|
from application.llm.base import BaseLLM |
|
import json |
|
import requests |
|
|
|
class DocsGPTAPILLM(BaseLLM): |
|
|
|
def __init__(self, *args, **kwargs): |
|
self.endpoint = "https://llm.docsgpt.co.uk" |
|
|
|
|
|
def gen(self, model, engine, messages, stream=False, **kwargs): |
|
context = messages[0]['content'] |
|
user_question = messages[-1]['content'] |
|
prompt = f"### Instruction \n {user_question} \n ### Context \n {context} \n ### Answer \n" |
|
|
|
response = requests.post( |
|
f"{self.endpoint}/answer", |
|
json={ |
|
"prompt": prompt, |
|
"max_new_tokens": 30 |
|
} |
|
) |
|
response_clean = response.json()['a'].split("###")[0] |
|
|
|
return response_clean |
|
|
|
def gen_stream(self, model, engine, messages, stream=True, **kwargs): |
|
context = messages[0]['content'] |
|
user_question = messages[-1]['content'] |
|
prompt = f"### Instruction \n {user_question} \n ### Context \n {context} \n ### Answer \n" |
|
|
|
|
|
response = requests.post( |
|
f"{self.endpoint}/stream", |
|
json={ |
|
"prompt": prompt, |
|
"max_new_tokens": 256 |
|
}, |
|
stream=True |
|
) |
|
|
|
for line in response.iter_lines(): |
|
if line: |
|
|
|
data_str = line.decode('utf-8') |
|
if data_str.startswith("data: "): |
|
data = json.loads(data_str[6:]) |
|
yield data['a'] |
|
|