fix tapestry instance increment, upgrade

Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 publish = true
 name = "llm-weaver"
-version = "0.1.92"
+version = "0.1.93"
 edition = "2021"
 description = "Manage long conversations with any LLM"
 readme = "README.md"

src/lib.rs

@@ -227,6 +227,11 @@ pub trait Config: Debug + Sized + Clone + Default + Send + Sync + 'static {
 	///
 	/// Defaults to `85%`
 	const TOKEN_THRESHOLD_PERCENTILE: BoundedU8<0, 100> = BoundedU8::new(85).unwrap();
+	/// Ensures that the maximum completion tokens is at least the minimum response length.
+	///
+	/// If the maximum completion tokens is less than the minimum response length, a summary
+	/// will be generated and a new tapestry fragment will be created.
+	const MINIMUM_RESPONSE_LENGTH: u64;
 
 	/// The LLM to use for generating responses to prompts.
 	type PromptModel: Llm<Self>;
@@ -396,15 +401,17 @@ pub trait Loom<T: Config> {
 		);
 		req_msgs.append(&mut ctx_msgs);
 
-		// New messages are not added here yet since we first calculate if the new messages would
+		// New messages are not added here yet since we first calculate if the new `msgs` would
 		// have the tapestry fragment exceed the maximum token limit and require a summary
 		// generation resulting in a new tapestry fragment.
 		//
 		// Either we are starting a new tapestry fragment with the instruction and summary messages
 		// or we are continuing the current tapestry fragment.
 		let msgs_tokens = Self::count_tokens_in_messages(msgs.iter());
 		let does_exceeding_max_token_limit = max_prompt_tokens_limit <=
-			current_tapestry_fragment.context_tokens.saturating_add(&msgs_tokens);
+			req_msgs.tokens.saturating_add(&msgs_tokens).saturating_add(
+				&PromptModelTokens::<T>::from_u64(T::MINIMUM_RESPONSE_LENGTH).unwrap(),
+			);
 
 		let (mut tapestry_fragment_to_persist, was_summary_generated) =
 			if does_exceeding_max_token_limit {
@@ -445,9 +452,7 @@ pub trait Loom<T: Config> {
 		req_msgs.extend(msgs.iter().map(|m| m.clone().into()).collect::<Vec<_>>());
 
 		// Tokens available for LLM response which would not exceed maximum token limit
-		let max_completion_tokens = max_prompt_tokens_limit
-			.saturating_sub(&tapestry_fragment_to_persist.context_tokens)
-			.saturating_sub(&req_msgs.tokens);
+		let max_completion_tokens = max_prompt_tokens_limit.saturating_sub(&req_msgs.tokens);
 
 		// Execute prompt to LLM
 		let response = prompt_llm_config
@@ -510,6 +515,8 @@ pub trait Loom<T: Config> {
 				.ctx_msgs_to_prompt_requests(tapestry_fragment.context_messages.as_slice()),
 		);
 
+		println!("prompting summary model");
+
 		let res = summary_model_config
 			.model
 			.prompt(

src/mock.rs

@@ -73,6 +73,8 @@ impl TapestryId for TestTapestryId {
 pub struct TestApp;
 impl Config for TestApp {
 	const TOKEN_THRESHOLD_PERCENTILE: BoundedU8<0, 100> = BoundedU8::new(70).unwrap();
+	const MINIMUM_RESPONSE_LENGTH: u64 = 300;
+
 	type PromptModel = TestLlm;
 	type SummaryModel = TestLlm;
 	type Chest = TestChest;

src/storage.rs

@@ -106,17 +106,16 @@ impl<T: Config> TapestryChestHandler<T> for TapestryChest {
 		let mut con = client.get_connection()?;
 		let base_key = &tapestry_id.base_key();
 
-		let tapestry_instance = verify_and_get_instance(&mut con, base_key, None).await?;
+		let mut tapestry_instance =
+			verify_and_get_instance(&mut con, base_key, None).await?.unwrap_or(0);
 
 		redis::transaction(&mut con, &[base_key], |con, pipe| {
-			let mut tapestry_instance = match tapestry_instance {
-				Some(instance) => instance,
-				None => {
-					pipe.hset(base_key, INSTANCE_COUNT, 1).ignore();
-					debug!("Saved \"instance_count\" member to {} key", base_key);
+			// If the tapestry does not exist (i.e. instance is at 0), then set it to 1
+			if tapestry_instance == 0 {
+				pipe.hset(base_key, INSTANCE_COUNT, 1).ignore();
+				debug!("Saved \"instance_count\" member to {} key", base_key);
 
-					1
-				},
+				tapestry_instance = 1
 			};
 
 			if increment {
@@ -149,7 +148,7 @@ impl<T: Config> TapestryChestHandler<T> for TapestryChest {
 			LoomError::from(StorageError::Redis(e))
 		})?;
 
-		Ok(tapestry_instance.unwrap_or(1))
+		Ok(tapestry_instance)
 	}
 
 	async fn save_tapestry_metadata<