sedibalm_client.py

import os
from huggingface_hub import InferenceClient

MODEL_ID = "Sedibaai/SedibaLM"
HF_TOKEN = os.getenv("HF_TOKEN") or os.getenv("LEOTSA_HF_TOKEN")
_client = InferenceClient(model=MODEL_ID, token=HF_TOKEN or None)

def sedibalm_generate(user_text: str, max_new_tokens: int = 128) -> str:
    messages = [{"role": "user", "content": user_text}]
    try:
        return _client.chat_completion(
            messages=messages,
            max_new_tokens=max_new_tokens,
            temperature=0.7,
            top_p=0.8,
            repetition_penalty=1.3,
        ).choices[0].message.content.strip()
    except Exception as e:
        return f"[inference error] {type(e).__name__}: {e}"