langhub_core/llms/
mod.rs

1use crate::types::*;
2use serde::{Deserialize, Serialize};
3use std::future::Future;
4use std::{collections::HashMap, pin::Pin};
5mod anthropic;
6mod deepseek;
7mod google;
8mod openai;
9
10pub use anthropic::{Anthropic, AnthropicModel};
11pub use deepseek::{DeepSeek, DeepSeekModel};
12pub use google::{GoogleAI, GoogleModel};
13pub use openai::{OpenAI, OpenAIModel};
14
15#[derive(Debug, Clone, Serialize, Deserialize)]
16pub struct LLMResult {
17    pub text: String,
18    pub metadata: Option<HashMap<String, serde_json::Value>>,
19}
20
21#[derive(Debug, Clone)]
22pub struct LLMOptions {
23    pub temperature: Option<f32>,
24    pub max_tokens: Option<u32>,
25    pub top_p: Option<f32>,
26    pub top_k: Option<u32>,
27    pub frequency_penalty: Option<f32>,
28    pub presence_penalty: Option<f32>,
29    pub stop_sequences: Option<Vec<String>>,
30    pub seed: Option<u64>,
31    pub response_format: Option<ResponseFormat>,
32}
33
34#[derive(Debug, Clone)]
35pub enum ResponseFormat {
36    Text,
37    Json,
38    JsonSchema { schema: serde_json::Value },
39}
40
41impl Default for LLMOptions {
42    fn default() -> Self {
43        Self {
44            temperature: Some(0.7),
45            max_tokens: Some(4096),
46            top_p: Some(0.95),
47            top_k: None,
48            frequency_penalty: None,
49            presence_penalty: None,
50            stop_sequences: None,
51            seed: None,
52            response_format: None,
53        }
54    }
55}
56
57pub trait LLM: Send + Sync {
58    fn generate(&self, prompt: &str) -> Pin<Box<dyn Future<Output = Result<LLMResult>> + Send + '_>>;
59
60    fn generate_with_options(
61        &self,
62        prompt: &str,
63        options: LLMOptions,
64    ) -> Pin<Box<dyn Future<Output = Result<LLMResult>> + Send + '_>>;
65
66    fn generate_batch(
67        &self,
68        prompts: Vec<String>,
69    ) -> Pin<Box<dyn Future<Output = Result<Vec<LLMResult>>> + Send + '_>> {
70        Box::pin(async move {
71            let mut results = Vec::new();
72            for prompt in prompts {
73                results.push(self.generate(&prompt).await?);
74            }
75            Ok(results)
76        })
77    }
78
79    fn chat(
80        &self,
81        messages: Vec<ChatMessage>,
82    ) -> Pin<Box<dyn Future<Output = Result<LLMResult>> + Send + '_>> {
83        Box::pin(async move {
84            let prompt = messages
85                .iter()
86                .map(|m| format!("{}: {}", m.role, m.content))
87                .collect::<Vec<_>>()
88                .join("\n");
89            self.generate(&prompt).await
90        })
91    }
92
93    fn get_model_name(&self) -> &str;
94    fn get_provider_name(&self) -> &str;
95    fn supports_function_calling(&self) -> bool {
96        false
97    }
98    fn supports_json_mode(&self) -> bool {
99        false
100    }
101    fn max_context_length(&self) -> Option<usize> {
102        None
103    }
104
105    fn get_provider_enum(&self) -> ModelProvider;
106}
107
108#[derive(Debug, Clone, Serialize, Deserialize)]
109pub struct ChatMessage {
110    pub role: String,
111    pub content: String,
112    pub name: Option<String>,
113    pub tool_calls: Option<Vec<ToolCall>>,
114}
115
116#[derive(Debug, Clone, Serialize, Deserialize)]
117pub struct ToolCall {
118    pub id: String,
119    pub r#type: String,
120    pub function: FunctionCall,
121}
122
123#[derive(Debug, Clone, Serialize, Deserialize)]
124pub struct FunctionCall {
125    pub name: String,
126    pub arguments: String,
127}
128
129impl ChatMessage {
130    pub fn user(content: &str) -> Self {
131        Self {
132            role: "user".to_string(),
133            content: content.to_string(),
134            name: None,
135            tool_calls: None,
136        }
137    }
138
139    pub fn assistant(content: &str) -> Self {
140        Self {
141            role: "assistant".to_string(),
142            content: content.to_string(),
143            name: None,
144            tool_calls: None,
145        }
146    }
147
148    pub fn system(content: &str) -> Self {
149        Self {
150            role: "system".to_string(),
151            content: content.to_string(),
152            name: None,
153            tool_calls: None,
154        }
155    }
156}
langhub_core/llms/mod.rs

langhub_core/llms/
mod.rs
