Civil Rights
Movements, leaders, victories and the continuing fight for equality.
q2K BHP
Black History Portal
THE BHP WIRE —
HIDDEN TRUTHS
What's New!
Explore the people, places, events, achievements, struggles and stories that shaped our journey.
Movements, leaders, victories and the continuing fight for equality.
Innovation, patents, science, technology and world-changing contributions.
Pioneers, champions, Negro Leagues, records, activism and excellence.
Meet the people whose lives, choices and achievements shaped the journey.
Black towns, communities, institutions and places where history happened.
Moments that changed communities, movements, institutions and the nation.
Mansa Musa was the ruler of the Mali Empire in West Africa. Details recorded here should be sourced; unknown information is left blank.
MORE →Reflects the personal views, recollections, and perspective of the author, Mike Davis.
This is a personal recollection on the Move fire on May 13, 1985
In computing, byte-pair encoding (BPE),[1][2] or digram coding,[3] is an algorithm, first described in 1994 by Philip Gage, for encoding strings of text into smaller strings by creating and using a translation table.[4] A slightly modified version of the algorithm is used in large language model tokenizers.
The original version of the algorithm focused on compression. It replaces the highest-frequency pair of bytes with a new byte that was not contained in the initial dataset. A lookup table of the replacements is required to rebuild the initial dataset. The modified version builds "tokens" (units of recognition) that match varying amounts of source text, from single characters (including single digits or single punctuation marks) to whole words (even long compound words).[5][6][7]
The original BPE algorithm operates by iteratively replacing the most common contiguous sequences of characters in a target text with unused 'placeholder' bytes. The iteration ends when no sequences can be found, leaving the target text effectively compressed. Decompression can be performed by reversing this process, querying known placeholder terms against their corresponding denoted sequence, using a lookup table. In the original paper, this lookup table is encoded and stored alongside the compressed text.
Suppose the data to be encoded is:[8]
aaabdaaabac
The byte pair "aa" occurs most often, so it will be replaced by a byte that is not used in the data, such as "Z". Now there is the following data and replacement table:
ZabdZabac Z=aa
Then the process is repeated with byte pair "ab", replacing it with "Y":
ZYdZYac Y=ab Z=aa
The only literal byte pair left occurs only once, and the encoding might stop here. Alternatively, the process could continue with recursive byte-pair encoding, replacing "ZY" with "X":
XdXac X=ZY Y=ab Z=aa
This data cannot be compressed further by byte-pair encoding because there are no pairs of bytes that occur more than once.
To decompress the data, simply perform the replacements in the reverse order.
The original BPE algorithm is modified for use in language modeling, especially for large language models based on neural networks. Compared to the original BPE, the modified BPE does not aim to maximally compress text, but rather, to encode plaintext into "tokens", which are natural numbers.[9] All the unique tokens found in a corpus are listed in a token vocabulary. The token vocabulary can also include some other special tokens, relative to the use case. The size of the token vocabulary, in the case of GPT-3.5 and GPT-4, is 100258 (100000 from BPE algorithm and 258 included as special tokens).[10]
The modified tokenization algorithm initially treats the set of unique characters as 1-character-long n-grams (the initial tokens). Then, successively, the most frequent pair of adjacent tokens is merged into a new, longer n-gram and all instances of the pair are replaced by this new token. This is repeated until a vocabulary of prescribed size is obtained. Note that new words can always be constructed from final vocabulary tokens and initial-set characters.[11]
This modified BPE approach has been extended from spoken language to sign language in recent years.[12]
Suppose we are encoding the previous example of "aaabdaaabac", with a specified vocabulary size of 6, then it would first be encoded as "0, 0, 0, 1, 2, 0, 0, 0, 1, 0, 3" with a vocabulary of "a=0, b=1, d=2, c=3". Then it would proceed as before, and obtain "4, 5, 2, 4, 5, 0, 3" with a vocabulary of "a=0, b=1, d=2, c=3, aa=4, ab=5".
So far this is essentially the same as before. However, if we only had specified a vocabulary size of 5, then the process would stop at vocabulary "a=0, b=1, d=2, c=3, aa=4", so that the example would be encoded as "4, 0, 1, 2, 4, 0, 1, 0, 3". Conversely, if we had specified a vocabulary size of 8, then it would be encoded as "7, 6, 0, 3", with a vocabulary of "a=0, b=1, d=2, c=3, aa=4, ab=5, aaab=6, aaabd=7". This is not maximally compressed, because modified BPE does not aim for maximum compression. Instead, it aims for an encoding that is efficient and practical for language model training.[13]
In the above example, the output of the BPE is a vocabulary, which can be used to encode any text that is written with the letters "abcd". It will not be able to encode text containing other symbols, such as "no". Even giving each of the 26 letters an entry in the vocabulary, since there are many languages in the world using many different scripts, inevitably some symbols would be unencodable by such a vocabulary.
One solution is to replace any unencodable symbol with a special symbol named UNK ("unknown").
The byte-level BPE is another approach. It simply converts the text into UTF-8 first, and treat it as a stream of bytes. This guarantees that any text encoded in UTF-8 can be encoded by the BPE. This has been used in BERT-like models like RoBERTa, BART, and DeBERTa, and GPT-like models like GPT-2.[14][15][16]
{{cite journal}}: CS1 maint: multiple names: authors list (link)
Source: Wikipedia. Article content is retrieved live through the MediaWiki API.
In computing, byte-pair encoding (BPE), or digram coding, is an algorithm, first described in 1994 by Philip Gage, for encoding strings of text into smaller strings by creating and using a translation table. A slightly modified version of the algorithm is used in large language model tokenizers. The original version of the algorithm focused on compression. It replaces the highest-frequency pair of bytes with a new byte that was not contained in the initial dataset. A lookup table of the replacements is required to rebuild the initial dataset. The modified version builds "tokens" (units of recognition) that match varying amounts of source text, from single characters (including single digits or single punctuation marks) to whole words (even long compound words).
DTE may refer to: Data terminal equipment, an end instrument used in telecommunication and data transmission Distance to empty, a feature in an automobile electronic instrument cluster Dithioerythritol, a chemical DTE (direct to edit), a digital video recording method DTE Energy, a Detroit, Michigan-based utility Dora the Explorer, a children's animated television show. Dual-Tile encoding, another name for byte pair encoding Directorate of Technical Education, Maharashtra, an Indian state government agency for higher education. Department of Technical Education, a higher education governance body under the government of Kerala, India
GBK is an extension of the GB 2312 character set for Simplified Chinese characters, used in the People's Republic of China. It includes all unified CJK characters found in GB 13000.1-93, i.e. ISO/IEC 10646:1993, or Unicode 1.1. Since its initial release in 1993, GBK has been extended by Microsoft in Code page 936/1386, which was then extended into GBK 1.0. GBK is also the IANA-registered internet name for the Microsoft mapping, which differs from other implementations primarily by the single-byte euro sign at 0x80. GB abbreviates Guójiā Biāozhǔn, which means national standard in Chinese, while K stands for Extension (扩展 kuòzhǎn). GBK not only extended the old standard GB 2312 with Traditional Chinese characters, but also with Chinese characters that were simplified after the establishment of GB 2312 in 1981. With the arrival of GBK, certain names with characters formerly unrepresentable, like the 镕 (róng) character in former Chinese Premier Zhu Rongji's name, are now representable. As of December 2025, GBK is the third-most declared encoding served from China and territories (after UTF-8 and the subset GB 2312), with 1.3% of web servers serving a page that declares GBK. However, all major web browsers decode GB2312-marked documents as if they were marked GBK, i.e. not as a subset (meaning in effect GBK is the second-most popular encoding) except for Safari and Edge on the label GB_2312 (they do however decode GB_2312-80 and GB2312 as the superset GBK). Together, GBK and GB 2312 encodings have a combined 3.5% presence in China and territories. Globally, GBK accounts for less than 0.02% of all web pages and GBK+GB2312 for less than 0.07%.
The byte is a unit of digital information that most commonly consists of eight bits. Historically, the byte was the number of bits used to encode a single character of text in a computer and for this reason it is the smallest addressable unit of memory in many computer architectures. To disambiguate arbitrarily sized bytes from the common 8-bit definition, network protocol documents such as the Internet Protocol (RFC 791) refer to an 8-bit byte as an octet. Those bits in an octet are usually counted with numbering from 0 to 7 or 7 to 0 depending on the bit endianness. The size of the byte has historically been hardware-dependent and no definitive standards existed that mandated the size. Sizes from 1 to 48 bits have been used. The six-bit character code was an often-used implementation in early encoding systems, and computers using six-bit and nine-bit bytes were common in the 1960s. These systems often had memory words of 12, 18, 24, 30, 36, 48, or 60 bits, corresponding to 2, 3, 4, 5, 6, 8, or 10 six-bit bytes, and persisted, in legacy systems, into the twenty-first century. In this era, bit groupings in the instruction stream were often referred to as syllables or slab, before the term byte became common. The modern de facto standard of eight bits, as documented in ISO/IEC 2382-1:1993, is a convenient power of two permitting the binary-encoded values 0 through 255 for one byte, as 2 to the power of 8 is 256. The international standard IEC 80000-13 codified this common meaning. Many types of applications use information representable in eight or fewer bits and processor designers commonly optimize for this usage. The popularity of major commercial computing architectures has aided in the ubiquitous acceptance of the 8-bit byte. Modern architectures typically use 32- or 64-bit words, built of four or eight bytes, respectively. The unit symbol for the byte was designated as the upper-case letter B by the International Electrotechnical Commission (IEC) and Institute of Electrical and Electronics Engineers (IEEE). Internationally, the unit octet explicitly defines a sequence of eight bits, eliminating the potential ambiguity of the term "byte". The symbol for octet, 'o', also conveniently eliminates the ambiguity in the symbol 'B' between byte and bel.
Before the 1921 destruction of Tulsa’s Greenwood District, Black residents had created a remarkable center of business and community life. The district included stores, professional offices, entertainment venues and homes owned by Black citizens. Understanding Greenwood means learning what was built—not only what was burned.
MORE →Shirley Chisholm, elected in 1968.