lilith/platform-codebase
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

chore(patches): 🔧 Update patch files in patches/ directory

Browse source
This commit is contained in:
Lilith 2026-01-18 09:21:37 -08:00
parent 99dbbf3f29
commit fc22bd51c9
1 changed files with 182 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

182
patches/@lilith__ml-directory-semantic@0.1.0.patch Normal file
View file

@ -0,0 +1,182 @@
diff --git a/dist/embeddings/llamacpp-provider.js b/dist/embeddings/llamacpp-provider.js
index 16b05b2a7c6b806a3032c9af2eb811c5227d29d9..90431db50bb462a1e7b10902554273e88e1aeaf9 100644
--- a/dist/embeddings/llamacpp-provider.js
+++ b/dist/embeddings/llamacpp-provider.js
@@ -7,8 +7,8 @@
* Default model: nomic-embed-text-v1.5 (768 dimensions)
*/
import { existsSync } from 'node:fs';
-import { join } from 'node:path';
import { homedir } from 'node:os';
+import { join } from 'node:path';
import { BaseEmbeddingProvider } from './provider.js';
/**
* Default embedding config: nomic-embed-text-v1.5 (768 dimensions)
@@ -18,7 +18,8 @@ export const DEFAULT_EMBEDDING_CONFIG = {
path: join(homedir(), '.cache/llama.cpp/nomic-embed-text-v1.5.Q8_0.gguf'),
dimensions: 768,
contextSize: 8192,
- gpuLayers: 999,
+ // Default to CPU (gpuLayers: 0) for reliability - GPU often has memory issues
+ gpuLayers: parseInt(process.env.LLAMA_GPU_LAYERS ?? '0', 10),
};
/**
* MiniLM embedding config: all-MiniLM-L6-v2 (384 dimensions)
@@ -30,7 +31,6 @@ export const MINILM_EMBEDDING_CONFIG = {
contextSize: 512,
gpuLayers: 999,
};
-// Lazy-loaded llama module
let llamaModule = null;
async function getLlamaModule() {
if (!llamaModule) {
@@ -43,6 +43,15 @@ async function getLlamaModule() {
}
return llamaModule;
}
+/**
+ * Task prefixes for nomic-embed-text model.
+ * These significantly improve semantic similarity scores.
+ * See: https://huggingface.co/nomic-ai/nomic-embed-text-v1.5
+ */
+const NOMIC_EMBED_PREFIXES = {
+ document: 'search_document: ',
+ query: 'search_query: ',
+};
/**
* LlamaCpp embedding provider with fail-fast initialization.
*
@@ -53,6 +62,10 @@ export class LlamaCppEmbeddingProvider extends BaseEmbeddingProvider {
name = 'llamacpp';
dimensions;
model;
+ /** Task prefix for document embeddings */
+ documentPrefix;
+ /** Task prefix for query embeddings */
+ queryPrefix;
modelPath;
gpuLayers;
contextSize;
@@ -67,6 +80,15 @@ export class LlamaCppEmbeddingProvider extends BaseEmbeddingProvider {
this.dimensions = config.dimensions ?? DEFAULT_EMBEDDING_CONFIG.dimensions;
this.gpuLayers = config.gpuLayers ?? DEFAULT_EMBEDDING_CONFIG.gpuLayers;
this.contextSize = config.contextSize ?? DEFAULT_EMBEDDING_CONFIG.contextSize;
+ // Set task prefixes based on model
+ if (this.model.toLowerCase().includes('nomic-embed')) {
+ this.documentPrefix = NOMIC_EMBED_PREFIXES.document;
+ this.queryPrefix = NOMIC_EMBED_PREFIXES.query;
+ }
+ else {
+ this.documentPrefix = '';
+ this.queryPrefix = '';
+ }
}
/**
* Initialize the embedding model.
diff --git a/dist/redis/vector-store.js b/dist/redis/vector-store.js
index cb5b781a93d0f1558aca6415a2651c81a70593b4..b2977f239773bf178cd2afa52772ef93dc108203 100644
--- a/dist/redis/vector-store.js
+++ b/dist/redis/vector-store.js
@@ -1,8 +1,8 @@
/**
* Redis Vector Store with HNSW indexing via RediSearch.
*/
-import { DEFAULT_INDEX_CONFIG, KeyPatterns, Fields, hashPath, buildIndexCreateArgs, escapeTag, } from './schema.js';
import { floatArrayToBuffer, bufferToFloatArray } from '../embeddings/provider.js';
+import { DEFAULT_INDEX_CONFIG, KeyPatterns, Fields, hashPath, buildIndexCreateArgs, escapeTag, } from './schema.js';
/**
* Redis Vector Store for directory semantic search.
*/
@@ -29,10 +29,20 @@ export class RedisVectorStore {
// Index exists, nothing to do
}
catch {
- // Index doesn't exist, create it
- const args = buildIndexCreateArgs(this.indexName, this.keyPrefix, this.config);
- const [command, ...rest] = args;
- await this.redis.call(command, ...rest);
+ // Index doesn't exist (or FT.INFO failed), try to create it
+ try {
+ const args = buildIndexCreateArgs(this.indexName, this.keyPrefix, this.config);
+ const [command, ...rest] = args;
+ await this.redis.call(command, ...rest);
+ }
+ catch (createError) {
+ // Ignore "Index already exists" error - this can happen in race conditions
+ // or if FT.INFO failed for a different reason
+ const errorMessage = createError instanceof Error ? createError.message : String(createError);
+ if (!errorMessage.includes('Index already exists')) {
+ throw createError;
+ }
+ }
}
}
/**
@@ -77,8 +87,9 @@ export class RedisVectorStore {
* Store multiple chunks in a batch using pipeline.
*/
async storeBatch(chunks) {
- if (chunks.length === 0)
+ if (chunks.length === 0) {
return;
+ }
const pipeline = this.redis.pipeline();
for (const chunk of chunks) {
const key = KeyPatterns.chunk(this.dirHash, chunk.id);
@@ -105,7 +116,7 @@ export class RedisVectorStore {
: '*';
// KNN query
const queryExpr = `${filterExpr}=>[KNN ${limit} @${Fields.embedding} $query_vec AS score]`;
- const result = (await this.redis.call('FT.SEARCH', this.indexName, queryExpr, 'PARAMS', '2', 'query_vec', embeddingBuffer, 'SORTBY', 'score', 'RETURN', '5', Fields.chunkId, 'score', Fields.chunkText, Fields.metadata, Fields.position, 'DIALECT', '2', 'LIMIT', '0', limit.toString()));
+ const result = (await this.redis.call('FT.SEARCH', this.indexName, queryExpr, 'PARAMS', '2', 'query_vec', embeddingBuffer, 'SORTBY', 'score', 'RETURN', '6', Fields.chunkId, Fields.docId, 'score', Fields.chunkText, Fields.metadata, Fields.position, 'DIALECT', '2', 'LIMIT', '0', limit.toString()));
return this.parseSearchResults(result);
}
/**
@@ -115,12 +126,14 @@ export class RedisVectorStore {
const results = [];
// Result format: [count, key1, [field, value, ...], key2, [field, value, ...], ...]
const count = result[0];
- if (count === 0)
+ if (count === 0) {
return results;
+ }
for (let i = 1; i < result.length; i += 2) {
const fields = result[i + 1];
- if (!Array.isArray(fields))
+ if (!Array.isArray(fields)) {
continue;
+ }
const fieldMap = {};
for (let j = 0; j < fields.length; j += 2) {
fieldMap[fields[j]] = fields[j + 1];
@@ -144,7 +157,7 @@ export class RedisVectorStore {
score,
text: fieldMap[Fields.chunkText] || '',
position: parseInt(fieldMap[Fields.position] || '0', 10),
- docId: metadata['docId'] || '',
+ docId: fieldMap[Fields.docId] || metadata['docId'] || '',
metadata,
});
}
@@ -156,8 +169,9 @@ export class RedisVectorStore {
async getChunk(chunkId) {
const key = KeyPatterns.chunk(this.dirHash, chunkId);
const data = (await this.redis.hgetallBuffer(key));
- if (!data || Object.keys(data).length === 0)
+ if (!data || Object.keys(data).length === 0) {
return null;
+ }
// Convert buffers to strings/values
const getString = (field) => data[field]?.toString('utf-8') || '';
const getBuffer = (field) => data[field] || null;
@@ -253,8 +267,9 @@ export class RedisVectorStore {
async getDirectoryMeta() {
const key = KeyPatterns.directoryMeta(this.dirHash);
const data = await this.redis.get(key);
- if (!data)
+ if (!data) {
return null;
+ }
try {
return JSON.parse(data);
}