Bloomz-3b Open-Source Multilingual Generation Model - Supports Text Generation in Multiple Languages and Programming Languages

Bloomz 3b

Developed by bigscience

BLOOMZ-3B1 is a multilingual generative model from the BLOOM series, supporting 46 languages and 13 programming languages, excelling in cross-lingual text generation tasks.

Large Language Model

Transformers

Supports Multiple LanguagesOpen Source License:Openrail #Multilingual Coreference Resolution #Cross-lingual Natural Inference #Low-resource Language Support

Downloads 6,022

Release Time : 10/8/2022

Model Overview

A multilingual generative model based on the BLOOM architecture, supporting various tasks such as sentiment analysis, Q&A, and text generation, with strong performance in coreference resolution and natural language inference tasks.

Model Features

Multilingual Support

Supports 46 human languages and 13 programming languages, covering major global language families

Cross-task Generalization

Consistent performance across diverse NLP tasks like coreference resolution and natural language inference

Cultural Adaptability

Specifically includes African, South Asian and other regional languages, reflecting cultural inclusivity

Model Capabilities

Multilingual text generation

Sentiment analysis

Technical Q&A

Fable creation

Coreference resolution

Natural language inference

Code generation

Use Cases

Content Creation

Multilingual Story Generation

Generates fables based on specified language and theme

Can produce complete fables in Spanish/Hindi and other language versions

Sentiment Analysis

Cross-lingual Comment Analysis

Determines sentiment polarity in Chinese/English comments

Accurately distinguishes positive/neutral/negative evaluations

Educational Assistance

Technical Concept Explanation

Explains professional terms in specified languages

E.g. explaining backpropagation principle in Telugu

🚀 Model Introduction

This is a model with multiple capabilities, supporting text generation tasks. It has been tested on various datasets and shows different performance in different tasks.

📦 Dataset and Model Information

📊 Dataset

The model uses the following dataset:

bigscience/xP3

📋 Model Details

Property	Details
Model Name	bloomz-3b1
License	bigscience-bloom-rail-1.0
Supported Languages	ak, ar, as, bm, bn, ca, code, en, es, eu, fon, fr, gu, hi, id, ig, ki, kn, lg, ln, ml, mr, ne, nso, ny, or, pa, pt, rn, rw, sn, st, sw, ta, te, tn, ts, tum, tw, ur, vi, wo, xh, yo, zh, zu
Programming Languages	C, C++, C#, Go, Java, JavaScript, Lua, PHP, Python, Ruby, Rust, Scala, TypeScript
Pipeline Tag	text-generation

🧪 Widget Examples

Here are some widget examples to showcase the model's capabilities:

Example Title	Text
"zh-en sentiment"	"A legendary beginning, an immortal myth. This is not just a movie, but a label for entering a new era, forever inscribed in the annals of history. Would you rate the previous review as positive, neutral or negative?"
"zh-zh sentiment"	"一个传奇的开端，一个不灭的神话，这不仅仅是一部电影，而是作为一个走进新时代的标签，永远彪炳史册。你认为这句话的立场是赞扬、中立还是批评？"
"vi-en query"	"Suggest at least five related search terms to "Mạng neural nhân tạo"."
"fr-fr query"	"Proposez au moins cinq mots clés concernant «Réseau de neurones artificiels»."
"te-en qa"	"Explain in a sentence in Telugu what is backpropagation in neural networks."
"en-en qa"	"Why is the sky blue?"
"es-en fable"	"Write a fairy tale about a troll saving a princess from a dangerous dragon. The fairy tale is a masterpiece that has achieved praise worldwide and its moral is "Heroes Come in All Shapes and Sizes". Story (in Spanish):"
"hi-en fable"	"Write a fable about wood elves living in a forest that is suddenly invaded by ogres. The fable is a masterpiece that has achieved praise worldwide and its moral is "Violence is the last refuge of the incompetent". Fable (in Hindi):"

📈 Model Performance

Coreference Resolution

Dataset	Name	Config	Split	Accuracy
winogrande	Winogrande XL (xl)	xl	validation	53.67
Muennighoff/xwinograd	XWinograd (en)	en	test	59.23
Muennighoff/xwinograd	XWinograd (fr)	fr	test	53.01
Muennighoff/xwinograd	XWinograd (jp)	jp	test	52.45
Muennighoff/xwinograd	XWinograd (pt)	pt	test	53.61
Muennighoff/xwinograd	XWinograd (ru)	ru	test	53.97
Muennighoff/xwinograd	XWinograd (zh)	zh	test	60.91

Natural Language Inference

Dataset	Name	Config	Split	Accuracy
anli	ANLI (r1)	r1	validation	40.1
anli	ANLI (r2)	r2	validation	36.8
anli	ANLI (r3)	r3	validation	40.0
super_glue	SuperGLUE (cb)	cb	validation	75.0
super_glue	SuperGLUE (rte)	rte	validation	76.17
xnli	XNLI (ar)	ar	validation	53.29
xnli	XNLI (bg)	bg	validation	43.82
xnli	XNLI (de)	de	validation	45.26
xnli	XNLI (el)	el	validation	42.61
xnli	XNLI (en)	en	validation	57.31
xnli	XNLI (es)	es	validation	56.14
xnli	XNLI (fr)	fr	validation	55.78
xnli	XNLI (hi)	hi	validation	51.49
xnli	XNLI (ru)	ru	validation	47.11
xnli	XNLI (sw)	sw	validation	47.83
xnli	XNLI (th)	th	validation	42.93
xnli	XNLI (tr)	tr	validation	37.23
xnli	XNLI (ur)	ur	validation	49.04
xnli	XNLI (vi)	vi	validation	53.98
xnli	XNLI (zh)	zh	validation	54.18

Program Synthesis

Dataset	Name	Config	Split	Pass@1	Pass@10	Pass@100
openai_humaneval	HumanEval	None	test	6.29	11.94	19.06

Sentence Completion

Dataset	Name	Config	Split	Accuracy
story_cloze	StoryCloze (2016)	"2016"	validation	87.33
super_glue	SuperGLUE (copa)	copa	validation	76.0
xcopa	XCOPA (et)	et	validation	53.0
xcopa	XCOPA (ht)	ht	validation	64.0
xcopa	XCOPA (id)	id	validation	70.0
xcopa	XCOPA (it)	it	validation	53.0
xcopa	XCOPA (qu)	qu	validation	56.0
xcopa	XCOPA (sw)	sw	validation	66.0
xcopa	XCOPA (ta)	ta	validation	59.0
xcopa	XCOPA (th)	th	validation	63.0
xcopa	XCOPA (tr)	tr	validation	61.0
xcopa	XCOPA (vi)	vi	validation	77.0
xcopa	XCOPA (zh)	zh	validation	73.0
Muennighoff/xstory_cloze	XStoryCloze (ar)	ar	validation	80.61
Muennighoff/xstory_cloze	XStoryCloze (es)	es	validation	85.9
Muennighoff/xstory_cloze	XStoryCloze (eu)	eu	validation	[Value missing in original]

📄 License

The model is released under the bigscience-bloom-rail-1.0 license.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご