Modelpublic

cstr/bert-base-NER-GGUF

sourceHugging Facemitupdated 2mo agoView on Hugging Face

0likes529downloads

Model Card

BERT Base NER — GGUF

GGUF conversion of dslim/bert-base-NER for use with CrispEmbed.

Fixed-label Named Entity Recognition on English text. BERT-base-cased (110M params) fine-tuned on CoNLL-03 with 9 IOB labels.

Labels

ID	Label	Description
0	O	Outside any entity
1	B-MISC	Beginning of miscellaneous entity
2	I-MISC	Inside miscellaneous entity
3	B-PER	Beginning of person name
4	I-PER	Inside person name
5	B-ORG	Beginning of organization
6	I-ORG	Inside organization
7	B-LOC	Beginning of location
8	I-LOC	Inside location

Available Formats

File	Format	Size
`bert-base-ner-f32.gguf`	Float32	412 MB
`bert-base-ner-q8_0.gguf`	Q8_0	111 MB
`bert-base-ner-q4_k.gguf`	Q4_K	70 MB

Usage

bash

crispembed -m bert-base-ner-q8_0.gguf --ner "Barack Obama was born in Hawaii"

python

from crispembed import CrispNER
ner = CrispNER("bert-base-ner-q8_0.gguf")
entities = ner.extract("Barack Obama was born in Hawaii")
# [{"text": "Barack Obama", "label": "PER", "start": 0, "end": 12, "score": 0.999},
#  {"text": "Hawaii", "label": "LOC", "start": 25, "end": 31, "score": 1.000}]

Auto-detected as BERT NER (vs GLiNER zero-shot) from ner.classifier.weight in GGUF.

Parity

Encoder output: cos_min=0.999971 vs HuggingFace transformers (F32).

Provenance and EU AI Act Art. 53 note

—Upstream model: dslim/bert-base-NER — published by dslim.
—Upstream licence: mit. This repository redistributes under the same terms; it grants no rights the upstream licence does not.
—What was done here: format conversion and/or quantisation only (GGUF). No training, no fine-tuning, no merging, no distillation, no change to architecture, vocabulary or capability. Only the numeric representation of the upstream weights differs.
—Training data: documented — where it is documented at all — by the upstream provider; see the upstream model card. No training data was used, added or selected by this repository.
—Provider status: under Regulation (EU) 2024/1689 the upstream authors remain the provider of this model. Converting the serialisation format does not make this repository the provider of a new general-purpose AI model, and no such claim is made. Questions about training content, copyright policy or model capability belong upstream.