divanvisagie / csep Goto Github PK

Fastembed looks like it is slower on a single embedding, however since it has a batch function, it might serve better for multiple embeds or at least the overhead might be worth he performance gain on larger context sets.

https://crates.io/crates/fastembed

✦ ❯ hyperfine --warmup 3 target/release/emby # Results on hello world string fastembed
Benchmark 1: target/release/emby
  Time (mean ± σ):     515.1 ms ±   2.8 ms    [User: 1100.6 ms, System: 210.6 ms]
  Range (min … max):   511.4 ms … 520.1 ms    10 runs


emby on  master [?] is 📦 v0.1.0 via 🦀 v1.79.0 took 6s
✦ ❯ hyperfine --warmup 3 ./test.sh # results on hello world string using curl to ollama
Benchmark 1: ./test.sh
  Time (mean ± σ):     115.4 ms ±  31.1 ms    [User: 12.0 ms, System: 13.8 ms]
  Range (min … max):    81.5 ms … 168.0 ms    36 runs

Add the option to select between using ollama and openai api

We have a lot of errors, one of which can be that it crashes the running instance of ollama.

The first thing we can look at is ignoring any files that are binary explicitly.

Description:
When the output of the csep command is generated, the similarity results should be listed in descending order of similarity rather than in what appears to be a random order. This will improve the usability by allowing users to see the most relevant results first without needing additional sorting.

Current Behavior:

• The similarity results appear to be listed in a random order when the command is executed.

Expected Behavior:

• The similarity results should be ordered from the highest similarity to the lowest, making it easier for users to identify the most relevant matches quickly.

Example:
If the user search query is "example query", the output should display results sorted by the highest similarity first:

Result 1: [highest similarity]
Result 2: [second highest similarity]
...
Result N: [lowest similarity]

Add a csep init command which will create a .csep directory which will store an embedidngs cache (inspired by the .git directory)