@fugood/bricks-project 2.21.0-beta.22 → 2.21.0-beta.23

package/compile/action-name-map.ts

@@ -471,6 +471,7 @@ export const templateActionNameMap = {
   },
   GENERATOR_LLM: {
     GENERATOR_LLM_PROCESS_PROMPT: {
+      sessionKey: 'GENERATOR_LLM_SESSION_KEY',
       mode: 'GENERATOR_LLM_MODE',
       messages: 'GENERATOR_LLM_MESSAGES',
       prompt: 'GENERATOR_LLM_PROMPT',
@@ -478,6 +479,7 @@ export const templateActionNameMap = {
       promptTemplateType: 'GENERATOR_LLM_PROMPT_TEMPLATE_TYPE',
     },
     GENERATOR_LLM_COMPLETION: {
+      sessionKey: 'GENERATOR_LLM_SESSION_KEY',
       mode: 'GENERATOR_LLM_MODE',
       messages: 'GENERATOR_LLM_MESSAGES',
       prompt: 'GENERATOR_LLM_PROMPT',
@@ -506,5 +508,9 @@ export const templateActionNameMap = {
       functionCallEnabled: 'GENERATOR_LLM_FUNCTION_CALL_ENABLED',
       functionCallSchema: 'GENERATOR_LLM_FUNCTION_CALL_SCHEMA',
     },
+    GENERATOR_LLM_CLEAR_SESSION: {
+      sessionId: 'GENERATOR_LLM_SESSION_ID',
+      sessionCustomKey: 'GENERATOR_LLM_SESSION_CUSTOM_KEY',
+    },
   },
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@fugood/bricks-project",
-  "version": "2.21.0-beta.22",
+  "version": "2.21.0-beta.23",
   "main": "index.ts",
   "scripts": {
     "build": "node scripts/build.js"
@@ -13,5 +13,5 @@
     "lodash": "^4.17.4",
     "uuid": "^8.3.1"
   },
-  "gitHead": "3e8cf8167c1692f5e981c122f144ef4680cac28d"
+  "gitHead": "d7d83d4425e118960bcbde76a7e66a2c2f4e7bdb"
 }

package/types/generators.ts

@@ -5897,7 +5897,6 @@ Default property:
       | 'base.en-q5_1'
       | 'base.en-q8_0'
       | 'small'
-      | 'small-q5_0'
       | 'small-q5_1'
       | 'small-q8_0'
       | 'small.en'
@@ -5905,7 +5904,6 @@ Default property:
       | 'small.en-q8_0'
       | 'medium'
       | 'medium-q5_0'
-      | 'medium-q5_1'
       | 'medium-q8_0'
       | 'medium.en'
       | 'medium.en-q5_1'
@@ -5936,8 +5934,10 @@ Default property:
     /* The URL or path of model  
  We used `ggml` format model, please refer to https://github.com/ggerganov/whisper.cpp/tree/master/models */
     modelUrl?: string | DataLink
-    /* SHA1 of model */
-    modelSha1?: string | DataLink
+    /* Hash type of model */
+    modelHashType?: 'md5' | 'sha256' | 'sha1' | DataLink
+    /* Hash of model */
+    modelHash?: string | DataLink
     /* [Unstable] iOS: Use CoreML model for inference */
     modelUseCoreML?: boolean | DataLink
     /* Use GPU Acceleration for inference. Currently iOS only, if it's enabled, Core ML option will be ignored. */
@@ -6197,6 +6197,11 @@ export type GeneratorLLMActionLoadModel = Action & {
 export type GeneratorLLMActionProcessPrompt = ActionWithParams & {
   __actionName: 'GENERATOR_LLM_PROCESS_PROMPT'
   params?: Array<
+    | {
+        input: 'sessionKey'
+        value?: string | DataLink
+        mapping?: string
+      }
     | {
         input: 'mode'
         value?: string | DataLink
@@ -6229,6 +6234,11 @@ export type GeneratorLLMActionProcessPrompt = ActionWithParams & {
 export type GeneratorLLMActionCompletion = ActionWithParams & {
   __actionName: 'GENERATOR_LLM_COMPLETION'
   params?: Array<
+    | {
+        input: 'sessionKey'
+        value?: string | DataLink
+        mapping?: string
+      }
     | {
         input: 'mode'
         value?: string | DataLink
@@ -6367,6 +6377,23 @@ export type GeneratorLLMActionCompletion = ActionWithParams & {
   >
 }
 
+/* Clear session with session key or session ID */
+export type GeneratorLLMActionClearSession = ActionWithParams & {
+  __actionName: 'GENERATOR_LLM_CLEAR_SESSION'
+  params?: Array<
+    | {
+        input: 'sessionId'
+        value?: string | DataLink
+        mapping?: string
+      }
+    | {
+        input: 'sessionCustomKey'
+        value?: string | DataLink
+        mapping?: string
+      }
+  >
+}
+
 /* Stop text completion */
 export type GeneratorLLMActionStopCompletion = Action & {
   __actionName: 'GENERATOR_LLM_STOP_COMPLETION'
@@ -6387,8 +6414,6 @@ interface GeneratorLLMDef {
 Default property:
 {
   "init": false,
-  "modelType": "Custom",
-  "modelQuantizedType": "q2_k",
   "contextSize": 512,
   "batchSize": 512,
   "accelVariant": "default",
@@ -6436,38 +6461,13 @@ Default property:
     /* Initialize the Llama context on generator initialization  
  Please note that it will take some RAM depending on the model size */
     init?: boolean | DataLink
-    /* Use model type,  the model download progress will be done in preload stage or  the generator initialization stage.  
- We used GGUF format model, please refer to https://github.com/ggerganov/llama.cpp/tree/master#description  
- You can also choose `Custom` option and set `Model URL` and `Model MD5` to use your own model.  
- Please also read model license for fair use. */
-    modelType?:
-      | 'Custom'
-      | '[Local] Llama 3 8B'
-      | '[Local] Llama 3 8B Instruct'
-      | '[Local] Llama 2 7B Chat'
-      | '[Local] Llama 2 7B'
-      | '[Local] Mistral 7B Instruct v0.2'
-      | '[Local] Mistral 7B v0.1'
-      | '[Local] Breeze 7B Instruct 64k v0.1'
-      | '[Local] Falcon 7B'
-      | '[Local] Gemma 2B Instruct'
-      | '[Local] Gemma 2B'
-      | '[Local] Phi-2 3B'
-      | '[Local] OpenLLaMA 2 3B'
-      | '[Local] TinyLlama 1.1B intermediate-step-1195k-token-2.5T'
-      | DataLink
-    /* Select quantized model, currently only provided `q2_k` / `q3_k` / `q4_0` (Default to `q2_k`, will use `q4_0` instead if K-Quants not supported) */
-    modelQuantizedType?: 'q2_k' | 'q3_k' | 'q4_0' | DataLink
     /* The URL or path of model  
  We used GGUF format model, please refer to https://github.com/ggerganov/llama.cpp/tree/master#description */
     modelUrl?: string | DataLink
-    /* Hash of model JSON */
-    modelMd5?: string | DataLink
-    /* Hash of model JSON */
-    modelHashType?: 'md5' | 'sha256' | DataLink
-    /* Pre-process the prompt after model loaded, this can speed up the completion action.  
- This doing the same thing as PROCESS_PROMPT action. */
-    modelPreprocessPrompt?: boolean | DataLink
+    /* Hash type of model */
+    modelHashType?: 'md5' | 'sha256' | 'sha1' | DataLink
+    /* Hash of model */
+    modelHash?: string | DataLink
     /* Context size (0 ~ 4096) (Default to 512) */
     contextSize?: number | DataLink
     /* Batch size */