Peter Tripp , Jason Mancuso , and Bennet created
Release Notes:
- Added support for OpenAI o1-mini and o1-preview models.
---------
Co-authored-by: Jason Mancuso <7891333+jvmncs@users.noreply.github.com>
Co-authored-by: Bennet <bennet@zed.dev>crates/language_model/src/model/cloud_model.rs | 2 +
crates/language_model/src/provider/open_ai.rs | 11 +++++--
crates/language_model/src/request.rs | 3 +
crates/open_ai/src/open_ai.rs | 27 ++++++++++++++-----
docs/src/assistant/configuration.md | 10 +++++-
5 files changed, 39 insertions(+), 14 deletions(-)
@@ -102,6 +102,8 @@ impl CloudModel {
| open_ai::Model::FourTurbo
| open_ai::Model::FourOmni
| open_ai::Model::FourOmniMini
+ | open_ai::Model::O1Mini
+ | open_ai::Model::O1Preview
| open_ai::Model::Custom { .. } => {
LanguageModelAvailability::RequiresPlan(Plan::ZedPro)
}
@@ -372,10 +372,13 @@ pub fn count_open_ai_tokens(
})
.collect::<Vec<_>>();
- if let open_ai::Model::Custom { .. } = model {
- tiktoken_rs::num_tokens_from_messages("gpt-4", &messages)
- } else {
- tiktoken_rs::num_tokens_from_messages(model.id(), &messages)
+ match model {
+ open_ai::Model::Custom { .. }
+ | open_ai::Model::O1Mini
+ | open_ai::Model::O1Preview => {
+ tiktoken_rs::num_tokens_from_messages("gpt-4", &messages)
+ }
+ _ => tiktoken_rs::num_tokens_from_messages(model.id(), &messages),
}
})
.boxed()
@@ -241,6 +241,7 @@ pub struct LanguageModelRequest {
impl LanguageModelRequest {
pub fn into_open_ai(self, model: String, max_output_tokens: Option<u32>) -> open_ai::Request {
+ let stream = !model.starts_with("o1-");
open_ai::Request {
model,
messages: self
@@ -259,7 +260,7 @@ impl LanguageModelRequest {
},
})
.collect(),
- stream: true,
+ stream,
stop: self.stop,
temperature: self.temperature,
max_tokens: max_output_tokens,
@@ -63,17 +63,22 @@ impl From<Role> for String {
#[cfg_attr(feature = "schemars", derive(schemars::JsonSchema))]
#[derive(Clone, Debug, Default, Serialize, Deserialize, PartialEq, EnumIter)]
pub enum Model {
- #[serde(rename = "gpt-3.5-turbo", alias = "gpt-3.5-turbo-0613")]
+ #[serde(rename = "gpt-3.5-turbo", alias = "gpt-3.5-turbo")]
ThreePointFiveTurbo,
- #[serde(rename = "gpt-4", alias = "gpt-4-0613")]
+ #[serde(rename = "gpt-4", alias = "gpt-4")]
Four,
- #[serde(rename = "gpt-4-turbo-preview", alias = "gpt-4-1106-preview")]
+ #[serde(rename = "gpt-4-turbo", alias = "gpt-4-turbo")]
FourTurbo,
- #[serde(rename = "gpt-4o", alias = "gpt-4o-2024-05-13")]
+ #[serde(rename = "gpt-4o", alias = "gpt-4o")]
#[default]
FourOmni,
- #[serde(rename = "gpt-4o-mini", alias = "gpt-4o-mini-2024-07-18")]
+ #[serde(rename = "gpt-4o-mini", alias = "gpt-4o-mini")]
FourOmniMini,
+ #[serde(rename = "o1-preview", alias = "o1-preview")]
+ O1Preview,
+ #[serde(rename = "o1-mini", alias = "o1-mini")]
+ O1Mini,
+
#[serde(rename = "custom")]
Custom {
name: String,
@@ -93,6 +98,8 @@ impl Model {
"gpt-4-turbo-preview" => Ok(Self::FourTurbo),
"gpt-4o" => Ok(Self::FourOmni),
"gpt-4o-mini" => Ok(Self::FourOmniMini),
+ "o1-preview" => Ok(Self::O1Preview),
+ "o1-mini" => Ok(Self::O1Mini),
_ => Err(anyhow!("invalid model id")),
}
}
@@ -101,9 +108,11 @@ impl Model {
match self {
Self::ThreePointFiveTurbo => "gpt-3.5-turbo",
Self::Four => "gpt-4",
- Self::FourTurbo => "gpt-4-turbo-preview",
+ Self::FourTurbo => "gpt-4-turbo",
Self::FourOmni => "gpt-4o",
Self::FourOmniMini => "gpt-4o-mini",
+ Self::O1Preview => "o1-preview",
+ Self::O1Mini => "o1-mini",
Self::Custom { name, .. } => name,
}
}
@@ -115,6 +124,8 @@ impl Model {
Self::FourTurbo => "gpt-4-turbo",
Self::FourOmni => "gpt-4o",
Self::FourOmniMini => "gpt-4o-mini",
+ Self::O1Preview => "o1-preview",
+ Self::O1Mini => "o1-mini",
Self::Custom {
name, display_name, ..
} => display_name.as_ref().unwrap_or(name),
@@ -123,11 +134,13 @@ impl Model {
pub fn max_token_count(&self) -> usize {
match self {
- Self::ThreePointFiveTurbo => 4096,
+ Self::ThreePointFiveTurbo => 16385,
Self::Four => 8192,
Self::FourTurbo => 128000,
Self::FourOmni => 128000,
Self::FourOmniMini => 128000,
+ Self::O1Preview => 128000,
+ Self::O1Mini => 128000,
Self::Custom { max_tokens, .. } => *max_tokens,
}
}
@@ -165,7 +165,7 @@ Zed will also use the `OPENAI_API_KEY` environment variable if it's defined.
#### OpenAI Custom Models {#openai-custom-models}
-The Zed Assistant comes pre-configured to use the latest version for common models (GPT-3.5 Turbo, GPT-4, GPT-4 Turbo, GPT-4o, GPT-4o mini). If you wish to use alternate models, perhaps a preview release or a dated model release, you can do so by adding the following to your Zed `settings.json`:
+The Zed Assistant comes pre-configured to use the latest version for common models (GPT-3.5 Turbo, GPT-4, GPT-4 Turbo, GPT-4o, GPT-4o mini). If you wish to use alternate models, perhaps a preview release or a dated model release or you wish to control the request parameters you can do so by adding the following to your Zed `settings.json`:
```json
{
@@ -176,6 +176,12 @@ The Zed Assistant comes pre-configured to use the latest version for common mode
"provider": "openai",
"name": "gpt-4o-2024-08-06",
"max_tokens": 128000
+ },
+ {
+ "name": "o1-mini",
+ "display_name": "o1-mini",
+ "max_tokens": 128000,
+ "max_completion_tokens": 20000
}
]
}
@@ -183,7 +189,7 @@ The Zed Assistant comes pre-configured to use the latest version for common mode
}
```
-You must provide the model's Context Window in the `max_tokens` parameter, this can be found [OpenAI Model Docs](https://platform.openai.com/docs/models). Custom models will be listed in the model dropdown in the assistant panel.
+You must provide the model's Context Window in the `max_tokens` parameter, this can be found [OpenAI Model Docs](https://platform.openai.com/docs/models). OpenAI `o1` models should set `max_completion_tokens` as well to avoid incurring high reasoning token costs. Custom models will be listed in the model dropdown in the assistant panel.
### Advanced configuration {#advanced-configuration}