Skip to content

Latest commit

 

History

History
138 lines (101 loc) · 5.34 KB

todo.md

File metadata and controls

138 lines (101 loc) · 5.34 KB
Raw

--- v0.0.1

[ok] - Test error handling when server fails [ok] - Create settings control (modal) [ok] - Edit README [ok] - Use form and state values in settings form [ok] - Check how to share state between backend and frontend [ok] - Make parameter selection form [ok] - Verify if Atom values for grid params can be scalar or always converted to lists (always convert to lists) [ok] - Disable submit when queries are running (maybe add spinning icon) [ok] - Improve collapsible that displays params, and other metadata [ok] - Segregate spinner into its own component [ok] - Add loader/spinner inside each result component [ok] - Add fixed header to results pannel with controls to expand/collapse metadata in all inferences [ok] - Add button to cancel/stop experiment [ok] - Add confirmation dialog when stopping experiment [ok] - Use Ollama server from settings in get_models and get inference [ok] - refetch get_models periodically or add icon to refetch -> every 30s [ok] - Use system prompt + user defined params and default params when invoking the completion API [ok] - Move commands and utility functions to commands.rs and lib.rs [ok] - Make desktop the default app, save old one [ok] - Rename "desktop" id in Cargo.toml to grid-search-desktop [ok] - Return full response from inference (after ollama-rs version bump) [ok] - Add response metadata to results [ok] - Verify if upstream issues with the "stop" option have been resolved (ommited until then) [ok] - Changing server_url should preserve form data and current results. Just reload available models and default to first [ok] - Format result's metadata in the frontend [ok] - Redo README (add pics!) [ok] - Test a multiplatform release

--- v0.0.2

[ok] - Add "tutorial" to start page [ok] - Add "repeat_last_n" param to inferences

--- v0.0.3

[ok] - In response metadata, use total duration to calculate throughput

--- v0.1.0 [ok] - Log experiments to JSON files

--- v0.1.1 [ok] - Use the "single-instance" plugin to prevent multiple concurrent instances of the app https://github.com/tauri-apps/plugins-workspace/tree/v1/plugins/single-instance

[ok] - Add option to display large textarea for prompt in a modal (check why it updates the experiment id in configs)

--- v0.1.2 [ok] - Add keyboard shortcuts [ok] - input validation should check for commas as separators. [ok] - "Full screen" prompt input

--- v0.2.0. [ok] - Added "request_timeout" parameter to settings

--- v0.2.1 [ok] - Supports user input for adicional parameters: tfs_z, mirostat, mirostat_eta and mirostat_tau.

--- v0.2.2 [ok] - Adds the application's version number to the settings interfac [ok] - Allows inspecting past experiments without downloading them

--- v0.3.0

[ok] - Calculate inference time and use this value to calc throughput. [ok] - Display Ollama's version next to the selected server [ok] - Check for slashes in server url [ok] - refetch version when config changes

[ok] - Allow multiple generations for each param combination

--- v0.4.0

[ok] - PRs: UI improvements [ok] - Add system prompt to main params form [ok] - Fix issue with date of the experiment being updated on re-renders [ok] - Present experiment inspection output in human readable format instead of JSON [ok] - Add versioning to the JSON log files, starting at this release's version.

--- v0.5.0

[ok] - Multiple prompts [ok] - Concurrent inferences [ok] - Hide/Show model names [ok] - Present experiment inspection output in human readable format instead of JSON [ok] - Add versioning to the JSON log files, starting at this release's version.

--- v0.5.1 [ok] - Fixes generation responses not returning metadata (like eval_duration, total_duration, eval_count). [ok] - Added Rust CI [ok] - Added Clippy checks when saving Rust code [ok] - Corrected existing Rust code to pass Clippy checks [ok] - Fixed padding in "Expand/Hide" buttons for params and metadate [ok] - Improved UI for component that displays inference parameters with collapsible prompts

--- v0.5.2 [ok] - Adds custom application icon [ok] - Handles Ollama version info not being correctly returned by the server

--- v0.5.3 [ok] - Handles Ollama servers using default ports (80 or 443)

  • Allow re-running past experiments Let's call this cloning an experiment. What happens if a model is not available anymore (skip/alert?)

    • alerting the model is not available, so the user can pull it again and retry before running.

    when hitting a "clone" button on an experiment entry, we just update the form and let the user change stuff or press submit without changing anything.

    This will result in a new experiment.

  • Add CI checks for typescript code (ESLint?)

  • Build options straight from config: https://github.com/pepperoni21/ollama-rs/blob/master/examples/options_from_json.rs

  • Add a damn icon.

  • Add feature to clear all existing log files.

  • Convert code snippets to markdown in responses? (react-markdown didn't work well to filter code)

  • Plan on saving results to SQLite

  • Add a "default prompt" setting to settings?

  • Add a way to filter visible results by params, or length or throughput?

  • Cancel async commands: tauri-apps/tauri#5894 tauri-apps/tauri#8351

  • Use new deserialize method to handle options: https://github.com/pepperoni21/ollama-rs/blob/master/examples/options_from_json.rs

  • Implement limited concurrency options (nope... ollama runs one request at a time)