Files
oak-gpui/crates/lmstudio/src/lmstudio.rs
T
Elijah McMorrisandPeter Tripp 52fa7ababb lmstudio: Fill max_tokens using the response from /models (#25606)
The info for `max_tokens` for the model is included in
`{api_url}/models`
I don't think this needs to be `.clamp` like in
`crates/ollama/src/ollama.rs` `get_max_tokens`, but it might need to be

## Before:
Every model shows 2k

![image](https://github.com/user-attachments/assets/676075c8-0ceb-44b1-ae27-72ed6a6d783c)

## After:

![image](https://github.com/user-attachments/assets/8291535b-976e-4601-b617-1a508bf44e12)

### Json from `{api_url}/models` with model not loaded
```json
  {
      "id": "qwen2.5-coder-1.5b-instruct-mlx",
      "object": "model",
      "type": "llm",
      "publisher": "lmstudio-community",
      "arch": "qwen2",
      "compatibility_type": "mlx",
      "quantization": "4bit",
      "state": "not-loaded",
      "max_context_length": 32768
    },
```

## Notes
The response from `{api_url}/models` seems to return the `max_tokens`
for the model, not the currently configured context length, but I think
showing the `max_tokens` for the model is better than setting 2k for
everything

`loaded_context_length` exists, but only if the model is loaded at the
startup of zed, which usually isn't the case

maybe `fetch_models` should be rerun when swapping lmstudio models

### Currently configured context
this isn't shown in `{api_url}/models`

![image](https://github.com/user-attachments/assets/8511cb9d-914b-4065-9eba-c0b086ad253b)

### Json from `{api_url}/models` with model loaded
```json
  {
     "id": "qwen2.5-coder-1.5b-instruct-mlx",
      "object": "model",
      "type": "llm",
      "publisher": "lmstudio-community",
      "arch": "qwen2",
      "compatibility_type": "mlx",
      "quantization": "4bit",
      "state": "loaded",
      "max_context_length": 32768,
      "loaded_context_length": 4096
    },
```

Release Notes:

- lmstudio: Fixed showing `max_tokens` in the assistant panel

---------

Co-authored-by: Peter Tripp <peter@zed.dev>
2025-06-06 20:21:23 +00:00

394 lines
11 KiB
Rust

use anyhow::{Context as _, Result};
use futures::{AsyncBufReadExt, AsyncReadExt, StreamExt, io::BufReader, stream::BoxStream};
use http_client::{AsyncBody, HttpClient, Method, Request as HttpRequest, http};
use serde::{Deserialize, Serialize};
use serde_json::Value;
use std::{convert::TryFrom, time::Duration};
pub const LMSTUDIO_API_URL: &str = "http://localhost:1234/api/v0";
#[derive(Clone, Copy, Serialize, Deserialize, Debug, Eq, PartialEq)]
#[serde(rename_all = "lowercase")]
pub enum Role {
User,
Assistant,
System,
Tool,
}
impl TryFrom<String> for Role {
type Error = anyhow::Error;
fn try_from(value: String) -> Result<Self> {
match value.as_str() {
"user" => Ok(Self::User),
"assistant" => Ok(Self::Assistant),
"system" => Ok(Self::System),
"tool" => Ok(Self::Tool),
_ => anyhow::bail!("invalid role '{value}'"),
}
}
}
impl From<Role> for String {
fn from(val: Role) -> Self {
match val {
Role::User => "user".to_owned(),
Role::Assistant => "assistant".to_owned(),
Role::System => "system".to_owned(),
Role::Tool => "tool".to_owned(),
}
}
}
#[cfg_attr(feature = "schemars", derive(schemars::JsonSchema))]
#[derive(Clone, Debug, Default, Serialize, Deserialize, PartialEq)]
pub struct Model {
pub name: String,
pub display_name: Option<String>,
pub max_tokens: usize,
pub supports_tool_calls: bool,
}
impl Model {
pub fn new(
name: &str,
display_name: Option<&str>,
max_tokens: Option<usize>,
supports_tool_calls: bool,
) -> Self {
Self {
name: name.to_owned(),
display_name: display_name.map(|s| s.to_owned()),
max_tokens: max_tokens.unwrap_or(2048),
supports_tool_calls,
}
}
pub fn id(&self) -> &str {
&self.name
}
pub fn display_name(&self) -> &str {
self.display_name.as_ref().unwrap_or(&self.name)
}
pub fn max_token_count(&self) -> usize {
self.max_tokens
}
pub fn supports_tool_calls(&self) -> bool {
self.supports_tool_calls
}
}
#[derive(Debug, Serialize, Deserialize)]
#[serde(untagged)]
pub enum ToolChoice {
Auto,
Required,
None,
Other(ToolDefinition),
}
#[derive(Clone, Deserialize, Serialize, Debug)]
#[serde(tag = "type", rename_all = "snake_case")]
pub enum ToolDefinition {
#[allow(dead_code)]
Function { function: FunctionDefinition },
}
#[derive(Clone, Debug, Serialize, Deserialize)]
pub struct FunctionDefinition {
pub name: String,
pub description: Option<String>,
pub parameters: Option<Value>,
}
#[derive(Serialize, Deserialize, Debug)]
#[serde(tag = "role", rename_all = "lowercase")]
pub enum ChatMessage {
Assistant {
#[serde(default)]
content: Option<String>,
#[serde(default, skip_serializing_if = "Vec::is_empty")]
tool_calls: Vec<ToolCall>,
},
User {
content: String,
},
System {
content: String,
},
Tool {
content: String,
tool_call_id: String,
},
}
#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
pub struct ToolCall {
pub id: String,
#[serde(flatten)]
pub content: ToolCallContent,
}
#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
#[serde(tag = "type", rename_all = "lowercase")]
pub enum ToolCallContent {
Function { function: FunctionContent },
}
#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
pub struct FunctionContent {
pub name: String,
pub arguments: String,
}
#[derive(Serialize, Debug)]
pub struct ChatCompletionRequest {
pub model: String,
pub messages: Vec<ChatMessage>,
pub stream: bool,
#[serde(skip_serializing_if = "Option::is_none")]
pub max_tokens: Option<i32>,
#[serde(skip_serializing_if = "Option::is_none")]
pub stop: Option<Vec<String>>,
#[serde(skip_serializing_if = "Option::is_none")]
pub temperature: Option<f32>,
#[serde(skip_serializing_if = "Vec::is_empty")]
pub tools: Vec<ToolDefinition>,
#[serde(skip_serializing_if = "Option::is_none")]
pub tool_choice: Option<ToolChoice>,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct ChatResponse {
pub id: String,
pub object: String,
pub created: u64,
pub model: String,
pub choices: Vec<ChoiceDelta>,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct ChoiceDelta {
pub index: u32,
pub delta: ResponseMessageDelta,
pub finish_reason: Option<String>,
}
#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
pub struct ToolCallChunk {
pub index: usize,
pub id: Option<String>,
// There is also an optional `type` field that would determine if a
// function is there. Sometimes this streams in with the `function` before
// it streams in the `type`
pub function: Option<FunctionChunk>,
}
#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
pub struct FunctionChunk {
pub name: Option<String>,
pub arguments: Option<String>,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct Usage {
pub prompt_tokens: u32,
pub completion_tokens: u32,
pub total_tokens: u32,
}
#[derive(Debug, Default, Clone, Deserialize, PartialEq)]
#[serde(transparent)]
pub struct Capabilities(Vec<String>);
impl Capabilities {
pub fn supports_tool_calls(&self) -> bool {
self.0.iter().any(|cap| cap == "tool_use")
}
}
#[derive(Serialize, Deserialize, Debug)]
#[serde(untagged)]
pub enum ResponseStreamResult {
Ok(ResponseStreamEvent),
Err { error: String },
}
#[derive(Serialize, Deserialize, Debug)]
pub struct ResponseStreamEvent {
pub created: u32,
pub model: String,
pub object: String,
pub choices: Vec<ChoiceDelta>,
pub usage: Option<Usage>,
}
#[derive(Deserialize)]
pub struct ListModelsResponse {
pub data: Vec<ModelEntry>,
}
#[derive(Clone, Debug, Deserialize, PartialEq)]
pub struct ModelEntry {
pub id: String,
pub object: String,
pub r#type: ModelType,
pub publisher: String,
pub arch: Option<String>,
pub compatibility_type: CompatibilityType,
pub quantization: Option<String>,
pub state: ModelState,
pub max_context_length: Option<usize>,
pub loaded_context_length: Option<usize>,
#[serde(default)]
pub capabilities: Capabilities,
}
#[derive(Clone, Debug, Serialize, Deserialize, PartialEq)]
#[serde(rename_all = "lowercase")]
pub enum ModelType {
Llm,
Embeddings,
Vlm,
}
#[derive(Clone, Debug, Serialize, Deserialize, PartialEq)]
#[serde(rename_all = "kebab-case")]
pub enum ModelState {
Loaded,
Loading,
NotLoaded,
}
#[derive(Clone, Debug, Serialize, Deserialize, PartialEq)]
#[serde(rename_all = "lowercase")]
pub enum CompatibilityType {
Gguf,
Mlx,
}
#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
pub struct ResponseMessageDelta {
pub role: Option<Role>,
pub content: Option<String>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub tool_calls: Option<Vec<ToolCallChunk>>,
}
pub async fn complete(
client: &dyn HttpClient,
api_url: &str,
request: ChatCompletionRequest,
) -> Result<ChatResponse> {
let uri = format!("{api_url}/chat/completions");
let request_builder = HttpRequest::builder()
.method(Method::POST)
.uri(uri)
.header("Content-Type", "application/json");
let serialized_request = serde_json::to_string(&request)?;
let request = request_builder.body(AsyncBody::from(serialized_request))?;
let mut response = client.send(request).await?;
if response.status().is_success() {
let mut body = Vec::new();
response.body_mut().read_to_end(&mut body).await?;
let response_message: ChatResponse = serde_json::from_slice(&body)?;
Ok(response_message)
} else {
let mut body = Vec::new();
response.body_mut().read_to_end(&mut body).await?;
let body_str = std::str::from_utf8(&body)?;
anyhow::bail!(
"Failed to connect to API: {} {}",
response.status(),
body_str
);
}
}
pub async fn stream_chat_completion(
client: &dyn HttpClient,
api_url: &str,
request: ChatCompletionRequest,
) -> Result<BoxStream<'static, Result<ResponseStreamEvent>>> {
let uri = format!("{api_url}/chat/completions");
let request_builder = http::Request::builder()
.method(Method::POST)
.uri(uri)
.header("Content-Type", "application/json");
let request = request_builder.body(AsyncBody::from(serde_json::to_string(&request)?))?;
let mut response = client.send(request).await?;
if response.status().is_success() {
let reader = BufReader::new(response.into_body());
Ok(reader
.lines()
.filter_map(|line| async move {
match line {
Ok(line) => {
let line = line.strip_prefix("data: ")?;
if line == "[DONE]" {
None
} else {
let result = serde_json::from_str(&line)
.context("Unable to parse chat completions response");
if let Err(ref e) = result {
eprintln!("Error parsing line: {e}\nLine content: '{line}'");
}
Some(result)
}
}
Err(e) => {
eprintln!("Error reading line: {e}");
Some(Err(e.into()))
}
}
})
.boxed())
} else {
let mut body = String::new();
response.body_mut().read_to_string(&mut body).await?;
anyhow::bail!(
"Failed to connect to LM Studio API: {} {}",
response.status(),
body,
);
}
}
pub async fn get_models(
client: &dyn HttpClient,
api_url: &str,
_: Option<Duration>,
) -> Result<Vec<ModelEntry>> {
let uri = format!("{api_url}/models");
let request_builder = HttpRequest::builder()
.method(Method::GET)
.uri(uri)
.header("Accept", "application/json");
let request = request_builder.body(AsyncBody::default())?;
let mut response = client.send(request).await?;
let mut body = String::new();
response.body_mut().read_to_string(&mut body).await?;
anyhow::ensure!(
response.status().is_success(),
"Failed to connect to LM Studio API: {} {}",
response.status(),
body,
);
let response: ListModelsResponse =
serde_json::from_str(&body).context("Unable to parse LM Studio models response")?;
Ok(response.data)
}