Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

/edit command adds opening file as instruction #1237

Closed
3 tasks done
5eqn opened this issue May 6, 2024 · 1 comment
Closed
3 tasks done

/edit command adds opening file as instruction #1237

5eqn opened this issue May 6, 2024 · 1 comment
Labels
bug Something isn't working

Comments

@5eqn
Copy link
Contributor

5eqn commented May 6, 2024

Before submitting your bug report

Relevant environment info

- OS: Arch Linux 6.8.8-arch1-1
- Continue: Latest preview branch (Commit 3c954dd8ceb0311c9dda95b0326494b12f04fdc4)
- IDE: VSCode 1.88.1
- Model: Ollama v0.1.32

Description

After PR #1231 , when using /edit, the content is appended with a MessagePart showing current opening file:

image

It does not get deleted in the prompt:

image

This can confuse the model.

Proposed fix

Use input instead of history[history.length - 1].content to get the content. I tested in my environment that this fix is valid.

This works because input is calculated by selecting the last text MessagePart (as in resolveInput).

Check out PR #1238 !

To reproduce

  1. Build extension from latest preview branch
  2. Use /edit
  3. Go to output and check the prompt

Log output

Settings:
contextLength: 8192
model: llama3-8b
maxTokens: 4096
stop: <|start_header_id|>,<|end_header_id|>,<|eot_id|>
raw: true

############################################

<|start_header_id|>user<|end_header_id|>

import { factorial, repeat } from "./factorial";
<START EDITING HERE>
function fib(n) {
    if (n <= 1) return n;
    return fib(n - 2) + fib(n - 1);
}
<STOP EDITING HERE>
```

Please rewrite the entire code block above, editing the portion below "<START EDITING HERE>" in order to satisfy the following request: "The following file is currently open. Don't reference it if it's not relevant to the user's message.

```example.ts
import { factorial, repeat } from "./factorial";

function fib(n) {
    if (n <= 1) return n;
    return fib(n - 2) + fib(n - 1);
}

let d = repeat(5, "a");
console.log(d);

let e = factorial(3);
console.log(e);
```

change this to O(n)". When you get to "<STOP EDITING HERE>", end your response.
<|eot_id|><|start_header_id|>assistant<|end_header_id|>
Sure! Here's the entire code block, including the rewritten portion:
```typescript
import { factorial, repeat } from "./factorial";
<START EDITING HERE>
@5eqn 5eqn added the bug Something isn't working label May 6, 2024
5eqn added a commit to 5eqn/continue that referenced this issue May 6, 2024
5eqn added a commit to 5eqn/continue that referenced this issue May 6, 2024
@5eqn
Copy link
Contributor Author

5eqn commented May 6, 2024

Fixed by #1238

@5eqn 5eqn closed this as completed May 6, 2024
sestinj added a commit that referenced this issue May 8, 2024
* Add abstraction for fetch to easily allow using request options (#1059)

* add fetch helper function with request options

* add support for request options for Jira context provider

* Add a new slash command to review code. (#1071)

* Add a new slash command to review code.

* clean code

* 🩹 add new starcoder artifact as stopword

* 💄 slight improvements to inline edit UI

* 🔖 update default models, bump gradle version

* 📝 recommend starcoder2

* 🐛 fix jetbrains encoding issue

* 🩹 don't index site-packages

* 🩹 error handling in JetBrains

* 🐛 fix copy to clipboard in jetbrains

* fix: cursor focus issue causing unwanted return to text area (#1086)

* 📝 mention autocomplete in jetbrains

* 📝 Tab-autocomplete README

* 🔥 remove note about custom ctx providers only being on VS Code

* 📝 docs about http context provider

* 👥 pull request template

* Update from Claude 2 to Claude 3 (#1078)

* 📝 add FAQ about single-line completions

* 📝 update autocomplete docs

* fix cursor focus issue causing unwanted return to text area

---------

Co-authored-by: Nate Sesti <sestinj@gmail.com>
Co-authored-by: Ty Dunn <ty@tydunn.com>
Co-authored-by: Nate Sesti <33237525+sestinj@users.noreply.github.com>

* Update tree-sitter-wasms to 0.1.11 (which includes Solidity)

* Make use of solidity tree-sitter parser

* 🔧 option to disable autocomplete from config.json

* ✨ option to disable streaming with anthropic

* ✅ Test to verify that files are packaged

* Add FIM template for CodeGemma (#1097)

Also pass stop tokens to llama.cpp.

* ✨ customizable rerankers (#1088)

* ✨ customizable rerankers

* 💄 fix early truncation button

* ⚡️ improvements to full text search + reranking

* ⚡️ only use starcoder2 stop words for starcoder2

* ⚡️ crawl code graph for call expressions

* 🚧 starcoder2-7b free trial

* 🚧 free trial client for embeddings and re-ranking

* 🚧 embeddings provider

* ✅ test for presence of files in CI

* 🐛 fixes to reranking

* ✨ new onboarding experience

* ✨ new onboarding experience

* 💄 small tweaks to onboarding

* 🩹 add stopAtLines filter to /edit

* 🐛 clean up vite build errors

* 👷 make vscode external in binary build

* 💄 improved models onboarding for existing users

* 💄 default indexing progress to 0.0

* 🐛 small fixes to reranking

* 👷 clear folders before prepackage

* 👷 say where .vsix is output

* 👷 also download arm packages outside of gh actions

* 🎨 add AbortSignal to indexing

* 🔧 starcoder, not 2 in config_schema

* 🚚 again, starcoder, not 2

* 🐛 fix bug when reranker undefined

* 🩹 fix binary tsc error

* ✨ configure context menu prompts

* 🐛 acknowledge useLegacyCompletionsEndpoint

* 🚑 fix keep existing config option

* 🔊 learn about selection

* ⚡️ improvements to indexing reporting when not in git repo

* 🥅 handle situation where git doesn't exist in workspace

* ✨ support for gemini 1.5 pro

* 🐛 handle embeddingProvider name not found

* ✨ Gemini 1.5 and GPT-4 Turbo

* 👷 fix os, arch undefined in prepackage.js

* ⚡️ better detection of terminal code blocks

* 🧑‍💻 solve tailwind css warnings

* ✨ cmd/ctrl+L to select terminal contents

* 🐛 correctly handle remotes not found

* ✨ allow templating for custom commands

* 🔥 temporarily remove cmd+L to select terminal contents

* 🐛 remove quotes around Ollama stop words

* ✨ add Cohere as Model Provider (#1119)

* 🩹 add gpt-4-turbo to list of chat_only models

* feat: use exponential backoff in llm chat (#1115)

Signed-off-by: inimaz <93inigo93@gmail.com>

* 🩹 update exponential backoff timing

* 💄 spell out Alt in keyboard shortcuts

* 🩹 don't set edit prompt for templateType "none"

* Adds additional ignores for C-fmilies langs (#1129)

Ignored:
- cache directory `.cache`, used by clangd
- dependency files `*o.d`, used by object files
- LLVM and GNU coverage files: `*.profraw`, `*.gcda` and `*.gcno`

* 🔥 temporarily remove problematic expandSnippet import

* 👷 add npx to prefix vsce in build

* 🐛 handle messages sent in multiple parts over stdin

* 🔖 update gradle version

* 🩹 for now, skip onboarding in jetbrains

* 🩹 temporarily don't show use codebase on jetbrains

* 🐛 use system certificates in binary

* 🔖 update jetbrains version

* 🩹 correctly contruct set of certs

* 🔖 bump intellij version to 0.0.45

* 🩹 update to support images for gpt-4-turbo

* 🐛 fix image support autodetection

* ⚡️ again, improve image support autodetection

* 🐛 set supportsCompletions based on useLegacyCompletionsEndpoint model setting

Closes #1132

* 📝 useLegacyCompletionsEndpoint within OpenAI docs

* 🔧 forceCompletionsEndpointType option

* Revert "🔧 forceCompletionsEndpointType option"

This reverts commit dd51fcb.

* 🩹 set default useLegacyCompletionsEndpoint to undefined

* 🩹 look for bedrock credentials in homedir

* 🩹 use title for autodetect

* 🐛 set supportsCompletions based on useLegacyCompletionsEndpoint model setting

Closes #1132

* 📝 useLegacyCompletionsEndpoint within OpenAI docs

* 🩹 set default useLegacyCompletionsEndpoint to undefined

* 🩹 look for bedrock credentials in homedir

* 🩹 use title for autodetect

* Fix slash command params loading

Existing slash commands expect an object named
"params" so mapping to "options" here caused
params to be undefined within the run scope. I
renamed from 'm' to 's' just to avoid potential
confusion with the model property mapping above.

* Add outputDir param to /share

* Enable basic tilde expansion for /share outputDir

* Add ability to specify workspace for /share

* Add datetimestamp to exported session filename

* Use `.`, `./`, or `.\` for current workspace

* Add description of outputDir param for /share

* Ensure replacement only at start of string

* Create user-specified directory if necessary

* Change "Continue" to "Assistant" in export

* Consolidate to single replace regex

* Reformat markdown code blocks

Currently, user-selected code blocks are formatted
with range in file (rif) info on the same line as
the triple backticks, which means that when
exported to markdown they don't have the language
info needed on that line for syntax highlighting.
This update moves the rif info to the following
line as a comment in the language of the file and
with the language info in the correct place.

Before:
```example.ts (3-6)

function fib(n) {
    if (n <= 1) return n;
    return fib(n - 2) + fib(n - 1);
}
```

After:
```ts
// example.ts (3-6)

function fib(n) {
    if (n <= 1) return n;
    return fib(n - 2) + fib(n - 1);
}
```

* Tidy regex to capture filename

* Tidy regex to capture filename

* Ensure adjacent codeblocks separated by newline

* Aesthetic tweaks to output format

* ✨ disableInFiles option for autocomplete

* feat(httpContextProvider): load AC on fetch client (#1150)

Co-authored-by: Bertrand Pinel <bertrand.pinel@pole-emploi.fr>

* ✨ global filewatcher for config.json/ts changes

* 🐛 retry webview requests so that first cmd+L works

* ✨ Improved onboarding experience (#1155)

* 🚸 onboarding improvements

* 🧑‍💻 keyboard shortcuts to toggle autocomplete and open config.json

* ⚡️ improve detection of terminal code blocks

* 🚧 onboarding improvements

* 🚧 more onboarding improvements

* 💄 last session button

* 🚸 show more fallback options in dropdown

* 💄 add sectioning to models page

* 💄 clean up delete model button

* 💄 make tooltip look nicer

* 🚸 download Ollama button

* 💄 local LLM onboarding

* 🐛 select correct terminal on "runCommand" message

* 💄 polish onboarding

* 💚 fix gui build errors

* 📝 add /v1 to OpenAI examples in docs

* 🚑 hotfix for not iterable error

* ✨ add Cohere as Embeddings Provider

* 💄 add llama3 to UI

* 🔥 remove disable indexing

* 🍱 update continue logo

* 🐛 fix language undefined bug

* 🐛 fix merge mistake

* 📝 rename googlepalmapi.md to googlegeminiapi.md

* 📝 update mistral models

* Rename to geminiapi & change filename this time

* ✨ global request options (#1153)

* ✨ global request options

* 🐛 fix jira context provider by injecting fetch

* ✨ request options for embeddings providers

* ✨ add Cohere as Reranker (#1159)

* ♻️ use custom requestOptions with CohereEmbeddingsProvider

* Update preIndexedDocs.ts (#1154)

Add WordPress and WooCommerce as preIndexedDocs.

* 🩹 remove example "outputDir" from default config

* Fix slash command params loading (#1084)

Existing slash commands expect an object named
"params" so mapping to "options" here caused
params to be undefined within the run scope. I
renamed from 'm' to 's' just to avoid potential
confusion with the model property mapping above.

* 🐛 don't index if no open workspace folders

* 💄 improve onboarding language

* 🚸 improve onboarding

* 🐛 stop loading when error

* 💄 replace text in input box

* Respect Retry-After header when available from 429 responses (#1182)

* 🩹 remove dead code for exponential backoff

This has been replaced by the withExponentialBackoff helper

* 🩹 respect Retry-After header when available

* 🚸 update inline tips language

* ✨ input box history

* 📌 update package-locks

* 🔊 log errors in prepackage

* 🐛 err to string

* 📌 pin llama-tokenizer-js

* 📌 update lockfile

* 🚚 change /docs to docs.

* 📦 package win-ca dependencies in binary

* 🔥 remove unpopular models from UI

* 🍱 new logo in jetbrains

* 🎨 use node-fetch everywhere

* 🚸 immediately select newly added models

* 🚸 spell out Alt instead of using symbol

* 🔥 remove config shortcut

* 🐛 fix changing model bug

* 🩹 de-duplicate before adding models

* 🔧 add embeddingsProvider specific request options

* 🎨 refactor to always use node-fetch from LLM

* 🔥 remove duplicate tokens generated

* 🔊 add timestamp to JetBrains logs

* 🎨 maxStopWords for Groq

* 🐛 fix groq provider calling /completions

* 🐛 correctly adhere to LanceDB table name spec

* 🐛 fix sqlite NOT NULL constraint failed error with custom model

* Fix issue where Accept/Reject All only accepts/rejects a single diff hunk. (#1197)

* Fix issues parsing Ollama /api/show endpoint payloads. (#1199)

* ✨ model role for inlineEdit

* 🩹 various small updates

* 🐛 fix openai image support

* 🔖 update gradle version

* 🍱 update jetbrains icon

* 🐛 fix autocomplete in notebook cells

* 🔥 remove unused media

* 🔥 remove unused files

* Fix schema to allow 'AUTODETECT' sentinel for model when provider is 'groq'. (#1203)

* 🐛 small improvements

* Fix issue with @codebase provider when n becomes odd due to a divide by 2 during the full text search portion of the query. (#1204)

* 🐛 add skipLines

* ✨ URLContextProvider

* 🥅 improved error handling for codebase indexing

* 🏷️ use official Git extension types

* ➕ declare vscode.git extension dependency

* ⚡️ use reranker for docs context provider

* 🚸 Use templating in default customCommand

* 🎨 use U+23CE

* 🚸 disable autocomplete in commit message box

* 🩹 add gems to default ignored paths

* ⚡️ format markdown blocks as comments in .ipynb completions

* 🐛 don't strip port in URL

* 🐛 fix "gemini" provider spacing issues

* 📦 update posthog version

* 🏷️ update types.ts

* 🐛 fix copy/paste/cut behavior in VS Code notebooks

* ✨ llama3 prompt template

* 🐛 fix undefined prefix, suffix and language for `/edit` (#1216)

* 🐛 add .bind to fix templating in systemMessage

* 🐛 small improvements to autocomplete

* Update DocsContextProvider.ts (#1217)

I fixed a bug where you were sending the query variable (which holds the base URL of the doc) to the rerank method, and it made no sense to rerank the chunks based on a URL. So I changed it to extras.fullInput because it should rerank based on the user input, which should provide better results.

* 📝 select-provider.md update

* 🐛 fix merge errors

* Nate/autocomplete-metrics (#1230)

* ⚡️ use context.selectedCompletionInfo, deduplicate logs

* ⚡️ don't reject if user keeps typing same as completion

* ⚡️ vscode autocomplete edge cases

* 🚧 WIP on vscode autocomplete

* ⚡️ better bracket handlng

* ⚡️ improved multi-line detection

* Active file default context (#1231)

* 🚸 include currently active file by default

* 🚸 warn if non-autocomplete model being used

* ✨ try hole filling template for gpt

* 💄 ui for no context

* ⚡️ leave out bottom of excessively large files

* 🚧 experimenting with perplexity style streaming

* 🐛 fix #1237

* 💚 fix type error

* ⚡️ improve LSP usage in autocomplete

* 🐛 fix content parsing regression in /edit

* add PySide6 docs to preindexed docs (#1236)

* CON-232 bring custom docs to top, alphabetize doc results, make scrol… (#1239)

* CON-232 bring custom docs to top, alphabetize doc results, make scrollable

* CON-232 cleanup

---------

Co-authored-by: Justin Milner <jmilner@jmilner-lt2.deka.local>

* 🚚 [Auxiliary -> Continue] Sidebar

* 🔊 log completion options in ~/.continue/sessions

* ⚡️ filter out completions that are only punctuation/space

* ⚡️ inject intellisense docs, no multi-line on comments

* ⚡️ crawl type definitions for autocomplete

* ⚡️ truncate function text

* ⚡️ cache LSP calls

* ⚡️ find recently edited ranges with perfect prefix match

* 🐛 fix gif paths

* ⚡️ bring back double new line stop words

* 📌 add yarn lock files

* 🐛 allow language keywords to be generated

* 💄 toggle on help button

* 🎨 defaultContext option

* 🐛 fix lancedb bug by upgrading

* 🐛 fix groq stop tokens

* 🐛 preventDefault to avoid double paste

* 🚸 don't repeatedly override cmd+J

* 🧑‍💻 fix npm run test in core

* 📝 change description

* 🐛 silence Ollama invalid server state warning

---------

Signed-off-by: inimaz <93inigo93@gmail.com>
Co-authored-by: Tobias Jung <102594442+tobiajung@users.noreply.github.com>
Co-authored-by: Jason Jacobs <nerfnerd@gmail.com>
Co-authored-by: Nithish <83941930+Nithishvb@users.noreply.github.com>
Co-authored-by: Ty Dunn <ty@tydunn.com>
Co-authored-by: Riccardo Schirone <riccardo.schirone@trailofbits.com>
Co-authored-by: postmasters <namnguyen@google.com>
Co-authored-by: Maxime Brunet <max@brnt.mx>
Co-authored-by: inimaz <49730431+inimaz@users.noreply.github.com>
Co-authored-by: SR_team <me@sr.team>
Co-authored-by: Roger Meier <r.meier@siemens.com>
Co-authored-by: Peter Zaback <pzaback@gmail.com>
Co-authored-by: Bertrand P <49525332+Berber31@users.noreply.github.com>
Co-authored-by: Bertrand Pinel <bertrand.pinel@pole-emploi.fr>
Co-authored-by: Jose Vega <bloguea.y.gana@gmail.com>
Co-authored-by: Nejc Habjan <hab.nejc@gmail.com>
Co-authored-by: Chad Yates <cyates@dynfxdigital.com>
Co-authored-by: 小颚虫 <123357481+5eqn@users.noreply.github.com>
Co-authored-by: 5eqn <491100866@qq.com>
Co-authored-by: Pixel <54180211+pixelsortr@users.noreply.github.com>
Co-authored-by: Justin Milner <42585006+justinmilner1@users.noreply.github.com>
Co-authored-by: Justin Milner <jmilner@jmilner-lt2.deka.local>
sestinj added a commit that referenced this issue May 17, 2024
* 🩹 for now, skip onboarding in jetbrains

* 🩹 temporarily don't show use codebase on jetbrains

* 🐛 use system certificates in binary

* 🔖 update jetbrains version

* 🩹 correctly contruct set of certs

* 🔖 bump intellij version to 0.0.45

* 🩹 update to support images for gpt-4-turbo

* 🐛 fix image support autodetection

* ⚡️ again, improve image support autodetection

* 🐛 set supportsCompletions based on useLegacyCompletionsEndpoint model setting

Closes #1132

* 📝 useLegacyCompletionsEndpoint within OpenAI docs

* 🔧 forceCompletionsEndpointType option

* Revert "🔧 forceCompletionsEndpointType option"

This reverts commit dd51fcb.

* 🩹 set default useLegacyCompletionsEndpoint to undefined

* 🩹 look for bedrock credentials in homedir

* 🩹 use title for autodetect

* 🐛 set supportsCompletions based on useLegacyCompletionsEndpoint model setting

Closes #1132

* 📝 useLegacyCompletionsEndpoint within OpenAI docs

* 🩹 set default useLegacyCompletionsEndpoint to undefined

* 🩹 look for bedrock credentials in homedir

* 🩹 use title for autodetect

* Fix slash command params loading

Existing slash commands expect an object named
"params" so mapping to "options" here caused
params to be undefined within the run scope. I
renamed from 'm' to 's' just to avoid potential
confusion with the model property mapping above.

* Add outputDir param to /share

* Enable basic tilde expansion for /share outputDir

* Add ability to specify workspace for /share

* Add datetimestamp to exported session filename

* Use `.`, `./`, or `.\` for current workspace

* Add description of outputDir param for /share

* Ensure replacement only at start of string

* Create user-specified directory if necessary

* Change "Continue" to "Assistant" in export

* Consolidate to single replace regex

* Reformat markdown code blocks

Currently, user-selected code blocks are formatted
with range in file (rif) info on the same line as
the triple backticks, which means that when
exported to markdown they don't have the language
info needed on that line for syntax highlighting.
This update moves the rif info to the following
line as a comment in the language of the file and
with the language info in the correct place.

Before:
```example.ts (3-6)

function fib(n) {
    if (n <= 1) return n;
    return fib(n - 2) + fib(n - 1);
}
```

After:
```ts
// example.ts (3-6)

function fib(n) {
    if (n <= 1) return n;
    return fib(n - 2) + fib(n - 1);
}
```

* Tidy regex to capture filename

* Tidy regex to capture filename

* Ensure adjacent codeblocks separated by newline

* Aesthetic tweaks to output format

* ✨ disableInFiles option for autocomplete

* feat(httpContextProvider): load AC on fetch client (#1150)

Co-authored-by: Bertrand Pinel <bertrand.pinel@pole-emploi.fr>

* ✨ global filewatcher for config.json/ts changes

* 🐛 retry webview requests so that first cmd+L works

* ✨ Improved onboarding experience (#1155)

* 🚸 onboarding improvements

* 🧑‍💻 keyboard shortcuts to toggle autocomplete and open config.json

* ⚡️ improve detection of terminal code blocks

* 🚧 onboarding improvements

* 🚧 more onboarding improvements

* 💄 last session button

* 🚸 show more fallback options in dropdown

* 💄 add sectioning to models page

* 💄 clean up delete model button

* 💄 make tooltip look nicer

* 🚸 download Ollama button

* 💄 local LLM onboarding

* 🐛 select correct terminal on "runCommand" message

* 💄 polish onboarding

* 💚 fix gui build errors

* 📝 add /v1 to OpenAI examples in docs

* 🚑 hotfix for not iterable error

* ✨ add Cohere as Embeddings Provider

* 💄 add llama3 to UI

* 🔥 remove disable indexing

* 🍱 update continue logo

* 🐛 fix language undefined bug

* 🐛 fix merge mistake

* 📝 rename googlepalmapi.md to googlegeminiapi.md

* 📝 update mistral models

* Rename to geminiapi & change filename this time

* ✨ global request options (#1153)

* ✨ global request options

* 🐛 fix jira context provider by injecting fetch

* ✨ request options for embeddings providers

* ✨ add Cohere as Reranker (#1159)

* ♻️ use custom requestOptions with CohereEmbeddingsProvider

* Update preIndexedDocs.ts (#1154)

Add WordPress and WooCommerce as preIndexedDocs.

* 🩹 remove example "outputDir" from default config

* Fix slash command params loading (#1084)

Existing slash commands expect an object named
"params" so mapping to "options" here caused
params to be undefined within the run scope. I
renamed from 'm' to 's' just to avoid potential
confusion with the model property mapping above.

* 🐛 don't index if no open workspace folders

* 💄 improve onboarding language

* 🚸 improve onboarding

* 🐛 stop loading when error

* 💄 replace text in input box

* Respect Retry-After header when available from 429 responses (#1182)

* 🩹 remove dead code for exponential backoff

This has been replaced by the withExponentialBackoff helper

* 🩹 respect Retry-After header when available

* 🚸 update inline tips language

* ✨ input box history

* 📌 update package-locks

* 🔊 log errors in prepackage

* 🐛 err to string

* 📌 pin llama-tokenizer-js

* 📌 update lockfile

* 🚚 change /docs to docs.

* 📦 package win-ca dependencies in binary

* 🔥 remove unpopular models from UI

* 🍱 new logo in jetbrains

* 🎨 use node-fetch everywhere

* 🚸 immediately select newly added models

* 🚸 spell out Alt instead of using symbol

* 🔥 remove config shortcut

* 🐛 fix changing model bug

* 🩹 de-duplicate before adding models

* 🔧 add embeddingsProvider specific request options

* 🎨 refactor to always use node-fetch from LLM

* 🔥 remove duplicate tokens generated

* 🔊 add timestamp to JetBrains logs

* 🎨 maxStopWords for Groq

* 🐛 fix groq provider calling /completions

* 🐛 correctly adhere to LanceDB table name spec

* 🐛 fix sqlite NOT NULL constraint failed error with custom model

* Fix issue where Accept/Reject All only accepts/rejects a single diff hunk. (#1197)

* Fix issues parsing Ollama /api/show endpoint payloads. (#1199)

* ✨ model role for inlineEdit

* 🩹 various small updates

* 🐛 fix openai image support

* 🔖 update gradle version

* 🍱 update jetbrains icon

* 🐛 fix autocomplete in notebook cells

* 🔥 remove unused media

* 🔥 remove unused files

* Fix schema to allow 'AUTODETECT' sentinel for model when provider is 'groq'. (#1203)

* 🐛 small improvements

* Fix issue with @codebase provider when n becomes odd due to a divide by 2 during the full text search portion of the query. (#1204)

* 🐛 add skipLines

* ✨ URLContextProvider

* 🥅 improved error handling for codebase indexing

* 🏷️ use official Git extension types

* ➕ declare vscode.git extension dependency

* ⚡️ use reranker for docs context provider

* 🚸 Use templating in default customCommand

* 🎨 use U+23CE

* 🚸 disable autocomplete in commit message box

* 🩹 add gems to default ignored paths

* ⚡️ format markdown blocks as comments in .ipynb completions

* 🐛 don't strip port in URL

* 🐛 fix "gemini" provider spacing issues

* 📦 update posthog version

* CON-1067: Failed state seems to be toggling as intended

* 🏷️ update types.ts

* 🐛 fix copy/paste/cut behavior in VS Code notebooks

* ✨ llama3 prompt template

* Rework for proper initialization on start up

* CON-1067 Clean-up

* CON-1067 more clean-up

* Add indexingNotLoaded state

* CON-1067 communicate progress to frontend

* 🐛 fix undefined prefix, suffix and language for `/edit` (#1216)

* 🐛 add .bind to fix templating in systemMessage

* CON-1067 clean up

* 🐛 small improvements to autocomplete

* Update DocsContextProvider.ts (#1217)

I fixed a bug where you were sending the query variable (which holds the base URL of the doc) to the rerank method, and it made no sense to rerank the chunks based on a URL. So I changed it to extras.fullInput because it should rerank based on the user input, which should provide better results.

* 📝 select-provider.md update

* 🐛 fix merge errors

* Nate/autocomplete-metrics (#1230)

* ⚡️ use context.selectedCompletionInfo, deduplicate logs

* ⚡️ don't reject if user keeps typing same as completion

* ⚡️ vscode autocomplete edge cases

* 🚧 WIP on vscode autocomplete

* ⚡️ better bracket handlng

* ⚡️ improved multi-line detection

* Active file default context (#1231)

* 🚸 include currently active file by default

* 🚸 warn if non-autocomplete model being used

* ✨ try hole filling template for gpt

* 💄 ui for no context

* ⚡️ leave out bottom of excessively large files

* 🚧 experimenting with perplexity style streaming

* 🐛 fix #1237

* 💚 fix type error

* ⚡️ improve LSP usage in autocomplete

* 🐛 fix content parsing regression in /edit

* add PySide6 docs to preindexed docs (#1236)

* CON-232 bring custom docs to top, alphabetize doc results, make scrol… (#1239)

* CON-232 bring custom docs to top, alphabetize doc results, make scrollable

* CON-232 cleanup

---------

Co-authored-by: Justin Milner <jmilner@jmilner-lt2.deka.local>

* CON-1067 condense some things

* 🚚 [Auxiliary -> Continue] Sidebar

* 🔊 log completion options in ~/.continue/sessions

* CON-1067 wrong ret val fix

* CON-1067: fixes from testing

* ⚡️ filter out completions that are only punctuation/space

* ⚡️ inject intellisense docs, no multi-line on comments

* ⚡️ crawl type definitions for autocomplete

* ⚡️ truncate function text

* ⚡️ cache LSP calls

* ⚡️ find recently edited ranges with perfect prefix match

* 🐛 fix gif paths

* ⚡️ bring back double new line stop words

* 📌 add yarn lock files

* 🐛 allow language keywords to be generated

* 💄 toggle on help button

* 🎨 defaultContext option

* 🐛 fix lancedb bug by upgrading

* 🐛 fix groq stop tokens

* 🐛 preventDefault to avoid double paste

* 🚸 don't repeatedly override cmd+J

* 🧑‍💻 fix npm run test in core

* 📝 change description

* 🐛 silence Ollama invalid server state warning

* ⚡️ more accurate getTerminalContents

* ⚡️ make getTerminalContents more accurate

* 🧑‍💻 use yarn instead of npm

* 👷 fix yarn add --no-save in prepackge

* 🐛 correctly read entire notebook file contents

* ➕ import handlebars

* 🔥 remove unnecessary migrations

* ⚡️ improve /comment prompt

* Add debug terminal context menu (#1261)

* Add --no-dependencies to vsce package (#1255)

This is not needed because we bundle first with esbuild, and
vsce pack has issues with modern package managers.

see: microsoft/vscode-vsce#421 (comment)

* ui: change line decoration color to use vscode theme (#1253)

* ui: change line decoration color to use vscode theme

to give user a more consistent experience by letting the decoration color to user the color defined in the theme.

* fix: incorrect color item

should be line background not text background
because the decoration is for the whole line

* 🎨 refactor indexing state into a single object

* CON-223 Correct diff streaming abort (#1263)

Co-authored-by: Justin Milner <jmilner@jmilner-lt2.deka.local>

* 📦 switch to pnpm (#1265)

* 🎨 use pnpm instead of yarn

* ➕ make dependencies explicit for pnpm

* 🐛 add powershell to extension mapping, and default to ext

* 🎨 make llamatokenizer commonjs compatible

* ➕ add esbuild optional deps

* 🚚 rename vendor/node_modules -> modules

* 🔖 update core version

* 🐛 fixes for transformers.js compatibility

* 🔖 update core version

* 🎨 set modelPath in constructor

* 🎨 fix transformers.js import

* 🎨 eslint enforce import extensions

* 🎨 require -> import

* 🚸 notify user if not diff in /commit

* 💄 Improve colors of the IntelliJ tool window icon (#1273)

Without this fix, the continue icon sticks out from the other toolwindow icons,
resulting in an inconsistent appearance of the whole IDE and creates a feeling
that the continue plugin "doesn't fit, something must be broken".

According to
https://plugins.jetbrains.com/docs/intellij/icons.html#new-ui-icon-colors
specific colors are needed to work nicely with dark and light modes. Bonus is
that the active tool window icon color then changes automatically to white.

Co-authored-by: Lukas Baron <LukasBaron@gmail.com>

* ✨ send Bearer token to Ollama if apiKey set

* 🐛 fix slash command bug

* 🧑‍💻 add onnxruntime (--save-dev) for types

* 🐛 don't apply to file when running code block in terminal

* 🐛 avoid double paste

* ✨ gpt-4o

* 🎨 pass uniqueId to constructor

* 🚸 focus without scrolling into vie

* 🎨 refactoring for continue-proxy LLM (#1277)

* 🐛 continue server client fixes

* 🎨 refactor OpenAI _getHeaders

* 🎨 pass ideSettings to llmFromDescription

* ⚡️ improve add docstring command

* 💚 ci updates

* 🐛 fix repeated paste bug

* 🐛 fix build script

* 🩹 various small improvements

* 📌 pin esbuild 0.17.19

* 🧑‍💻 pnpm i gui in prepackage

* 🐛 show all diff changes in vscode

* 🩹 getMetaKeyName

* 🐛 fix reading of unopened ipynb files

* ⚡️ gpt-4o system prompt

* 💄 make font size configurable in config.json ui.fontSize

* 🩹 properly dispose of diff handler

* 🐛 fix indexing status display

* ⚡️ context pruning

* 🎨 update free trial models

* fix: remove some backup files generated by pkg if present (#1287)

---------

Co-authored-by: Roger Meier <r.meier@siemens.com>
Co-authored-by: Peter Zaback <pzaback@gmail.com>
Co-authored-by: Bertrand P <49525332+Berber31@users.noreply.github.com>
Co-authored-by: Bertrand Pinel <bertrand.pinel@pole-emploi.fr>
Co-authored-by: Maxime Brunet <max@brnt.mx>
Co-authored-by: Jose Vega <bloguea.y.gana@gmail.com>
Co-authored-by: Nejc Habjan <hab.nejc@gmail.com>
Co-authored-by: Chad Yates <cyates@dynfxdigital.com>
Co-authored-by: Justin Milner <jmilner@jmilner-lt2.deka.local>
Co-authored-by: 小颚虫 <123357481+5eqn@users.noreply.github.com>
Co-authored-by: 5eqn <491100866@qq.com>
Co-authored-by: Pixel <54180211+pixelsortr@users.noreply.github.com>
Co-authored-by: Justin Milner <42585006+justinmilner1@users.noreply.github.com>
Co-authored-by: Devin Gould <djgould0628@gmail.com>
Co-authored-by: Dipta Mahardhika <146386738+d-mahard@users.noreply.github.com>
Co-authored-by: tnglemongrass <113173292+tnglemongrass@users.noreply.github.com>
Co-authored-by: Lukas Baron <LukasBaron@gmail.com>
Co-authored-by: Fernando <fernando.sanchez.jr@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working
Projects
None yet
Development

No branches or pull requests

1 participant