@fugood/bricks-project 2.21.10 → 2.21.12

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@fugood/bricks-project",
-  "version": "2.21.10",
+  "version": "2.21.12",
   "main": "index.ts",
   "scripts": {
     "build": "node scripts/build.js"

package/types/generators.ts

@@ -6622,6 +6622,8 @@ Default property:
     useMmap?: boolean | DataLink
     /* Use Flash Attention for inference (Recommended with GPU enabled) */
     useFlashAttn?: boolean | DataLink
+    /* Use full-size SWA cache. May improve performance for multiple sequences but uses more memory. */
+    useSwaFull?: boolean | DataLink
     /* KV cache data type for the K (Default: f16) */
     cacheKType?: 'f16' | 'f32' | 'q8_0' | 'q4_0' | 'q4_1' | 'iq4_nl' | 'q5_0' | 'q5_1' | DataLink
     /* KV cache data type for the V (Default: f16) */
@@ -6858,6 +6860,7 @@ Default property:
   "vocoderUrl": "https://huggingface.co/ggml-org/WavTokenizer/resolve/main/WavTokenizer-Large-75-F16.gguf",
   "vocoderHashType": "sha256",
   "vocoderHash": "2356baa8631cc2995ea3465196a017a2733600d849a91180c0f97fa7fb375bbe",
+  "vocoderBatchSize": 4096,
   "outputType": "play",
   "cacheGenerated": true,
   "autoInferEnable": false,
@@ -6868,10 +6871,11 @@ Default property:
   "completionTopK": 40,
   "completionTopP": 0.9,
   "completionMinP": 0.05,
-  "useGuideToken": false,
-  "contextSize": 4096,
-  "batchSize": 512,
-  "maxThreads": 1,
+  "useGuideToken": true,
+  "contextSize": 8192,
+  "batchSize": 8192,
+  "microBatchSize": 512,
+  "maxThreads": 2,
   "accelVariant": "default",
   "mainGpu": 0,
   "gpuLayers": 0,
@@ -6896,6 +6900,8 @@ Default property:
     vocoderHashType?: 'md5' | 'sha256' | 'sha1' | DataLink
     /* Hash of vocoder model */
     vocoderHash?: string | DataLink
+    /* Batch size of vocoder model */
+    vocoderBatchSize?: number | DataLink
     /* Output mode */
     outputType?: 'play' | 'file' | DataLink
     /* Enable cache for generated audio */
@@ -6922,12 +6928,16 @@ Default property:
     completionMinP?: number | DataLink
     /* Set the random number generator (RNG) seed (default: -1, -1 = random seed) */
     completionSeed?: number | DataLink
+    /* Number of tokens to predict */
+    completionPredict?: number | DataLink
     /* Enable guide token to help prevent hallucinations by forcing the TTS to use the correct words. */
     useGuideToken?: boolean | DataLink
     /* Context size, for OutTTS recommended 4096 ~ 8192 (Default to 4096) */
     contextSize?: number | DataLink
     /* Logical batch size for prompt processing */
     batchSize?: number | DataLink
+    /* Physical batch size for prompt processing */
+    microBatchSize?: number | DataLink
     /* Number of threads */
     maxThreads?: number | DataLink
     /* Accelerator variant (Only for desktop)  

package/utils/event-props.ts

@@ -841,6 +841,7 @@ export const templateEventPropsMap = {
     ],
     onCompletionFunctionCall: [
       'GENERATOR_LLM_COMPLETION_SESSION_KEY', // type: string
+      'GENERATOR_LLM_COMPLETION_FUNCTION_CALL_ID', // type: string
       'GENERATOR_LLM_COMPLETION_FUNCTION_CALL_NAME', // type: string
       'GENERATOR_LLM_COMPLETION_FUNCTION_ARGUMENTS', // type: object
       'GENERATOR_LLM_COMPLETION_FUNCTION_DETAILS', // type: object