You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
langchain/libs/community/langchain_community/llms/bittensor.py

175 lines
6.1 KiB
Python

import http.client
import json
import ssl
from typing import Any, List, Mapping, Optional
from langchain_core.callbacks import CallbackManagerForLLMRun
from langchain_core.language_models.llms import LLM
class NIBittensorLLM(LLM):
"""NIBittensor LLMs
NIBittensorLLM is created by Neural Internet (https://neuralinternet.ai/),
powered by Bittensor, a decentralized network full of different AI models.
To analyze API_KEYS and logs of your usage visit
https://api.neuralinternet.ai/api-keys
https://api.neuralinternet.ai/logs
Example:
.. code-block:: python
from langchain_community.llms import NIBittensorLLM
llm = NIBittensorLLM()
"""
system_prompt: Optional[str]
"""Provide system prompt that you want to supply it to model before every prompt"""
top_responses: Optional[int] = 0
"""Provide top_responses to get Top N miner responses on one request.May get delayed
Don't use in Production"""
@property
def _llm_type(self) -> str:
return "NIBittensorLLM"
def _call(
self,
prompt: str,
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> str:
"""
Wrapper around the bittensor top miner models. Its built by Neural Internet.
Call the Neural Internet's BTVEP Server and return the output.
Parameters (optional):
system_prompt(str): A system prompt defining how your model should respond.
top_responses(int): Total top miner responses to retrieve from Bittensor
protocol.
Return:
The generated response(s).
Example:
.. code-block:: python
from langchain_community.llms import NIBittensorLLM
llm = NIBittensorLLM(system_prompt="Act like you are programmer with \
5+ years of experience.")
"""
# Creating HTTPS connection with SSL
context = ssl.create_default_context()
context.check_hostname = True
conn = http.client.HTTPSConnection("test.neuralinternet.ai", context=context)
# Sanitizing User Input before passing to API.
if isinstance(self.top_responses, int):
top_n = min(100, self.top_responses)
else:
top_n = 0
default_prompt = "You are an assistant which is created by Neural Internet(NI) \
in decentralized network named as a Bittensor."
if self.system_prompt is None:
system_prompt = (
default_prompt
+ " Your task is to provide accurate response based on user prompt"
)
else:
system_prompt = default_prompt + str(self.system_prompt)
# Retrieving API KEY to pass into header of each request
conn.request("GET", "/admin/api-keys/")
api_key_response = conn.getresponse()
api_keys_data = (
api_key_response.read().decode("utf-8").replace("\n", "").replace("\t", "")
)
api_keys_json = json.loads(api_keys_data)
api_key = api_keys_json[0]["api_key"]
# Creating Header and getting top benchmark miner uids
headers = {
"Content-Type": "application/json",
"Authorization": f"Bearer {api_key}",
"Endpoint-Version": "2023-05-19",
}
conn.request("GET", "/top_miner_uids", headers=headers)
miner_response = conn.getresponse()
miner_data = (
miner_response.read().decode("utf-8").replace("\n", "").replace("\t", "")
)
uids = json.loads(miner_data)
# Condition for benchmark miner response
if isinstance(uids, list) and uids and not top_n:
for uid in uids:
try:
payload = json.dumps(
{
"uids": [uid],
"messages": [
{"role": "system", "content": system_prompt},
{"role": "user", "content": prompt},
],
}
)
conn.request("POST", "/chat", payload, headers)
init_response = conn.getresponse()
init_data = (
init_response.read()
.decode("utf-8")
.replace("\n", "")
.replace("\t", "")
)
init_json = json.loads(init_data)
if "choices" not in init_json:
continue
reply = init_json["choices"][0]["message"]["content"]
conn.close()
return reply
except Exception:
continue
# For top miner based on bittensor response
try:
payload = json.dumps(
{
"top_n": top_n,
"messages": [
{"role": "system", "content": system_prompt},
{"role": "user", "content": prompt},
],
}
)
conn.request("POST", "/chat", payload, headers)
response = conn.getresponse()
utf_string = (
response.read().decode("utf-8").replace("\n", "").replace("\t", "")
)
if top_n:
conn.close()
return utf_string
json_resp = json.loads(utf_string)
reply = json_resp["choices"][0]["message"]["content"]
conn.close()
return reply
except Exception as e:
conn.request("GET", f"/error_msg?e={e}&p={prompt}", headers=headers)
return "Sorry I am unable to provide response now, Please try again later."
@property
def _identifying_params(self) -> Mapping[str, Any]:
"""Get the identifying parameters."""
return {
"system_prompt": self.system_prompt,
"top_responses": self.top_responses,
}