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

Add Semantic Caching #16

Merged
merged 1 commit into from
Jul 10, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,12 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0

### Fixed

## [0.4.0]

### Added
- Added a launcher function `launch` to make it easier to launch the app.
- Semantic caching enabled by SemanticCaches.jl. You can change it by setting `cached=false` in the `launch()` function.

## [0.3.0]

### Added
Expand Down
8 changes: 6 additions & 2 deletions Project.toml
Original file line number Diff line number Diff line change
@@ -1,21 +1,25 @@
name = "ProToPortal"
uuid = "f9496bd6-a3bb-4afc-927d-7268532ebfa9"
authors = ["J S <49557684+svilupp@users.noreply.github.com> and contributors"]
version = "0.3.0"
version = "0.4.0"

[deps]
Base64 = "2a0f44e3-6c83-55bd-87e4-b1978d98bd5f"
Dates = "ade2ca70-3891-5945-98fb-dc099432e06a"
GenieFramework = "a59fdf5c-6bf0-4f5d-949c-a137c9e2f353"
GenieSession = "03cc5b98-4f21-4eb6-99f2-22eced81f962"
HTTP = "cd3eb016-35fb-5094-929b-558a96fad6f3"
PromptingTools = "670122d1-24a8-4d70-bfce-740807c42192"
SemanticCaches = "03ba8f0e-aaaa-4626-a19b-56297996781b"

[compat]
Aqua = "0.7"
Dates = "<0.0.1, 1"
GenieFramework = "2.1"
GenieSession = "1"
PromptingTools = "0.33"
HTTP = "1"
PromptingTools = "0.37.1"
SemanticCaches = "0.2"
Test = "<0.0.1, 1"
julia = "1.10"

Expand Down
6 changes: 5 additions & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -32,11 +32,15 @@ using Pkg; Pkg.activate("."); Pkg.instantiate(".")
```julia
# as a quick hack if you don't have your environment variables set up, run the below line with your OpenAI key
# ENV["OPENAI_API_KEY"] = "<your_openai_api_key>"
include("main.jl")
ENV["DATADEPS_ALWAYS_ACCEPT"] = "true" # required for caching
using ProToPortal
launch(; cached = true)
```

Then head to your browser and go to [http://127.0.0.1:8000](http://127.0.0.1:8000) to see the app.

It will now cache similar LLM requests by default (disable with `cached=false` in `launch()` function).

For the purists: simply run `julia --project -t auto main.jl` in your terminal (once installed)!

How to start? Type `Say hi!` in the question box on the Chat tab and click Submit (or press CTRL+ENTER).
Expand Down
7 changes: 5 additions & 2 deletions docs/src/index.md
Original file line number Diff line number Diff line change
Expand Up @@ -54,8 +54,11 @@ It's the first Julia-focused GUI (evaluate Julia code, fix it, critique it - or
Clone ProToPortal, instantiate it, enable your desired settings, and streamline your LLM interactions right away:

```julia
using Pkg; Pkg.activate("."); Pkg.instantiate(".")
include("main.jl")
# as a quick hack if you don't have your environment variables set up, run the below line with your OpenAI key
# ENV["OPENAI_API_KEY"] = "<your_openai_api_key>"
ENV["DATADEPS_ALWAYS_ACCEPT"] = "true" # required for caching
using ProToPortal
launch(; cached = true)
```

Then head to your browser and go to [http://127.0.0.1:8000](http://127.0.0.1:8000) to see the app.
Expand Down
6 changes: 5 additions & 1 deletion docs/src/introduction.md
Original file line number Diff line number Diff line change
Expand Up @@ -27,11 +27,15 @@ using Pkg; Pkg.activate("."); Pkg.instantiate(".")
```julia
# as a quick hack if you don't have your environment variables set up, run the below line with your OpenAI key
# ENV["OPENAI_API_KEY"] = "<your_openai_api_key>"
include("main.jl")
ENV["DATADEPS_ALWAYS_ACCEPT"] = "true" # required for caching
using ProToPortal
launch(; cached = true)
```

Then head to your browser and go to [http://127.0.0.1:8000](http://127.0.0.1:8000) to see the app.

It will now cache similar LLM requests by default (disable with `cached=false` in `launch()` function).

For the purists: simply run `julia --project -t auto main.jl` in your terminal (once installed)!

How to start? Type `Say hi!` in the question box on the Chat tab and click Submit (or press CTRL+ENTER).
Expand Down
Binary file added docs/src/videos/screen-capture-code-fixing.gif
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added docs/src/videos/screen-capture-plain.webm
Binary file not shown.
11 changes: 4 additions & 7 deletions main.jl
Original file line number Diff line number Diff line change
@@ -1,9 +1,6 @@
using Pkg
Pkg.activate(".")
using GenieFramework
ENV["GENIE_HOST"] = "127.0.0.1"
ENV["PORT"] = "8000"
## ENV["GENIE_ENV"] = "prod"
include("app.jl") # hack for hot-reloading when fixing things
Genie.loadapp();
up(async = true);
## Required to support semantic caching
ENV["DATADEPS_ALWAYS_ACCEPT"] = "true"
using ProToPortal
ProToPortal.launch(8000, "0.0.0.0"; async = false, cached = true, cache_verbose = true)
81 changes: 81 additions & 0 deletions src/CacheLayer.jl
Original file line number Diff line number Diff line change
@@ -0,0 +1,81 @@
## Define the new caching mechanism as a layer for HTTP
## See documentation [here](https://juliaweb.github.io/HTTP.jl/stable/client/#Quick-Examples)
"""
CacheLayer

A module providing caching of LLM requests for ProToPortal.

It caches 3 URL paths:
- `/v1/chat/completions` (for OpenAI API)
- `/v1/embeddings` (for OpenAI API)
- `/v1/rerank` (for Cohere API)

# How to use
You can use the layer directly
`CacheLayer.get(req)`

You can push the layer globally in all HTTP.jl requests
`HTTP.pushlayer!(CacheLayer.cache_layer)`

You can remove the layer later
`HTTP.poplayer!()`

"""
module CacheLayer

using SemanticCaches, HTTP
using PromptingTools: JSON3

const SEM_CACHE = SemanticCache()
const HASH_CACHE = HashCache()

function cache_layer(handler)
return function (req; kw...)
VERBOSE = Base.get(ENV, "CACHES_VERBOSE", "true") == "true"
if req.method == "POST" && !isempty(req.body)
body = JSON3.read(copy(req.body))
## chat/completions is for OpenAI, v1/messages is for Anthropic
if occursin("v1/chat/completions", req.target) ||
occursin("v1/messages", req.target)
## We're in chat completion endpoint
temperature_str = haskey(body, :temperature) ? body[:temperature] : "-"
cache_key = string("chat-", body[:model], "-", temperature_str)
input = join([m["content"] for m in body[:messages]], " ")
elseif occursin("v1/embeddings", req.target)
cache_key = string("emb-", body[:model])
## We're in embedding endpoint
input = join(body[:input], " ")
elseif occursin("v1/rerank", req.target)
cache_key = string("rerank-", body[:model], "-", body[:top_n])
input = join([body[:query], body[:documents]...], " ")
else
## Skip, unknown API
VERBOSE && @info "Skipping cache for $(req.method) $(req.target)"
return handler(req; kw...)
end
## Check the cache

VERBOSE && @info "Check if we can cache this request ($(length(input)) chars)"
active_cache = length(input) > 5000 ? HASH_CACHE : SEM_CACHE
item = active_cache(cache_key, input; verbose = 2 * VERBOSE) # change verbosity to 0 to disable detailed logs
if !isvalid(item)
VERBOSE && @info "Cache miss! Pinging the API"
# pass the request along to the next layer by calling `cache_layer` arg `handler`
resp = handler(req; kw...)
item.output = resp
# Let's remember it for the next time
push!(active_cache, item)
end
## Return the calculated or cached result
return item.output
end
# pass the request along to the next layer by calling `cache_layer` arg `handler`
# also pass along the trailing keyword args `kw...`
return handler(req; kw...)
end
end

# Create a new client with the auth layer added
HTTP.@client [cache_layer]

end # module
5 changes: 5 additions & 0 deletions src/ProToPortal.jl
Original file line number Diff line number Diff line change
Expand Up @@ -53,6 +53,11 @@ include("llm.jl")
export meta_prompt_step!
include("meta_prompting.jl")

include("CacheLayer.jl")

export launch
include("server.jl")

function __init__()
## Load extra templates
PT.load_templates!(joinpath(@__DIR__, "..", "templates"); remember_path = true) # add our custom ones
Expand Down
38 changes: 38 additions & 0 deletions src/server.jl
Original file line number Diff line number Diff line change
@@ -0,0 +1,38 @@
"""
launch(
port::Int = get(ENV, "PORT", 8000), host::String = get(
ENV, "GENIE_HOST", "127.0.0.1");
async::Bool = true, cached::Bool = true, cache_verbose::Bool = false)

Launches ProToPortal in the browser.

Defaults to: `http://127.0.0.1:8000`.
This is a convenience wrapper around `Genie.up`, to customize the server configuration use `Genie.up()` and `Genie.config`.

# Arguments
- `port::Union{Int, String} = get(ENV, "PORT", "8000")`: The port to launch the server on.
- `host::String = get(ENV, "GENIE_HOST", "127.0.0.1")`: The host to launch the server on.
- `async::Bool = true`: Whether to launch the server asynchronously, ie, in the background.
- `cached::Bool = true`: Whether to use semantic caching of the requests.
- `cache_verbose::Bool = true`: Whether to print verbose information about the caching process.

If you want to remove the cache layer later, you can use `import HTTP; HTTP.poplayer!()`.
"""
function launch(
port::Union{Int, String} = get(ENV, "PORT", "8000"),
host::String = get(ENV, "GENIE_HOST", "127.0.0.1");
async::Bool = true, cached::Bool = true, cache_verbose::Bool = true)
## Loads app.jl in the root directory
Genie.loadapp(pkgdir(ProToPortal))

## Enables caching
ENV["CACHES_VERBOSE"] = cache_verbose ? "true" : "false"
if cached
@info "Caching enabled globally (for all requests, see `CacheLayer` module for details). Remove with `HTTP.poplayer!()`"
HTTP.pushlayer!(CacheLayer.cache_layer)
end
## Convert to INT
port_ = port isa Integer ? port : tryparse(Int, port)
@assert port_ isa Integer "Port must be an integer. Provided: $port"
up(port_, host; async)
end
Loading