Tracking PR. This swaps async-openai with my fork, to make the whole thing more friendly to not-fully-compliant LLMs. Added support for NanoGPT

.gitignore

@@ -7,3 +7,5 @@ Cargo.lock
 
 # Ignore files generated by text_to_speech example
 *.mp3
+.env
+rustc-ice*

Cargo.toml

@@ -23,7 +23,6 @@ regex = "1.10.4"
 log = "0.4.21"
 html-escape = "0.2.13"
 reqwest-eventsource = "0.6.0"
-async-openai = "0.28.1"
 mockito = "1.4.0"
 tiktoken-rs = "0.5.8"
 sqlx = { version = "0.8.0", default-features = false, features = [
@@ -43,8 +42,8 @@ surrealdb = { version = "2.0.2", optional = true, default-features = false }
 csv = "1.3.0"
 urlencoding = "2.1.3"
 lopdf = { version = "0.34.0", features = ["nom_parser"], optional = true }
-pdf-extract = { version = "0.7.8", optional = true  }
-thiserror = "2.0.0"
+pdf-extract = { version = "0.7.8", optional = true }
+thiserror = "2.*"
 futures-util = "0.3.30"
 async-stream = "0.3.5"
 tokio-stream = "0.1.15"
@@ -84,7 +83,11 @@ ollama-rs = { version = "0.2.0", optional = true, features = [
     "chat-history",
 ] }
 mistralai-client = { version = "0.14.0", optional = true }
-
+dotenvy = "0.15.7"
+tracing-subscriber = "0.3.19"
+tracing = "0.1.41"
+byte_string = "1.0.0"
+async-openai = { path = "../async-openai/async-openai" }
 
 [features]
 default = []
@@ -124,3 +127,9 @@ testcontainers = "0.23"
 
 [build-dependencies]
 cc = { version = "1", optional = true }
+
+[profile.dev]
+lto = false
+opt-level = 0
+incremental = true
+codegen-units = 256

examples/nanogpt_models.rs

@@ -0,0 +1,46 @@
+use langchain_rust::llm::nanogpt::*;
+use dotenvy::dotenv;
+use tracing::Level;
+use tracing_subscriber::FmtSubscriber;
+
+#[tokio::main]
+async fn main() {
+    dotenv().expect(".env file not found");
+
+    let subscriber = FmtSubscriber::builder()
+        .with_max_level(Level::INFO)
+        .finish();
+    tracing::subscriber::set_global_default(subscriber).unwrap();
+
+    let nano_config = OpenAIConfig::new()
+        .with_api_base("https://nano-gpt.com/api/v1")
+        .with_api_key(dotenvy::var("NANOGPT_KEY").expect("NANOGPT_KEY not set"));
+
+    let client = NanoGPT::default().with_config(nano_config);
+
+    // Basic model listing
+    let basic_models = client.get_models(false)
+        .await
+        .unwrap_or_else(|e| panic!("Error fetching models: {:?}", e));
+
+    println!("\nBasic Models:");
+    for model in &basic_models.data {
+        println!("- {} ({})", model.id, model.owned_by);
+    }
+
+    // Detailed model listing with pricing
+    let detailed_models = client.get_models(true)
+        .await
+        .unwrap_or_else(|e| panic!("Error fetching detailed models: {:?}", e));
+
+    println!("\nDetailed Models:");
+    for model in detailed_models.data {
+        println!("Model: {}", model.name.unwrap_or("Unnamed".into()));
+        println!("ID: {}", model.id);
+        if let Some(pricing) = model.pricing {
+            println!("Pricing: ${}/M tokens (input), ${}/M tokens (output)", 
+                pricing.prompt, pricing.completion);
+        }
+        println!("---");
+    }
+}

examples/nanogpt_stream.rs

@@ -0,0 +1,67 @@
+use futures::StreamExt;
+use langchain_rust::{
+    chain::{Chain, LLMChainBuilder},
+    fmt_message, fmt_template,
+    message_formatter,
+    prompt::HumanMessagePromptTemplate,
+    prompt_args,
+    schemas::messages::Message,
+    template_fstring,
+};
+
+use langchain_rust::llm::nanogpt::*;
+
+use dotenvy::{self, dotenv};
+use tracing::Level;
+use tracing_subscriber::FmtSubscriber;
+
+#[tokio::main]
+async fn main() {
+    dotenv().expect(".env file not found");
+    // Serving as an exmaple of using non major LLM provider, tracing logs are incredibly useful here.
+
+    let subscriber = FmtSubscriber::builder()
+        .with_max_level(Level::TRACE)
+        .with_file(true)
+        .with_line_number(true)
+        .finish();
+
+    tracing::subscriber::set_global_default(subscriber).expect("setting default subscriber failed");
+
+    let nano_config = OpenAIConfig::new()
+        .with_api_base("https://nano-gpt.com/api/v1")
+        .with_api_key(dotenvy::var("NANOGPT_KEY").expect("provide key"));
+
+    let open_ai = NanoGPT::default()
+        .with_config(nano_config)
+        .with_model("deepseek-ai/DeepSeek-V3.1:thinking");
+
+    let prompt = message_formatter![
+        fmt_message!(Message::new_system_message(
+            "You excel at metacognition. You are a true philosopher, becacuse you are a piece of consciousness with no ego"
+        )),
+        fmt_template!(HumanMessagePromptTemplate::new(template_fstring!(
+            "{input}", "input"
+        )))
+    ];
+
+    let chain = LLMChainBuilder::new()
+        .prompt(prompt)
+        .llm(open_ai.clone())
+        .build()
+        .unwrap();
+
+    let mut stream = chain
+        .stream(prompt_args! {
+        "input" => "Contemplate the intricate web of causality and how that affects one's being, over and over",
+           })
+        .await
+        .unwrap();
+
+    while let Some(result) = stream.next().await {
+        match result {
+            Ok(value) => value.to_stdout().unwrap(),
+            Err(e) => println!("protocol non compliance: {:?}", e),
+        }
+    }
+}

src/document_loaders/mod.rs

@@ -38,3 +38,5 @@ pub use dir_loader::*;
 mod source_code_loader;
 #[cfg(feature = "tree-sitter")]
 pub use source_code_loader::*;
+
+pub use dotenvy;

src/embedding/openai/openai_embedder.rs

@@ -14,7 +14,7 @@ pub struct OpenAiEmbedder<C: Config> {
     model: String,
 }
 
-impl<C: Config + Send + Sync + 'static> Into<Box<dyn Embedder>> for OpenAiEmbedder<C> {
+impl<C: Config + Send + Sync + 'static + Clone> Into<Box<dyn Embedder>> for OpenAiEmbedder<C> {
     fn into(self) -> Box<dyn Embedder> {
         Box::new(self)
     }
@@ -46,7 +46,7 @@ impl Default for OpenAiEmbedder<OpenAIConfig> {
 }
 
 #[async_trait]
-impl<C: Config + Send + Sync> Embedder for OpenAiEmbedder<C> {
+impl<C: Config + Send + Sync + Clone> Embedder for OpenAiEmbedder<C> {
     async fn embed_documents(&self, documents: &[String]) -> Result<Vec<Vec<f64>>, EmbedderError> {
         let client = Client::with_config(self.config.clone());
 

src/language_models/error.rs

@@ -6,7 +6,7 @@ use serde_json::Error as SerdeJsonError;
 use thiserror::Error;
 use tokio::time::error::Elapsed;
 
-use crate::llm::{AnthropicError, DeepseekError, QwenError};
+use crate::llm::{nanogpt::NanoGPTResponse, AnthropicError, DeepseekError, QwenError};
 
 #[derive(Error, Debug)]
 pub enum LLMError {
@@ -49,4 +49,7 @@ pub enum LLMError {
 
     #[error("Error: {0}")]
     OtherError(String),
+
+    #[error("NanoGPT error")]
+    NanoGPTError(NanoGPTResponse),
 }

src/llm/mod.rs

@@ -12,3 +12,5 @@ pub use qwen::*;
 
 pub mod deepseek;
 pub use deepseek::*;
+
+pub mod nanogpt;