node-llama-cpp 3.0.0-beta.15 → 3.0.0-beta.17

package/dist/chatWrappers/utils/resolveChatWrapper.js

@@ -1,5 +1,6 @@
 import { parseModelFileName } from "../../utils/parseModelFileName.js";
-import { LlamaChatWrapper } from "../LlamaChatWrapper.js";
+import { Llama3ChatWrapper } from "../Llama3ChatWrapper.js";
+import { Llama2ChatWrapper } from "../Llama2ChatWrapper.js";
 import { ChatMLChatWrapper } from "../ChatMLChatWrapper.js";
 import { GeneralChatWrapper } from "../GeneralChatWrapper.js";
 import { FalconChatWrapper } from "../FalconChatWrapper.js";
@@ -11,7 +12,7 @@ import { TemplateChatWrapper } from "../generic/TemplateChatWrapper.js";
 import { getConsoleLogPrefix } from "../../utils/getConsoleLogPrefix.js";
 import { isJinjaTemplateEquivalentToSpecializedChatWrapper } from "./isJinjaTemplateEquivalentToSpecializedChatWrapper.js";
 export const specializedChatWrapperTypeNames = Object.freeze([
-    "general", "llamaChat", "alpacaChat", "functionary", "chatML", "falconChat", "gemma"
+    "general", "llama3Chat", "llama2Chat", "alpacaChat", "functionary", "chatML", "falconChat", "gemma"
 ]);
 export const templateChatWrapperTypeNames = Object.freeze([
     "template", "jinjaTemplate"
@@ -23,7 +24,8 @@ export const resolvableChatWrapperTypeNames = Object.freeze([
 ]);
 const chatWrappers = {
     "general": GeneralChatWrapper,
-    "llamaChat": LlamaChatWrapper,
+    "llama3Chat": Llama3ChatWrapper,
+    "llama2Chat": Llama2ChatWrapper,
     "alpacaChat": AlpacaChatWrapper,
     "functionary": FunctionaryChatWrapper,
     "chatML": ChatMLChatWrapper,
@@ -100,7 +102,7 @@ export function resolveChatWrapper({ type = "auto", bosString, filename, fileInf
         }
     }
     const modelJinjaTemplate = customWrapperSettings?.jinjaTemplate?.template ?? fileInfo?.metadata?.tokenizer?.chat_template;
-    if (!noJinja && modelJinjaTemplate != null && modelJinjaTemplate.trim() !== "") {
+    if (modelJinjaTemplate != null && modelJinjaTemplate.trim() !== "") {
         const jinjaTemplateChatWrapperOptions = {
             ...(customWrapperSettings?.jinjaTemplate ?? {}),
             template: modelJinjaTemplate
@@ -121,13 +123,15 @@ export function resolveChatWrapper({ type = "auto", bosString, filename, fileInf
                     return new Wrapper(testChatWrapperSettings);
             }
         }
-        if (!fallbackToOtherWrappersOnJinjaError)
-            return new JinjaTemplateChatWrapper(jinjaTemplateChatWrapperOptions);
-        try {
-            return new JinjaTemplateChatWrapper(jinjaTemplateChatWrapperOptions);
-        }
-        catch (err) {
-            console.error(getConsoleLogPrefix() + "Error creating Jinja template chat wrapper. Falling back to resolve other chat wrappers. Error:", err);
+        if (!noJinja) {
+            if (!fallbackToOtherWrappersOnJinjaError)
+                return new JinjaTemplateChatWrapper(jinjaTemplateChatWrapperOptions);
+            try {
+                return new JinjaTemplateChatWrapper(jinjaTemplateChatWrapperOptions);
+            }
+            catch (err) {
+                console.error(getConsoleLogPrefix() + "Error creating Jinja template chat wrapper. Falling back to resolve other chat wrappers. Error:", err);
+            }
         }
     }
     // try to find a pattern in the Jinja template to resolve to a specialized chat wrapper,
@@ -136,9 +140,11 @@ export function resolveChatWrapper({ type = "auto", bosString, filename, fileInf
         if (modelJinjaTemplate.includes("<|im_start|>"))
             return createSpecializedChatWrapper(ChatMLChatWrapper);
         else if (modelJinjaTemplate.includes("[INST]"))
-            return createSpecializedChatWrapper(LlamaChatWrapper, {
+            return createSpecializedChatWrapper(Llama2ChatWrapper, {
                 addSpaceBeforeEos: modelJinjaTemplate.includes("' ' + eos_token")
             });
+        else if (modelJinjaTemplate.includes("<|start_header_id|>") && modelJinjaTemplate.includes("<|end_header_id|>"))
+            return createSpecializedChatWrapper(Llama3ChatWrapper);
         else if (modelJinjaTemplate.includes("<start_of_turn>"))
             return createSpecializedChatWrapper(GemmaChatWrapper);
     }
@@ -151,19 +157,21 @@ export function resolveChatWrapper({ type = "auto", bosString, filename, fileInf
             const firstSplitLowercaseSubType = splitLowercaseSubType[0];
             if (lowercaseName === "llama") {
                 if (splitLowercaseSubType.includes("chat"))
-                    return createSpecializedChatWrapper(LlamaChatWrapper);
+                    return createSpecializedChatWrapper(Llama2ChatWrapper);
                 return createSpecializedChatWrapper(GeneralChatWrapper);
             }
+            else if (lowercaseName === "codellama")
+                return createSpecializedChatWrapper(GeneralChatWrapper);
             else if (lowercaseName === "yarn" && firstSplitLowercaseSubType === "llama")
-                return createSpecializedChatWrapper(LlamaChatWrapper);
+                return createSpecializedChatWrapper(Llama2ChatWrapper);
             else if (lowercaseName === "orca")
                 return createSpecializedChatWrapper(ChatMLChatWrapper);
             else if (lowercaseName === "phind" && lowercaseSubType === "codellama")
-                return createSpecializedChatWrapper(LlamaChatWrapper);
+                return createSpecializedChatWrapper(Llama2ChatWrapper);
             else if (lowercaseName === "mistral")
                 return createSpecializedChatWrapper(GeneralChatWrapper);
             else if (firstSplitLowercaseSubType === "llama")
-                return createSpecializedChatWrapper(LlamaChatWrapper);
+                return createSpecializedChatWrapper(Llama2ChatWrapper);
             else if (lowercaseSubType === "alpaca")
                 return createSpecializedChatWrapper(AlpacaChatWrapper);
             else if (lowercaseName === "functionary")
@@ -179,14 +187,14 @@ export function resolveChatWrapper({ type = "auto", bosString, filename, fileInf
     if (fileInfo != null) {
         const arch = fileInfo.metadata.general?.architecture;
         if (arch === "llama")
-            return createSpecializedChatWrapper(LlamaChatWrapper);
+            return createSpecializedChatWrapper(GeneralChatWrapper);
         else if (arch === "falcon")
             return createSpecializedChatWrapper(FalconChatWrapper);
     }
     if (bosString === "" || bosString == null)
         return null;
     if ("<s>[INST] <<SYS>>\n".startsWith(bosString)) {
-        return createSpecializedChatWrapper(LlamaChatWrapper);
+        return createSpecializedChatWrapper(Llama2ChatWrapper);
     }
     else if ("<|im_start|>system\n".startsWith(bosString)) {
         return createSpecializedChatWrapper(ChatMLChatWrapper);

package/dist/chatWrappers/utils/resolveChatWrapper.js.map

@@ -1 +1 @@
-{"version":3,"file":"resolveChatWrapper.js","sourceRoot":"","sources":["../../../src/chatWrappers/utils/resolveChatWrapper.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,kBAAkB,EAAC,MAAM,mCAAmC,CAAC;AACrE,OAAO,EAAC,gBAAgB,EAAC,MAAM,wBAAwB,CAAC;AACxD,OAAO,EAAC,iBAAiB,EAAC,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAC,kBAAkB,EAAC,MAAM,0BAA0B,CAAC;AAC5D,OAAO,EAAC,iBAAiB,EAAC,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAC,sBAAsB,EAAC,MAAM,8BAA8B,CAAC;AACpE,OAAO,EAAC,iBAAiB,EAAC,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAC,gBAAgB,EAAC,MAAM,wBAAwB,CAAC;AACxD,OAAO,EAAC,wBAAwB,EAAkC,MAAM,wCAAwC,CAAC;AACjH,OAAO,EAAC,mBAAmB,EAAC,MAAM,mCAAmC,CAAC;AACtE,OAAO,EAAC,mBAAmB,EAAC,MAAM,oCAAoC,CAAC;AAEvE,OAAO,EAAC,iDAAiD,EAAC,MAAM,wDAAwD,CAAC;AAIzH,MAAM,CAAC,MAAM,+BAA+B,GAAG,MAAM,CAAC,MAAM,CAAC;IACzD,SAAS,EAAE,WAAW,EAAE,YAAY,EAAE,aAAa,EAAE,QAAQ,EAAE,YAAY,EAAE,OAAO;CAC9E,CAAC,CAAC;AAGZ,MAAM,CAAC,MAAM,4BAA4B,GAAG,MAAM,CAAC,MAAM,CAAC;IACtD,UAAU,EAAE,eAAe;CACrB,CAAC,CAAC;AAGZ,MAAM,CAAC,MAAM,8BAA8B,GAAG,MAAM,CAAC,MAAM,CAAC;IACxD,MAAM;IACN,GAAG,+BAA+B;IAClC,GAAG,4BAA4B;CACzB,CAAC,CAAC;AAGZ,MAAM,YAAY,GAAG;IACjB,SAAS,EAAE,kBAAkB;IAC7B,WAAW,EAAE,gBAAgB;IAC7B,YAAY,EAAE,iBAAiB;IAC/B,aAAa,EAAE,sBAAsB;IACrC,QAAQ,EAAE,iBAAiB;IAC3B,YAAY,EAAE,iBAAiB;IAC/B,OAAO,EAAE,gBAAgB;IACzB,UAAU,EAAE,mBAAmB;IAC/B,eAAe,EAAE,wBAAwB;CACiD,CAAC;AAC/F,MAAM,uBAAuB,GAAG,IAAI,GAAG,CACnC,MAAM,CAAC,OAAO,CAAC,YAAY,CAAC;KACvB,GAAG,CAAC,CAAC,CAAC,UAAU,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAC5B,CAAC,OAAO,EAAE,UAAuC,CAAC,CACrD,CAAC,CACT,CAAC;AA2BF;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,kBAAkB,CAAC,EAC/B,IAAI,GAAG,MAAM,EACb,SAAS,EACT,QAAQ,EACR,QAAQ,EACR,SAAS,EACT,qBAAqB,EACrB,WAAW,GAAG,IAAI,EAClB,mCAAmC,GAAG,IAAI,EAC1C,OAAO,GAAG,KAAK,EACS;IACxB,SAAS,4BAA4B,CACjC,sBAAyB,EACzB,kBAA+C,EAAE;QAEjD,MAAM,qBAAqB,GAAG,uBAAuB,CAAC,GAAG,CAAC,sBAAsB,CAAmC,CAAC;QACpH,MAAM,mBAAmB,GAAG,qBAAqB,EAAE,CAAC,qBAAqB,CAAC,CAAC;QAE3E,OAAO,IAAK,sBAA8B,CAAC;YACvC,GAAG,CAAC,eAAe,IAAI,EAAE,CAAC;YAC1B,GAAG,CAAC,mBAAmB,IAAI,EAAE,CAAC;SACjC,CAAC,CAAC;IACP,CAAC;IAED,IAAI,IAAI,KAAK,MAAM,IAAI,IAAI,IAAI,IAAI,EAAE;QACjC,IAAI,yBAAyB,CAAC,IAAI,CAAC,EAAE;YACjC,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC;YAEnC,IAAI,gBAAgB,CAAC,OAAO,EAAE,mBAAmB,CAAC,EAAE;gBAChD,MAAM,eAAe,GAAG,qBAAqB,EAAE,QAAQ,CAAC;gBACxD,IAAI,eAAe,IAAI,IAAI,IAAI,eAAe,EAAE,QAAQ,IAAI,IAAI,IAAI,eAAe,EAAE,eAAe,IAAI,IAAI;oBACxG,eAAe,EAAE,aAAa,IAAI,IAAI,IAAI,eAAe,EAAE,YAAY,IAAI,IAAI,EACjF;oBACE,IAAI,WAAW;wBACX,OAAO,CAAC,IAAI,CAAC,mBAAmB,EAAE,GAAG,0JAA0J,CAAC,CAAC;iBACxM;;oBACG,OAAO,IAAI,mBAAmB,CAAC,eAAe,CAAC,CAAC;aACvD;iBAAM,IAAI,gBAAgB,CAAC,OAAO,EAAE,wBAAwB,CAAC,EAAE;gBAC5D,MAAM,aAAa,GAAG,qBAAqB,EAAE,aAAa,EAAE,QAAQ,IAAI,QAAQ,EAAE,QAAQ,EAAE,SAAS,EAAE,aAAa,CAAC;gBAErH,IAAI,aAAa,IAAI,IAAI,EAAE;oBACvB,IAAI,WAAW;wBACX,OAAO,CAAC,IAAI,CAAC,mBAAmB,EAAE,GAAG,qGAAqG,CAAC,CAAC;iBACnJ;qBAAM;oBACH,IAAI;wBACA,OAAO,IAAI,wBAAwB,CAAC;4BAChC,GAAG,CAAC,qBAAqB,EAAE,aAAa,IAAI,EAAE,CAAC;4BAC/C,QAAQ,EAAE,aAAa;yBAC1B,CAAC,CAAC;qBACN;oBAAC,OAAO,GAAG,EAAE;wBACV,IAAI,CAAC,mCAAmC;4BACpC,MAAM,GAAG,CAAC;6BACT,IAAI,WAAW;4BAChB,OAAO,CAAC,KAAK,CAAC,mBAAmB,EAAE,GAAG,iGAAiG,EAAE,GAAG,CAAC,CAAC;qBACrJ;iBACJ;aACJ;;gBACG,KAAK,CAAC,OAAuB,CAAC,CAAC;SACtC;aAAM,IAAI,MAAM,CAAC,MAAM,CAAC,YAAY,EAAE,IAAI,CAAC,EAAE;YAC1C,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC;YACnC,MAAM,eAAe,GACjB,qBAAqB,EAAE,CAAC,IAAI,CAAC,CAAC;YAElC,OAAO,IAAK,OAAe,CAAC,eAAe,CAAC,CAAC;SAChD;KACJ;IAED,MAAM,kBAAkB,GAAG,qBAAqB,EAAE,aAAa,EAAE,QAAQ,IAAI,QAAQ,EAAE,QAAQ,EAAE,SAAS,EAAE,aAAa,CAAC;IAE1H,IAAI,CAAC,OAAO,IAAI,kBAAkB,IAAI,IAAI,IAAI,kBAAkB,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE;QAC5E,MAAM,+BAA+B,GAAoC;YACrE,GAAG,CAAC,qBAAqB,EAAE,aAAa,IAAI,EAAE,CAAC;YAC/C,QAAQ,EAAE,kBAAkB;SAC/B,CAAC;QAEF,KAAK,MAAM,8BAA8B,IAAI,+BAA+B,EAAE;YAC1E,MAAM,OAAO,GAAG,YAAY,CAAC,8BAA8B,CAAC,CAAC;YAC7D,MAAM,eAAe,GAAG,qBAAqB,EAAE,CAAC,8BAA8B,CAAC,CAAC;YAEhF,MAAM,wBAAwB,GAAG,OAAO,CAAC,yDAAyD,EAAE,EAAE,IAAI,EAAE,CAAC;YAC7G,IAAI,wBAAwB,CAAC,MAAM,KAAK,CAAC;gBACrC,wBAAwB,CAAC,IAAI,CAAC,EAAS,CAAC,CAAC;YAE7C,KAAK,MAAM,iBAAiB,IAAI,wBAAwB,EAAE;gBACtD,MAAM,uBAAuB,GAAG;oBAC5B,GAAG,CAAC,eAAe,IAAI,EAAE,CAAC;oBAC1B,GAAG,CAAC,iBAAiB,IAAI,EAAE,CAAC;iBAC/B,CAAC;gBACF,MAAM,WAAW,GAAG,IAAK,OAAe,CAAC,uBAAuB,CAAC,CAAC;gBAElE,IAAI,iDAAiD,CAAC,+BAA+B,EAAE,WAAW,EAAE,SAAS,CAAC;oBAC1G,OAAO,IAAK,OAAe,CAAC,uBAAuB,CAAC,CAAC;aAC5D;SACJ;QAED,IAAI,CAAC,mCAAmC;YACpC,OAAO,IAAI,wBAAwB,CAAC,+BAA+B,CAAC,CAAC;QAEzE,IAAI;YACA,OAAO,IAAI,wBAAwB,CAAC,+BAA+B,CAAC,CAAC;SACxE;QAAC,OAAO,GAAG,EAAE;YACV,OAAO,CAAC,KAAK,CAAC,mBAAmB,EAAE,GAAG,iGAAiG,EAAE,GAAG,CAAC,CAAC;SACjJ;KACJ;IAED,wFAAwF;IACxF,sFAAsF;IACtF,IAAI,kBAAkB,IAAI,IAAI,IAAI,kBAAkB,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE;QAChE,IAAI,kBAAkB,CAAC,QAAQ,CAAC,cAAc,CAAC;YAC3C,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;aACtD,IAAI,kBAAkB,CAAC,QAAQ,CAAC,QAAQ,CAAC;YAC1C,OAAO,4BAA4B,CAAC,gBAAgB,EAAE;gBAClD,iBAAiB,EAAE,kBAAkB,CAAC,QAAQ,CAAC,iBAAiB,CAAC;aACpE,CAAC,CAAC;aACF,IAAI,kBAAkB,CAAC,QAAQ,CAAC,iBAAiB,CAAC;YACnD,OAAO,4BAA4B,CAAC,gBAAgB,CAAC,CAAC;KAC7D;IAED,IAAI,QAAQ,IAAI,IAAI,EAAE;QAClB,MAAM,EAAC,IAAI,EAAE,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAC,GAAG,kBAAkB,CAAC,QAAQ,CAAC,CAAC;QAE1E,IAAI,QAAQ,EAAE,WAAW,EAAE,KAAK,MAAM,EAAE;YACpC,MAAM,aAAa,GAAG,IAAI,EAAE,WAAW,EAAE,CAAC;YAC1C,MAAM,gBAAgB,GAAG,OAAO,EAAE,WAAW,EAAE,CAAC;YAChD,MAAM,qBAAqB,GAAG,CAAC,gBAAgB,EAAE,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CACrE,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAC5C,CAAC;YACF,MAAM,0BAA0B,GAAG,qBAAqB,CAAC,CAAC,CAAC,CAAC;YAE5D,IAAI,aAAa,KAAK,OAAO,EAAE;gBAC3B,IAAI,qBAAqB,CAAC,QAAQ,CAAC,MAAM,CAAC;oBACtC,OAAO,4BAA4B,CAAC,gBAAgB,CAAC,CAAC;gBAE1D,OAAO,4BAA4B,CAAC,kBAAkB,CAAC,CAAC;aAC3D;iBAAM,IAAI,aAAa,KAAK,MAAM,IAAI,0BAA0B,KAAK,OAAO;gBACzE,OAAO,4BAA4B,CAAC,gBAAgB,CAAC,CAAC;iBACrD,IAAI,aAAa,KAAK,MAAM;gBAC7B,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;iBACtD,IAAI,aAAa,KAAK,OAAO,IAAI,gBAAgB,KAAK,WAAW;gBAClE,OAAO,4BAA4B,CAAC,gBAAgB,CAAC,CAAC;iBACrD,IAAI,aAAa,KAAK,SAAS;gBAChC,OAAO,4BAA4B,CAAC,kBAAkB,CAAC,CAAC;iBACvD,IAAI,0BAA0B,KAAK,OAAO;gBAC3C,OAAO,4BAA4B,CAAC,gBAAgB,CAAC,CAAC;iBACrD,IAAI,gBAAgB,KAAK,QAAQ;gBAClC,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;iBACtD,IAAI,aAAa,KAAK,aAAa;gBACpC,OAAO,4BAA4B,CAAC,sBAAsB,CAAC,CAAC;iBAC3D,IAAI,aAAa,KAAK,SAAS,IAAI,qBAAqB,CAAC,QAAQ,CAAC,SAAS,CAAC;gBAC7E,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;iBACtD,IAAI,aAAa,KAAK,OAAO;gBAC9B,OAAO,4BAA4B,CAAC,gBAAgB,CAAC,CAAC;iBACrD,IAAI,qBAAqB,CAAC,QAAQ,CAAC,QAAQ,CAAC;gBAC7C,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;SAC9D;KACJ;IAED,IAAI,QAAQ,IAAI,IAAI,EAAE;QAClB,MAAM,IAAI,GAAG,QAAQ,CAAC,QAAQ,CAAC,OAAO,EAAE,YAAY,CAAC;QAErD,IAAI,IAAI,KAAK,OAAO;YAChB,OAAO,4BAA4B,CAAC,gBAAgB,CAAC,CAAC;aACrD,IAAI,IAAI,KAAK,QAAQ;YACtB,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;KAC9D;IAED,IAAI,SAAS,KAAK,EAAE,IAAI,SAAS,IAAI,IAAI;QACrC,OAAO,IAAI,CAAC;IAEhB,IAAI,qBAAqB,CAAC,UAAU,CAAC,SAAS,CAAC,EAAE;QAC7C,OAAO,4BAA4B,CAAC,gBAAgB,CAAC,CAAC;KACzD;SAAM,IAAI,sBAAsB,CAAC,UAAU,CAAC,SAAS,CAAC,EAAE;QACrD,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;KAC1D;IAED,OAAO,IAAI,CAAC;AAChB,CAAC;AAED,MAAM,UAAU,4BAA4B,CAAC,IAAY;IACrD,OAAO,+BAA+B,CAAC,QAAQ,CAAC,IAAW,CAAC,CAAC;AACjE,CAAC;AAED,MAAM,UAAU,yBAAyB,CAAC,IAAY;IAClD,OAAO,4BAA4B,CAAC,QAAQ,CAAC,IAAW,CAAC,CAAC;AAC9D,CAAC;AAED,0FAA0F;AAC1F,SAAS,gBAAgB,CAAI,KAAU,EAAE,cAAiB;IACtD,OAAO,KAAK,KAAK,cAAc,CAAC;AACpC,CAAC"}
+{"version":3,"file":"resolveChatWrapper.js","sourceRoot":"","sources":["../../../src/chatWrappers/utils/resolveChatWrapper.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,kBAAkB,EAAC,MAAM,mCAAmC,CAAC;AACrE,OAAO,EAAC,iBAAiB,EAAC,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAC,iBAAiB,EAAC,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAC,iBAAiB,EAAC,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAC,kBAAkB,EAAC,MAAM,0BAA0B,CAAC;AAC5D,OAAO,EAAC,iBAAiB,EAAC,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAC,sBAAsB,EAAC,MAAM,8BAA8B,CAAC;AACpE,OAAO,EAAC,iBAAiB,EAAC,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAC,gBAAgB,EAAC,MAAM,wBAAwB,CAAC;AACxD,OAAO,EAAC,wBAAwB,EAAkC,MAAM,wCAAwC,CAAC;AACjH,OAAO,EAAC,mBAAmB,EAAC,MAAM,mCAAmC,CAAC;AACtE,OAAO,EAAC,mBAAmB,EAAC,MAAM,oCAAoC,CAAC;AAEvE,OAAO,EAAC,iDAAiD,EAAC,MAAM,wDAAwD,CAAC;AAIzH,MAAM,CAAC,MAAM,+BAA+B,GAAG,MAAM,CAAC,MAAM,CAAC;IACzD,SAAS,EAAE,YAAY,EAAE,YAAY,EAAE,YAAY,EAAE,aAAa,EAAE,QAAQ,EAAE,YAAY,EAAE,OAAO;CAC7F,CAAC,CAAC;AAGZ,MAAM,CAAC,MAAM,4BAA4B,GAAG,MAAM,CAAC,MAAM,CAAC;IACtD,UAAU,EAAE,eAAe;CACrB,CAAC,CAAC;AAGZ,MAAM,CAAC,MAAM,8BAA8B,GAAG,MAAM,CAAC,MAAM,CAAC;IACxD,MAAM;IACN,GAAG,+BAA+B;IAClC,GAAG,4BAA4B;CACzB,CAAC,CAAC;AAGZ,MAAM,YAAY,GAAG;IACjB,SAAS,EAAE,kBAAkB;IAC7B,YAAY,EAAE,iBAAiB;IAC/B,YAAY,EAAE,iBAAiB;IAC/B,YAAY,EAAE,iBAAiB;IAC/B,aAAa,EAAE,sBAAsB;IACrC,QAAQ,EAAE,iBAAiB;IAC3B,YAAY,EAAE,iBAAiB;IAC/B,OAAO,EAAE,gBAAgB;IACzB,UAAU,EAAE,mBAAmB;IAC/B,eAAe,EAAE,wBAAwB;CACiD,CAAC;AAC/F,MAAM,uBAAuB,GAAG,IAAI,GAAG,CACnC,MAAM,CAAC,OAAO,CAAC,YAAY,CAAC;KACvB,GAAG,CAAC,CAAC,CAAC,UAAU,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAC5B,CAAC,OAAO,EAAE,UAAuC,CAAC,CACrD,CAAC,CACT,CAAC;AA2BF;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,kBAAkB,CAAC,EAC/B,IAAI,GAAG,MAAM,EACb,SAAS,EACT,QAAQ,EACR,QAAQ,EACR,SAAS,EACT,qBAAqB,EACrB,WAAW,GAAG,IAAI,EAClB,mCAAmC,GAAG,IAAI,EAC1C,OAAO,GAAG,KAAK,EACS;IACxB,SAAS,4BAA4B,CACjC,sBAAyB,EACzB,kBAA+C,EAAE;QAEjD,MAAM,qBAAqB,GAAG,uBAAuB,CAAC,GAAG,CAAC,sBAAsB,CAAmC,CAAC;QACpH,MAAM,mBAAmB,GAAG,qBAAqB,EAAE,CAAC,qBAAqB,CAAC,CAAC;QAE3E,OAAO,IAAK,sBAA8B,CAAC;YACvC,GAAG,CAAC,eAAe,IAAI,EAAE,CAAC;YAC1B,GAAG,CAAC,mBAAmB,IAAI,EAAE,CAAC;SACjC,CAAC,CAAC;IACP,CAAC;IAED,IAAI,IAAI,KAAK,MAAM,IAAI,IAAI,IAAI,IAAI,EAAE;QACjC,IAAI,yBAAyB,CAAC,IAAI,CAAC,EAAE;YACjC,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC;YAEnC,IAAI,gBAAgB,CAAC,OAAO,EAAE,mBAAmB,CAAC,EAAE;gBAChD,MAAM,eAAe,GAAG,qBAAqB,EAAE,QAAQ,CAAC;gBACxD,IAAI,eAAe,IAAI,IAAI,IAAI,eAAe,EAAE,QAAQ,IAAI,IAAI,IAAI,eAAe,EAAE,eAAe,IAAI,IAAI;oBACxG,eAAe,EAAE,aAAa,IAAI,IAAI,IAAI,eAAe,EAAE,YAAY,IAAI,IAAI,EACjF;oBACE,IAAI,WAAW;wBACX,OAAO,CAAC,IAAI,CAAC,mBAAmB,EAAE,GAAG,0JAA0J,CAAC,CAAC;iBACxM;;oBACG,OAAO,IAAI,mBAAmB,CAAC,eAAe,CAAC,CAAC;aACvD;iBAAM,IAAI,gBAAgB,CAAC,OAAO,EAAE,wBAAwB,CAAC,EAAE;gBAC5D,MAAM,aAAa,GAAG,qBAAqB,EAAE,aAAa,EAAE,QAAQ,IAAI,QAAQ,EAAE,QAAQ,EAAE,SAAS,EAAE,aAAa,CAAC;gBAErH,IAAI,aAAa,IAAI,IAAI,EAAE;oBACvB,IAAI,WAAW;wBACX,OAAO,CAAC,IAAI,CAAC,mBAAmB,EAAE,GAAG,qGAAqG,CAAC,CAAC;iBACnJ;qBAAM;oBACH,IAAI;wBACA,OAAO,IAAI,wBAAwB,CAAC;4BAChC,GAAG,CAAC,qBAAqB,EAAE,aAAa,IAAI,EAAE,CAAC;4BAC/C,QAAQ,EAAE,aAAa;yBAC1B,CAAC,CAAC;qBACN;oBAAC,OAAO,GAAG,EAAE;wBACV,IAAI,CAAC,mCAAmC;4BACpC,MAAM,GAAG,CAAC;6BACT,IAAI,WAAW;4BAChB,OAAO,CAAC,KAAK,CAAC,mBAAmB,EAAE,GAAG,iGAAiG,EAAE,GAAG,CAAC,CAAC;qBACrJ;iBACJ;aACJ;;gBACG,KAAK,CAAC,OAAuB,CAAC,CAAC;SACtC;aAAM,IAAI,MAAM,CAAC,MAAM,CAAC,YAAY,EAAE,IAAI,CAAC,EAAE;YAC1C,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC;YACnC,MAAM,eAAe,GACjB,qBAAqB,EAAE,CAAC,IAAI,CAAC,CAAC;YAElC,OAAO,IAAK,OAAe,CAAC,eAAe,CAAC,CAAC;SAChD;KACJ;IAED,MAAM,kBAAkB,GAAG,qBAAqB,EAAE,aAAa,EAAE,QAAQ,IAAI,QAAQ,EAAE,QAAQ,EAAE,SAAS,EAAE,aAAa,CAAC;IAE1H,IAAI,kBAAkB,IAAI,IAAI,IAAI,kBAAkB,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE;QAChE,MAAM,+BAA+B,GAAoC;YACrE,GAAG,CAAC,qBAAqB,EAAE,aAAa,IAAI,EAAE,CAAC;YAC/C,QAAQ,EAAE,kBAAkB;SAC/B,CAAC;QAEF,KAAK,MAAM,8BAA8B,IAAI,+BAA+B,EAAE;YAC1E,MAAM,OAAO,GAAG,YAAY,CAAC,8BAA8B,CAAC,CAAC;YAC7D,MAAM,eAAe,GAAG,qBAAqB,EAAE,CAAC,8BAA8B,CAAC,CAAC;YAEhF,MAAM,wBAAwB,GAAG,OAAO,CAAC,yDAAyD,EAAE,EAAE,IAAI,EAAE,CAAC;YAC7G,IAAI,wBAAwB,CAAC,MAAM,KAAK,CAAC;gBACrC,wBAAwB,CAAC,IAAI,CAAC,EAAS,CAAC,CAAC;YAE7C,KAAK,MAAM,iBAAiB,IAAI,wBAAwB,EAAE;gBACtD,MAAM,uBAAuB,GAAG;oBAC5B,GAAG,CAAC,eAAe,IAAI,EAAE,CAAC;oBAC1B,GAAG,CAAC,iBAAiB,IAAI,EAAE,CAAC;iBAC/B,CAAC;gBACF,MAAM,WAAW,GAAG,IAAK,OAAe,CAAC,uBAAuB,CAAC,CAAC;gBAElE,IAAI,iDAAiD,CAAC,+BAA+B,EAAE,WAAW,EAAE,SAAS,CAAC;oBAC1G,OAAO,IAAK,OAAe,CAAC,uBAAuB,CAAC,CAAC;aAC5D;SACJ;QAED,IAAI,CAAC,OAAO,EAAE;YACV,IAAI,CAAC,mCAAmC;gBACpC,OAAO,IAAI,wBAAwB,CAAC,+BAA+B,CAAC,CAAC;YAEzE,IAAI;gBACA,OAAO,IAAI,wBAAwB,CAAC,+BAA+B,CAAC,CAAC;aACxE;YAAC,OAAO,GAAG,EAAE;gBACV,OAAO,CAAC,KAAK,CAAC,mBAAmB,EAAE,GAAG,iGAAiG,EAAE,GAAG,CAAC,CAAC;aACjJ;SACJ;KACJ;IAED,wFAAwF;IACxF,sFAAsF;IACtF,IAAI,kBAAkB,IAAI,IAAI,IAAI,kBAAkB,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE;QAChE,IAAI,kBAAkB,CAAC,QAAQ,CAAC,cAAc,CAAC;YAC3C,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;aACtD,IAAI,kBAAkB,CAAC,QAAQ,CAAC,QAAQ,CAAC;YAC1C,OAAO,4BAA4B,CAAC,iBAAiB,EAAE;gBACnD,iBAAiB,EAAE,kBAAkB,CAAC,QAAQ,CAAC,iBAAiB,CAAC;aACpE,CAAC,CAAC;aACF,IAAI,kBAAkB,CAAC,QAAQ,CAAC,qBAAqB,CAAC,IAAI,kBAAkB,CAAC,QAAQ,CAAC,mBAAmB,CAAC;YAC3G,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;aACtD,IAAI,kBAAkB,CAAC,QAAQ,CAAC,iBAAiB,CAAC;YACnD,OAAO,4BAA4B,CAAC,gBAAgB,CAAC,CAAC;KAC7D;IAED,IAAI,QAAQ,IAAI,IAAI,EAAE;QAClB,MAAM,EAAC,IAAI,EAAE,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAC,GAAG,kBAAkB,CAAC,QAAQ,CAAC,CAAC;QAE1E,IAAI,QAAQ,EAAE,WAAW,EAAE,KAAK,MAAM,EAAE;YACpC,MAAM,aAAa,GAAG,IAAI,EAAE,WAAW,EAAE,CAAC;YAC1C,MAAM,gBAAgB,GAAG,OAAO,EAAE,WAAW,EAAE,CAAC;YAChD,MAAM,qBAAqB,GAAG,CAAC,gBAAgB,EAAE,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CACrE,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAC5C,CAAC;YACF,MAAM,0BAA0B,GAAG,qBAAqB,CAAC,CAAC,CAAC,CAAC;YAE5D,IAAI,aAAa,KAAK,OAAO,EAAE;gBAC3B,IAAI,qBAAqB,CAAC,QAAQ,CAAC,MAAM,CAAC;oBACtC,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;gBAE3D,OAAO,4BAA4B,CAAC,kBAAkB,CAAC,CAAC;aAC3D;iBAAM,IAAI,aAAa,KAAK,WAAW;gBACpC,OAAO,4BAA4B,CAAC,kBAAkB,CAAC,CAAC;iBACvD,IAAI,aAAa,KAAK,MAAM,IAAI,0BAA0B,KAAK,OAAO;gBACvE,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;iBACtD,IAAI,aAAa,KAAK,MAAM;gBAC7B,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;iBACtD,IAAI,aAAa,KAAK,OAAO,IAAI,gBAAgB,KAAK,WAAW;gBAClE,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;iBACtD,IAAI,aAAa,KAAK,SAAS;gBAChC,OAAO,4BAA4B,CAAC,kBAAkB,CAAC,CAAC;iBACvD,IAAI,0BAA0B,KAAK,OAAO;gBAC3C,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;iBACtD,IAAI,gBAAgB,KAAK,QAAQ;gBAClC,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;iBACtD,IAAI,aAAa,KAAK,aAAa;gBACpC,OAAO,4BAA4B,CAAC,sBAAsB,CAAC,CAAC;iBAC3D,IAAI,aAAa,KAAK,SAAS,IAAI,qBAAqB,CAAC,QAAQ,CAAC,SAAS,CAAC;gBAC7E,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;iBACtD,IAAI,aAAa,KAAK,OAAO;gBAC9B,OAAO,4BAA4B,CAAC,gBAAgB,CAAC,CAAC;iBACrD,IAAI,qBAAqB,CAAC,QAAQ,CAAC,QAAQ,CAAC;gBAC7C,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;SAC9D;KACJ;IAED,IAAI,QAAQ,IAAI,IAAI,EAAE;QAClB,MAAM,IAAI,GAAG,QAAQ,CAAC,QAAQ,CAAC,OAAO,EAAE,YAAY,CAAC;QAErD,IAAI,IAAI,KAAK,OAAO;YAChB,OAAO,4BAA4B,CAAC,kBAAkB,CAAC,CAAC;aACvD,IAAI,IAAI,KAAK,QAAQ;YACtB,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;KAC9D;IAED,IAAI,SAAS,KAAK,EAAE,IAAI,SAAS,IAAI,IAAI;QACrC,OAAO,IAAI,CAAC;IAEhB,IAAI,qBAAqB,CAAC,UAAU,CAAC,SAAS,CAAC,EAAE;QAC7C,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;KAC1D;SAAM,IAAI,sBAAsB,CAAC,UAAU,CAAC,SAAS,CAAC,EAAE;QACrD,OAAO,4BAA4B,CAAC,iBAAiB,CAAC,CAAC;KAC1D;IAED,OAAO,IAAI,CAAC;AAChB,CAAC;AAED,MAAM,UAAU,4BAA4B,CAAC,IAAY;IACrD,OAAO,+BAA+B,CAAC,QAAQ,CAAC,IAAW,CAAC,CAAC;AACjE,CAAC;AAED,MAAM,UAAU,yBAAyB,CAAC,IAAY;IAClD,OAAO,4BAA4B,CAAC,QAAQ,CAAC,IAAW,CAAC,CAAC;AAC9D,CAAC;AAED,0FAA0F;AAC1F,SAAS,gBAAgB,CAAI,KAAU,EAAE,cAAiB;IACtD,OAAO,KAAK,KAAK,cAAc,CAAC;AACpC,CAAC"}

package/dist/cli/commands/BuildCommand.js

@@ -30,7 +30,7 @@ export const BuildCommand = {
             .option("nodeTarget", {
             alias: "t",
             type: "string",
-            description: "The Node.js version to compile llama.cpp for. Example: v18.0.0"
+            description: "The Node.js version to compile llama.cpp for. Example: `v18.0.0`"
         })
             .option("gpu", {
             type: "string",

package/dist/cli/commands/BuildCommand.js.map

@@ -1 +1 @@
-{"version":3,"file":"BuildCommand.js","sourceRoot":"","sources":["../../../src/cli/commands/BuildCommand.ts"],"names":[],"mappings":"AAAA,OAAO,OAAO,MAAM,SAAS,CAAC;AAE9B,OAAO,KAAK,MAAM,OAAO,CAAC;AAC1B,OAAO,EAAC,eAAe,EAAC,MAAM,yCAAyC,CAAC;AACxE,OAAO,OAAO,MAAM,wBAAwB,CAAC;AAC7C,OAAO,EAAC,eAAe,EAAC,MAAM,gCAAgC,CAAC;AAC/D,OAAO,EAAC,yBAAyB,EAAE,sBAAsB,EAAE,IAAI,EAAE,yBAAyB,EAAC,MAAM,iBAAiB,CAAC;AACnH,OAAO,EAAC,qBAAqB,EAAC,MAAM,sBAAsB,CAAC;AAC3D,OAAO,cAAc,MAAM,+BAA+B,CAAC;AAC3D,OAAO,EAAC,8BAA8B,EAAC,MAAM,wDAAwD,CAAC;AACtG,OAAO,EAAC,WAAW,EAAC,MAAM,qCAAqC,CAAC;AAChE,OAAO,EAAC,yBAAyB,EAAC,MAAM,mDAAmD,CAAC;AAC5F,OAAO,EAAC,gCAAgC,EAAE,oBAAoB,EAAC,MAAM,2CAA2C,CAAC;AACjH,OAAO,EAAyB,sBAAsB,EAAE,0BAA0B,EAAC,MAAM,yBAAyB,CAAC;AACnH,OAAO,EAAC,oBAAoB,EAAC,MAAM,kCAAkC,CAAC;AACtE,OAAO,EAAC,yBAAyB,EAAC,MAAM,mDAAmD,CAAC;AAC5F,OAAO,EAAC,mBAAmB,EAAC,MAAM,oCAAoC,CAAC;AACvE,OAAO,EAAC,qBAAqB,EAAC,MAAM,0BAA0B,CAAC;AAC/D,OAAO,EAAC,eAAe,EAAC,MAAM,yCAAyC,CAAC;AAYxE,MAAM,CAAC,MAAM,YAAY,GAAwC;IAC7D,OAAO,EAAE,OAAO;IAChB,QAAQ,EAAE,4CAA4C;IACtD,OAAO,CAAC,KAAK;QACT,OAAO,KAAK;aACP,MAAM,CAAC,MAAM,EAAE;YACZ,KAAK,EAAE,GAAG;YACV,IAAI,EAAE,QAAQ;YACd,MAAM,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK;YACxB,WAAW,EAAE,2CAA2C;SAC3D,CAAC;aACD,MAAM,CAAC,YAAY,EAAE;YAClB,KAAK,EAAE,GAAG;YACV,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,gEAAgE;SAChF,CAAC;aACD,MAAM,CAAC,KAAK,EAAE;YACX,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,yBAAyB;YAElC,2FAA2F;YAC3F,OAAO,EAAE,sBAAwF;YACjG,MAAM,EAAE,0BAA0B;YAClC,WAAW,EAAE,wDAAwD;SACxE,CAAC;aACD,MAAM,CAAC,gBAAgB,EAAE;YACtB,KAAK,EAAE,IAAI;YACX,IAAI,EAAE,SAAS;YACf,OAAO,EAAE,KAAK;YACd,WAAW,EAAE,+CAA+C;SAC/D,CAAC;aACD,MAAM,CAAC,6CAA6C,EAAE;YACnD,IAAI,EAAE,SAAS;YACf,MAAM,EAAE,IAAI;YACZ,OAAO,EAAE,KAAK;YACd,WAAW,EAAE,+DAA+D;SAC/E,CAAC,CAAC;IACX,CAAC;IACD,OAAO,EAAE,oBAAoB;CAChC,CAAC;AAEF,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,EACvC,IAAI,GAAG,SAAS,EAChB,UAAU,GAAG,SAAS,EACtB,GAAG,GAAG,yBAAyB,EAC/B,cAAc,GAAG,KAAK,EACtB,2CAA2C,GAAG,KAAK,EACxC;IACX,IAAI,CAAC,CAAC,MAAM,oBAAoB,EAAE,CAAC,EAAE;QACjC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,yEAAyE,CAAC,CAAC,CAAC;QAClG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;KACnB;IAED,MAAM,qCAAqC,GAAG,CAAC,2CAA2C,IAAI,CAAC,IAAI,CAAC;IAEpG,MAAM,6BAA6B,GAAG,MAAM,gCAAgC,EAAE,CAAC;IAE/E,MAAM,QAAQ,GAAG,WAAW,EAAE,CAAC;IAC/B,MAAM,YAAY,GAAG,MAAM,eAAe,EAAE,CAAC;IAC7C,MAAM,kBAAkB,GAAG,yBAAyB,EAAE,CAAC;IACvD,MAAM,cAAc,GAAe,MAAM,yBAAyB,CAAC,GAAG,EAAE,EAAC,QAAQ,EAAE,IAAI,EAAC,CAAC,CAAC;IAC1F,IAAI,eAAe,GAAG,KAAK,CAAC;IAE5B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,cAAc,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE;QAC5C,MAAM,QAAQ,GAAG,cAAc,CAAC,CAAC,CAAC,CAAC;QACnC,MAAM,UAAU,GAAG,CAAC,KAAK,cAAc,CAAC,MAAM,GAAG,CAAC,CAAC;QAEnD,oBAAoB,CAAC,QAAQ,CAAC,CAAC;QAE/B,IAAI,CAAC,eAAe,EAAE;YAClB,MAAM,qBAAqB,CAAC,IAAI,CAAC,CAAC;YAClC,eAAe,GAAG,IAAI,CAAC;SAC1B;QAED,MAAM,YAAY,GAAiB;YAC/B,kBAAkB;YAClB,YAAY,EAAE,IAAI;YAClB,QAAQ;YACR,YAAY;YACZ,IAAI,EAAE,IAAI;gBACN,CAAC,CAAC,IAA2B;gBAC7B,CAAC,CAAC,OAAO,CAAC,IAAI;YAClB,GAAG,EAAE,QAAQ;YACb,QAAQ,EAAE;gBACN,IAAI,EAAE,6BAA6B,EAAE,kBAAkB,IAAI,yBAAyB;gBACpF,OAAO,EAAE,6BAA6B,EAAE,GAAG,IAAI,sBAAsB;aACxE;SACJ,CAAC;QAEF,IAAI;YACA,MAAM,cAAc,CAAC;gBACjB,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,qBAAqB,CAAC;gBAC1C,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC;gBACzC,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,6BAA6B,CAAC;aAClD,EAAE,KAAK,IAAI,EAAE;gBACV,MAAM,eAAe,CAAC,YAAY,EAAE;oBAChC,UAAU,EAAE,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS;oBAC/C,mBAAmB,EAAE,IAAI;oBACzB,qBAAqB,EAAE,KAAK;oBAC5B,0BAA0B,EAAE,KAAK;oBACjC,qCAAqC;iBACxC,CAAC,CAAC;YACP,CAAC,CAAC,CAAC;SACN;QAAC,OAAO,GAAG,EAAE;YACV,OAAO,CAAC,KAAK,CACT,mBAAmB,EAAE;gBACrB,kCAAkC,qBAAqB,CAAC,QAAQ,CAAC,YAAY;gBAC7E,CACI,CAAC,UAAU;oBACP,CAAC,CAAC,2CAA2C,qBAAqB,CAAC,cAAc,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,YAAY;oBACrG,CAAC,CAAC,EAAE,CACX;gBACD,QAAQ,EACR,GAAG,CACN,CAAC;YAEF,IAAI,UAAU;gBACV,MAAM,GAAG,CAAC;YAEd,SAAS;SACZ;QAED,MAAM,OAAO,CAAC;YACV,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,0BAA0B,CAAC;YAC/C,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,yBAAyB,CAAC;YAC9C,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,kCAAkC,CAAC;SACvD,EAAE,KAAK,IAAI,EAAE;YACV,MAAM,eAAe,EAAE,CAAC;QAC5B,CAAC,CAAC,CAAC;QAEH,IAAI,CAAC,cAAc,EAAE;YACjB,OAAO,CAAC,GAAG,EAAE,CAAC;YACd,8BAA8B,CAAC,YAAY,EAAE,GAAG,KAAK,MAAM,EAAE,IAAI,CAAC,CAAC;YACnE,OAAO,CAAC,GAAG,EAAE,CAAC;SACjB;QAED,MAAM;KACT;AACL,CAAC"}
+{"version":3,"file":"BuildCommand.js","sourceRoot":"","sources":["../../../src/cli/commands/BuildCommand.ts"],"names":[],"mappings":"AAAA,OAAO,OAAO,MAAM,SAAS,CAAC;AAE9B,OAAO,KAAK,MAAM,OAAO,CAAC;AAC1B,OAAO,EAAC,eAAe,EAAC,MAAM,yCAAyC,CAAC;AACxE,OAAO,OAAO,MAAM,wBAAwB,CAAC;AAC7C,OAAO,EAAC,eAAe,EAAC,MAAM,gCAAgC,CAAC;AAC/D,OAAO,EAAC,yBAAyB,EAAE,sBAAsB,EAAE,IAAI,EAAE,yBAAyB,EAAC,MAAM,iBAAiB,CAAC;AACnH,OAAO,EAAC,qBAAqB,EAAC,MAAM,sBAAsB,CAAC;AAC3D,OAAO,cAAc,MAAM,+BAA+B,CAAC;AAC3D,OAAO,EAAC,8BAA8B,EAAC,MAAM,wDAAwD,CAAC;AACtG,OAAO,EAAC,WAAW,EAAC,MAAM,qCAAqC,CAAC;AAChE,OAAO,EAAC,yBAAyB,EAAC,MAAM,mDAAmD,CAAC;AAC5F,OAAO,EAAC,gCAAgC,EAAE,oBAAoB,EAAC,MAAM,2CAA2C,CAAC;AACjH,OAAO,EAAyB,sBAAsB,EAAE,0BAA0B,EAAC,MAAM,yBAAyB,CAAC;AACnH,OAAO,EAAC,oBAAoB,EAAC,MAAM,kCAAkC,CAAC;AACtE,OAAO,EAAC,yBAAyB,EAAC,MAAM,mDAAmD,CAAC;AAC5F,OAAO,EAAC,mBAAmB,EAAC,MAAM,oCAAoC,CAAC;AACvE,OAAO,EAAC,qBAAqB,EAAC,MAAM,0BAA0B,CAAC;AAC/D,OAAO,EAAC,eAAe,EAAC,MAAM,yCAAyC,CAAC;AAYxE,MAAM,CAAC,MAAM,YAAY,GAAwC;IAC7D,OAAO,EAAE,OAAO;IAChB,QAAQ,EAAE,4CAA4C;IACtD,OAAO,CAAC,KAAK;QACT,OAAO,KAAK;aACP,MAAM,CAAC,MAAM,EAAE;YACZ,KAAK,EAAE,GAAG;YACV,IAAI,EAAE,QAAQ;YACd,MAAM,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK;YACxB,WAAW,EAAE,2CAA2C;SAC3D,CAAC;aACD,MAAM,CAAC,YAAY,EAAE;YAClB,KAAK,EAAE,GAAG;YACV,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,kEAAkE;SAClF,CAAC;aACD,MAAM,CAAC,KAAK,EAAE;YACX,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,yBAAyB;YAElC,2FAA2F;YAC3F,OAAO,EAAE,sBAAwF;YACjG,MAAM,EAAE,0BAA0B;YAClC,WAAW,EAAE,wDAAwD;SACxE,CAAC;aACD,MAAM,CAAC,gBAAgB,EAAE;YACtB,KAAK,EAAE,IAAI;YACX,IAAI,EAAE,SAAS;YACf,OAAO,EAAE,KAAK;YACd,WAAW,EAAE,+CAA+C;SAC/D,CAAC;aACD,MAAM,CAAC,6CAA6C,EAAE;YACnD,IAAI,EAAE,SAAS;YACf,MAAM,EAAE,IAAI;YACZ,OAAO,EAAE,KAAK;YACd,WAAW,EAAE,+DAA+D;SAC/E,CAAC,CAAC;IACX,CAAC;IACD,OAAO,EAAE,oBAAoB;CAChC,CAAC;AAEF,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,EACvC,IAAI,GAAG,SAAS,EAChB,UAAU,GAAG,SAAS,EACtB,GAAG,GAAG,yBAAyB,EAC/B,cAAc,GAAG,KAAK,EACtB,2CAA2C,GAAG,KAAK,EACxC;IACX,IAAI,CAAC,CAAC,MAAM,oBAAoB,EAAE,CAAC,EAAE;QACjC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,yEAAyE,CAAC,CAAC,CAAC;QAClG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;KACnB;IAED,MAAM,qCAAqC,GAAG,CAAC,2CAA2C,IAAI,CAAC,IAAI,CAAC;IAEpG,MAAM,6BAA6B,GAAG,MAAM,gCAAgC,EAAE,CAAC;IAE/E,MAAM,QAAQ,GAAG,WAAW,EAAE,CAAC;IAC/B,MAAM,YAAY,GAAG,MAAM,eAAe,EAAE,CAAC;IAC7C,MAAM,kBAAkB,GAAG,yBAAyB,EAAE,CAAC;IACvD,MAAM,cAAc,GAAe,MAAM,yBAAyB,CAAC,GAAG,EAAE,EAAC,QAAQ,EAAE,IAAI,EAAC,CAAC,CAAC;IAC1F,IAAI,eAAe,GAAG,KAAK,CAAC;IAE5B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,cAAc,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE;QAC5C,MAAM,QAAQ,GAAG,cAAc,CAAC,CAAC,CAAC,CAAC;QACnC,MAAM,UAAU,GAAG,CAAC,KAAK,cAAc,CAAC,MAAM,GAAG,CAAC,CAAC;QAEnD,oBAAoB,CAAC,QAAQ,CAAC,CAAC;QAE/B,IAAI,CAAC,eAAe,EAAE;YAClB,MAAM,qBAAqB,CAAC,IAAI,CAAC,CAAC;YAClC,eAAe,GAAG,IAAI,CAAC;SAC1B;QAED,MAAM,YAAY,GAAiB;YAC/B,kBAAkB;YAClB,YAAY,EAAE,IAAI;YAClB,QAAQ;YACR,YAAY;YACZ,IAAI,EAAE,IAAI;gBACN,CAAC,CAAC,IAA2B;gBAC7B,CAAC,CAAC,OAAO,CAAC,IAAI;YAClB,GAAG,EAAE,QAAQ;YACb,QAAQ,EAAE;gBACN,IAAI,EAAE,6BAA6B,EAAE,kBAAkB,IAAI,yBAAyB;gBACpF,OAAO,EAAE,6BAA6B,EAAE,GAAG,IAAI,sBAAsB;aACxE;SACJ,CAAC;QAEF,IAAI;YACA,MAAM,cAAc,CAAC;gBACjB,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,qBAAqB,CAAC;gBAC1C,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC;gBACzC,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,6BAA6B,CAAC;aAClD,EAAE,KAAK,IAAI,EAAE;gBACV,MAAM,eAAe,CAAC,YAAY,EAAE;oBAChC,UAAU,EAAE,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS;oBAC/C,mBAAmB,EAAE,IAAI;oBACzB,qBAAqB,EAAE,KAAK;oBAC5B,0BAA0B,EAAE,KAAK;oBACjC,qCAAqC;iBACxC,CAAC,CAAC;YACP,CAAC,CAAC,CAAC;SACN;QAAC,OAAO,GAAG,EAAE;YACV,OAAO,CAAC,KAAK,CACT,mBAAmB,EAAE;gBACrB,kCAAkC,qBAAqB,CAAC,QAAQ,CAAC,YAAY;gBAC7E,CACI,CAAC,UAAU;oBACP,CAAC,CAAC,2CAA2C,qBAAqB,CAAC,cAAc,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,YAAY;oBACrG,CAAC,CAAC,EAAE,CACX;gBACD,QAAQ,EACR,GAAG,CACN,CAAC;YAEF,IAAI,UAAU;gBACV,MAAM,GAAG,CAAC;YAEd,SAAS;SACZ;QAED,MAAM,OAAO,CAAC;YACV,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,0BAA0B,CAAC;YAC/C,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,yBAAyB,CAAC;YAC9C,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,kCAAkC,CAAC;SACvD,EAAE,KAAK,IAAI,EAAE;YACV,MAAM,eAAe,EAAE,CAAC;QAC5B,CAAC,CAAC,CAAC;QAEH,IAAI,CAAC,cAAc,EAAE;YACjB,OAAO,CAAC,GAAG,EAAE,CAAC;YACd,8BAA8B,CAAC,YAAY,EAAE,GAAG,KAAK,MAAM,EAAE,IAAI,CAAC,CAAC;YACnE,OAAO,CAAC,GAAG,EAAE,CAAC;SACjB;QAED,MAAM;KACT;AACL,CAAC"}

package/dist/cli/commands/ChatCommand.d.ts

@@ -1,8 +1,11 @@
 import { CommandModule } from "yargs";
 import { LlamaGrammar } from "../../evaluator/LlamaGrammar.js";
+import { BuildGpu } from "../../bindings/types.js";
 import { SpecializedChatWrapperTypeName } from "../../chatWrappers/utils/resolveChatWrapper.js";
 type ChatCommand = {
-    model: string;
+    modelPath?: string;
+    header?: string[];
+    gpu?: BuildGpu | "auto";
     systemInfo: boolean;
     systemPrompt: string;
     systemPromptFile?: string;

package/dist/cli/commands/ChatCommand.js

@@ -11,7 +11,7 @@ import { getLlama } from "../../bindings/getLlama.js";
 import { LlamaGrammar } from "../../evaluator/LlamaGrammar.js";
 import { LlamaChatSession } from "../../evaluator/LlamaChatSession/LlamaChatSession.js";
 import { LlamaJsonSchemaGrammar } from "../../evaluator/LlamaJsonSchemaGrammar.js";
-import { LlamaLogLevel, LlamaLogLevelGreaterThan } from "../../bindings/types.js";
+import { LlamaLogLevel, LlamaLogLevelGreaterThan, nodeLlamaCppGpuOptions, parseNodeLlamaCppGpuOption } from "../../bindings/types.js";
 import withOra from "../../utils/withOra.js";
 import { TokenMeter } from "../../evaluator/TokenMeter.js";
 import { printInfoLine } from "../utils/printInfoLine.js";
@@ -19,25 +19,42 @@ import { resolveChatWrapper, specializedChatWrapperTypeNames } from "../../chatW
 import { GeneralChatWrapper } from "../../chatWrappers/GeneralChatWrapper.js";
 import { printCommonInfoLines } from "../utils/printCommonInfoLines.js";
 import { resolveCommandGgufPath } from "../utils/resolveCommandGgufPath.js";
+import { withProgressLog } from "../../utils/withProgressLog.js";
+import { resolveHeaderFlag } from "../utils/resolveHeaderFlag.js";
 export const ChatCommand = {
     command: "chat [modelPath]",
     describe: "Chat with a Llama model",
     builder(yargs) {
         const isInDocumentationMode = getIsInDocumentationMode();
         return yargs
-            .option("model", {
-            alias: ["m", "modelPath"],
+            .option("modelPath", {
+            alias: ["m", "model", "path", "url"],
             type: "string",
-            demandOption: true,
-            description: "Llama model file to use for the chat",
-            group: "Required:"
+            description: "Llama model file to use for the chat. Can be a path to a local file or a URL of a model file to download"
+        })
+            .option("header", {
+            alias: ["H"],
+            type: "string",
+            array: true,
+            description: "Headers to use when downloading a model from a URL, in the format `key: value`. You can pass this option multiple times to add multiple headers."
+        })
+            .option("gpu", {
+            type: "string",
+            // yargs types don't support passing `false` as a choice, although it is supported by yargs
+            choices: nodeLlamaCppGpuOptions,
+            coerce: (value) => {
+                if (value == null || value == "")
+                    return undefined;
+                return parseNodeLlamaCppGpuOption(value);
+            },
+            defaultDescription: "Uses the latest local build, and fallbacks to \"auto\"",
+            description: "Compute layer implementation type to use for llama.cpp. If omitted, uses the latest local build, and fallbacks to \"auto\""
         })
             .option("systemInfo", {
             alias: "i",
             type: "boolean",
             default: false,
-            description: "Print llama.cpp system info",
-            group: "Optional:"
+            description: "Print llama.cpp system info"
         })
             .option("systemPrompt", {
             alias: "s",
@@ -45,195 +62,167 @@ export const ChatCommand = {
             default: defaultChatSystemPrompt,
             defaultDescription: " ",
             description: "System prompt to use against the model" +
-                (isInDocumentationMode ? "" : (". [default value: " + defaultChatSystemPrompt.split("\n").join(" ") + "]")),
-            group: "Optional:"
+                (isInDocumentationMode ? "" : (". [default value: " + defaultChatSystemPrompt.split("\n").join(" ") + "]"))
         })
             .option("systemPromptFile", {
             type: "string",
-            description: "Path to a file to load text from and use as as the model system prompt",
-            group: "Optional:"
+            description: "Path to a file to load text from and use as as the model system prompt"
         })
             .option("prompt", {
             type: "string",
-            description: "First prompt to automatically send to the model when starting the chat",
-            group: "Optional:"
+            description: "First prompt to automatically send to the model when starting the chat"
         })
             .option("promptFile", {
             type: "string",
-            description: "Path to a file to load text from and use as a first prompt to automatically send to the model when starting the chat",
-            group: "Optional:"
+            description: "Path to a file to load text from and use as a first prompt to automatically send to the model when starting the chat"
         })
             .option("wrapper", {
             alias: "w",
             type: "string",
             default: "auto",
             choices: ["auto", ...specializedChatWrapperTypeNames],
-            description: "Chat wrapper to use. Use `auto` to automatically select a wrapper based on the model's BOS token",
-            group: "Optional:"
+            description: "Chat wrapper to use. Use `auto` to automatically select a wrapper based on the model's BOS token"
         })
             .option("noJinja", {
             type: "boolean",
             default: false,
-            description: "Don't use a Jinja wrapper, even if it's the best option for the model",
-            group: "Optional:"
+            description: "Don't use a Jinja wrapper, even if it's the best option for the model"
         })
             .option("contextSize", {
             alias: "c",
             type: "number",
             description: "Context size to use for the model context",
             default: -1,
-            defaultDescription: "Automatically determined based on the available VRAM",
-            group: "Optional:"
+            defaultDescription: "Automatically determined based on the available VRAM"
         })
             .option("batchSize", {
             alias: "b",
             type: "number",
-            description: "Batch size to use for the model context. The default value is the context size",
-            group: "Optional:"
+            description: "Batch size to use for the model context. The default value is the context size"
         })
             .option("noTrimWhitespace", {
             type: "boolean",
             alias: ["noTrim"],
             default: false,
-            description: "Don't trim whitespaces from the model response",
-            group: "Optional:"
+            description: "Don't trim whitespaces from the model response"
         })
             .option("grammar", {
             alias: "g",
             type: "string",
             default: "text",
             choices: ["text", "json", "list", "arithmetic", "japanese", "chess"],
-            description: "Restrict the model response to a specific grammar, like JSON for example",
-            group: "Optional:"
+            description: "Restrict the model response to a specific grammar, like JSON for example"
         })
             .option("jsonSchemaGrammarFile", {
             alias: ["jsgf"],
             type: "string",
-            description: "File path to a JSON schema file, to restrict the model response to only generate output that conforms to the JSON schema",
-            group: "Optional:"
+            description: "File path to a JSON schema file, to restrict the model response to only generate output that conforms to the JSON schema"
         })
             .option("threads", {
             type: "number",
             default: 6,
-            description: "Number of threads to use for the evaluation of tokens",
-            group: "Optional:"
+            description: "Number of threads to use for the evaluation of tokens"
         })
             .option("temperature", {
             alias: "t",
             type: "number",
             default: 0,
-            description: "Temperature is a hyperparameter that controls the randomness of the generated text. It affects the probability distribution of the model's output tokens. A higher temperature (e.g., 1.5) makes the output more random and creative, while a lower temperature (e.g., 0.5) makes the output more focused, deterministic, and conservative. The suggested temperature is 0.8, which provides a balance between randomness and determinism. At the extreme, a temperature of 0 will always pick the most likely next token, leading to identical outputs in each run. Set to `0` to disable.",
-            group: "Optional:"
+            description: "Temperature is a hyperparameter that controls the randomness of the generated text. It affects the probability distribution of the model's output tokens. A higher temperature (e.g., 1.5) makes the output more random and creative, while a lower temperature (e.g., 0.5) makes the output more focused, deterministic, and conservative. The suggested temperature is 0.8, which provides a balance between randomness and determinism. At the extreme, a temperature of 0 will always pick the most likely next token, leading to identical outputs in each run. Set to `0` to disable."
         })
             .option("minP", {
             alias: "mp",
             type: "number",
             default: 0,
-            description: "From the next token candidates, discard the percentage of tokens with the lowest probability. For example, if set to `0.05`, 5% of the lowest probability tokens will be discarded. This is useful for generating more high-quality results when using a high temperature. Set to a value between `0` and `1` to enable. Only relevant when `temperature` is set to a value greater than `0`.",
-            group: "Optional:"
+            description: "From the next token candidates, discard the percentage of tokens with the lowest probability. For example, if set to `0.05`, 5% of the lowest probability tokens will be discarded. This is useful for generating more high-quality results when using a high temperature. Set to a value between `0` and `1` to enable. Only relevant when `temperature` is set to a value greater than `0`."
         })
             .option("topK", {
             alias: "k",
             type: "number",
             default: 40,
-            description: "Limits the model to consider only the K most likely next tokens for sampling at each step of sequence generation. An integer number between `1` and the size of the vocabulary. Set to `0` to disable (which uses the full vocabulary). Only relevant when `temperature` is set to a value greater than 0.",
-            group: "Optional:"
+            description: "Limits the model to consider only the K most likely next tokens for sampling at each step of sequence generation. An integer number between `1` and the size of the vocabulary. Set to `0` to disable (which uses the full vocabulary). Only relevant when `temperature` is set to a value greater than 0."
         })
             .option("topP", {
             alias: "p",
             type: "number",
             default: 0.95,
-            description: "Dynamically selects the smallest set of tokens whose cumulative probability exceeds the threshold P, and samples the next token only from this set. A float number between `0` and `1`. Set to `1` to disable. Only relevant when `temperature` is set to a value greater than `0`.",
-            group: "Optional:"
+            description: "Dynamically selects the smallest set of tokens whose cumulative probability exceeds the threshold P, and samples the next token only from this set. A float number between `0` and `1`. Set to `1` to disable. Only relevant when `temperature` is set to a value greater than `0`."
         })
             .option("gpuLayers", {
             alias: "gl",
             type: "number",
             description: "number of layers to store in VRAM",
             default: -1,
-            defaultDescription: "Automatically determined based on the available VRAM",
-            group: "Optional:"
+            defaultDescription: "Automatically determined based on the available VRAM"
         })
             .option("repeatPenalty", {
             alias: "rp",
             type: "number",
             default: 1.1,
-            description: "Prevent the model from repeating the same token too much. Set to `1` to disable.",
-            group: "Optional:"
+            description: "Prevent the model from repeating the same token too much. Set to `1` to disable."
         })
             .option("lastTokensRepeatPenalty", {
             alias: "rpn",
             type: "number",
             default: 64,
-            description: "Number of recent tokens generated by the model to apply penalties to repetition of",
-            group: "Optional:"
+            description: "Number of recent tokens generated by the model to apply penalties to repetition of"
         })
             .option("penalizeRepeatingNewLine", {
             alias: "rpnl",
             type: "boolean",
             default: true,
-            description: "Penalize new line tokens. set \"--no-penalizeRepeatingNewLine\" or \"--no-rpnl\" to disable",
-            group: "Optional:"
+            description: "Penalize new line tokens. set `--no-penalizeRepeatingNewLine` or `--no-rpnl` to disable"
         })
             .option("repeatFrequencyPenalty", {
             alias: "rfp",
             type: "number",
-            description: "For n time a token is in the `punishTokens` array, lower its probability by `n * repeatFrequencyPenalty`. Set to a value between `0` and `1` to enable.",
-            group: "Optional:"
+            description: "For n time a token is in the `punishTokens` array, lower its probability by `n * repeatFrequencyPenalty`. Set to a value between `0` and `1` to enable."
         })
             .option("repeatPresencePenalty", {
             alias: "rpp",
             type: "number",
-            description: "Lower the probability of all the tokens in the `punishTokens` array by `repeatPresencePenalty`. Set to a value between `0` and `1` to enable.",
-            group: "Optional:"
+            description: "Lower the probability of all the tokens in the `punishTokens` array by `repeatPresencePenalty`. Set to a value between `0` and `1` to enable."
         })
             .option("maxTokens", {
             alias: "mt",
             type: "number",
             default: 0,
-            description: "Maximum number of tokens to generate in responses. Set to `0` to disable. Set to `-1` to set to the context size",
-            group: "Optional:"
+            description: "Maximum number of tokens to generate in responses. Set to `0` to disable. Set to `-1` to set to the context size"
         })
             .option("noHistory", {
             alias: "nh",
             type: "boolean",
             default: false,
-            description: "Don't load or save chat history",
-            group: "Optional:"
+            description: "Don't load or save chat history"
         })
             .option("environmentFunctions", {
             alias: "ef",
             type: "boolean",
             default: false,
-            description: "Provide access to environment functions like `getDate` and `getTime`",
-            group: "Optional:"
+            description: "Provide access to environment functions like `getDate` and `getTime`"
         })
             .option("debug", {
             alias: "d",
             type: "boolean",
             default: false,
-            description: "Print llama.cpp info and debug logs",
-            group: "Optional:"
+            description: "Print llama.cpp info and debug logs"
         })
             .option("meter", {
             type: "boolean",
             default: false,
-            description: "Log how many tokens were used as input and output for each response",
-            group: "Optional:"
+            description: "Print how many tokens were used as input and output for each response"
         })
             .option("printTimings", {
             alias: "pt",
             type: "boolean",
             default: false,
-            description: "Print llama.cpp timings after each response",
-            group: "Optional:"
+            description: "Print llama.cpp timings after each response"
         });
     },
-    async handler({ model, systemInfo, systemPrompt, systemPromptFile, prompt, promptFile, wrapper, noJinja, contextSize, batchSize, noTrimWhitespace, grammar, jsonSchemaGrammarFile, threads, temperature, minP, topK, topP, gpuLayers, repeatPenalty, lastTokensRepeatPenalty, penalizeRepeatingNewLine, repeatFrequencyPenalty, repeatPresencePenalty, maxTokens, noHistory, environmentFunctions, debug, meter, printTimings }) {
+    async handler({ modelPath, header, gpu, systemInfo, systemPrompt, systemPromptFile, prompt, promptFile, wrapper, noJinja, contextSize, batchSize, noTrimWhitespace, grammar, jsonSchemaGrammarFile, threads, temperature, minP, topK, topP, gpuLayers, repeatPenalty, lastTokensRepeatPenalty, penalizeRepeatingNewLine, repeatFrequencyPenalty, repeatPresencePenalty, maxTokens, noHistory, environmentFunctions, debug, meter, printTimings }) {
         try {
             await RunChat({
-                model, systemInfo, systemPrompt, systemPromptFile, prompt, promptFile, wrapper, noJinja, contextSize, batchSize,
-                noTrimWhitespace, grammar, jsonSchemaGrammarFile, threads, temperature, minP, topK, topP, gpuLayers,
+                modelPath, header, gpu, systemInfo, systemPrompt, systemPromptFile, prompt, promptFile, wrapper, noJinja, contextSize,
+                batchSize, noTrimWhitespace, grammar, jsonSchemaGrammarFile, threads, temperature, minP, topK, topP, gpuLayers,
                 lastTokensRepeatPenalty, repeatPenalty, penalizeRepeatingNewLine, repeatFrequencyPenalty, repeatPresencePenalty, maxTokens,
                 noHistory, environmentFunctions, debug, meter, printTimings
             });
@@ -245,22 +234,28 @@ export const ChatCommand = {
         }
     }
 };
-async function RunChat({ model: modelArg, systemInfo, systemPrompt, systemPromptFile, prompt, promptFile, wrapper, noJinja, contextSize, batchSize, noTrimWhitespace, grammar: grammarArg, jsonSchemaGrammarFile: jsonSchemaGrammarFilePath, threads, temperature, minP, topK, topP, gpuLayers, lastTokensRepeatPenalty, repeatPenalty, penalizeRepeatingNewLine, repeatFrequencyPenalty, repeatPresencePenalty, maxTokens, noHistory, environmentFunctions, debug, meter, printTimings }) {
+async function RunChat({ modelPath: modelArg, header: headerArg, gpu, systemInfo, systemPrompt, systemPromptFile, prompt, promptFile, wrapper, noJinja, contextSize, batchSize, noTrimWhitespace, grammar: grammarArg, jsonSchemaGrammarFile: jsonSchemaGrammarFilePath, threads, temperature, minP, topK, topP, gpuLayers, lastTokensRepeatPenalty, repeatPenalty, penalizeRepeatingNewLine, repeatFrequencyPenalty, repeatPresencePenalty, maxTokens, noHistory, environmentFunctions, debug, meter, printTimings }) {
     if (contextSize === -1)
         contextSize = undefined;
     if (gpuLayers === -1)
         gpuLayers = undefined;
+    const headers = resolveHeaderFlag(headerArg);
     const trimWhitespace = !noTrimWhitespace;
     if (debug)
         console.info(`${chalk.yellow("Log level:")} debug`);
-    const resolvedModelPath = await resolveCommandGgufPath(modelArg);
     const llamaLogLevel = debug
         ? LlamaLogLevel.debug
         : LlamaLogLevel.warn;
-    const llama = await getLlama("lastBuild", {
-        logLevel: llamaLogLevel
-    });
+    const llama = gpu == null
+        ? await getLlama("lastBuild", {
+            logLevel: llamaLogLevel
+        })
+        : await getLlama({
+            gpu,
+            logLevel: llamaLogLevel
+        });
     const logBatchSize = batchSize != null;
+    const resolvedModelPath = await resolveCommandGgufPath(modelArg, llama, headers);
     if (systemInfo)
         console.log(llama.systemInfo);
     if (systemPromptFile != null && systemPromptFile !== "") {
@@ -278,18 +273,34 @@ async function RunChat({ model: modelArg, systemInfo, systemPrompt, systemPrompt
         batchSize = contextSize;
     }
     let initialPrompt = prompt ?? null;
-    const model = await withOra({
-        loading: chalk.blue("Loading model"),
-        success: chalk.blue("Model loaded"),
-        fail: chalk.blue("Failed to load model"),
-        useStatusLogs: debug
-    }, async () => {
+    const model = await withProgressLog({
+        loadingText: chalk.blue.bold("Loading model"),
+        successText: chalk.blue("Model loaded"),
+        failText: chalk.blue("Failed to load model"),
+        liveUpdates: !debug,
+        noProgress: debug,
+        liveCtrlCSendsAbortSignal: true
+    }, async (progressUpdater) => {
         try {
             return await llama.loadModel({
                 modelPath: resolvedModelPath,
-                gpuLayers: gpuLayers != null ? gpuLayers : undefined
+                gpuLayers: gpuLayers != null
+                    ? gpuLayers
+                    : contextSize != null
+                        ? { fitContext: { contextSize } }
+                        : undefined,
+                ignoreMemorySafetyChecks: gpuLayers != null,
+                onLoadProgress(loadProgress) {
+                    progressUpdater.setProgress(loadProgress);
+                },
+                loadSignal: progressUpdater.abortSignal
             });
         }
+        catch (err) {
+            if (err === progressUpdater.abortSignal?.reason)
+                process.exit(0);
+            throw err;
+        }
         finally {
             if (llama.logLevel === LlamaLogLevel.debug) {
                 await new Promise((accept) => setTimeout(accept, 0)); // wait for logs to finish printing
@@ -307,7 +318,8 @@ async function RunChat({ model: modelArg, systemInfo, systemPrompt, systemPrompt
             return await model.createContext({
                 contextSize: contextSize != null ? contextSize : undefined,
                 batchSize: batchSize != null ? batchSize : undefined,
-                threads
+                threads,
+                ignoreMemorySafetyChecks: gpuLayers != null || contextSize != null
             });
         }
         finally {
@@ -327,7 +339,7 @@ async function RunChat({ model: modelArg, systemInfo, systemPrompt, systemPrompt
         bosString: model.tokens.bosString,
         filename: model.filename,
         fileInfo: model.fileInfo,
-        tokenizer: model.tokenize,
+        tokenizer: model.tokenizer,
         noJinja
     }) ?? new GeneralChatWrapper();
     const contextSequence = context.getSequence();