Commit Graph
6 Commits
Author SHA1 Message Date
Richard Feldman d7004030b3 Code block evals (#29619)
Add a targeted eval for code block formatting, and revise the system
prompt accordingly.

### Eval before, n=8

<img width="728" alt="eval before"
src="https://github.com/user-attachments/assets/552b6146-3d26-4eaa-86f9-9fc36c0cadf2"
/>

### Eval after prompt change, n=8 (excluding the new evals, so just
testing the prompt change)

<img width="717" alt="eval after"
src="https://github.com/user-attachments/assets/c78c7a54-4c65-470c-b135-8691584cd73e"
/>

Release Notes:

- N/A
2025-04-29 18:52:09 -04:00
Agus Zubiaga fd17f2d8ae agent: Enrich grep tool output with syntax information (#29601)
The `grep` tool used to include 4 lines of context around the match, but
the lines included would often be unhelpful. This PR improves this
behavior by using the range of the parent syntax node that contains the
full line(s) matched.

The match headers will also now include symbol breadcrumbs so that the
model can already gather code structure before/without reading files.

````md
### impl GitRepository for RealGitRepository › fn compare_checkpoints › L1278-1284
```rust
                let result = git
                    .run(&[
                        "diff-tree",
                        "--quiet",
                        &left.commit_sha.to_string(),
                        &right.commit_sha.to_string(),
                    ])
```
````

This positively impacts the `add_arg_to_trait_method` eval example with
better diff output, fewer tool failures, and reduced total turns.

Note: We have some plans to use a an "elision" approach where we would
combine all matches for a given file, skipping lines between them while
keeping symbol declaration lines. The theory is that this would be map
more closely to the expected input for edits. For now, this PR is a
significant improvement.

Release Notes:

- Agent: Enrich `grep` tool output with syntax information
2025-04-29 17:03:02 +00:00
Max Brunsfeld f125353b6f Add tree-sitter example to the eval (#29321)
Interesting things about this example:
* It's a useful, non-trivial change I made with the agent in Tree-sitter
* It runs fast
* It frequently showcases edit file errors
* It occasionally completely errors out due to errors parsing tool call
input JSON

Release Notes:

- N/A
2025-04-23 18:46:38 -07:00
Agus Zubiaga 8b5835de17 agent: Improve initial file search quality (#29317)
This PR significantly improves the quality of the initial file search
that occurs when the model doesn't yet know the full path to a file it
needs to read/edit.

Previously, the assertions in file_search often failed on main as the
model attempted to guess full file paths. On this branch, it reliably
calls `find_path` (previously `path_search`) before reading files.

After getting the model to find paths first, I noticed it would try
using `grep` instead of `path_search`. This motivated renaming
`path_search` to `find_path` (continuing the analogy to unix commands)
and adding system prompt instructions about proper tool selection.

Note: I know the command is just called `find`, but that seemed too
general.

In my eval runs, the `file_search` example improved from 40% ± 10% to
98% ± 2%. The only assertion I'm seeing occasionally fail is "glob
starts with `**` or project". We can probably add some instructions in
that regard.

Release Notes:

- N/A
2025-04-23 21:24:41 -03:00
Agus ZubiagaandRichard Feldman 45d3f5168a eval: New add_arg_to_trait_method example (#29297)
Release Notes:

- N/A

---------

Co-authored-by: Richard Feldman <oss@rtfeldman.com>
2025-04-23 18:46:39 +00:00
ce1a674eba eval: Fine-grained assertions (#29246)
- Support programmatic examples
([example](https://github.com/zed-industries/zed/blob/17feb260a0919e102ae4e220669c467885bf4b57/crates/eval/src/examples/file_search.rs))
- Combine data-driven example declarations into a single `.toml` file
([example](https://github.com/zed-industries/zed/blob/17feb260a0919e102ae4e220669c467885bf4b57/crates/eval/src/examples/find_and_replace_diff_card.toml))
- Run judge on individual assertions (previously called "criteria")
- Report judge and programmatic assertions in one combined table

Note: We still need to work on concept naming 

<img width=400
src="https://github.com/user-attachments/assets/fc719c93-467f-412b-8d47-68821bd8a5f5">

Release Notes:

- N/A

---------

Co-authored-by: Richard Feldman <oss@rtfeldman.com>
Co-authored-by: Max Brunsfeld <maxbrunsfeld@gmail.com>
Co-authored-by: Thomas Mickley-Doyle <tmickleydoyle@gmail.com>
2025-04-22 23:58:58 -03:00