Latpull docs, demo, browserify, countTokens, tokenStats #30
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
I don't actually want to encode into tokens for my use case, quickly count to check my request won't exceed the limit.
This should be faster since we don't initialize the memory for the output array.
```
const crypto = require('crypto');
// Generate a random string of a given length
function generateRandomString(length) {
return crypto.randomBytes(length).toString('hex');
}
const {encode, decode, countTokens} = require('gpt-3-encoder')
let str = 'This is an example sentence to try encoding out on!'
// let now = Date.now();
let encoded = encode(str)
console.log('Encoded this string looks like: ', encoded)
console.log('We can look at each token and what it represents)
let tokencount = 0;
for(let token of encoded){
tokencount ++;
console.log({token, string: decode([token])})
}
console.log("there are n tokens: ", tokencount);
let decoded = decode(encoded)
console.log('We can decode it back into:\n', decoded)
let now = Date.now();
// todo: write an benchmark for the above method vs int countTokens(str)
str = generateRandomString(10000);
console.time('fencode');
encoded = encode(str);
console.log(`First encode to cache string n stuff in mem`);
console.timeEnd('fencode');
console.log(`Original string length: ${str.length}`);
// Benchmark the encode function
console.time('encode');
encoded = encode(str);
console.log(`Encoded string length: ${encoded.length}`);
console.timeEnd('encode');
// Benchmark the countTokens function
console.time('countTokens');
let tokenCount = countTokens(str);
console.log(`Number of tokens: ${tokenCount}`);
console.timeEnd('countTokens');
console.log(`Original string length: ${str.length}`);
console.log(`Encoded string length: ${encoded.length}`);
console.log(`Number of tokens: ${tokenCount}`);
```
```
We can decode it back into:
This is an example sentence to try encoding out on!
First encode to cache string n stuff in mem
fencode: 163.57ms
Original string length: 20000
Encoded string length: 11993
encode: 124.265ms
Number of tokens: 11993
countTokens: 29.2ms
Original string length: 20000
Encoded string length: 11993
Number of tokens: 11993
```
Co-authored-by: Andrew Healey <[email protected]>
I don't actually want to encode into tokens for my use case, quickly count to check my request won't exceed the limit.
This should be faster since we don't initialize the memory for the output array.
```
const crypto = require('crypto');
// Generate a random string of a given length
function generateRandomString(length) {
return crypto.randomBytes(length).toString('hex');
}
const {encode, decode, countTokens} = require('gpt-3-encoder')
let str = 'This is an example sentence to try encoding out on!'
// let now = Date.now();
let encoded = encode(str)
console.log('Encoded this string looks like: ', encoded)
console.log('We can look at each token and what it represents)
let tokencount = 0;
for(let token of encoded){
tokencount ++;
console.log({token, string: decode([token])})
}
console.log("there are n tokens: ", tokencount);
let decoded = decode(encoded)
console.log('We can decode it back into:\n', decoded)
let now = Date.now();
// todo: write an benchmark for the above method vs int countTokens(str)
str = generateRandomString(10000);
console.time('fencode');
encoded = encode(str);
console.log(`First encode to cache string n stuff in mem`);
console.timeEnd('fencode');
console.log(`Original string length: ${str.length}`);
// Benchmark the encode function
console.time('encode');
encoded = encode(str);
console.log(`Encoded string length: ${encoded.length}`);
console.timeEnd('encode');
// Benchmark the countTokens function
console.time('countTokens');
let tokenCount = countTokens(str);
console.log(`Number of tokens: ${tokenCount}`);
console.timeEnd('countTokens');
console.log(`Original string length: ${str.length}`);
console.log(`Encoded string length: ${encoded.length}`);
console.log(`Number of tokens: ${tokenCount}`);
```
```
We can decode it back into:
This is an example sentence to try encoding out on!
First encode to cache string n stuff in mem
fencode: 163.57ms
Original string length: 20000
Encoded string length: 11993
encode: 124.265ms
Number of tokens: 11993
countTokens: 29.2ms
Original string length: 20000
Encoded string length: 11993
Number of tokens: 11993
```
Co-authored-by: Kier <[email protected]>
# Conflicts: # Encoder.js
merge back the one other change and follow nick with moving to version 1.2 for feature add.
…e updates maybe we should host jsdoc on github pages also performed npm audit and updates for security vulnerabilities.
It seems to work fine
By using the Buffer class in this way, you can encode and decode strings as UTF-8 without using the TextEncoder and TextDecoder classes. This can be useful if you need to support earlier versions of Node.js that do not have these classes in the util module.
…eaking change but i think its actually mor widely compatible and should be fine.
…by webpaking... todo need to test browser version This precomputes bpe_ranks so that it is it the from needed make sure to rebuild if you change the bpe dump file
…ll need to deploy docs to github pages for demo to work
This was referenced Jan 18, 2023
Open
syonfox
commented
Jan 27, 2023
| @@ -16,7 +16,7 @@ jobs: | |||
|
|
|||
| strategy: | |||
| matrix: | |||
| node-version: [12.x, 14.x] | |||
| node-version: [12.x, 14.x, 16.x, 18.x] | |||
|
|
||
| export function decode(tokens: number[]): string; | ||
|
|
||
| export function countTokens(text: string): number; |
| @@ -1,6 +1,8 @@ | |||
| const { encode, decode } = require("./Encoder"); | |||
| const { encode, decode, countTokens, tokenStats } = require("./Encoder"); | |||
|
|
|||
There was a problem hiding this comment.
Also added functions to node lib index
| } | ||
| "browserify": "^17.0.0", | ||
| "jest": "^26.4.2", | ||
| "jsdoc": "^4.0.0" |
There was a problem hiding this comment.
only dev dependencies added for building docs and browserifying a version for client side execution (2.5M including data)
| }, | ||
| "repository": { | ||
| "type": "git", | ||
| "url": "git+https://github.com/AIDungeon/GPT-3-Encoder.git" | ||
| "url": "git+https://github.com/latitudegames/GPT-3-Encoder.git" |
There was a problem hiding this comment.
Were you guys renaming the pakage to aidungon
Encoder.js
Outdated
| const encoder = JSON.parse(fs.readFileSync(path.join(__dirname, './encoder.json'))); | ||
| const bpe_file = fs.readFileSync(path.join(__dirname, './vocab.bpe'), 'utf-8'); | ||
| // const encoder = loader('./encoder.json'); | ||
|
|
There was a problem hiding this comment.
We could erase these comments but in order to support loading into a browser module i have loaded the data into encoder.js and bpe_ranks.js
These are pre-computed a build time and this removes some of the work the client has to do other then load the data into memory. Todo maybe we should clean up these comments. but theoretical you could switch it back if you only wanted nodejs.
|
I don't know if your fork is stable or not, but I can't open issues to there... Simply adding this to a next repo (rc3), and calling the tokenStats seems to be broken... Also, it would be nice to type out the tokenstats return not just an any but something like an (didn't checked if this is actually correct or not): interface TokenStats {
count: number,
unique: number,
frequency: Record<number, number>,
positions: Record<number, number[]>,
tokens: number[],
} |
|
The pre 1.4rcX should be stable but doesn't have it all. I made some more extreme changes namely browser support. It does pass the tests and seems to work for a few people. needs a bit of elbow grease to clean up some stuff. I can look into the type def. I also enabled issues in my fork. |
…460069823
Add types let me know if this is the correct way to implement it.
And here are some additional useful stats that could be added to the stats object:
averageWordLength: The average length of the words in the tokens array.
mostFrequentWords: An array of the most frequently occurring words in the tokens array, ordered by frequency.
leastFrequentWords: An array of the least frequently occurring words in the tokens array, ordered by frequency.
wordPositions: An object that maps each word in the tokens array to its positions (indices) in the array.
wordLengths: An array of the lengths of the words in the tokens array.
# Conflicts: # docs/Encoder.js.html # docs/global.html # docs/index.html # package-lock.json
syonfox
commented
Mar 8, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I have reverted npm branching and changed syonfox to latitudegames for a possible pull. you can check it out and test in the syonfox fork