Yehowshua/ZIm
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Add support for queuing status updates in cloud language model provider (#29818)

Browse source 
This sets us up to display queue position information to the user, once
our language model backend is updated to support request queuing.

The JSON returned by the LLM backend will need to look like this:

```json
{"queue": {"status": "queued", "position": 1}}
{"queue": {"status": "started"}}
{"event": {"THE_UPSTREAM_MODEL_PROVIDER_EVENT": "..."}} 
```

Release Notes:

- N/A

---------

Co-authored-by: Marshall Bowers <git@maxdeviant.com>
This commit is contained in:
Max Brunsfeld 2025-05-02 13:36:39 -07:00 • committed by GitHub
parent 4d1df7bcd7
commit 04772bf17d
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
9 changed files with 492 additions and 430 deletions
Download patch file Download diff file Expand all files Collapse all files

31
crates/agent/src/active_thread.rs
View file

@ -4,8 +4,8 @@ use crate::context_store::ContextStore;
use crate::context_strip::{ContextStrip, ContextStripEvent, SuggestContextKind};
use crate::message_editor::insert_message_creases;
use crate::thread::{
LastRestoreCheckpoint, MessageCrease, MessageId, MessageSegment, Thread, ThreadError,
ThreadEvent, ThreadFeedback,
LastRestoreCheckpoint, MessageCrease, MessageId, MessageSegment, QueueState, Thread,
ThreadError, ThreadEvent, ThreadFeedback,
};
use crate::thread_store::{RulesLoadingError, ThreadStore};
use crate::tool_use::{PendingToolUseStatus, ToolUse};
@ -1733,8 +1733,27 @@ impl ActiveThread {
let show_feedback = thread.is_turn_end(ix);
let generating_label = (is_generating && is_last_message)
.then(|| AnimatedLabel::new("Generating").size(LabelSize::Small));
let generating_label = is_last_message
.then(|| match (thread.queue_state(), is_generating) {
(Some(QueueState::Sending), _) => Some(
AnimatedLabel::new("Sending")
.size(LabelSize::Small)
.into_any_element(),
),
(Some(QueueState::Queued { position }), _) => Some(
Label::new(format!("Queue position: {position}"))
.size(LabelSize::Small)
.color(Color::Muted)
.into_any_element(),
),
(_, true) => Some(
AnimatedLabel::new("Generating")
.size(LabelSize::Small)
.into_any_element(),
),
_ => None,
})
.flatten();
let editing_message_state = self
.editing_message
@ -2105,7 +2124,7 @@ impl ActiveThread {
parent.child(self.render_rules_item(cx))
})
.child(styled_message)
.when(generating_label.is_some(), |this| {
.when_some(generating_label, |this, generating_label| {
this.child(
h_flex()
.h_8()
@ -2113,7 +2132,7 @@ impl ActiveThread {
.mb_4()
.ml_4()
.py_1p5()
.child(generating_label.unwrap()),
.child(generating_label),
)
})
.when(show_feedback, move |parent| {

29
crates/agent/src/thread.rs
View file

@ -320,6 +320,13 @@ fn default_completion_mode(cx: &App) -> CompletionMode {
}
}
#[derive(Debug, Clone, Copy)]
pub enum QueueState {
Sending,
Queued { position: usize },
Started,
}
/// A thread of conversation with the LLM.
pub struct Thread {
id: ThreadId,
@ -625,6 +632,12 @@ impl Thread {
!self.pending_completions.is_empty() || !self.all_tools_finished()
}
pub fn queue_state(&self) -> Option<QueueState> {
self.pending_completions
.first()
.map(|pending_completion| pending_completion.queue_state)
}
pub fn tools(&self) -> &Entity<ToolWorkingSet> {
&self.tools
}
@ -1470,6 +1483,20 @@ impl Thread {
});
}
}
LanguageModelCompletionEvent::QueueUpdate(queue_event) => {
if let Some(completion) = thread
.pending_completions
.iter_mut()
.find(|completion| completion.id == pending_completion_id)
{
completion.queue_state = match queue_event {
language_model::QueueState::Queued { position } => {
QueueState::Queued { position }
}
language_model::QueueState::Started => QueueState::Started,
}
}
}
}
thread.touch_updated_at();
@ -1590,6 +1617,7 @@ impl Thread {
self.pending_completions.push(PendingCompletion {
id: pending_completion_id,
queue_state: QueueState::Sending,
_task: task,
});
}
@ -2499,6 +2527,7 @@ impl EventEmitter<ThreadEvent> for Thread {}
struct PendingCompletion {
id: usize,
queue_state: QueueState,
_task: Task<()>,
}

1
crates/assistant_context_editor/src/context.rs
View file

@ -2371,6 +2371,7 @@ impl AssistantContext {
});
match event {
LanguageModelCompletionEvent::QueueUpdate { .. } => {}
LanguageModelCompletionEvent::StartMessage { .. } => {}
LanguageModelCompletionEvent::Stop(reason) => {
stop_reason = reason;

4
crates/eval/src/instance.rs
View file

@ -1017,7 +1017,8 @@ pub fn response_events_to_markdown(
}
Ok(
LanguageModelCompletionEvent::UsageUpdate(_)
| LanguageModelCompletionEvent::StartMessage { .. },
| LanguageModelCompletionEvent::StartMessage { .. }
| LanguageModelCompletionEvent::QueueUpdate { .. },
) => {}
Err(error) => {
flush_buffers(&mut response, &mut text_buffer, &mut thinking_buffer);
@ -1092,6 +1093,7 @@ impl ThreadDialog {
// Skip these
Ok(LanguageModelCompletionEvent::UsageUpdate(_))
| Ok(LanguageModelCompletionEvent::QueueUpdate { .. })
| Ok(LanguageModelCompletionEvent::StartMessage { .. })
| Ok(LanguageModelCompletionEvent::Stop(_)) => {}

9
crates/language_model/src/language_model.rs
View file

@ -64,9 +64,17 @@ pub struct LanguageModelCacheConfiguration {
pub min_total_token: usize,
}
#[derive(Debug, PartialEq, Clone, Copy, Serialize, Deserialize)]
#[serde(tag = "status", rename_all = "snake_case")]
pub enum QueueState {
Queued { position: usize },
Started,
}
/// A completion event from a language model.
#[derive(Debug, PartialEq, Clone, Serialize, Deserialize)]
pub enum LanguageModelCompletionEvent {
QueueUpdate(QueueState),
Stop(StopReason),
Text(String),
Thinking {
@ -349,6 +357,7 @@ pub trait LanguageModel: Send + Sync {
let last_token_usage = last_token_usage.clone();
async move {
match result {
Ok(LanguageModelCompletionEvent::QueueUpdate { .. }) => None,
Ok(LanguageModelCompletionEvent::StartMessage { .. }) => None,
Ok(LanguageModelCompletionEvent::Text(text)) => Some(Ok(text)),
Ok(LanguageModelCompletionEvent::Thinking { .. }) => None,

369
crates/language_models/src/provider/anthropic.rs
View file

@ -469,7 +469,7 @@ impl LanguageModel for AnthropicModel {
Ok(anthropic_err) => anthropic_err_to_anyhow(anthropic_err),
Err(err) => anyhow!(err),
})?;
Ok(map_to_language_model_completion_events(response))
Ok(AnthropicEventMapper::new().map_stream(response))
});
async move { Ok(future.await?.boxed()) }.boxed()
}
@ -629,215 +629,186 @@ pub fn into_anthropic(
}
}
pub fn map_to_language_model_completion_events(
events: Pin<Box<dyn Send + Stream<Item = Result<Event, AnthropicError>>>>,
) -> impl Stream<Item = Result<LanguageModelCompletionEvent, LanguageModelCompletionError>> {
struct RawToolUse {
id: String,
name: String,
input_json: String,
}
pub struct AnthropicEventMapper {
tool_uses_by_index: HashMap<usize, RawToolUse>,
usage: Usage,
stop_reason: StopReason,
}
struct State {
events: Pin<Box<dyn Send + Stream<Item = Result<Event, AnthropicError>>>>,
tool_uses_by_index: HashMap<usize, RawToolUse>,
usage: Usage,
stop_reason: StopReason,
}
futures::stream::unfold(
State {
events,
impl AnthropicEventMapper {
pub fn new() -> Self {
Self {
tool_uses_by_index: HashMap::default(),
usage: Usage::default(),
stop_reason: StopReason::EndTurn,
},
|mut state| async move {
while let Some(event) = state.events.next().await {
match event {
Ok(event) => match event {
Event::ContentBlockStart {
index,
content_block,
} => match content_block {
ResponseContent::Text { text } => {
return Some((
vec![Ok(LanguageModelCompletionEvent::Text(text))],
state,
));
}
ResponseContent::Thinking { thinking } => {
return Some((
vec![Ok(LanguageModelCompletionEvent::Thinking {
text: thinking,
signature: None,
})],
state,
));
}
ResponseContent::RedactedThinking { .. } => {
// Redacted thinking is encrypted and not accessible to the user, see:
// https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking#suggestions-for-handling-redacted-thinking-in-production
}
ResponseContent::ToolUse { id, name, .. } => {
state.tool_uses_by_index.insert(
index,
RawToolUse {
id,
name,
input_json: String::new(),
},
);
}
},
Event::ContentBlockDelta { index, delta } => match delta {
ContentDelta::TextDelta { text } => {
return Some((
vec![Ok(LanguageModelCompletionEvent::Text(text))],
state,
));
}
ContentDelta::ThinkingDelta { thinking } => {
return Some((
vec![Ok(LanguageModelCompletionEvent::Thinking {
text: thinking,
signature: None,
})],
state,
));
}
ContentDelta::SignatureDelta { signature } => {
return Some((
vec![Ok(LanguageModelCompletionEvent::Thinking {
text: "".to_string(),
signature: Some(signature),
})],
state,
));
}
ContentDelta::InputJsonDelta { partial_json } => {
if let Some(tool_use) = state.tool_uses_by_index.get_mut(&index) {
tool_use.input_json.push_str(&partial_json);
}
}
// Try to convert invalid (incomplete) JSON into
// valid JSON that serde can accept, e.g. by closing
// unclosed delimiters. This way, we can update the
// UI with whatever has been streamed back so far.
if let Ok(input) = serde_json::Value::from_str(
&partial_json_fixer::fix_json(&tool_use.input_json),
) {
return Some((
vec![Ok(LanguageModelCompletionEvent::ToolUse(
LanguageModelToolUse {
id: tool_use.id.clone().into(),
name: tool_use.name.clone().into(),
is_input_complete: false,
raw_input: tool_use.input_json.clone(),
input,
},
))],
state,
));
}
}
}
},
Event::ContentBlockStop { index } => {
if let Some(tool_use) = state.tool_uses_by_index.remove(&index) {
let input_json = tool_use.input_json.trim();
let input_value = if input_json.is_empty() {
Ok(serde_json::Value::Object(serde_json::Map::default()))
} else {
serde_json::Value::from_str(input_json)
};
let event_result = match input_value {
Ok(input) => Ok(LanguageModelCompletionEvent::ToolUse(
LanguageModelToolUse {
id: tool_use.id.into(),
name: tool_use.name.into(),
is_input_complete: true,
input,
raw_input: tool_use.input_json.clone(),
},
)),
Err(json_parse_err) => {
Err(LanguageModelCompletionError::BadInputJson {
id: tool_use.id.into(),
tool_name: tool_use.name.into(),
raw_input: input_json.into(),
json_parse_error: json_parse_err.to_string(),
})
}
};
pub fn map_stream(
mut self,
events: Pin<Box<dyn Send + Stream<Item = Result<Event, AnthropicError>>>>,
) -> impl Stream<Item = Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
{
events.flat_map(move |event| {
futures::stream::iter(match event {
Ok(event) => self.map_event(event),
Err(error) => vec![Err(LanguageModelCompletionError::Other(anyhow!(error)))],
})
})
}
return Some((vec![event_result], state));
}
pub fn map_event(
&mut self,
event: Event,
) -> Vec<Result<LanguageModelCompletionEvent, LanguageModelCompletionError>> {
match event {
Event::ContentBlockStart {
index,
content_block,
} => match content_block {
ResponseContent::Text { text } => {
vec![Ok(LanguageModelCompletionEvent::Text(text))]
}
ResponseContent::Thinking { thinking } => {
vec![Ok(LanguageModelCompletionEvent::Thinking {
text: thinking,
signature: None,
})]
}
ResponseContent::RedactedThinking { .. } => {
// Redacted thinking is encrypted and not accessible to the user, see:
// https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking#suggestions-for-handling-redacted-thinking-in-production
Vec::new()
}
ResponseContent::ToolUse { id, name, .. } => {
self.tool_uses_by_index.insert(
index,
RawToolUse {
id,
name,
input_json: String::new(),
},
);
Vec::new()
}
},
Event::ContentBlockDelta { index, delta } => match delta {
ContentDelta::TextDelta { text } => {
vec![Ok(LanguageModelCompletionEvent::Text(text))]
}
ContentDelta::ThinkingDelta { thinking } => {
vec![Ok(LanguageModelCompletionEvent::Thinking {
text: thinking,
signature: None,
})]
}
ContentDelta::SignatureDelta { signature } => {
vec![Ok(LanguageModelCompletionEvent::Thinking {
text: "".to_string(),
signature: Some(signature),
})]
}
ContentDelta::InputJsonDelta { partial_json } => {
if let Some(tool_use) = self.tool_uses_by_index.get_mut(&index) {
tool_use.input_json.push_str(&partial_json);
// Try to convert invalid (incomplete) JSON into
// valid JSON that serde can accept, e.g. by closing
// unclosed delimiters. This way, we can update the
// UI with whatever has been streamed back so far.
if let Ok(input) = serde_json::Value::from_str(
&partial_json_fixer::fix_json(&tool_use.input_json),
) {
return vec![Ok(LanguageModelCompletionEvent::ToolUse(
LanguageModelToolUse {
id: tool_use.id.clone().into(),
name: tool_use.name.clone().into(),
is_input_complete: false,
raw_input: tool_use.input_json.clone(),
input,
},
))];
}
Event::MessageStart { message } => {
update_usage(&mut state.usage, &message.usage);
return Some((
vec![
Ok(LanguageModelCompletionEvent::UsageUpdate(convert_usage(
&state.usage,
))),
Ok(LanguageModelCompletionEvent::StartMessage {
message_id: message.id,
}),
],
state,
));
}
Event::MessageDelta { delta, usage } => {
update_usage(&mut state.usage, &usage);
if let Some(stop_reason) = delta.stop_reason.as_deref() {
state.stop_reason = match stop_reason {
"end_turn" => StopReason::EndTurn,
"max_tokens" => StopReason::MaxTokens,
"tool_use" => StopReason::ToolUse,
_ => {
log::error!(
"Unexpected anthropic stop_reason: {stop_reason}"
);
StopReason::EndTurn
}
};
}
return Some((
vec![Ok(LanguageModelCompletionEvent::UsageUpdate(
convert_usage(&state.usage),
))],
state,
));
}
Event::MessageStop => {
return Some((
vec![Ok(LanguageModelCompletionEvent::Stop(state.stop_reason))],
state,
));
}
Event::Error { error } => {
return Some((
vec![Err(LanguageModelCompletionError::Other(anyhow!(
AnthropicError::ApiError(error)
)))],
state,
));
}
_ => {}
},
Err(err) => {
return Some((
vec![Err(LanguageModelCompletionError::Other(anyhow!(err)))],
state,
));
}
return vec![];
}
},
Event::ContentBlockStop { index } => {
if let Some(tool_use) = self.tool_uses_by_index.remove(&index) {
let input_json = tool_use.input_json.trim();
let input_value = if input_json.is_empty() {
Ok(serde_json::Value::Object(serde_json::Map::default()))
} else {
serde_json::Value::from_str(input_json)
};
let event_result = match input_value {
Ok(input) => Ok(LanguageModelCompletionEvent::ToolUse(
LanguageModelToolUse {
id: tool_use.id.into(),
name: tool_use.name.into(),
is_input_complete: true,
input,
raw_input: tool_use.input_json.clone(),
},
)),
Err(json_parse_err) => Err(LanguageModelCompletionError::BadInputJson {
id: tool_use.id.into(),
tool_name: tool_use.name.into(),
raw_input: input_json.into(),
json_parse_error: json_parse_err.to_string(),
}),
};
vec![event_result]
} else {
Vec::new()
}
}
Event::MessageStart { message } => {
update_usage(&mut self.usage, &message.usage);
vec![
Ok(LanguageModelCompletionEvent::UsageUpdate(convert_usage(
&self.usage,
))),
Ok(LanguageModelCompletionEvent::StartMessage {
message_id: message.id,
}),
]
}
Event::MessageDelta { delta, usage } => {
update_usage(&mut self.usage, &usage);
if let Some(stop_reason) = delta.stop_reason.as_deref() {
self.stop_reason = match stop_reason {
"end_turn" => StopReason::EndTurn,
"max_tokens" => StopReason::MaxTokens,
"tool_use" => StopReason::ToolUse,
_ => {
log::error!("Unexpected anthropic stop_reason: {stop_reason}");
StopReason::EndTurn
}
};
}
vec![Ok(LanguageModelCompletionEvent::UsageUpdate(
convert_usage(&self.usage),
))]
}
Event::MessageStop => {
vec![Ok(LanguageModelCompletionEvent::Stop(self.stop_reason))]
}
Event::Error { error } => {
vec![Err(LanguageModelCompletionError::Other(anyhow!(
AnthropicError::ApiError(error)
)))]
}
_ => Vec::new(),
}
}
}
None
},
)
.flat_map(futures::stream::iter)
struct RawToolUse {
id: String,
name: String,
input_json: String,
}
pub fn anthropic_err_to_anyhow(err: AnthropicError) -> anyhow::Error {

94
crates/language_models/src/provider/cloud.rs
View file

@ -1,11 +1,10 @@
use anthropic::{AnthropicError, AnthropicModelMode, parse_prompt_too_long};
use anthropic::{AnthropicModelMode, parse_prompt_too_long};
use anyhow::{Result, anyhow};
use client::{Client, UserStore, zed_urls};
use collections::BTreeMap;
use feature_flags::{FeatureFlagAppExt, LlmClosedBetaFeatureFlag, ZedProFeatureFlag};
use futures::{
AsyncBufReadExt, FutureExt, Stream, StreamExt, TryStreamExt as _, future::BoxFuture,
stream::BoxStream,
AsyncBufReadExt, FutureExt, Stream, StreamExt, future::BoxFuture, stream::BoxStream,
};
use gpui::{AnyElement, AnyView, App, AsyncApp, Context, Entity, Subscription, Task};
use http_client::{AsyncBody, HttpClient, Method, Response, StatusCode};
@ -14,7 +13,7 @@ use language_model::{
LanguageModelCompletionError, LanguageModelId, LanguageModelKnownError, LanguageModelName,
LanguageModelProviderId, LanguageModelProviderName, LanguageModelProviderState,
LanguageModelProviderTosView, LanguageModelRequest, LanguageModelToolSchemaFormat,
ModelRequestLimitReachedError, RateLimiter, RequestUsage, ZED_CLOUD_PROVIDER_ID,
ModelRequestLimitReachedError, QueueState, RateLimiter, RequestUsage, ZED_CLOUD_PROVIDER_ID,
};
use language_model::{
LanguageModelAvailability, LanguageModelCompletionEvent, LanguageModelProvider, LlmApiToken,
@ -26,6 +25,7 @@ use serde::{Deserialize, Serialize, de::DeserializeOwned};
use settings::{Settings, SettingsStore};
use smol::Timer;
use smol::io::{AsyncReadExt, BufReader};
use std::pin::Pin;
use std::str::FromStr as _;
use std::{
sync::{Arc, LazyLock},
@ -41,9 +41,9 @@ use zed_llm_client::{
};
use crate::AllLanguageModelSettings;
use crate::provider::anthropic::{count_anthropic_tokens, into_anthropic};
use crate::provider::google::into_google;
use crate::provider::open_ai::{count_open_ai_tokens, into_open_ai};
use crate::provider::anthropic::{AnthropicEventMapper, count_anthropic_tokens, into_anthropic};
use crate::provider::google::{GoogleEventMapper, into_google};
use crate::provider::open_ai::{OpenAiEventMapper, count_open_ai_tokens, into_open_ai};
pub const PROVIDER_NAME: &str = "Zed";
@ -518,7 +518,7 @@ impl CloudLanguageModel {
client: Arc<Client>,
llm_api_token: LlmApiToken,
body: CompletionBody,
) -> Result<(Response<AsyncBody>, Option<RequestUsage>)> {
) -> Result<(Response<AsyncBody>, Option<RequestUsage>, bool)> {
let http_client = &client.http_client();
let mut token = llm_api_token.acquire(&client).await?;
@ -536,13 +536,18 @@ impl CloudLanguageModel {
let request = request_builder
.header("Content-Type", "application/json")
.header("Authorization", format!("Bearer {token}"))
.header("x-zed-client-supports-queueing", "true")
.body(serde_json::to_string(&body)?.into())?;
let mut response = http_client.send(request).await?;
let status = response.status();
if status.is_success() {
let includes_queue_events = response
.headers()
.get("x-zed-server-supports-queueing")
.is_some();
let usage = RequestUsage::from_headers(response.headers()).ok();
return Ok((response, usage));
return Ok((response, usage, includes_queue_events));
} else if response
.headers()
.get(EXPIRED_LLM_TOKEN_HEADER_NAME)
@ -782,7 +787,7 @@ impl LanguageModel for CloudLanguageModel {
let client = self.client.clone();
let llm_api_token = self.llm_api_token.clone();
let future = self.request_limiter.stream_with_usage(async move {
let (response, usage) = Self::perform_llm_completion(
let (response, usage, includes_queue_events) = Self::perform_llm_completion(
client.clone(),
llm_api_token,
CompletionBody {
@ -811,9 +816,11 @@ impl LanguageModel for CloudLanguageModel {
Err(err) => anyhow!(err),
})?;
let mut mapper = AnthropicEventMapper::new();
Ok((
crate::provider::anthropic::map_to_language_model_completion_events(
Box::pin(response_lines(response).map_err(AnthropicError::Other)),
map_cloud_completion_events(
Box::pin(response_lines(response, includes_queue_events)),
move |event| mapper.map_event(event),
),
usage,
))
@ -829,7 +836,7 @@ impl LanguageModel for CloudLanguageModel {
let request = into_open_ai(request, model, model.max_output_tokens());
let llm_api_token = self.llm_api_token.clone();
let future = self.request_limiter.stream_with_usage(async move {
let (response, usage) = Self::perform_llm_completion(
let (response, usage, includes_queue_events) = Self::perform_llm_completion(
client.clone(),
llm_api_token,
CompletionBody {
@ -842,9 +849,12 @@ impl LanguageModel for CloudLanguageModel {
},
)
.await?;
let mut mapper = OpenAiEventMapper::new();
Ok((
crate::provider::open_ai::map_to_language_model_completion_events(
Box::pin(response_lines(response)),
map_cloud_completion_events(
Box::pin(response_lines(response, includes_queue_events)),
move |event| mapper.map_event(event),
),
usage,
))
@ -860,7 +870,7 @@ impl LanguageModel for CloudLanguageModel {
let request = into_google(request, model.id().into());
let llm_api_token = self.llm_api_token.clone();
let future = self.request_limiter.stream_with_usage(async move {
let (response, usage) = Self::perform_llm_completion(
let (response, usage, includes_queue_events) = Self::perform_llm_completion(
client.clone(),
llm_api_token,
CompletionBody {
@ -873,10 +883,12 @@ impl LanguageModel for CloudLanguageModel {
},
)
.await?;
let mut mapper = GoogleEventMapper::new();
Ok((
crate::provider::google::map_to_language_model_completion_events(Box::pin(
response_lines(response),
)),
map_cloud_completion_events(
Box::pin(response_lines(response, includes_queue_events)),
move |event| mapper.map_event(event),
),
usage,
))
});
@ -890,16 +902,54 @@ impl LanguageModel for CloudLanguageModel {
}
}
#[derive(Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum CloudCompletionEvent<T> {
Queue(QueueState),
Event(T),
}
fn map_cloud_completion_events<T, F>(
stream: Pin<Box<dyn Stream<Item = Result<CloudCompletionEvent<T>>> + Send>>,
mut map_callback: F,
) -> BoxStream<'static, Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
where
T: DeserializeOwned + 'static,
F: FnMut(T) -> Vec<Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
+ Send
+ 'static,
{
stream
.flat_map(move |event| {
futures::stream::iter(match event {
Err(error) => {
vec![Err(LanguageModelCompletionError::Other(error))]
}
Ok(CloudCompletionEvent::Queue(event)) => {
vec![Ok(LanguageModelCompletionEvent::QueueUpdate(event))]
}
Ok(CloudCompletionEvent::Event(event)) => map_callback(event),
})
})
.boxed()
}
fn response_lines<T: DeserializeOwned>(
response: Response<AsyncBody>,
) -> impl Stream<Item = Result<T>> {
includes_queue_events: bool,
) -> impl Stream<Item = Result<CloudCompletionEvent<T>>> {
futures::stream::try_unfold(
(String::new(), BufReader::new(response.into_body())),
move |(mut line, mut body)| async {
move |(mut line, mut body)| async move {
match body.read_line(&mut line).await {
Ok(0) => Ok(None),
Ok(_) => {
let event: T = serde_json::from_str(&line)?;
let event = if includes_queue_events {
serde_json::from_str::<CloudCompletionEvent<T>>(&line)?
} else {
CloudCompletionEvent::Event(serde_json::from_str::<T>(&line)?)
};
line.clear();
Ok(Some((event, (line, body))))
}

185
crates/language_models/src/provider/google.rs
View file

@ -24,7 +24,10 @@ use schemars::JsonSchema;
use serde::{Deserialize, Serialize};
use settings::{Settings, SettingsStore};
use std::pin::Pin;
use std::sync::Arc;
use std::sync::{
Arc,
atomic::{self, AtomicU64},
};
use strum::IntoEnumIterator;
use theme::ThemeSettings;
use ui::{Icon, IconName, List, Tooltip, prelude::*};
@ -371,7 +374,7 @@ impl LanguageModel for GoogleLanguageModel {
let response = request
.await
.map_err(|err| LanguageModelCompletionError::Other(anyhow!(err)))?;
Ok(map_to_language_model_completion_events(response))
Ok(GoogleEventMapper::new().map_stream(response))
});
async move { Ok(future.await?.boxed()) }.boxed()
}
@ -486,108 +489,98 @@ pub fn into_google(
}
}
pub fn map_to_language_model_completion_events(
events: Pin<Box<dyn Send + Stream<Item = Result<GenerateContentResponse>>>>,
) -> impl Stream<Item = Result<LanguageModelCompletionEvent, LanguageModelCompletionError>> {
use std::sync::atomic::{AtomicU64, Ordering};
pub struct GoogleEventMapper {
usage: UsageMetadata,
stop_reason: StopReason,
}
static TOOL_CALL_COUNTER: AtomicU64 = AtomicU64::new(0);
struct State {
events: Pin<Box<dyn Send + Stream<Item = Result<GenerateContentResponse>>>>,
usage: UsageMetadata,
stop_reason: StopReason,
}
futures::stream::unfold(
State {
events,
impl GoogleEventMapper {
pub fn new() -> Self {
Self {
usage: UsageMetadata::default(),
stop_reason: StopReason::EndTurn,
},
|mut state| async move {
if let Some(event) = state.events.next().await {
match event {
Ok(event) => {
let mut events: Vec<_> = Vec::new();
let mut wants_to_use_tool = false;
if let Some(usage_metadata) = event.usage_metadata {
update_usage(&mut state.usage, &usage_metadata);
events.push(Ok(LanguageModelCompletionEvent::UsageUpdate(
convert_usage(&state.usage),
)))
}
if let Some(candidates) = event.candidates {
for candidate in candidates {
if let Some(finish_reason) = candidate.finish_reason.as_deref() {
state.stop_reason = match finish_reason {
"STOP" => StopReason::EndTurn,
"MAX_TOKENS" => StopReason::MaxTokens,
_ => {
log::error!(
"Unexpected google finish_reason: {finish_reason}"
);
StopReason::EndTurn
}
};
}
candidate
.content
.parts
.into_iter()
.for_each(|part| match part {
Part::TextPart(text_part) => events.push(Ok(
LanguageModelCompletionEvent::Text(text_part.text),
)),
Part::InlineDataPart(_) => {}
Part::FunctionCallPart(function_call_part) => {
wants_to_use_tool = true;
let name: Arc<str> =
function_call_part.function_call.name.into();
let next_tool_id =
TOOL_CALL_COUNTER.fetch_add(1, Ordering::SeqCst);
let id: LanguageModelToolUseId =
format!("{}-{}", name, next_tool_id).into();
}
}
events.push(Ok(LanguageModelCompletionEvent::ToolUse(
LanguageModelToolUse {
id,
name,
is_input_complete: true,
raw_input: function_call_part
.function_call
.args
.to_string(),
input: function_call_part.function_call.args,
},
)));
}
Part::FunctionResponsePart(_) => {}
});
}
}
pub fn map_stream(
mut self,
events: Pin<Box<dyn Send + Stream<Item = Result<GenerateContentResponse>>>>,
) -> impl Stream<Item = Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
{
events.flat_map(move |event| {
futures::stream::iter(match event {
Ok(event) => self.map_event(event),
Err(error) => vec![Err(LanguageModelCompletionError::Other(anyhow!(error)))],
})
})
}
// Even when Gemini wants to use a Tool, the API
// responds with `finish_reason: STOP`
if wants_to_use_tool {
state.stop_reason = StopReason::ToolUse;
pub fn map_event(
&mut self,
event: GenerateContentResponse,
) -> Vec<Result<LanguageModelCompletionEvent, LanguageModelCompletionError>> {
static TOOL_CALL_COUNTER: AtomicU64 = AtomicU64::new(0);
let mut events: Vec<_> = Vec::new();
let mut wants_to_use_tool = false;
if let Some(usage_metadata) = event.usage_metadata {
update_usage(&mut self.usage, &usage_metadata);
events.push(Ok(LanguageModelCompletionEvent::UsageUpdate(
convert_usage(&self.usage),
)))
}
if let Some(candidates) = event.candidates {
for candidate in candidates {
if let Some(finish_reason) = candidate.finish_reason.as_deref() {
self.stop_reason = match finish_reason {
"STOP" => StopReason::EndTurn,
"MAX_TOKENS" => StopReason::MaxTokens,
_ => {
log::error!("Unexpected google finish_reason: {finish_reason}");
StopReason::EndTurn
}
events.push(Ok(LanguageModelCompletionEvent::Stop(state.stop_reason)));
return Some((events, state));
}
Err(err) => {
return Some((
vec![Err(LanguageModelCompletionError::Other(anyhow!(err)))],
state,
));
}
};
}
}
candidate
.content
.parts
.into_iter()
.for_each(|part| match part {
Part::TextPart(text_part) => {
events.push(Ok(LanguageModelCompletionEvent::Text(text_part.text)))
}
Part::InlineDataPart(_) => {}
Part::FunctionCallPart(function_call_part) => {
wants_to_use_tool = true;
let name: Arc<str> = function_call_part.function_call.name.into();
let next_tool_id =
TOOL_CALL_COUNTER.fetch_add(1, atomic::Ordering::SeqCst);
let id: LanguageModelToolUseId =
format!("{}-{}", name, next_tool_id).into();
None
},
)
.flat_map(futures::stream::iter)
events.push(Ok(LanguageModelCompletionEvent::ToolUse(
LanguageModelToolUse {
id,
name,
is_input_complete: true,
raw_input: function_call_part.function_call.args.to_string(),
input: function_call_part.function_call.args,
},
)));
}
Part::FunctionResponsePart(_) => {}
});
}
}
// Even when Gemini wants to use a Tool, the API
// responds with `finish_reason: STOP`
if wants_to_use_tool {
self.stop_reason = StopReason::ToolUse;
}
events.push(Ok(LanguageModelCompletionEvent::Stop(self.stop_reason)));
events
}
}
pub fn count_google_tokens(

200
crates/language_models/src/provider/open_ai.rs
View file

@ -330,8 +330,11 @@ impl LanguageModel for OpenAiLanguageModel {
> {
let request = into_open_ai(request, &self.model, self.max_output_tokens());
let completions = self.stream_completion(request, cx);
async move { Ok(map_to_language_model_completion_events(completions.await?).boxed()) }
.boxed()
async move {
let mapper = OpenAiEventMapper::new();
Ok(mapper.map_stream(completions.await?).boxed())
}
.boxed()
}
}
@ -422,123 +425,108 @@ pub fn into_open_ai(
}
}
pub fn map_to_language_model_completion_events(
events: Pin<Box<dyn Send + Stream<Item = Result<ResponseStreamEvent>>>>,
) -> impl Stream<Item = Result<LanguageModelCompletionEvent, LanguageModelCompletionError>> {
#[derive(Default)]
struct RawToolCall {
id: String,
name: String,
arguments: String,
}
pub struct OpenAiEventMapper {
tool_calls_by_index: HashMap<usize, RawToolCall>,
}
struct State {
events: Pin<Box<dyn Send + Stream<Item = Result<ResponseStreamEvent>>>>,
tool_calls_by_index: HashMap<usize, RawToolCall>,
}
futures::stream::unfold(
State {
events,
impl OpenAiEventMapper {
pub fn new() -> Self {
Self {
tool_calls_by_index: HashMap::default(),
},
|mut state| async move {
if let Some(event) = state.events.next().await {
match event {
Ok(event) => {
let Some(choice) = event.choices.first() else {
return Some((
vec![Err(LanguageModelCompletionError::Other(anyhow!(
"Response contained no choices"
)))],
state,
));
};
}
}
let mut events = Vec::new();
if let Some(content) = choice.delta.content.clone() {
events.push(Ok(LanguageModelCompletionEvent::Text(content)));
}
pub fn map_stream(
mut self,
events: Pin<Box<dyn Send + Stream<Item = Result<ResponseStreamEvent>>>>,
) -> impl Stream<Item = Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
{
events.flat_map(move |event| {
futures::stream::iter(match event {
Ok(event) => self.map_event(event),
Err(error) => vec![Err(LanguageModelCompletionError::Other(anyhow!(error)))],
})
})
}
if let Some(tool_calls) = choice.delta.tool_calls.as_ref() {
for tool_call in tool_calls {
let entry = state
.tool_calls_by_index
.entry(tool_call.index)
.or_default();
pub fn map_event(
&mut self,
event: ResponseStreamEvent,
) -> Vec<Result<LanguageModelCompletionEvent, LanguageModelCompletionError>> {
let Some(choice) = event.choices.first() else {
return vec![Err(LanguageModelCompletionError::Other(anyhow!(
"Response contained no choices"
)))];
};
if let Some(tool_id) = tool_call.id.clone() {
entry.id = tool_id;
}
let mut events = Vec::new();
if let Some(content) = choice.delta.content.clone() {
events.push(Ok(LanguageModelCompletionEvent::Text(content)));
}
if let Some(function) = tool_call.function.as_ref() {
if let Some(name) = function.name.clone() {
entry.name = name;
}
if let Some(tool_calls) = choice.delta.tool_calls.as_ref() {
for tool_call in tool_calls {
let entry = self.tool_calls_by_index.entry(tool_call.index).or_default();
if let Some(arguments) = function.arguments.clone() {
entry.arguments.push_str(&arguments);
}
}
}
}
if let Some(tool_id) = tool_call.id.clone() {
entry.id = tool_id;
}
match choice.finish_reason.as_deref() {
Some("stop") => {
events.push(Ok(LanguageModelCompletionEvent::Stop(
StopReason::EndTurn,
)));
}
Some("tool_calls") => {
events.extend(state.tool_calls_by_index.drain().map(
|(_, tool_call)| match serde_json::Value::from_str(
&tool_call.arguments,
) {
Ok(input) => Ok(LanguageModelCompletionEvent::ToolUse(
LanguageModelToolUse {
id: tool_call.id.clone().into(),
name: tool_call.name.as_str().into(),
is_input_complete: true,
input,
raw_input: tool_call.arguments.clone(),
},
)),
Err(error) => {
Err(LanguageModelCompletionError::BadInputJson {
id: tool_call.id.into(),
tool_name: tool_call.name.as_str().into(),
raw_input: tool_call.arguments.into(),
json_parse_error: error.to_string(),
})
}
},
));
events.push(Ok(LanguageModelCompletionEvent::Stop(
StopReason::ToolUse,
)));
}
Some(stop_reason) => {
log::error!("Unexpected OpenAI stop_reason: {stop_reason:?}",);
events.push(Ok(LanguageModelCompletionEvent::Stop(
StopReason::EndTurn,
)));
}
None => {}
}
return Some((events, state));
if let Some(function) = tool_call.function.as_ref() {
if let Some(name) = function.name.clone() {
entry.name = name;
}
Err(err) => {
return Some((vec![Err(LanguageModelCompletionError::Other(err))], state));
if let Some(arguments) = function.arguments.clone() {
entry.arguments.push_str(&arguments);
}
}
}
}
None
},
)
.flat_map(futures::stream::iter)
match choice.finish_reason.as_deref() {
Some("stop") => {
events.push(Ok(LanguageModelCompletionEvent::Stop(StopReason::EndTurn)));
}
Some("tool_calls") => {
events.extend(self.tool_calls_by_index.drain().map(|(_, tool_call)| {
match serde_json::Value::from_str(&tool_call.arguments) {
Ok(input) => Ok(LanguageModelCompletionEvent::ToolUse(
LanguageModelToolUse {
id: tool_call.id.clone().into(),
name: tool_call.name.as_str().into(),
is_input_complete: true,
input,
raw_input: tool_call.arguments.clone(),
},
)),
Err(error) => Err(LanguageModelCompletionError::BadInputJson {
id: tool_call.id.into(),
tool_name: tool_call.name.as_str().into(),
raw_input: tool_call.arguments.into(),
json_parse_error: error.to_string(),
}),
}
}));
events.push(Ok(LanguageModelCompletionEvent::Stop(StopReason::ToolUse)));
}
Some(stop_reason) => {
log::error!("Unexpected OpenAI stop_reason: {stop_reason:?}",);
events.push(Ok(LanguageModelCompletionEvent::Stop(StopReason::EndTurn)));
}
None => {}
}
events
}
}
#[derive(Default)]
struct RawToolCall {
id: String,
name: String,
arguments: String,
}
pub fn count_open_ai_tokens(