Extract completion provider crate (#14823)
We will soon need `semantic_index` to be able to use
`CompletionProvider`. This is currently impossible due to a cyclic crate
dependency, because `CompletionProvider` lives in the `assistant` crate,
which depends on `semantic_index`.
This PR breaks the dependency cycle by extracting two crates out of
`assistant`: `language_model` and `completion`.
Only one piece of logic changed: [this
code](922fcaf5a6 (diff-3857b3707687a4d585f1200eec4c34a7a079eae8d303b4ce5b4fce46234ace9fR61-R69)).
* As of https://github.com/zed-industries/zed/pull/13276, whenever we
ask a given completion provider for its available models, OpenAI
providers would go and ask the global assistant settings whether the
user had configured an `available_models` setting, and if so, return
that.
* This PR changes it so that instead of eagerly asking the assistant
settings for this info (the new crate must not depend on `assistant`, or
else the dependency cycle would be back), OpenAI completion providers
now store the user-configured settings as part of their struct, and
whenever the settings change, we update the provider.
In theory, this change should not change user-visible behavior...but
since it's the only change in this large PR that's more than just moving
code around, I'm mentioning it here in case there's an unexpected
regression in practice! (cc @amtoaer in case you'd like to try out this
branch and verify that the feature is still working the way you expect.)
Release Notes:
- N/A
---------
Co-authored-by: Marshall Bowers <elliott.codes@gmail.com>
This commit is contained in:
Richard Feldman
GitHub
parent
b9a53ffa0b
commit
ec487d8f64
30 changed files with 820 additions and 610 deletions
|
|
@ -1,166 +1,19 @@
|
|||
use std::fmt;
|
||||
use std::{sync::Arc, time::Duration};
|
||||
|
||||
use crate::{preprocess_anthropic_request, LanguageModel, LanguageModelRequest};
|
||||
pub use anthropic::Model as AnthropicModel;
|
||||
use gpui::Pixels;
|
||||
pub use ollama::Model as OllamaModel;
|
||||
pub use open_ai::Model as OpenAiModel;
|
||||
use schemars::{
|
||||
schema::{InstanceType, Metadata, Schema, SchemaObject},
|
||||
JsonSchema,
|
||||
};
|
||||
use serde::{
|
||||
de::{self, Visitor},
|
||||
Deserialize, Deserializer, Serialize, Serializer,
|
||||
use anthropic::Model as AnthropicModel;
|
||||
use client::Client;
|
||||
use completion::{
|
||||
AnthropicCompletionProvider, CloudCompletionProvider, CompletionProvider,
|
||||
LanguageModelCompletionProvider, OllamaCompletionProvider, OpenAiCompletionProvider,
|
||||
};
|
||||
use gpui::{AppContext, Pixels};
|
||||
use language_model::{CloudModel, LanguageModel};
|
||||
use ollama::Model as OllamaModel;
|
||||
use open_ai::Model as OpenAiModel;
|
||||
use parking_lot::RwLock;
|
||||
use schemars::{schema::Schema, JsonSchema};
|
||||
use serde::{Deserialize, Serialize};
|
||||
use settings::{Settings, SettingsSources};
|
||||
use strum::{EnumIter, IntoEnumIterator};
|
||||
|
||||
#[derive(Clone, Debug, Default, PartialEq, EnumIter)]
|
||||
pub enum CloudModel {
|
||||
Gpt3Point5Turbo,
|
||||
Gpt4,
|
||||
Gpt4Turbo,
|
||||
#[default]
|
||||
Gpt4Omni,
|
||||
Gpt4OmniMini,
|
||||
Claude3_5Sonnet,
|
||||
Claude3Opus,
|
||||
Claude3Sonnet,
|
||||
Claude3Haiku,
|
||||
Gemini15Pro,
|
||||
Gemini15Flash,
|
||||
Custom(String),
|
||||
}
|
||||
|
||||
impl Serialize for CloudModel {
|
||||
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
|
||||
where
|
||||
S: Serializer,
|
||||
{
|
||||
serializer.serialize_str(self.id())
|
||||
}
|
||||
}
|
||||
|
||||
impl<'de> Deserialize<'de> for CloudModel {
|
||||
fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
|
||||
where
|
||||
D: Deserializer<'de>,
|
||||
{
|
||||
struct ZedDotDevModelVisitor;
|
||||
|
||||
impl<'de> Visitor<'de> for ZedDotDevModelVisitor {
|
||||
type Value = CloudModel;
|
||||
|
||||
fn expecting(&self, formatter: &mut fmt::Formatter) -> fmt::Result {
|
||||
formatter.write_str("a string for a ZedDotDevModel variant or a custom model")
|
||||
}
|
||||
|
||||
fn visit_str<E>(self, value: &str) -> Result<Self::Value, E>
|
||||
where
|
||||
E: de::Error,
|
||||
{
|
||||
let model = CloudModel::iter()
|
||||
.find(|model| model.id() == value)
|
||||
.unwrap_or_else(|| CloudModel::Custom(value.to_string()));
|
||||
Ok(model)
|
||||
}
|
||||
}
|
||||
|
||||
deserializer.deserialize_str(ZedDotDevModelVisitor)
|
||||
}
|
||||
}
|
||||
|
||||
impl JsonSchema for CloudModel {
|
||||
fn schema_name() -> String {
|
||||
"ZedDotDevModel".to_owned()
|
||||
}
|
||||
|
||||
fn json_schema(_generator: &mut schemars::gen::SchemaGenerator) -> Schema {
|
||||
let variants = CloudModel::iter()
|
||||
.filter_map(|model| {
|
||||
let id = model.id();
|
||||
if id.is_empty() {
|
||||
None
|
||||
} else {
|
||||
Some(id.to_string())
|
||||
}
|
||||
})
|
||||
.collect::<Vec<_>>();
|
||||
Schema::Object(SchemaObject {
|
||||
instance_type: Some(InstanceType::String.into()),
|
||||
enum_values: Some(variants.iter().map(|s| s.clone().into()).collect()),
|
||||
metadata: Some(Box::new(Metadata {
|
||||
title: Some("ZedDotDevModel".to_owned()),
|
||||
default: Some(CloudModel::default().id().into()),
|
||||
examples: variants.into_iter().map(Into::into).collect(),
|
||||
..Default::default()
|
||||
})),
|
||||
..Default::default()
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
impl CloudModel {
|
||||
pub fn id(&self) -> &str {
|
||||
match self {
|
||||
Self::Gpt3Point5Turbo => "gpt-3.5-turbo",
|
||||
Self::Gpt4 => "gpt-4",
|
||||
Self::Gpt4Turbo => "gpt-4-turbo-preview",
|
||||
Self::Gpt4Omni => "gpt-4o",
|
||||
Self::Gpt4OmniMini => "gpt-4o-mini",
|
||||
Self::Claude3_5Sonnet => "claude-3-5-sonnet",
|
||||
Self::Claude3Opus => "claude-3-opus",
|
||||
Self::Claude3Sonnet => "claude-3-sonnet",
|
||||
Self::Claude3Haiku => "claude-3-haiku",
|
||||
Self::Gemini15Pro => "gemini-1.5-pro",
|
||||
Self::Gemini15Flash => "gemini-1.5-flash",
|
||||
Self::Custom(id) => id,
|
||||
}
|
||||
}
|
||||
|
||||
pub fn display_name(&self) -> &str {
|
||||
match self {
|
||||
Self::Gpt3Point5Turbo => "GPT 3.5 Turbo",
|
||||
Self::Gpt4 => "GPT 4",
|
||||
Self::Gpt4Turbo => "GPT 4 Turbo",
|
||||
Self::Gpt4Omni => "GPT 4 Omni",
|
||||
Self::Gpt4OmniMini => "GPT 4 Omni Mini",
|
||||
Self::Claude3_5Sonnet => "Claude 3.5 Sonnet",
|
||||
Self::Claude3Opus => "Claude 3 Opus",
|
||||
Self::Claude3Sonnet => "Claude 3 Sonnet",
|
||||
Self::Claude3Haiku => "Claude 3 Haiku",
|
||||
Self::Gemini15Pro => "Gemini 1.5 Pro",
|
||||
Self::Gemini15Flash => "Gemini 1.5 Flash",
|
||||
Self::Custom(id) => id.as_str(),
|
||||
}
|
||||
}
|
||||
|
||||
pub fn max_token_count(&self) -> usize {
|
||||
match self {
|
||||
Self::Gpt3Point5Turbo => 2048,
|
||||
Self::Gpt4 => 4096,
|
||||
Self::Gpt4Turbo | Self::Gpt4Omni => 128000,
|
||||
Self::Gpt4OmniMini => 128000,
|
||||
Self::Claude3_5Sonnet
|
||||
| Self::Claude3Opus
|
||||
| Self::Claude3Sonnet
|
||||
| Self::Claude3Haiku => 200000,
|
||||
Self::Gemini15Pro => 128000,
|
||||
Self::Gemini15Flash => 32000,
|
||||
Self::Custom(_) => 4096, // TODO: Make this configurable
|
||||
}
|
||||
}
|
||||
|
||||
pub fn preprocess_request(&self, request: &mut LanguageModelRequest) {
|
||||
match self {
|
||||
Self::Claude3Opus | Self::Claude3Sonnet | Self::Claude3Haiku => {
|
||||
preprocess_anthropic_request(request)
|
||||
}
|
||||
_ => {}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#[derive(Copy, Clone, Default, Debug, Serialize, Deserialize, JsonSchema)]
|
||||
#[serde(rename_all = "snake_case")]
|
||||
|
|
@ -620,6 +473,124 @@ fn merge<T>(target: &mut T, value: Option<T>) {
|
|||
}
|
||||
}
|
||||
|
||||
pub fn update_completion_provider_settings(
|
||||
provider: &mut CompletionProvider,
|
||||
version: usize,
|
||||
cx: &mut AppContext,
|
||||
) {
|
||||
let updated = match &AssistantSettings::get_global(cx).provider {
|
||||
AssistantProvider::ZedDotDev { model } => provider
|
||||
.update_current_as::<_, CloudCompletionProvider>(|provider| {
|
||||
provider.update(model.clone(), version);
|
||||
}),
|
||||
AssistantProvider::OpenAi {
|
||||
model,
|
||||
api_url,
|
||||
low_speed_timeout_in_seconds,
|
||||
available_models,
|
||||
} => provider.update_current_as::<_, OpenAiCompletionProvider>(|provider| {
|
||||
provider.update(
|
||||
choose_openai_model(&model, &available_models),
|
||||
api_url.clone(),
|
||||
low_speed_timeout_in_seconds.map(Duration::from_secs),
|
||||
version,
|
||||
);
|
||||
}),
|
||||
AssistantProvider::Anthropic {
|
||||
model,
|
||||
api_url,
|
||||
low_speed_timeout_in_seconds,
|
||||
} => provider.update_current_as::<_, AnthropicCompletionProvider>(|provider| {
|
||||
provider.update(
|
||||
model.clone(),
|
||||
api_url.clone(),
|
||||
low_speed_timeout_in_seconds.map(Duration::from_secs),
|
||||
version,
|
||||
);
|
||||
}),
|
||||
AssistantProvider::Ollama {
|
||||
model,
|
||||
api_url,
|
||||
low_speed_timeout_in_seconds,
|
||||
} => provider.update_current_as::<_, OllamaCompletionProvider>(|provider| {
|
||||
provider.update(
|
||||
model.clone(),
|
||||
api_url.clone(),
|
||||
low_speed_timeout_in_seconds.map(Duration::from_secs),
|
||||
version,
|
||||
cx,
|
||||
);
|
||||
}),
|
||||
};
|
||||
|
||||
// Previously configured provider was changed to another one
|
||||
if updated.is_none() {
|
||||
provider.update_provider(|client| create_provider_from_settings(client, version, cx));
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) fn create_provider_from_settings(
|
||||
client: Arc<Client>,
|
||||
settings_version: usize,
|
||||
cx: &mut AppContext,
|
||||
) -> Arc<RwLock<dyn LanguageModelCompletionProvider>> {
|
||||
match &AssistantSettings::get_global(cx).provider {
|
||||
AssistantProvider::ZedDotDev { model } => Arc::new(RwLock::new(
|
||||
CloudCompletionProvider::new(model.clone(), client.clone(), settings_version, cx),
|
||||
)),
|
||||
AssistantProvider::OpenAi {
|
||||
model,
|
||||
api_url,
|
||||
low_speed_timeout_in_seconds,
|
||||
available_models,
|
||||
} => Arc::new(RwLock::new(OpenAiCompletionProvider::new(
|
||||
choose_openai_model(&model, &available_models),
|
||||
api_url.clone(),
|
||||
client.http_client(),
|
||||
low_speed_timeout_in_seconds.map(Duration::from_secs),
|
||||
settings_version,
|
||||
available_models.clone(),
|
||||
))),
|
||||
AssistantProvider::Anthropic {
|
||||
model,
|
||||
api_url,
|
||||
low_speed_timeout_in_seconds,
|
||||
} => Arc::new(RwLock::new(AnthropicCompletionProvider::new(
|
||||
model.clone(),
|
||||
api_url.clone(),
|
||||
client.http_client(),
|
||||
low_speed_timeout_in_seconds.map(Duration::from_secs),
|
||||
settings_version,
|
||||
))),
|
||||
AssistantProvider::Ollama {
|
||||
model,
|
||||
api_url,
|
||||
low_speed_timeout_in_seconds,
|
||||
} => Arc::new(RwLock::new(OllamaCompletionProvider::new(
|
||||
model.clone(),
|
||||
api_url.clone(),
|
||||
client.http_client(),
|
||||
low_speed_timeout_in_seconds.map(Duration::from_secs),
|
||||
settings_version,
|
||||
cx,
|
||||
))),
|
||||
}
|
||||
}
|
||||
|
||||
/// Choose which model to use for openai provider.
|
||||
/// If the model is not available, try to use the first available model, or fallback to the original model.
|
||||
fn choose_openai_model(
|
||||
model: &::open_ai::Model,
|
||||
available_models: &[::open_ai::Model],
|
||||
) -> ::open_ai::Model {
|
||||
available_models
|
||||
.iter()
|
||||
.find(|&m| m == model)
|
||||
.or_else(|| available_models.first())
|
||||
.unwrap_or_else(|| model)
|
||||
.clone()
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use gpui::{AppContext, UpdateGlobal};
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue