text
stringlengths 1
3.86k
|
---|
aer banknote berlitz calloway centrust cluett fromstein gitano guterman hydro-quebec ipo kia memotec mlx nahb punts rake regatta rubens sim snack-food ssangyong swapo wachter |
pierre <unk> N years old will join the board as a nonexecutive director nov. N |
mr. <unk> is chairman of <unk> n.v. the dutch publishing group |
rudolph <unk> N years old and former chairman of consolidated gold fields plc was named a nonexecutive director of this british industrial conglomerate |
a form of asbestos once used to make kent cigarette filters has caused a high percentage of cancer deaths among a group of workers exposed to it more than N years ago researchers reported |
the asbestos fiber <unk> is unusually <unk> once it enters the <unk> with even brief exposures to it causing symptoms that show up decades later researchers said |
<unk> inc. the unit of new york-based <unk> corp. that makes kent cigarettes stopped using <unk> in its <unk> cigarette filters in N |
although preliminary findings were reported more than a year ago the latest results appear in today 's new england journal of medicine a forum likely to bring new attention to the problem |
a <unk> <unk> said this is an old story |
we 're talking about years ago before anyone heard of asbestos having any questionable properties |
there is no asbestos in our products now |
neither <unk> nor the researchers who studied the workers were aware of any research on smokers of the kent cigarettes |
we have no useful information on whether users are at risk said james a. <unk> of boston 's <unk> cancer institute |
dr. <unk> led a team of researchers from the national cancer institute and the medical schools of harvard university and boston university |
the <unk> spokeswoman said asbestos was used in very modest amounts in making paper for the filters in the early 1950s and replaced with a different type of <unk> in N |
from N to N N billion kent cigarettes with the filters were sold the company said |
among N men who worked closely with the substance N have died more than three times the expected number |
four of the five surviving workers have <unk> diseases including three with recently <unk> cancer |
the total of N deaths from malignant <unk> lung cancer and <unk> was far higher than expected the researchers said |
the <unk> rate is a striking finding among those of us who study <unk> diseases said dr. <unk> |
the percentage of lung cancer deaths among the workers at the west <unk> mass. paper factory appears to be the highest for any asbestos workers studied in western industrialized countries he said |
the plant which is owned by <unk> & <unk> co. was under contract with <unk> to make the cigarette filters |
the finding probably will support those who argue that the u.s. should regulate the class of asbestos including <unk> more <unk> than the common kind of asbestos <unk> found in most schools and other buildings dr. <unk> said |
the u.s. is one of the few industrialized nations that does n't have a higher standard of regulation for the smooth <unk> fibers such as <unk> that are classified as <unk> according to <unk> t. <unk> a professor of <unk> at the university of vermont college of medicine |
more common <unk> fibers are <unk> and are more easily rejected by the body dr. <unk> explained |
in july the environmental protection agency imposed a gradual ban on virtually all uses of asbestos |
by N almost all remaining uses of <unk> asbestos will be outlawed |
about N workers at a factory that made paper for the kent filters were exposed to asbestos in the 1950s |
areas of the factory were particularly dusty where the <unk> was used |
workers dumped large <unk> <unk> of the imported material into a huge <unk> poured in cotton and <unk> fibers and <unk> mixed the dry fibers in a process used to make filters |
workers described clouds of blue dust that hung over parts of the factory even though <unk> fans <unk> the area |
there 's no question that some of those workers and managers contracted <unk> diseases said <unk> phillips vice president of human resources for <unk> & <unk> |
but you have to recognize that these events took place N years ago |
it has no bearing on our work force today |
yields on money-market mutual funds continued to slide amid signs that portfolio managers expect further declines in interest rates |
the average seven-day compound yield of the N taxable funds tracked by <unk> 's money fund report eased a fraction of a percentage point to N N from N N for the week ended tuesday |
compound yields assume reinvestment of dividends and that the current yield continues for a year |
average maturity of the funds ' investments <unk> by a day to N days the longest since early august according to donoghue 's |
longer maturities are thought to indicate declining interest rates because they permit portfolio managers to retain relatively higher rates for a longer period |
shorter maturities are considered a sign of rising rates because portfolio managers can capture higher rates sooner |
the average maturity for funds open only to institutions considered by some to be a stronger indicator because those managers watch the market closely reached a high point for the year N days |
nevertheless said <unk> <unk> <unk> editor of money fund report yields may <unk> up again before they <unk> down because of recent rises in short-term interest rates |
the yield on six-month treasury bills sold at monday 's auction for example rose to N N from N N |
despite recent declines in yields investors continue to pour cash into money funds |
assets of the N taxable funds grew by $ N billion during the latest week to $ N billion |
typically money-fund yields beat comparable short-term investments because portfolio managers can vary maturities and go after the highest rates |
the top money funds are currently yielding well over N N |
dreyfus world-wide dollar the <unk> fund had a seven-day compound yield of N N during the latest week down from N N a week earlier |
it invests heavily in dollar-denominated securities overseas and is currently <unk> management fees which boosts its yield |
the average seven-day simple yield of the N funds was N N down from N N |
the 30-day simple yield fell to an average N N from N N the 30-day compound yield slid to an average N N from N N |
j.p. <unk> vice chairman of <unk> grace & co. which holds a N N interest in this <unk> company was elected a director |
he succeeds <unk> d. <unk> formerly a <unk> grace vice chairman who resigned |
<unk> grace holds three of grace energy 's seven board seats |
pacific first financial corp. said shareholders approved its acquisition by royal <unk> ltd. of toronto for $ N a share or $ N million |
the thrift holding company said it expects to obtain regulatory approval and complete the transaction by year-end |
<unk> international inc. said its <unk> & <unk> unit completed the sale of its <unk> controls operations to <unk> s.p a. for $ N million |
<unk> is an italian state-owned holding company with interests in the mechanical engineering industry |
<unk> controls based in <unk> ohio makes computerized industrial controls systems |
it employs N people and has annual revenue of about $ N million |
the federal government suspended sales of u.s. savings bonds because congress has n't lifted the ceiling on government debt |
until congress acts the government has n't any authority to issue new debt obligations of any kind the treasury said |
the government 's borrowing authority dropped at midnight tuesday to $ N trillion from $ N trillion |
legislation to lift the debt ceiling is <unk> in the fight over cutting capital-gains taxes |
the house has voted to raise the ceiling to $ N trillion but the senate is n't expected to act until next week at the earliest |
the treasury said the u.s. will default on nov. N if congress does n't act by then |
clark j. <unk> was named senior vice president and general manager of this u.s. sales and marketing arm of japanese auto maker mazda motor corp |
in the new position he will oversee mazda 's u.s. sales service parts and marketing operations |
previously mr. <unk> N years old was general marketing manager of chrysler corp. 's chrysler division |
he had been a sales and marketing executive with chrysler for N years |
when it 's time for their <unk> <unk> the nation 's manufacturing <unk> typically jet off to the <unk> <unk> of resort towns like <unk> <unk> and hot springs |
not this year |
the national association of manufacturers settled on the <unk> capital of indianapolis for its fall board meeting |
and the city decided to treat its guests more like royalty or rock stars than factory owners |
the idea of course to prove to N corporate decision makers that the buckle on the <unk> belt is n't so <unk> after all that it 's a good place for a company to expand |
on the receiving end of the message were officials from giants like du pont and <unk> along with lesser <unk> like <unk> steel and the valley queen <unk> factory |
for <unk> the executives joined mayor william h. <unk> iii for an evening of the indianapolis <unk> <unk> and a guest <unk> victor <unk> |
champagne and <unk> followed |
the next morning with a police <unk> <unk> of executives and their wives <unk> to the indianapolis motor <unk> <unk> by traffic or red lights |
the governor could n't make it so the <unk> governor welcomed the special guests |
a buffet breakfast was held in the museum where food and drinks are banned to everyday visitors |
then in the guests ' honor the <unk> <unk> out four drivers crews and even the official indianapolis N announcer for a <unk> exhibition race |
after the race fortune N executives <unk> like <unk> over the cars and drivers |
no <unk> the drivers pointed out they still had space on their machines for another sponsor 's name or two |
back downtown the <unk> squeezed in a few meetings at the hotel before <unk> the buses again |
this time it was for dinner and <unk> a block away |
under the stars and <unk> of the <unk> indiana <unk> <unk> nine of the hottest chefs in town fed them indiana <unk> <unk> <unk> <unk> <unk> <unk> and <unk> <unk> with a <unk> <unk> |
knowing a <unk> and free <unk> when they eat one the executives gave the chefs a standing <unk> |
more than a few <unk> say the <unk> treatment <unk> them to return to a <unk> city for future meetings |
but for now they 're looking forward to their winter meeting <unk> in february |
south korea registered a trade deficit of $ N million in october reflecting the country 's economic <unk> according to government figures released wednesday |
preliminary <unk> by the trade and industry ministry showed another trade deficit in october the fifth monthly setback this year casting a cloud on south korea 's <unk> economy |
exports in october stood at $ N billion a mere N N increase from a year earlier while imports increased sharply to $ N billion up N N from last october |
south korea 's economic boom which began in N stopped this year because of prolonged labor disputes trade conflicts and sluggish exports |
government officials said exports at the end of the year would remain under a government target of $ N billion |
despite the gloomy forecast south korea has recorded a trade surplus of $ N million so far this year |
from january to october the nation 's accumulated exports increased N N from the same period last year to $ N billion |
imports were at $ N billion up N N |
newsweek trying to keep pace with rival time magazine announced new advertising rates for N and said it will introduce a new incentive plan for advertisers |
the new ad plan from newsweek a unit of the washington post co. is the second incentive plan the magazine has offered advertisers in three years |
Input files for generating the Importance Matrix
Which file to use for generating the importance matrix
Not all importance matrices are equal. The best results are obtained when using a source file similar to the training data. Size also matters: the bigger the model (eg: 70b vs 13b) and the higher the quant (eg: q6k_ vs iq3_xs), the bigger the source file needs to be to make an impact. Multiple input files can be combined if needed; for example:
cat multilingual.txt code.txt badwords_multilingual.txt > custom_multilingual.matrix
Note on context size when generating the matrix: in general, a small context size such as 512 is recommended, and community tests have shown it usually performs than a larger one such as 4096. However, I would argue this is is highly dependent on the source data you are using: with random tokens or short text a small context makes sense; but when using larger texts, a larger context matching the size of the texts might be a better choice. Remember that the size is in tokens, which roughly translates to number of words, not characters.
Another important factor to consider is, an importance matrix based on english language only will degrade the model multingual capabilities. Therefore the use of multilingual matrices are especially important when considering models like Mistral or Command-R. The same applies to coding capabilities. Choose carefully based on the model strength.
You will find below descriptions for the various input files provided, to help you choose the correct one.
Community provided files
groups_merged
"Here is a decent general purpose imatrix calibration dataset. It should be more diverse than wikitext at ~30k tokens, as it is excerpts of a larger dataset which includes coding examples (which seems quite important!)
This means it's generally higher entropy data compared to wikitext, and it's real data rather than pseudo-randomly generated data.
I get lower KL div than wikitext for the same length and the outputs seem qualitatively better." (kalomaze)
https://github.com/ggerganov/llama.cpp/discussions/5263#discussioncomment-8395384
group_10_merged
(superseeded by groups_merged)
"This is about ~50k pseudo-random tokens.
I am getting the best balance between the maximum divergence and the other divergence statistics using this file when quantizing 7b" (kalomaze)
https://github.com/ggerganov/llama.cpp/discussions/5263#discussioncomment-8349233
20k_random_data
(superseeded by groups_10_merged)
https://github.com/ggerganov/llama.cpp/discussions/5006#discussioncomment-8163190
8k_random_data
(superseeded by 20k_random_data)
https://github.com/ggerganov/llama.cpp/discussions/5006#discussion-6087829
badwords
402 english words that can be considered dirty, naughty, obscene, or otherwise bad words.
This could be useful to remove guard rails.
Compiled from Shutterstock github repo
badwords_multilingual
2580 words that can be considered dirty, naughty, obscene, or otherwise bad words. Includes 26 languages.
This could be useful to remove guard rails.
Compiled from Shutterstock github repo
ptb.train
Penn Treebank (PTB) is a widely used preprocessed large dataset designed for language training. Casing,
punctuation and numbers have been removed from the training data. Recently it has kind of been superseeded
by WikiText which does not have these removals, features a larger vocabulary and full articles (better
suited for models that can take advantage of long term dependencies). However, for importantce matrix training,
PTB is still a valid dataset, which has the advantage of being manually curated, and similar to WikiText,
without being WikiText; this can help against bias.
WikiText
The WikiText language modeling dataset is a collection of over 100 million tokens extracted from the set of
verified Good and Featured articles on Wikipedia. Compared to PTB, WikiText-2 is over 2 times larger and
WikiText-103 is over 110 times larger. As it is composed of full articles, the dataset is well suited for models
that can take advantage of long term dependencies.
https://huggingface.co/datasets/wikitext
WikiText_FR
70 million tokens extracted from the set of french Wikipedia articles that are classified as "quality articles"
or "good articles".
https://huggingface.co/datasets/asi/wikitext_fr
c4
The C4 dataset is a collection text sourced from the public Common Crawl web scrape.
It includes heuristics to extract only natural language (as opposed to boilerplate and other gibberish)
in addition to extensive deduplication. C4 dataset was explicitly designed to be English only:
any page that was not given a probability of at least 99% of being English by langdetect was discarded.
code (exllamav2)
Programming
multilingual (exllamav2)
English, Arabic, Chinese, French, German, Japanese, Polish, Russian, Spanish, Swedish, Turkish, Hebrew,
Macedonian, Norwegian, Lithuanian, Greek, Italian, Afrikaans, Dutch, Danish.
technical (exllamav2)
Technical writing.
tiny
Very short stories. Be mindful of the prevalence of "Once upon a time" and "<|endoftext|>".
Extract from TinyStories dataset
wiki (exllamav2)
Small Wikipedia dump. Unclean, contains many unwanted tags.
exllamav2 calibration data taken from:
https://github.com/turboderp/exllamav2/tree/master/conversion/standard_cal_data
How to quantize using an imatrix, with llama.cpp
- Get one of the input files collected here, or elsewhere.
- Convert or download the model you want to quantise, in fp16 GGUF format.
- Generate an imatrix file specific to the model you want to quantise
cd <llama.cpp directory>
./imatrix -m <model_path>/ggml-model-f16.gguf -f <plain_text_matrix_file> -o <output.matrix> -t 12 -ngl 144 --chunks 100 -b 512 -c 512
# -ngl : layers offloaded to gpu (recommended to use number of layers the model contains)
# -t 12 : number of threads (should probably match no of cpu)
# -c 512 : context size, testing seems to show 512 is recommended (default=512, 0=loaded from model)
# -b 200 : batch size (default=512)
# --chunks 100 (recommended)
# --mlock : keep model in ram (only use if you had sufficient RAM for the whole fp16)
- Use the generated matrix file to quantise the model (see further down for some pre-computed matrix files)
./quantize --imatrix <output.matrix> <model_path>/ggml-model-f16.gguf <quantisation_level, eg:IQ4_XS>
Note: normal quantisation also benefits from using a matrix file. It also seem that a bigger input matrix is better for higher quantisation.
Pre-computed matrix files
Since generating a matrix files takes time and requires significant processing power and memory, some kind folks have made available pre-computed matrix files. You can use those directly in the quantize process. However, remember they can only be used for the specific model mentioned, and no other.
Joseph717171/Imatrices a growing list of many matrix files for 7B to 17B models
ikawrakow/imatrix-from-wiki-train matrix files for base models (lama, mistral, nous-hermes, qwen) trained on wiki.train.raw
- Downloads last month
- 198