Files
tkmind_go/crates/goose/src/providers/groq.rs
T

153 lines
5.1 KiB
Rust

use super::errors::ProviderError;
use crate::message::Message;
use crate::model::ModelConfig;
use crate::providers::base::{ConfigKey, Provider, ProviderMetadata, ProviderUsage, Usage};
use crate::providers::formats::openai::{create_request, get_usage, response_to_message};
use crate::providers::utils::get_model;
use anyhow::Result;
use async_trait::async_trait;
use mcp_core::Tool;
use reqwest::{Client, StatusCode};
use serde_json::Value;
use std::time::Duration;
pub const GROQ_API_HOST: &str = "https://api.groq.com";
pub const GROQ_DEFAULT_MODEL: &str = "llama-3.3-70b-versatile";
pub const GROQ_KNOWN_MODELS: &[&str] = &["gemma2-9b-it", "llama-3.3-70b-versatile"];
pub const GROQ_DOC_URL: &str = "https://console.groq.com/docs/models";
#[derive(serde::Serialize)]
pub struct GroqProvider {
#[serde(skip)]
client: Client,
host: String,
api_key: String,
model: ModelConfig,
}
impl Default for GroqProvider {
fn default() -> Self {
let model = ModelConfig::new(GroqProvider::metadata().default_model);
GroqProvider::from_env(model).expect("Failed to initialize Groq provider")
}
}
impl GroqProvider {
pub fn from_env(model: ModelConfig) -> Result<Self> {
let config = crate::config::Config::global();
let api_key: String = config.get_secret("GROQ_API_KEY")?;
let host: String = config
.get("GROQ_HOST")
.unwrap_or_else(|_| GROQ_API_HOST.to_string());
let client = Client::builder()
.timeout(Duration::from_secs(600))
.build()?;
Ok(Self {
client,
host,
api_key,
model,
})
}
async fn post(&self, payload: Value) -> anyhow::Result<Value, ProviderError> {
let url = format!(
"{}/openai/v1/chat/completions",
self.host.trim_end_matches('/')
);
let response = self
.client
.post(&url)
.header("Authorization", format!("Bearer {}", self.api_key))
.json(&payload)
.send()
.await?;
let status = response.status();
let payload: Option<Value> = response.json().await.ok();
match status {
StatusCode::OK => payload.ok_or_else( || ProviderError::RequestFailed("Response body is not valid JSON".to_string()) ),
StatusCode::UNAUTHORIZED | StatusCode::FORBIDDEN => {
Err(ProviderError::Authentication(format!("Authentication failed. Please ensure your API keys are valid and have the required permissions. \
Status: {}. Response: {:?}", status, payload)))
}
StatusCode::PAYLOAD_TOO_LARGE => {
Err(ProviderError::ContextLengthExceeded(format!("{:?}", payload)))
}
StatusCode::TOO_MANY_REQUESTS => {
Err(ProviderError::RateLimitExceeded(format!("{:?}", payload)))
}
StatusCode::INTERNAL_SERVER_ERROR | StatusCode::SERVICE_UNAVAILABLE => {
Err(ProviderError::ServerError(format!("{:?}", payload)))
}
_ => {
tracing::debug!(
"{}", format!("Provider request failed with status: {}. Payload: {:?}", status, payload)
);
Err(ProviderError::RequestFailed(format!("Request failed with status: {}", status)))
}
}
}
}
#[async_trait]
impl Provider for GroqProvider {
fn metadata() -> ProviderMetadata {
ProviderMetadata::new(
"groq",
"Groq",
"Fast inference with Groq hardware",
GROQ_DEFAULT_MODEL,
GROQ_KNOWN_MODELS.iter().map(|&s| s.to_string()).collect(),
GROQ_DOC_URL,
vec![
ConfigKey::new("GROQ_API_KEY", true, true, None),
ConfigKey::new("GROQ_HOST", false, false, Some(GROQ_API_HOST)),
],
)
}
fn get_model_config(&self) -> ModelConfig {
self.model.clone()
}
#[tracing::instrument(
skip(self, system, messages, tools),
fields(model_config, input, output, input_tokens, output_tokens, total_tokens)
)]
async fn complete(
&self,
system: &str,
messages: &[Message],
tools: &[Tool],
) -> anyhow::Result<(Message, ProviderUsage), ProviderError> {
let payload = create_request(
&self.model,
system,
messages,
tools,
&super::utils::ImageFormat::OpenAi,
)?;
let response = self.post(payload.clone()).await?;
let message = response_to_message(response.clone())?;
let usage = match get_usage(&response) {
Ok(usage) => usage,
Err(ProviderError::UsageError(e)) => {
tracing::warn!("Failed to get usage data: {}", e);
Usage::default()
}
Err(e) => return Err(e),
};
let model = get_model(&response);
super::utils::emit_debug_trace(self, &payload, &response, &usage);
Ok((message, ProviderUsage::new(model, usage)))
}
}