
'claude-3-sonnet': claude_3_sonnet,
'claude-3-5-sonnet': claude_3_5_sonnet,
'claude-3-haiku': claude_3_haiku,
+
+ 'claude-3-opus': claude_3,
+ 'claude-3-5-sonnet': claude_3_5,
+
+
### Reka AI ###
'reka': reka_core,
### NVIDIA ###
- 'nemotron-4-340b': nemotron_4_340b_instruct,
+ 'nemotron-4-340b-instruct': nemotron_4_340b_instruct,
### Blackbox ###
'blackbox': blackbox,
@@ -584,14 +574,14 @@ class ModelUtils:
'deepseek-chat': deepseek_chat,
### Qwen ###
- 'Qwen2-7b': Qwen2_7B_Instruct,
+ 'Qwen2-7b-instruct': Qwen2_7B_instruct,
### Zhipu AI ###
- 'glm4-9b': glm4_9B_chat,
+ 'glm4-9b-chat': glm4_9B_chat,
'chatglm3-6b': chatglm3_6B,
### 01-ai ###
- 'Yi-1.5-9b': Yi_1_5_9B_Chat,
+ 'Yi-1.5-9b-chat': Yi_1_5_9B_chat,
# Other
'pi': pi,