Implementation:Hpcaitech ColossalAI PanGuLLM

Knowledge Sources	Hpcaitech_ColossalAI
Domains	NLP, LLM Integration, Cloud Services
Last Updated	2026-02-09 00:00 GMT

Overview

Pangu (PanGu LLM) is a LangChain-compatible LLM wrapper that integrates with Huawei's PanGu large language models hosted on the Huawei Cloud platform via REST API.

Description

This class extends LangChain's LLM base class to provide access to Huawei's PanGu models. It authenticates through Huawei Cloud's IAM token system using username, password, and domain credentials, and communicates with the model endpoint via HTTPS. The class supports both text completion and chat model interfaces, with configurable generation parameters including max_tokens, temperature, and top-p sampling. Stop-word truncation is applied to the generated text before returning.

Usage

Use Pangu when you need to integrate a Huawei PanGu model into a LangChain-based ColossalQA pipeline. You must set the URL, USERNAME, PASSWORD, and DOMAIN_NAME environment variables or pass them explicitly through set_auth_config.

Code Reference

Source Location

• Repository: Hpcaitech_ColossalAI
• File: applications/ColossalQA/colossalqa/local/pangu_llm.py
• Lines: 1-150

Signature

class Pangu(LLM):
    n: int
    gen_config: dict = None
    auth_config: dict = None

    def __init__(self, gen_config=None, **kwargs):
        ...

    def _call(self, prompt: str, stop: Optional[List[str]] = None, **kwargs) -> str:
        ...

    def set_auth_config(self, **kwargs):
        ...

    def get_latest_auth_token(self, region, username, password, domain_name):
        ...

    def text_completion(self, text, gen_config, auth_config):
        ...

    def chat_model(self, messages, gen_config, auth_config):
        ...

Import

from colossalqa.local.pangu_llm import Pangu

I/O Contract

Inputs

Name	Type	Required	Description
n	int	Yes	Instance identifier parameter
gen_config	dict	No	Generation configuration dictionary (default: {"user": "User", "max_tokens": 50, "temperature": 0.95, "n": 1})
prompt	str	Yes	The text prompt to send to the PanGu model (used in _call)
stop	Optional[List[str]]	No	List of stop words to truncate the generated response
messages	list	Yes (for chat_model)	List of message dictionaries for the chat model interface

Outputs

Name	Type	Description
_call return	str	The generated text from the PanGu model, optionally truncated at stop words
text_completion return	dict	Raw API response dictionary containing choices with generated text
chat_model return	dict	Raw API response dictionary from the chat endpoint
get_latest_auth_token return	str	IAM authentication token (X-Subject-Token) for Huawei Cloud API

Environment Variables

Variable	Description
URL	The full endpoint URL for the PanGu model service
USERNAME	IAM username for Huawei Cloud authentication
PASSWORD	IAM user password for Huawei Cloud authentication
DOMAIN_NAME	The Huawei Cloud account domain name (parent account managing IAM users)

Usage Examples

import os
from colossalqa.local.pangu_llm import Pangu

# Set Huawei Cloud credentials
os.environ["URL"] = "https://pangu.cn-north-4.myhuaweicloud.com/v1/completions"
os.environ["USERNAME"] = "my_iam_user"
os.environ["PASSWORD"] = "my_password"
os.environ["DOMAIN_NAME"] = "my_domain"

# Create and configure the PanGu LLM
pg = Pangu(n=1)
pg.set_auth_config()

# Text completion
response = pg("What is deep learning?")
print(response)

# Chat model usage
messages = [{"role": "user", "content": "Explain transformers."}]
chat_response = pg.chat_model(
    messages=messages,
    gen_config=pg.gen_config,
    auth_config=pg.auth_config,
)

Related Pages

• Environment:Hpcaitech_ColossalAI_ColossalQA_RAG_Environment