summaryrefslogtreecommitdiffstats
path: root/g4f/Provider/HuggingChat.py
diff options
context:
space:
mode:
Diffstat (limited to 'g4f/Provider/HuggingChat.py')
-rw-r--r--g4f/Provider/HuggingChat.py62
1 files changed, 16 insertions, 46 deletions
diff --git a/g4f/Provider/HuggingChat.py b/g4f/Provider/HuggingChat.py
index b2cf9793..f1cec775 100644
--- a/g4f/Provider/HuggingChat.py
+++ b/g4f/Provider/HuggingChat.py
@@ -1,6 +1,6 @@
from __future__ import annotations
-import json
+import json, uuid
from aiohttp import ClientSession
@@ -12,7 +12,7 @@ class HuggingChat(AsyncGeneratorProvider):
url = "https://huggingface.co/chat"
needs_auth = True
working = True
- model = "OpenAssistant/oasst-sft-6-llama-30b-xor"
+ model = "meta-llama/Llama-2-70b-chat-hf"
@classmethod
async def create_async_generator(
@@ -37,55 +37,25 @@ class HuggingChat(AsyncGeneratorProvider):
cookies=cookies,
headers=headers
) as session:
- async with session.post(f"{cls.url}/conversation", proxy=proxy, json={"model": model}) as response:
+ async with session.post(f"{cls.url}/conversation", json={"model": model}, proxy=proxy) as response:
conversation_id = (await response.json())["conversationId"]
send = {
+ "id": str(uuid.uuid4()),
"inputs": format_prompt(messages),
- "parameters": {
- "temperature": 0.2,
- "truncate": 1000,
- "max_new_tokens": 1024,
- "stop": ["</s>"],
- "top_p": 0.95,
- "repetition_penalty": 1.2,
- "top_k": 50,
- "return_full_text": False,
- **kwargs
- },
- "stream": stream,
- "options": {
- "id": "9e9b8bc4-6604-40c6-994e-8eb78fa32e37",
- "response_id": "04ce2602-3bea-45e8-8efc-cef00680376a",
- "is_retry": False,
- "use_cache": False,
- "web_search_id": ""
- }
+ "is_retry": False,
+ "response_id": str(uuid.uuid4()),
+ "web_search": False
}
- async with session.post(f"{cls.url}/conversation/{conversation_id}", proxy=proxy, json=send) as response:
- if not stream:
- data = await response.json()
- if "error" in data:
- raise RuntimeError(data["error"])
- elif isinstance(data, list):
- yield data[0]["generated_text"].strip()
- else:
- raise RuntimeError(f"Response: {data}")
- else:
- start = "data:"
- first = True
- async for line in response.content:
- line = line.decode("utf-8")
- if line.startswith(start):
- line = json.loads(line[len(start):-1])
- if "token" not in line:
- raise RuntimeError(f"Response: {line}")
- if not line["token"]["special"]:
- if first:
- yield line["token"]["text"].lstrip()
- first = False
- else:
- yield line["token"]["text"]
+ async with session.post(f"{cls.url}/conversation/{conversation_id}", json=send, proxy=proxy) as response:
+ async for line in response.content:
+ line = json.loads(line[:-1])
+ if "type" not in line:
+ raise RuntimeError(f"Response: {line}")
+ elif line["type"] == "stream":
+ yield line["token"]
+ elif line["type"] == "finalAnswer":
+ break
async with session.delete(f"{cls.url}/conversation/{conversation_id}", proxy=proxy) as response:
response.raise_for_status()