improve rag

orca-core/src/llm/quantized.rs

@@ -122,11 +122,17 @@ impl Quantized {
         self.sample_len = sample_len;
         self
     }
+
     pub fn with_model(mut self, model: Model) -> Self {
         self.which = model;
         self
     }
 
+    pub fn with_seed(mut self, seed: u64) -> Self {
+        self.seed = seed;
+        self
+    }
+
     fn tokenizer(&self) -> anyhow::Result<Tokenizer> {
         let tokenizer_path = match &self.tokenizer {
             Some(config) => std::path::PathBuf::from(config),
@@ -314,7 +320,7 @@ impl LLM for Quantized {
             Quantized::format_chat_prompt(prompt.to_chat()?)
         };
 
-        log::info!("prompt:\n{}", &prompt);
+        log::debug!("prompt:\n{}", &prompt);
         let mut result = String::new();
         let tokens = tokenizer.encode(prompt, true).map_err(anyhow::Error::msg)?;
         if log::log_enabled!(log::Level::Debug) {

orca-core/src/memory.rs

@@ -84,6 +84,10 @@ impl ChatBuffer {
             memory: ChatPrompt(Vec::new()),
         }
     }
+
+    pub fn from_chat(chat: &ChatPrompt) -> Self {
+        Self { memory: chat.clone() }
+    }
 }
 
 impl Memory for ChatBuffer {

orca-core/src/pipeline/simple.rs

@@ -201,6 +201,7 @@ impl<M: LLM + Clone + 'static> Pipeline for LLMPipeline<M> {
             let mut locked_memory = memory.lock().await; // Lock the memory
             let mem = locked_memory.memory();
             mem.save(prompt);
+            log::debug!("Memory: {}", mem);
             self.llm.generate(mem.clone_prompt()).await?
         } else {
             self.llm.generate(prompt.clone_prompt()).await?

orca-core/src/prompt/mod.rs

@@ -235,14 +235,15 @@ pub trait Prompt: Sync + Send + Display {
     /// ```
     fn save(&mut self, _data: Box<dyn Prompt>) {
         unimplemented!("save not implemented for this prompt type");
+        // Err(anyhow::anyhow!("save not implemented for this prompt type"))
     }
 
     /// Convert the current prompt to a `ChatPrompt`.
     ///
     /// # Returns
     /// * `Result<ChatPrompt>` - The `ChatPrompt` representation of the prompt or an error.
     fn to_chat(&self) -> Result<ChatPrompt> {
-        unimplemented!("Unable to convert prompt to ChatPrompt");
+        Err(anyhow::anyhow!("Unable to convert prompt to ChatPrompt"))
     }
 
     /// Clone the current prompt into a Boxed trait object.