Merge branch 'master' of https://github.com/Stillerman/huggingface-vs…

…code-attribution

README.md

@@ -1,18 +1,37 @@
 # Project for testing open source code completion models
 
-It was forked from [tabnine-vscode](https://github.com/codota/tabnine-vscode) & modified for making it compatible with open sorurce code models on hf.co/models
+It was forked from [tabnine-vscode](https://github.com/codota/tabnine-vscode) & modified for making it compatible with open source code models on [hf.co/models](https://huggingface.co/models)
 
 ## Installing
 
 Install just like any other [vscode extension](https://marketplace.visualstudio.com/items?itemName=HuggingFace.huggingface-vscode).
 
+By default, this extension is using [bigcode/starcoder](https://huggingface.co/bigcode/starcoder) & [Hugging Face Inference API](https://huggingface.co/inference-api) for the inference. However, you can [configure](#configuring) to make inference requests to your custom endpoint that is not Hugging Face Inference API. Thus, if you are using the default Hugging Face Inference AP inference, you'd need to provide [HF API Token](#hf-api-token).
+
+#### HF API token
+
+You can supply your HF API token ([hf.co/settings/token](https://hf.co/settings/token)) with this command:
+1. `Cmd/Ctrl+Shift+P` to open VSCode command palette
+2. Type: `Hugging Face Code: Set API token`
+
+<img src="https://github.com/huggingface/huggingface-vscode/raw/master/assets/set-api-token.png" width="800px">
+
 ## Testing
 
 1. Create a new python file
 2. Try typing `def main():`
 
 <img src="https://github.com/huggingface/huggingface-vscode/raw/master/assets/ext-working.png" width="800px">
 
+#### Checking if the generated code in in [The Stack](https://huggingface.co/datasets/bigcode/the-stack)
+
+Hit `Ctrl+Esc` to check if the generated code is in in [The Stack](https://huggingface.co/datasets/bigcode/the-stack).
+This is a rapid first-pass attribution check using [stack.dataportraits.org](https://stack.dataportraits.org).
+We check for sequences of at least 50 characters that match a Bloom filter.
+This means false positives are possible and long enough surrounding context is necesssary (see the [paper](https://dataportraits.org/) for details on n-gram striding and sequence length).
+[The dedicated Stack search tool](https://hf.co/spaces/bigcode/search) is a full dataset index and can be used for a complete second pass. 
+
+
 ## Developing
 Make sure you've [installed yarn](https://yarnpkg.com/getting-started/install) on your system.
 1. Clone this repo: `git clone https://github.com/huggingface/huggingface-vscode`
@@ -34,12 +53,6 @@ You can configure: endpoint to where request will be sent and special tokens.
 
 <img src="https://github.com/huggingface/huggingface-vscode/raw/master/assets/set-configs.png" width="800px">
 
-#### HF API token
-
-You can supply your HF API token (hf.co/settings/token) with this command:
-
-<img src="https://github.com/huggingface/huggingface-vscode/raw/master/assets/set-api-token.png" width="800px">
-
 Example:
 
 Let's say your current code is this:

package.json

@@ -1,7 +1,7 @@
 {
  "publisher": "HuggingFace",
  "name": "huggingface-vscode",
- "version": "0.0.12",
+ "version": "0.0.24",
  "displayName": "HF Code Autocomplete",
  "description": "AI Autocomplete for OSS code-gen models",
  "icon": "small_logo.png",

src/globals/consts.ts

@@ -18,7 +18,7 @@ export const STATUS_NAME = "Tabnine";
 
 export const INSTRUMENTATION_KEY = "<INSTRUMENTATION_KEY>";
 
-export const CHAR_LIMIT = 100_000;
+export const CHAR_LIMIT = 4_000;
 export const MAX_NUM_RESULTS = 5;
 export const CONSECUTIVE_RESTART_THRESHOLD = 100;
 export const REQUEST_FAILURES_THRESHOLD = 20;

src/inlineSuggestions/highlightStackAttributions.ts

@@ -49,9 +49,10 @@ export default async function highlightStackAttributions(): Promise<void> {
  headers: { "Content-Type": "application/json" },
  });
 
- if(!resp.ok){
- return;
- }
+ if(!resp.ok){
+ return;
+ }
+
  const json = await resp.json() as any as {spans: [number, number][]} 
  const {spans} = json
 

src/runCompletion.ts

@@ -1,4 +1,4 @@
-import { Position, Range, TextDocument, WorkspaceConfiguration, workspace } from "vscode";
+import { Position, Range, TextDocument, WorkspaceConfiguration, workspace, window, env, Uri } from "vscode";
 import {URL} from "url";
 import fetch from "node-fetch";
 import { AutocompleteResult, ResultEntry } from "./binary/requests/requests";
@@ -10,6 +10,8 @@ import { getTabnineExtensionContext } from "./globals/tabnineExtensionContext";
 
 export type CompletionType = "normal" | "snippet";
 
+let didShowTokenWarning = false;
+
 export default async function runCompletion(
  document: TextDocument,
  position: Position,
@@ -19,23 +21,11 @@ export default async function runCompletion(
  setLoadingStatus(FULL_BRAND_REPRESENTATION);
  const offset = document.offsetAt(position);
  const beforeStartOffset = Math.max(0, offset - CHAR_LIMIT);
- const afterEndOffset = offset + CHAR_LIMIT;
+ // const afterEndOffset = offset + CHAR_LIMIT;
  const beforeStart = document.positionAt(beforeStartOffset);
- const afterEnd = document.positionAt(afterEndOffset);
+ // const afterEnd = document.positionAt(afterEndOffset);
  const prefix = document.getText(new Range(beforeStart, position)) + currentSuggestionText;
- const suffix = document.getText(new Range(position, afterEnd));
- // const requestData = {
- // filename: getFileNameWithExtension(document),
- // prefix,
- // suffix,
- // region_includes_beginning: beforeStartOffset === 0,
- // region_includes_end: document.offsetAt(afterEnd) !== afterEndOffset,
- // max_num_results: getMaxResults(),
- // offset,
- // line: position.line,
- // character: position.character,
- // indentation_size: getTabSize(),
- // };
+ // const suffix = document.getText(new Range(position, afterEnd));
 
  type Config = WorkspaceConfiguration & {
  modelIdOrEndpoint: string;
@@ -47,26 +37,37 @@ export default async function runCompletion(
  temperature: number;
  };
  const config: Config = workspace.getConfiguration("HuggingFaceCode") as Config;
- const { modelIdOrEndpoint, isFillMode, startToken, middleToken, endToken, stopToken, temperature } = config;
+ const { modelIdOrEndpoint, stopToken, temperature } = config;
+
+ const context = getTabnineExtensionContext();
+ const apiToken = await context?.secrets.get("apiToken");
 
  let endpoint = ""
  try{
  new URL(modelIdOrEndpoint);
  endpoint = modelIdOrEndpoint;
  }catch(e){
  endpoint = `https://api-inference.huggingface.co/models/${modelIdOrEndpoint}`
- }
 
- let inputs = `${startToken}${prefix}`;
- if(isFillMode){
- inputs += `${middleToken}${suffix}`;
+ // if user hasn't supplied API Token yet, ask user to supply one
+ if(!apiToken && !didShowTokenWarning){
+ didShowTokenWarning = true;
+ void window.showInformationMessage(`In order to use "${modelIdOrEndpoint}" through Hugging Face API Inference, you'd need Hugging Face API Token`,
+ "Get your token"
+ ).then(clicked => {
+ if (clicked) {
+ void env.openExternal(Uri.parse("https://github.com/huggingface/huggingface-vscode#hf-api-token"));
+ }
+ });
+ }
  }
- inputs += endToken;
+
+ const inputs = prefix;
 
  const data = {
  inputs,
  parameters: {
- max_new_tokens: 256,
+ max_new_tokens: 60,
  temperature,
  do_sample: temperature > 0,
  top_p: 0.95,
@@ -76,9 +77,6 @@ export default async function runCompletion(
 
  logInput(inputs, data.parameters);
 
- const context = getTabnineExtensionContext();
- const apiToken = await context?.secrets.get("apiToken");
-
  const headers = {
  "Content-Type": "application/json",
  "Authorization": "",
@@ -95,14 +93,15 @@ export default async function runCompletion(
 
  if(!res.ok){
  console.error("Error sending a request", res.status, res.statusText);
+ setDefaultStatus();
  return null;
  }
 
  const generatedTextRaw = getGeneratedText(await res.json());
+
  let generatedText = generatedTextRaw.replace(stopToken, "");
- const indexEndToken = generatedText.indexOf(endToken)
- if(indexEndToken !== -1){
- generatedText = generatedText.slice(indexEndToken+endToken.length).trim();
+ if(generatedText.slice(0, inputs.length) === inputs){
+ generatedText = generatedText.slice(inputs.length);
  }
 
  const resultEntry: ResultEntry = {