The traceback points to the LlamaIndex/Hugging Face integration importing huggingface_hub.inference._types. Changing the quantization configuration alone would not address that import.
An archival LlamaIndex v0.10.0 integration file contains the exact line:
from huggingface_hub.inference._types import ConversationalOutput
We checked the published huggingface-hub==0.27.0 wheel against its pinned v0.27.0 source (huggingface/huggingface_hub commit 6bfa5dd44d0472909dc778d30fbaac92e84ff96e, under src/). Their complete Python-file layouts match, and that module path is absent from both. For these selected snapshots, the integration expects an unavailable module; the wheel hasn’t simply omitted a Python file present in the matching source.
This is an older question, and the installed versions weren’t supplied. These archival snapshots don’t establish your original environment or a working replacement combination. This reply was prepared with AI assistance; we used our Yumi import-layout checker for the static comparison, without installing or executing package code.
To choose a repair, first capture versions from the notebook’s actual Python environment:
import sys
import subprocess
subprocess.run([
sys.executable, "-m", "pip", "show",
"huggingface-hub", "llama-index-llms-huggingface", "llama-index-core"
])
subprocess.run([sys.executable, "-m", "pip", "check"])
Use the integration’s dependency constraints and documentation to select compatible integration/Hub versions, restart the kernel after installation changes, and test the import before the model. A clean dependency check alone doesn’t prove the import works, and this traceback doesn’t justify a blanket Hub downgrade.