MeiliSearch/crates/meilisearch/src/routes/chats/utils.rs

253 lines
9.3 KiB
Rust

use std::cell::RefCell;
use std::sync::RwLock;
use actix_web_lab::sse::{self, Event};
use async_openai::types::{
ChatChoiceStream, ChatCompletionMessageToolCall, ChatCompletionMessageToolCallChunk,
ChatCompletionRequestAssistantMessage, ChatCompletionRequestMessage,
ChatCompletionStreamResponseDelta, ChatCompletionToolType, CreateChatCompletionStreamResponse,
FunctionCall, FunctionCallStream, Role,
};
use bumpalo::Bump;
use meilisearch_types::error::{Code, ResponseError};
use meilisearch_types::heed::RoTxn;
use meilisearch_types::milli::index::ChatConfig;
use meilisearch_types::milli::prompt::{Prompt, PromptData};
use meilisearch_types::milli::update::new::document::DocumentFromDb;
use meilisearch_types::milli::{
DocumentId, FieldIdMapWithMetadata, GlobalFieldsIdsMap, MetadataBuilder,
};
use meilisearch_types::{Document, Index};
use serde::Serialize;
use tokio::sync::mpsc::error::SendError;
use tokio::sync::mpsc::Sender;
use super::errors::StreamErrorEvent;
use super::MEILI_APPEND_CONVERSATION_MESSAGE_NAME;
use crate::routes::chats::{MEILI_SEARCH_PROGRESS_NAME, MEILI_SEARCH_SOURCES_NAME};
pub struct SseEventSender(Sender<Event>);
impl SseEventSender {
pub fn new(sender: Sender<Event>) -> Self {
Self(sender)
}
/// Ask the front-end user to append this tool *call* to the conversation
pub async fn append_tool_call_conversation_message(
&self,
resp: CreateChatCompletionStreamResponse,
call_id: String,
function_name: String,
function_arguments: String,
) -> Result<(), SendError<Event>> {
#[allow(deprecated)] // function_call
let message =
ChatCompletionRequestMessage::Assistant(ChatCompletionRequestAssistantMessage {
content: None,
refusal: None,
name: None,
audio: None,
tool_calls: Some(vec![ChatCompletionMessageToolCall {
id: call_id,
r#type: Some(ChatCompletionToolType::Function),
function: FunctionCall { name: function_name, arguments: function_arguments },
}]),
function_call: None,
});
self.append_conversation_message(resp, &message).await
}
/// Ask the front-end user to append this tool to the conversation
pub async fn append_conversation_message(
&self,
mut resp: CreateChatCompletionStreamResponse,
message: &ChatCompletionRequestMessage,
) -> Result<(), SendError<Event>> {
let call_text = serde_json::to_string(message).unwrap();
let tool_call = ChatCompletionMessageToolCallChunk {
index: 0,
id: Some(uuid::Uuid::new_v4().to_string()),
r#type: Some(ChatCompletionToolType::Function),
function: Some(FunctionCallStream {
name: Some(MEILI_APPEND_CONVERSATION_MESSAGE_NAME.to_string()),
arguments: Some(call_text),
}),
};
resp.choices[0] = ChatChoiceStream {
index: 0,
#[allow(deprecated)] // function_call
delta: ChatCompletionStreamResponseDelta {
content: None,
function_call: None,
tool_calls: Some(vec![tool_call]),
role: Some(Role::Assistant),
refusal: None,
},
finish_reason: None,
logprobs: None,
};
self.send_json(&resp).await
}
pub async fn report_search_progress(
&self,
mut resp: CreateChatCompletionStreamResponse,
call_id: &str,
function_name: &str,
function_arguments: &str,
) -> Result<(), SendError<Event>> {
#[derive(Debug, Clone, Serialize)]
/// Provides information about the current Meilisearch search operation.
struct MeiliSearchProgress<'a> {
/// The call ID to track the sources of the search.
call_id: &'a str,
/// The name of the function we are executing.
function_name: &'a str,
/// The arguments of the function we are executing, encoded in JSON.
function_arguments: &'a str,
}
let progress = MeiliSearchProgress { call_id, function_name, function_arguments };
let call_text = serde_json::to_string(&progress).unwrap();
let tool_call = ChatCompletionMessageToolCallChunk {
index: 0,
id: Some(uuid::Uuid::new_v4().to_string()),
r#type: Some(ChatCompletionToolType::Function),
function: Some(FunctionCallStream {
name: Some(MEILI_SEARCH_PROGRESS_NAME.to_string()),
arguments: Some(call_text),
}),
};
resp.choices[0] = ChatChoiceStream {
index: 0,
#[allow(deprecated)] // function_call
delta: ChatCompletionStreamResponseDelta {
content: None,
function_call: None,
tool_calls: Some(vec![tool_call]),
role: Some(Role::Assistant),
refusal: None,
},
finish_reason: None,
logprobs: None,
};
self.send_json(&resp).await
}
pub async fn report_sources(
&self,
mut resp: CreateChatCompletionStreamResponse,
call_id: &str,
documents: &[Document],
) -> Result<(), SendError<Event>> {
#[derive(Debug, Clone, Serialize)]
/// Provides sources of the search.
struct MeiliSearchSources<'a> {
/// The call ID to track the original search associated to those sources.
call_id: &'a str,
/// The documents associated with the search (call_id).
/// Only the displayed attributes of the documents are returned.
sources: &'a [Document],
}
let sources = MeiliSearchSources { call_id, sources: documents };
let call_text = serde_json::to_string(&sources).unwrap();
let tool_call = ChatCompletionMessageToolCallChunk {
index: 0,
id: Some(uuid::Uuid::new_v4().to_string()),
r#type: Some(ChatCompletionToolType::Function),
function: Some(FunctionCallStream {
name: Some(MEILI_SEARCH_SOURCES_NAME.to_string()),
arguments: Some(call_text),
}),
};
resp.choices[0] = ChatChoiceStream {
index: 0,
#[allow(deprecated)] // function_call
delta: ChatCompletionStreamResponseDelta {
content: None,
function_call: None,
tool_calls: Some(vec![tool_call]),
role: Some(Role::Assistant),
refusal: None,
},
finish_reason: None,
logprobs: None,
};
self.send_json(&resp).await
}
pub async fn forward_response(
&self,
resp: &CreateChatCompletionStreamResponse,
) -> Result<(), SendError<Event>> {
self.send_json(resp).await
}
pub async fn send_error(&self, error: &StreamErrorEvent) -> Result<(), SendError<Event>> {
self.send_json(error).await
}
pub async fn stop(self) -> Result<(), SendError<Event>> {
// It is the way OpenAI sends a correct end of stream
// <https://platform.openai.com/docs/api-reference/assistants-streaming/events>
const DONE_DATA: &str = "[DONE]";
self.0.send(Event::Data(sse::Data::new(DONE_DATA))).await
}
async fn send_json<S: Serialize>(&self, data: &S) -> Result<(), SendError<Event>> {
self.0.send(Event::Data(sse::Data::new_json(data).unwrap())).await
}
}
/// Format documents based on the provided template and maximum bytes.
///
/// This formatting function is usually used to generate a summary of the documents for LLMs.
pub fn format_documents<'doc>(
rtxn: &RoTxn<'_>,
index: &Index,
doc_alloc: &'doc Bump,
internal_docids: Vec<DocumentId>,
) -> Result<Vec<&'doc str>, ResponseError> {
let ChatConfig { prompt: PromptData { template, max_bytes }, .. } = index.chat_config(rtxn)?;
let prompt = Prompt::new(template, max_bytes).unwrap();
let fid_map = index.fields_ids_map(rtxn)?;
let metadata_builder = MetadataBuilder::from_index(index, rtxn)?;
let fid_map_with_meta = FieldIdMapWithMetadata::new(fid_map.clone(), metadata_builder);
let global = RwLock::new(fid_map_with_meta);
let gfid_map = RefCell::new(GlobalFieldsIdsMap::new(&global));
let external_ids: Vec<String> = index
.external_id_of(rtxn, internal_docids.iter().copied())?
.into_iter()
.collect::<Result<_, _>>()?;
let mut renders = Vec::new();
for (docid, external_docid) in internal_docids.into_iter().zip(external_ids) {
let document = match DocumentFromDb::new(docid, rtxn, index, &fid_map)? {
Some(doc) => doc,
None => unreachable!("Document with internal ID {docid} not found"),
};
let text = match prompt.render_document(&external_docid, document, &gfid_map, doc_alloc) {
Ok(text) => text,
Err(err) => {
return Err(ResponseError::from_msg(
err.to_string(),
Code::InvalidChatSettingDocumentTemplate,
))
}
};
renders.push(text);
}
Ok(renders)
}