goby-9-ie-vi

Joint entity + relation extraction for Vietnamese. The best-calibrated model in the goby family: higher precision at the same recall than its siblings, not just a different threshold.

Fine-tuned from knowledgator/gliner-relex-large-v1.0. Released by Belumind. Apache-2.0.

Try it in the browser - no install, runs on ZeroGPU.

Which goby do I want?

Your input	Use
Clean Vietnamese prose, you want the best accuracy	goby-9-ie-vi (this one)
OCR output, ALL CAPS, missing diacritics	goby-7-ie-vi
You need maximum relation recall	goby-4-ie-vi

goby-9 is not robust to degraded text - see Known limitations. That is not a small caveat, read it before choosing.

What changed

Same training data as goby-4. The only difference is the loss configuration: negative sampling ratio 1.0 -> 2.5, focal loss alpha 0.75 -> 0.5. No new data, no augmentation.

This matters because it moves the precision/recall curve rather than sliding along it:

at recall	goby-4 precision	goby-9 precision
~77	46.4	50.3
~74	53.8	57.9
~64	67.3	67.7

If the two models only differed by an operating point, precision at matched recall would be identical. It is not: +3.9 and +4.1 in the range most people run at.

Results

Benchmark: 450 held-out Vietnamese sentences, 1,328 hand-audited entity spans. All figures at threshold=0.5 unless stated.

Typed NER

	Precision	Recall	F1
goby-4-ie-vi	50.5	76.6	60.9
goby-9-ie-vi	57.9	73.3	64.7

Paired bootstrap, goby-9 minus goby-4: +3.8 F1, 95% CI [+2.7, +4.8], 100% of resamples positive.

Threshold sweep:

threshold	P	R	F1
0.3	45.9	77.9	57.8
0.4	50.3	76.7	60.7
0.5	57.9	73.3	64.7
0.6	67.7	64.4	66.0
0.7	80.8	52.0	63.3

It also fires far less: 1,672 predictions at 0.5 against goby-4 at 2,005, for higher recall-adjusted precision.

Per entity type (F1)

Type	goby-4	goby-9
ngay (date)	89.3	92.6
dia diem (location)	66.8	69.8
van ban (document)	66.7	68.8
to chuc (organisation)	55.4	59.6
nguoi (person)	53.2	57.1
san pham (product)	59.0	51.1
giai thuong (award)	42.3	45.5
su kien (event)	40.0	37.5
tac pham (work)	52.6	33.3
chuc vu (job title)	21.7	31.2

goby-9 wins on 7 of 10. It loses on tac pham (52.6 -> 33.3), san pham (59.0 -> 51.1) and su kien (40.0 -> 37.5) - the sharper decision boundary costs recall on the types with the fewest training examples.

Relations

	strict	partial
goby-4-ie-vi	40.7	57.7
goby-9-ie-vi	41.6	56.7

Relation quality is essentially unchanged (-1.0 partial, +0.9 strict).

Usage

pip install gliner==0.2.29

from gliner import GLiNER

model = GLiNER.from_pretrained("belumind/goby-9-ie-vi").to("cuda")

KEYS = {
    "to_chuc": "tổ chức",   "van_ban": "văn bản",
    "nguoi": "người",       "dia_diem": "địa điểm",
    "ngay": "ngày",         "chuc_vu": "chức vụ",
}
INV = {v: k for k, v in KEYS.items()}

ents, rels = model.inference(
    texts=["Công ty Cổ phần Giải trí Galaxy được thành lập ngày 13 / 9 / 2003 ."],
    labels=list(KEYS.values()),
    relations=["thành lập ngày", "đặt trụ sở tại"],
    threshold=0.5, relation_threshold=0.5,
    return_relations=True, flat_ner=True,
)
out = [{"text": x["text"], "type": INV[x["label"]]} for x in ents[0]]

Pass the Vietnamese labels with diacritics. GLiNER encodes the label string with its text encoder, so to_chuc is effectively an unseen label: it costs about 25 F1. Map to snake_case keys after inference, as above.

Labels must keep their Vietnamese diacritics

GLiNER encodes the label STRING with the same text encoder as the input, so a de-accented label is a different vector. Measured on goby-7, threshold 0.5:

labels passed to the model	NER F1
Vietnamese, with diacritics	57.2
ASCII with spaces (to chuc)	36.1
snake_case ASCII (to_chuc)	32.3

About 25 F1, lost with no error and no warning. This repo ships goby_labels.py so the mistake fails loudly instead of silently:

import os, sys
from huggingface_hub import hf_hub_download

sys.path.insert(0, os.path.dirname(hf_hub_download("REPO_ID", "goby_labels.py")))
from goby_labels import extract, check_labels, LabelError

ents, rels = extract(model, [text], json_keys=True)

extract() calls the model with the Vietnamese labels and renames the types to to_chuc, van_ban, ... only in the RETURNED dicts - free, and JSON-safe keys. Passing a de-accented label raises LabelError; an unknown ASCII label warns. check_labels(labels) is the bare assertion if you keep your own inference call.

Weights ship as both model.safetensors (what gliner loads by preference) and pytorch_model.bin; the two are byte-identical, tensor for tensor.

Known limitations

• Degraded text destroys this model. Relation F1 on the same 267 sentences: clean 56.7, without diacritics 6.0, ALL CAPS 6.8. goby-7 scores 28.5 and 30.0 on the same inputs. If your text is not clean and correctly cased, use goby-7, not this.
• chuc vu (job titles): do not trust the 31.2 above. On a separate benchmark of 55 hand-annotated, correctly-delimited job titles, goby-9 gets only 5 exactly right (F1 13.3) - and the base model beats every model in this family there (13 exact, F1 18.3). Our fine-tuning data carries a systematic truncation of Vietnamese job titles, and every goby model inherited it. We publish this because the main benchmark hides it.
• tac pham regressed against goby-4 (52.6 -> 33.3).
• ky (signed document) and ke nhiem (successor of) remain the weakest relations.
• Trained on Vietnamese Wikipedia-style prose. Legal, medical and conversational text are out of distribution.

Citation

@misc{goby9ievi2026,
  title  = {goby-9-ie-vi: a recalibrated joint IE model for Vietnamese},
  author = {Belumind},
  year   = {2026},
  url    = {https://huggingface.co/belumind/goby-9-ie-vi}
}