CoolFace
Apppublic

DawitShibabaw/CBE-pms

sourceHugging Faceupdated 3mo agoView on Hugging Face
0likes
App README

CBE Performance Management System

AI-powered performance management system for Commercial Bank of Ethiopia.

Runs Llama 3.2 3B Instruct locally from DawitShibabaw/llama-3.2-3b-instruct using GPU acceleration (ZeroGPU on HF Spaces).

Features

  • —📊 Dashboard with KPI overview
  • —📤 Data ingestion (BSC, JD, LOS documents)
  • —🎯 AI-generated SMART objectives (3-step LLM pipeline)
  • —📈 Objective tracking & progress monitoring
  • —⭐ Performance appraisal management
  • —💬 Feedback & coaching records

Setup

Required secret

VariableDescription
HF_TOKENYour HuggingFace token — set in Space → Settings → Variables and secrets. Used to download the model from your private repo.

Hardware

This Space requires GPU to run efficiently. On HuggingFace Spaces:

  1. 1.Go to Settings → Hardware
  2. 2.Select ZeroGPU or a persistent GPU tier (T4, A10G, A100)
  3. 3.Save and restart the Space

The @spaces.GPU decorator automatically allocates GPU for generation requests and releases it when idle.

Optional overrides

VariableDescription
PMS_LLM_BACKENDOverride backend: local_transformers (default), groq, hf_inference, ollama, vllm
PMS_LOCAL_MODEL_IDOverride the model repo (default: DawitShibabaw/llama-3.2-3b-instruct)
PMS_LOCAL_MODEL_DEVICEOverride device (default: auto = GPU if available, else CPU)
PMS_LOCAL_MODEL_DTYPEOverride dtype (default: bfloat16 for GPU; use float32 for CPU)
GROQ_API_KEYOptional: use Groq's cloud API instead (PMS_LLM_BACKEND=groq)

Running locally

bash
# Start the backend
cd Back_End
uvicorn scripts.API.main:app --reload

# In another terminal, start the UI
python gradio_app.py