mirror of
https://github.com/NicolasBohn/NexQuant.git
synced 2026-07-28 16:07:46 +00:00
c6833b0858
Co-authored-by: xuyang1 <xuyang1@microsoft.com>
28 lines
654 B
Bash
28 lines
654 B
Bash
# Global configs:
|
|
USE_AZURE=True
|
|
MAX_RETRY=10
|
|
RETRY_WAIT_SECONDS=20
|
|
DUMP_CHAT_CACHE=True
|
|
USE_CHAT_CACHE=True
|
|
DUMP_EMBEDDING_CACHE=True
|
|
USE_EMBEDDING_CACHE=True
|
|
LOG_LLM_CHAT_CONTENT=False
|
|
CHAT_FREQUENCY_PENALTY=0.0
|
|
CHAT_PRESENCE_PENALTY=0.0
|
|
|
|
# embedding model configs:
|
|
EMBEDDING_OPENAI_API_KEY=your_api_key
|
|
EMBEDDING_AZURE_API_BASE=your_api_base
|
|
EMBEDDING_AZURE_API_VERSION=your_api_version
|
|
EMBEDDING_MODEL=text-embedding-3-small
|
|
|
|
|
|
# chat model configs:
|
|
CHAT_OPENAI_API_KEY=your_api_key # 5c
|
|
CHAT_AZURE_API_BASE=your_api_base
|
|
CHAT_AZURE_API_VERSION=your_api_version
|
|
CHAT_MODEL=your_model_version
|
|
CHAT_MAX_TOKENS=3000
|
|
CHAT_TEMPERATURE=0.7
|
|
CHAT_STREAM=True
|