node-llama-cpp 3.0.0-beta.15 → 3.0.0-beta.16

package/dist/gguf/insights/GgufInsightsConfigurationResolver.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"GgufInsightsConfigurationResolver.js","sourceRoot":"","sources":["../../../src/gguf/insights/GgufInsightsConfigurationResolver.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,IAAI,CAAC;AAIpB,OAAO,EAAC,0BAA0B,EAAC,MAAM,8CAA8C,CAAC;AACxF,OAAO,EAAC,2BAA2B,EAAC,MAAM,wCAAwC,CAAC;AACnF,OAAO,EAAC,+BAA+B,EAAC,MAAM,4CAA4C,CAAC;AAC3F,OAAO,EAAC,WAAW,EAAC,MAAM,wBAAwB,CAAC;AAInD,MAAM,OAAO,iCAAiC;IAC1C,gBAAgB,CAAkB,aAAa,CAAe;IAE9D,YAAoB,YAA0B;QAC1C,IAAI,CAAC,aAAa,GAAG,YAAY,CAAC;IACtC,CAAC;IAED,IAAW,YAAY;QACnB,OAAO,IAAI,CAAC,aAAa,CAAC;IAC9B,CAAC;IAED;;;;;;;;;;;;OAYG;IACI,oCAAoC,CAAC,EACxC,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,IAAI,CAAC,aAAa,CAAC,gBAAgB,IAAI,IAAI,CAAC,EACzE,gBAAgB,GAAG,KAAK,KAIxB,EAAE,EAAE,EACJ,YAAY,GAAG,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,iBAAiB,CAAC,cAAc,EAAE,CAAC,EACnF,WAAW,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC,EAAC,KAAK,EAAE,EAAE,CAAC,QAAQ,EAAE,EAAE,IAAI,EAAE,EAAE,CAAC,OAAO,EAAE,EAAC,CAAC,CAAC,EAClE,oBAAoB,GAAG,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,eAAe,EAChE,QAAQ,GAAG,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,GAAG,EACxC,0BAA0B,GAAG,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,qBAAqB,KAO5E,EAAE;QAiCF,MAAM,uBAAuB,GAAG,IAAI,CAAC;QAErC,MAAM,SAAS,GAAG,YAAY,EAAE,CAAC;QACjC,MAAM,QAAQ,GAAG,WAAW,EAAE,CAAC;QAC/B,MAAM,iBAAiB,GAAG,IAAI,CAAC,qBAAqB,CAAC,MAAM,EAAE;YACzD,YAAY,EAAE,GAAG,EAAE,CAAC,SAAS;YAC7B,oBAAoB;YACpB,QAAQ;YACR,0BAA0B;SAC7B,CAAC,CAAC;QACH,MAAM,SAAS,GAAG,0BAA0B,IAAI,QAAQ,KAAK,KAAK,CAAC;QACnE,MAAM,2BAA2B,GAAG,IAAI,CAAC,aAAa,CAAC,iCAAiC,CAAC;YACrF,SAAS,EAAE,iBAAiB;SAC/B,CAAC,CAAC;QAEH,MAAM,mBAAmB,GAAG,IAAI,CAAC,yBAAyB,CAAC,MAAM,EAAE;YAC/D,YAAY,EAAE,GAAG,EAAE,CAAC,CAAC,EAAC,KAAK,EAAE,SAAS,CAAC,KAAK,EAAE,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,CAAC,IAAI,GAAG,2BAA2B,CAAC,OAAO,CAAC,EAAC,CAAC;YACvH,QAAQ;YACR,kBAAkB,EAAE,gBAAgB;YACpC,cAAc,EAAE,iBAAiB;YACjC,qBAAqB,EAAE,IAAI,CAAC,aAAa,CAAC,gBAAgB,IAAI,uBAAuB;SACxF,CAAC,CAAC;QACH,MAAM,6BAA6B,GAAG,IAAI,CAAC,aAAa,CAAC,mCAAmC,CAAC;YACzF,WAAW,EAAE,mBAAmB;YAChC,kBAAkB,EAAE,gBAAgB;YACpC,cAAc,EAAE,iBAAiB;SACpC,CAAC,CAAC;QAEH,MAAM,UAAU,GAAG;YACf,SAAS,EAAE,EAAE;YACb,qBAAqB,EAAE,EAAE;YACzB,WAAW,EAAE,EAAE;YACf,iBAAiB,EAAE,EAAE;YACrB,qBAAqB,EAAE,EAAE;YACzB,gBAAgB,EAAE,EAAE;SACd,CAAC;QAEX,MAAM,eAAe,GAAG,UAAU,CAAC,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,iBAAiB,GAAG,IAAI,CAAC,aAAa,CAAC,WAAW,CAAC,CAAC;QAC/G,MAAM,2BAA2B,GAAG,UAAU,CAAC,qBAAqB,GAAG,CACnE,iBAAiB,KAAK,IAAI,CAAC,aAAa,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAC/D,CAAC;QACF,MAAM,iBAAiB,GAAG,UAAU,CAAC,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,mBAAmB,GAAG,WAAW,CAAC,CAAC;QAClG,MAAM,uBAAuB,GAAG,UAAU,CAAC,iBAAiB,GAAG,CAC3D,2BAA2B,CAAC,MAAM,IAAI,QAAQ,CAAC,IAAI;YAC/C,CAAC,CAAC,CAAC;YACH,CAAC,CAAC,2BAA2B,CAAC,MAAM,IAAI,QAAQ,CAAC,KAAK;gBAClD,CAAC,CAAC,GAAG;gBACL,CAAC,CAAC,CACE,GAAG,GAAG,IAAI,CAAC,GAAG,CACV,GAAG,EACH,GAAG,GAAG,CACF,CAAC,2BAA2B,CAAC,MAAM,GAAG,QAAQ,CAAC,KAAK,CAAC,GAAG,QAAQ,CAAC,KAAK,CACzE,CACJ,CACJ,CACZ,CAAC;QACF,MAAM,sBAAsB,GAAG,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,mBAAmB,GAAG,WAAW,CAAC,GAAG,WAAW,CAAC,CAAC;QAE9G,MAAM,kBAAkB,GAAG,SAAS;YAChC,CAAC,CAAC,CACE,CAAC,eAAe,GAAG,2BAA2B,GAAG,iBAAiB,GAAG,uBAAuB,CAAC;gBAC7F,CAAC,UAAU,CAAC,SAAS,GAAG,UAAU,CAAC,qBAAqB,GAAG,UAAU,CAAC,WAAW,GAAG,UAAU,CAAC,iBAAiB,CAAC,CACpH;YACD,CAAC,CAAC,CACE,CAAC,iBAAiB,GAAG,uBAAuB,GAAG,6BAA6B,CAAC,IAAI,CAAC,aAAa,CAAC,SAAS,CAAC,CAAC;gBAC3G,CAAC,UAAU,CAAC,WAAW,GAAG,UAAU,CAAC,iBAAiB,GAAG,UAAU,CAAC,qBAAqB,CAAC,CAAC,CAAC;QACpG,MAAM,UAAU,GAAG,sBAAsB,GAAG,UAAU,CAAC,gBAAgB,CAAC;QAExE,OAAO;YACH,kBAAkB;YAClB,UAAU;YACV,UAAU,EAAE,kBAAkB,GAAG,UAAU;YAE3C,cAAc,EAAE;gBACZ,SAAS,EAAE,iBAAiB;gBAC5B,WAAW,EAAE,mBAAmB;gBAEhC,aAAa,EAAE,2BAA2B,CAAC,MAAM;gBACjD,eAAe,EAAE,6BAA6B,CAAC,MAAM;gBACrD,aAAa,EAAE,2BAA2B,CAAC,MAAM,GAAG,6BAA6B,CAAC,MAAM;gBAExF,cAAc,EAAE,2BAA2B,CAAC,OAAO;gBACnD,gBAAgB,EAAE,6BAA6B,CAAC,OAAO;gBACvD,cAAc,EAAE,2BAA2B,CAAC,OAAO,GAAG,6BAA6B,CAAC,OAAO;aAC9F;SACJ,CAAC;IACN,CAAC;IAEM,qBAAqB,CAAC,SAAyC,EAAE,EACpE,wBAAwB,GAAG,KAAK,EAChC,YAAY,GAAG,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,iBAAiB,CAAC,cAAc,EAAE,CAAC,EACnF,oBAAoB,GAAG,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,eAAe,EAAE,QAAQ,GAAG,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,GAAG,EAC1G,0BAA0B,GAAG,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,qBAAqB,KAI5E,EAAE;QACF,OAAO,2BAA2B,CAAC,SAAS,EAAE;YAC1C,YAAY,EAAE,IAAI,CAAC,aAAa;YAChC,wBAAwB;YACxB,YAAY;YACZ,oBAAoB;YACpB,QAAQ;YACR,0BAA0B;SAC7B,CAAC,CAAC;IACP,CAAC;IAEM,yBAAyB,CAAC,WAA+C,EAAE,EAC9E,cAAc,EACd,SAAS,EACT,SAAS,GAAG,0BAA0B,EAAE,EACxC,qBAAqB,EACrB,YAAY,GAAG,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,iBAAiB,CAAC,cAAc,EAAE,CAAC,EACnF,QAAQ,GAAG,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,GAAG,EACxC,wBAAwB,GAAG,KAAK,EAChC,kBAAkB,GAAG,KAAK,EAU7B;QACG,OAAO,+BAA+B,CAAC;YACnC,WAAW;YACX,SAAS;YACT,SAAS;YACT,iBAAiB,EAAE,IAAI,CAAC,aAAa;YACrC,cAAc;YACd,qBAAqB;YACrB,YAAY;YACZ,QAAQ;YACR,wBAAwB;YACxB,kBAAkB;SACrB,CAAC,CAAC;IACP,CAAC;IAED,gBAAgB;IACT,MAAM,CAAC,OAAO,CAAC,YAA0B;QAC5C,OAAO,IAAI,iCAAiC,CAAC,YAAY,CAAC,CAAC;IAC/D,CAAC;CACJ;AAED,SAAS,6BAA6B,CAAC,SAAiB;IACpD,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;IAC/B,OAAO,EAAE,GAAG,WAAW,CAAC,SAAS,EAAE,CAAC;YAChC,KAAK,EAAE,IAAI;YACX,GAAG,EAAE,IAAI,GAAG,GAAG;YACf,MAAM,EAAE,EAAE;SACb,EAAE;YACC,KAAK,EAAE,IAAI,GAAG,GAAG;YACjB,GAAG,EAAE,IAAI,GAAG,CAAC;YACb,MAAM,EAAE,EAAE;SACb,EAAE;YACC,KAAK,EAAE,IAAI,GAAG,CAAC;YACf,MAAM,EAAE,CAAC;SACZ,CAAC,CAAC,CAAC;AACR,CAAC"}

package/dist/gguf/insights/utils/resolveContextContextSizeOption.d.ts

@@ -0,0 +1,18 @@
+import { LlamaContextOptions } from "../../../evaluator/LlamaContext/types.js";
+import { GgufInsights } from "../GgufInsights.js";
+import { BuildGpu } from "../../../bindings/types.js";
+export declare function resolveContextContextSizeOption({ contextSize, batchSize, sequences, modelFileInsights, modelGpuLayers, modelTrainContextSize, getVramState, llamaGpu, ignoreMemorySafetyChecks, isEmbeddingContext }: {
+    contextSize?: LlamaContextOptions["contextSize"];
+    batchSize?: LlamaContextOptions["batchSize"];
+    sequences: number;
+    modelFileInsights: GgufInsights;
+    modelGpuLayers: number;
+    modelTrainContextSize: number;
+    getVramState(): {
+        total: number;
+        free: number;
+    };
+    llamaGpu: BuildGpu;
+    ignoreMemorySafetyChecks?: boolean;
+    isEmbeddingContext?: boolean;
+}): number;

package/dist/gguf/insights/utils/resolveContextContextSizeOption.js

@@ -0,0 +1,52 @@
+import { minAllowedContextSizeInCalculations } from "../../../config.js";
+import { getDefaultContextBatchSize, getDefaultModelContextSize } from "../../../evaluator/LlamaContext/LlamaContext.js";
+export function resolveContextContextSizeOption({ contextSize, batchSize, sequences, modelFileInsights, modelGpuLayers, modelTrainContextSize, getVramState, llamaGpu, ignoreMemorySafetyChecks = false, isEmbeddingContext = false }) {
+    if (contextSize == null)
+        contextSize = "auto";
+    if (typeof contextSize === "number") {
+        const resolvedContextSize = Math.max(1, Math.floor(contextSize));
+        if (ignoreMemorySafetyChecks)
+            return resolvedContextSize;
+        const vramState = getVramState();
+        const contextVram = modelFileInsights.estimateContextResourceRequirements({
+            contextSize: resolvedContextSize,
+            batchSize: batchSize ?? getDefaultContextBatchSize({ contextSize: resolvedContextSize, sequences }),
+            modelGpuLayers: modelGpuLayers,
+            sequences,
+            isEmbeddingContext
+        }).gpuVram;
+        if (contextVram > vramState.free)
+            throw new Error(`The context size of ${resolvedContextSize}${sequences > 1 ? ` with ${sequences} sequences` : ""} is too large for the available VRAM`);
+        return resolvedContextSize;
+    }
+    else if (contextSize === "auto" || typeof contextSize === "object") {
+        if (llamaGpu === false)
+            return modelTrainContextSize;
+        const vramState = getVramState();
+        if (vramState.total === 0)
+            return modelTrainContextSize;
+        const freeVram = vramState.free;
+        const maxContextSize = contextSize === "auto"
+            ? getDefaultModelContextSize({ trainContextSize: modelTrainContextSize })
+            : Math.min(contextSize.max ?? getDefaultModelContextSize({ trainContextSize: modelTrainContextSize }), getDefaultModelContextSize({ trainContextSize: modelTrainContextSize }));
+        const minContextSize = contextSize === "auto"
+            ? minAllowedContextSizeInCalculations
+            : Math.max(contextSize.min ?? minAllowedContextSizeInCalculations, minAllowedContextSizeInCalculations);
+        for (let testContextSize = maxContextSize; testContextSize >= minContextSize; testContextSize--) {
+            const contextVram = modelFileInsights.estimateContextResourceRequirements({
+                contextSize: testContextSize,
+                batchSize: batchSize ?? getDefaultContextBatchSize({ contextSize: testContextSize, sequences }),
+                modelGpuLayers: modelGpuLayers,
+                sequences,
+                isEmbeddingContext
+            }).gpuVram;
+            if (contextVram <= freeVram)
+                return testContextSize;
+        }
+        if (ignoreMemorySafetyChecks)
+            return minContextSize;
+        throw new Error(`The available VRAM is too small to fit the context size of ${maxContextSize}${sequences > 1 ? ` with ${sequences} sequences` : ""}`);
+    }
+    throw new Error(`Invalid context size: "${contextSize}"`);
+}
+//# sourceMappingURL=resolveContextContextSizeOption.js.map

package/dist/gguf/insights/utils/resolveContextContextSizeOption.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"resolveContextContextSizeOption.js","sourceRoot":"","sources":["../../../../src/gguf/insights/utils/resolveContextContextSizeOption.ts"],"names":[],"mappings":"AAGA,OAAO,EAAC,mCAAmC,EAAC,MAAM,oBAAoB,CAAC;AACvE,OAAO,EAAC,0BAA0B,EAAE,0BAA0B,EAAC,MAAM,iDAAiD,CAAC;AAEvH,MAAM,UAAU,+BAA+B,CAAC,EAC5C,WAAW,EAAE,SAAS,EAAE,SAAS,EAAE,iBAAiB,EAAE,cAAc,EAAE,qBAAqB,EAAE,YAAY,EAAE,QAAQ,EACnH,wBAAwB,GAAG,KAAK,EAAE,kBAAkB,GAAG,KAAK,EAY/D;IACG,IAAI,WAAW,IAAI,IAAI;QACnB,WAAW,GAAG,MAAM,CAAC;IAEzB,IAAI,OAAO,WAAW,KAAK,QAAQ,EAAE;QACjC,MAAM,mBAAmB,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC;QAEjE,IAAI,wBAAwB;YACxB,OAAO,mBAAmB,CAAC;QAE/B,MAAM,SAAS,GAAG,YAAY,EAAE,CAAC;QACjC,MAAM,WAAW,GAAG,iBAAiB,CAAC,mCAAmC,CAAC;YACtE,WAAW,EAAE,mBAAmB;YAChC,SAAS,EAAE,SAAS,IAAI,0BAA0B,CAAC,EAAC,WAAW,EAAE,mBAAmB,EAAE,SAAS,EAAC,CAAC;YACjG,cAAc,EAAE,cAAc;YAC9B,SAAS;YACT,kBAAkB;SACrB,CAAC,CAAC,OAAO,CAAC;QAEX,IAAI,WAAW,GAAG,SAAS,CAAC,IAAI;YAC5B,MAAM,IAAI,KAAK,CAAC,uBAAuB,mBAAmB,GAAG,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,SAAS,YAAY,CAAC,CAAC,CAAC,EAAE,sCAAsC,CAAC,CAAC;QAE5J,OAAO,mBAAmB,CAAC;KAC9B;SAAM,IAAI,WAAW,KAAK,MAAM,IAAI,OAAO,WAAW,KAAK,QAAQ,EAAE;QAClE,IAAI,QAAQ,KAAK,KAAK;YAClB,OAAO,qBAAqB,CAAC;QAEjC,MAAM,SAAS,GAAG,YAAY,EAAE,CAAC;QAEjC,IAAI,SAAS,CAAC,KAAK,KAAK,CAAC;YACrB,OAAO,qBAAqB,CAAC;QAEjC,MAAM,QAAQ,GAAG,SAAS,CAAC,IAAI,CAAC;QAEhC,MAAM,cAAc,GAAG,WAAW,KAAK,MAAM;YACzC,CAAC,CAAC,0BAA0B,CAAC,EAAC,gBAAgB,EAAE,qBAAqB,EAAC,CAAC;YACvE,CAAC,CAAC,IAAI,CAAC,GAAG,CACN,WAAW,CAAC,GAAG,IAAI,0BAA0B,CAAC,EAAC,gBAAgB,EAAE,qBAAqB,EAAC,CAAC,EACxF,0BAA0B,CAAC,EAAC,gBAAgB,EAAE,qBAAqB,EAAC,CAAC,CACxE,CAAC;QAEN,MAAM,cAAc,GAAG,WAAW,KAAK,MAAM;YACzC,CAAC,CAAC,mCAAmC;YACrC,CAAC,CAAC,IAAI,CAAC,GAAG,CACN,WAAW,CAAC,GAAG,IAAI,mCAAmC,EACtD,mCAAmC,CACtC,CAAC;QAEN,KAAK,IAAI,eAAe,GAAG,cAAc,EAAE,eAAe,IAAI,cAAc,EAAE,eAAe,EAAE,EAAE;YAC7F,MAAM,WAAW,GAAG,iBAAiB,CAAC,mCAAmC,CAAC;gBACtE,WAAW,EAAE,eAAe;gBAC5B,SAAS,EAAE,SAAS,IAAI,0BAA0B,CAAC,EAAC,WAAW,EAAE,eAAe,EAAE,SAAS,EAAC,CAAC;gBAC7F,cAAc,EAAE,cAAc;gBAC9B,SAAS;gBACT,kBAAkB;aACrB,CAAC,CAAC,OAAO,CAAC;YAEX,IAAI,WAAW,IAAI,QAAQ;gBACvB,OAAO,eAAe,CAAC;SAC9B;QAED,IAAI,wBAAwB;YACxB,OAAO,cAAc,CAAC;QAE1B,MAAM,IAAI,KAAK,CAAC,8DAA8D,cAAc,GAAG,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,SAAS,YAAY,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;KACzJ;IAED,MAAM,IAAI,KAAK,CAAC,0BAA0B,WAAW,GAAG,CAAC,CAAC;AAC9D,CAAC"}

package/dist/gguf/insights/utils/resolveModelGpuLayersOption.d.ts

@@ -0,0 +1,14 @@
+import { LlamaModelOptions } from "../../../evaluator/LlamaModel.js";
+import { BuildGpu } from "../../../bindings/types.js";
+import type { GgufInsights } from "../GgufInsights.js";
+export declare function resolveModelGpuLayersOption(gpuLayers: LlamaModelOptions["gpuLayers"], { ggufInsights, ignoreMemorySafetyChecks, getVramState, llamaVramPaddingSize, llamaGpu, llamaSupportsGpuOffloading }: {
+    ggufInsights: GgufInsights;
+    ignoreMemorySafetyChecks?: boolean;
+    getVramState(): {
+        total: number;
+        free: number;
+    };
+    llamaVramPaddingSize: number;
+    llamaGpu: BuildGpu;
+    llamaSupportsGpuOffloading: boolean;
+}): number;

package/dist/gguf/insights/utils/resolveModelGpuLayersOption.js

@@ -0,0 +1,177 @@
+import { InsufficientMemoryError } from "../../../utils/InsufficientMemoryError.js";
+import { findBestOption } from "../../../utils/findBestOption.js";
+import { getDefaultContextBatchSize, getDefaultModelContextSize } from "../../../evaluator/LlamaContext/LlamaContext.js";
+import { minAllowedContextSizeInCalculations } from "../../../config.js";
+import { scoreLevels } from "./scoreLevels.js";
+const fitContextExtraMemoryPaddingPercentage = 0.5;
+export function resolveModelGpuLayersOption(gpuLayers, { ggufInsights, ignoreMemorySafetyChecks = false, getVramState, llamaVramPaddingSize, llamaGpu, llamaSupportsGpuOffloading }) {
+    if (gpuLayers == null)
+        gpuLayers = "auto";
+    if (!llamaSupportsGpuOffloading)
+        return 0;
+    if (gpuLayers === "max" || typeof gpuLayers === "number") {
+        const resolvedGpuLayers = typeof gpuLayers === "number"
+            ? Math.max(0, Math.min(ggufInsights.totalLayers, gpuLayers))
+            : ggufInsights.totalLayers;
+        if (ignoreMemorySafetyChecks)
+            return resolvedGpuLayers;
+        const vramState = getVramState();
+        const maxLayersRequirements = getVramRequiredForGpuLayers({
+            gpuLayers: resolvedGpuLayers,
+            ggufInsights,
+            currentVram: vramState.free
+        });
+        if (maxLayersRequirements == null)
+            throw new InsufficientMemoryError("Not enough VRAM to fit the model with the specified settings");
+        return resolvedGpuLayers;
+    }
+    else if (gpuLayers === "auto" || typeof gpuLayers === "object") {
+        if (llamaGpu === false)
+            return 0;
+        const vramState = getVramState();
+        if (vramState.total === 0)
+            return 0;
+        let freeVram = vramState.free;
+        if (typeof gpuLayers === "object" && gpuLayers.fitContext?.contextSize != null) {
+            freeVram -= llamaVramPaddingSize * fitContextExtraMemoryPaddingPercentage;
+            if (freeVram < 0)
+                freeVram = 0;
+        }
+        const bestGpuLayersOption = getBestGpuLayersForFreeVram({
+            ggufInsights,
+            freeVram,
+            fitContext: typeof gpuLayers === "object"
+                ? gpuLayers.fitContext
+                : undefined,
+            minGpuLayers: typeof gpuLayers === "object"
+                ? gpuLayers.min
+                : undefined,
+            maxGpuLayers: typeof gpuLayers === "object"
+                ? gpuLayers.max
+                : undefined
+        });
+        const hasGpuLayersRequirements = typeof gpuLayers === "object" &&
+            (gpuLayers.min != null || gpuLayers.max != null || gpuLayers.fitContext?.contextSize != null);
+        if (!ignoreMemorySafetyChecks && bestGpuLayersOption == null && hasGpuLayersRequirements)
+            throw new InsufficientMemoryError("Not enough VRAM to fit the model with the specified settings");
+        return bestGpuLayersOption ?? 0;
+    }
+    throw new Error(`Invalid gpuLayers value: ${gpuLayers}`);
+}
+function getBestGpuLayersForFreeVram({ ggufInsights, freeVram, fitContext, minGpuLayers, maxGpuLayers }) {
+    return findBestOption({
+        *generator() {
+            const minLayers = Math.floor(Math.max(0, minGpuLayers ?? 0));
+            const maxLayers = Math.floor(Math.min(ggufInsights.totalLayers, maxGpuLayers ?? ggufInsights.totalLayers));
+            for (let layers = maxLayers; layers >= minLayers; layers--) {
+                yield {
+                    gpuLayers: layers
+                };
+            }
+        },
+        score(option) {
+            const layersRequirements = getVramRequiredForGpuLayers({
+                gpuLayers: option.gpuLayers,
+                ggufInsights,
+                currentVram: freeVram,
+                fitContext
+            });
+            if (layersRequirements == null)
+                return null;
+            return scoreGpuLayersAndContextCombination({ gpuLayers: option.gpuLayers, contextSize: layersRequirements.contextSize }, {
+                totalGpuLayers: ggufInsights.totalLayers,
+                trainContextSize: getDefaultModelContextSize({ trainContextSize: ggufInsights.trainContextSize })
+            });
+        }
+    })?.gpuLayers ?? null;
+}
+function scoreGpuLayersAndContextCombination({ gpuLayers, contextSize }, { totalGpuLayers, trainContextSize }) {
+    function scoreGpuLayers() {
+        return scoreLevels(gpuLayers, [{
+                start: 0,
+                points: 4
+            }, {
+                start: 1,
+                points: 26
+            }, {
+                start: totalGpuLayers,
+                points: 14,
+                end: totalGpuLayers
+            }]);
+    }
+    function scoreContextSize() {
+        const gpuLayersPercentage = gpuLayers / totalGpuLayers;
+        return scoreLevels(contextSize, [{
+                start: 0,
+                points: 2
+            }, {
+                start: 1024,
+                points: 4
+            }, {
+                start: 2048,
+                points: gpuLayersPercentage < 0.1 ? 1 : 8
+            }, {
+                start: 4096,
+                points: gpuLayersPercentage < 0.3 ? 4 : 16
+            }, {
+                start: 8192,
+                points: gpuLayersPercentage < 0.6 ? 1 : 8,
+                end: Math.max(trainContextSize, 16384)
+            }]);
+    }
+    return scoreGpuLayers() + scoreContextSize();
+}
+function getVramRequiredForGpuLayers({ gpuLayers, ggufInsights, currentVram, fitContext }) {
+    const modelVram = ggufInsights.estimateModelResourceRequirements({ gpuLayers }).gpuVram;
+    if (modelVram > currentVram)
+        return null;
+    if (fitContext != null && fitContext.contextSize != null) {
+        const contextVram = ggufInsights.estimateContextResourceRequirements({
+            contextSize: fitContext.contextSize,
+            batchSize: getDefaultContextBatchSize({ contextSize: fitContext.contextSize, sequences: 1 }),
+            modelGpuLayers: gpuLayers,
+            sequences: 1,
+            isEmbeddingContext: fitContext.embeddingContext ?? false
+        }).gpuVram;
+        const totalVram = modelVram + contextVram;
+        if (totalVram > currentVram)
+            return null;
+        return {
+            contextSize: fitContext.contextSize,
+            contextVram,
+            totalVram
+        };
+    }
+    const maxContext = findMaxPossibleContextSizeForVram({
+        gpuLayers,
+        ggufInsights,
+        vram: currentVram - modelVram,
+        isEmbeddingContext: fitContext?.embeddingContext ?? false
+    });
+    if (maxContext == null || modelVram + maxContext.vram > currentVram)
+        return null;
+    return {
+        contextSize: maxContext.contextSize,
+        contextVram: maxContext.vram,
+        totalVram: modelVram + maxContext.vram
+    };
+}
+function findMaxPossibleContextSizeForVram({ gpuLayers, ggufInsights, vram, isEmbeddingContext }) {
+    const maxContextSize = getDefaultModelContextSize({ trainContextSize: ggufInsights.trainContextSize });
+    for (let contextSize = maxContextSize; contextSize >= minAllowedContextSizeInCalculations; contextSize--) {
+        const contextVram = ggufInsights.estimateContextResourceRequirements({
+            contextSize,
+            batchSize: getDefaultContextBatchSize({ contextSize, sequences: 1 }),
+            modelGpuLayers: gpuLayers,
+            sequences: 1,
+            isEmbeddingContext
+        }).gpuVram;
+        if (contextVram <= vram)
+            return {
+                contextSize,
+                vram: contextVram
+            };
+    }
+    return null;
+}
+//# sourceMappingURL=resolveModelGpuLayersOption.js.map

package/dist/gguf/insights/utils/resolveModelGpuLayersOption.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"resolveModelGpuLayersOption.js","sourceRoot":"","sources":["../../../../src/gguf/insights/utils/resolveModelGpuLayersOption.ts"],"names":[],"mappings":"AAEA,OAAO,EAAC,uBAAuB,EAAC,MAAM,2CAA2C,CAAC;AAClF,OAAO,EAAC,cAAc,EAAC,MAAM,kCAAkC,CAAC;AAChE,OAAO,EAAC,0BAA0B,EAAE,0BAA0B,EAAC,MAAM,iDAAiD,CAAC;AACvH,OAAO,EAAC,mCAAmC,EAAC,MAAM,oBAAoB,CAAC;AACvE,OAAO,EAAC,WAAW,EAAC,MAAM,kBAAkB,CAAC;AAG7C,MAAM,sCAAsC,GAAG,GAAG,CAAC;AAEnD,MAAM,UAAU,2BAA2B,CAAC,SAAyC,EAAE,EACnF,YAAY,EAAE,wBAAwB,GAAG,KAAK,EAAE,YAAY,EAAE,oBAAoB,EAClF,QAAQ,EAAE,0BAA0B,EAIvC;IACG,IAAI,SAAS,IAAI,IAAI;QACjB,SAAS,GAAG,MAAM,CAAC;IAEvB,IAAI,CAAC,0BAA0B;QAC3B,OAAO,CAAC,CAAC;IAEb,IAAI,SAAS,KAAK,KAAK,IAAI,OAAO,SAAS,KAAK,QAAQ,EAAE;QACtD,MAAM,iBAAiB,GAAG,OAAO,SAAS,KAAK,QAAQ;YACnD,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,YAAY,CAAC,WAAW,EAAE,SAAS,CAAC,CAAC;YAC5D,CAAC,CAAC,YAAY,CAAC,WAAW,CAAC;QAE/B,IAAI,wBAAwB;YACxB,OAAO,iBAAiB,CAAC;QAE7B,MAAM,SAAS,GAAG,YAAY,EAAE,CAAC;QACjC,MAAM,qBAAqB,GAAG,2BAA2B,CAAC;YACtD,SAAS,EAAE,iBAAiB;YAC5B,YAAY;YACZ,WAAW,EAAE,SAAS,CAAC,IAAI;SAC9B,CAAC,CAAC;QAEH,IAAI,qBAAqB,IAAI,IAAI;YAC7B,MAAM,IAAI,uBAAuB,CAAC,8DAA8D,CAAC,CAAC;QAEtG,OAAO,iBAAiB,CAAC;KAC5B;SAAM,IAAI,SAAS,KAAK,MAAM,IAAI,OAAO,SAAS,KAAK,QAAQ,EAAE;QAC9D,IAAI,QAAQ,KAAK,KAAK;YAClB,OAAO,CAAC,CAAC;QAEb,MAAM,SAAS,GAAG,YAAY,EAAE,CAAC;QACjC,IAAI,SAAS,CAAC,KAAK,KAAK,CAAC;YACrB,OAAO,CAAC,CAAC;QAEb,IAAI,QAAQ,GAAG,SAAS,CAAC,IAAI,CAAC;QAC9B,IAAI,OAAO,SAAS,KAAK,QAAQ,IAAI,SAAS,CAAC,UAAU,EAAE,WAAW,IAAI,IAAI,EAAE;YAC5E,QAAQ,IAAI,oBAAoB,GAAG,sCAAsC,CAAC;YAE1E,IAAI,QAAQ,GAAG,CAAC;gBACZ,QAAQ,GAAG,CAAC,CAAC;SACpB;QAED,MAAM,mBAAmB,GAAG,2BAA2B,CAAC;YACpD,YAAY;YACZ,QAAQ;YACR,UAAU,EAAE,OAAO,SAAS,KAAK,QAAQ;gBACrC,CAAC,CAAC,SAAS,CAAC,UAAU;gBACtB,CAAC,CAAC,SAAS;YACf,YAAY,EAAE,OAAO,SAAS,KAAK,QAAQ;gBACvC,CAAC,CAAC,SAAS,CAAC,GAAG;gBACf,CAAC,CAAC,SAAS;YACf,YAAY,EAAE,OAAO,SAAS,KAAK,QAAQ;gBACvC,CAAC,CAAC,SAAS,CAAC,GAAG;gBACf,CAAC,CAAC,SAAS;SAClB,CAAC,CAAC;QAEH,MAAM,wBAAwB,GAAG,OAAO,SAAS,KAAK,QAAQ;YAC1D,CAAC,SAAS,CAAC,GAAG,IAAI,IAAI,IAAI,SAAS,CAAC,GAAG,IAAI,IAAI,IAAI,SAAS,CAAC,UAAU,EAAE,WAAW,IAAI,IAAI,CAAC,CAAC;QAElG,IAAI,CAAC,wBAAwB,IAAI,mBAAmB,IAAI,IAAI,IAAI,wBAAwB;YACpF,MAAM,IAAI,uBAAuB,CAAC,8DAA8D,CAAC,CAAC;QAEtG,OAAO,mBAAmB,IAAI,CAAC,CAAC;KACnC;IAED,MAAM,IAAI,KAAK,CAAC,4BAA4B,SAAS,EAAE,CAAC,CAAC;AAC7D,CAAC;AAED,SAAS,2BAA2B,CAAC,EACjC,YAAY,EACZ,QAAQ,EACR,UAAU,EACV,YAAY,EACZ,YAAY,EAOf;IACG,OAAO,cAAc,CAAC;QAClB,CAAC,SAAS;YACN,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,YAAY,IAAI,CAAC,CAAC,CAAC,CAAC;YAC7D,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,YAAY,CAAC,WAAW,EAAE,YAAY,IAAI,YAAY,CAAC,WAAW,CAAC,CAAC,CAAC;YAE3G,KAAK,IAAI,MAAM,GAAG,SAAS,EAAE,MAAM,IAAI,SAAS,EAAE,MAAM,EAAE,EAAE;gBACxD,MAAM;oBACF,SAAS,EAAE,MAAM;iBACpB,CAAC;aACL;QACL,CAAC;QACD,KAAK,CAAC,MAAM;YACR,MAAM,kBAAkB,GAAG,2BAA2B,CAAC;gBACnD,SAAS,EAAE,MAAM,CAAC,SAAS;gBAC3B,YAAY;gBACZ,WAAW,EAAE,QAAQ;gBACrB,UAAU;aACb,CAAC,CAAC;YAEH,IAAI,kBAAkB,IAAI,IAAI;gBAC1B,OAAO,IAAI,CAAC;YAEhB,OAAO,mCAAmC,CAAC,EAAC,SAAS,EAAE,MAAM,CAAC,SAAS,EAAE,WAAW,EAAE,kBAAkB,CAAC,WAAW,EAAC,EAAE;gBACnH,cAAc,EAAE,YAAY,CAAC,WAAW;gBACxC,gBAAgB,EAAE,0BAA0B,CAAC,EAAC,gBAAgB,EAAE,YAAY,CAAC,gBAAgB,EAAC,CAAC;aAClG,CAAC,CAAC;QACP,CAAC;KACJ,CAAC,EAAE,SAAS,IAAI,IAAI,CAAC;AAC1B,CAAC;AAED,SAAS,mCAAmC,CAAC,EAAC,SAAS,EAAE,WAAW,EAA2C,EAAE,EAC7G,cAAc,EAAE,gBAAgB,EAGnC;IACG,SAAS,cAAc;QACnB,OAAO,WAAW,CAAC,SAAS,EAAE,CAAC;gBAC3B,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;aACZ,EAAE;gBACC,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,EAAE;aACb,EAAE;gBACC,KAAK,EAAE,cAAc;gBACrB,MAAM,EAAE,EAAE;gBACV,GAAG,EAAE,cAAc;aACtB,CAAC,CAAC,CAAC;IACR,CAAC;IAED,SAAS,gBAAgB;QACrB,MAAM,mBAAmB,GAAG,SAAS,GAAG,cAAc,CAAC;QAEvD,OAAO,WAAW,CAAC,WAAW,EAAE,CAAC;gBAC7B,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;aACZ,EAAE;gBACC,KAAK,EAAE,IAAI;gBACX,MAAM,EAAE,CAAC;aACZ,EAAE;gBACC,KAAK,EAAE,IAAI;gBACX,MAAM,EAAE,mBAAmB,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;aAC5C,EAAE;gBACC,KAAK,EAAE,IAAI;gBACX,MAAM,EAAE,mBAAmB,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE;aAC7C,EAAE;gBACC,KAAK,EAAE,IAAI;gBACX,MAAM,EAAE,mBAAmB,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;gBACzC,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,gBAAgB,EAAE,KAAK,CAAC;aACzC,CAAC,CAAC,CAAC;IACR,CAAC;IAED,OAAO,cAAc,EAAE,GAAG,gBAAgB,EAAE,CAAC;AACjD,CAAC;AAED,SAAS,2BAA2B,CAAC,EACjC,SAAS,EAAE,YAAY,EAAE,WAAW,EAAE,UAAU,EAGnD;IACG,MAAM,SAAS,GAAG,YAAY,CAAC,iCAAiC,CAAC,EAAC,SAAS,EAAC,CAAC,CAAC,OAAO,CAAC;IAEtF,IAAI,SAAS,GAAG,WAAW;QACvB,OAAO,IAAI,CAAC;IAEhB,IAAI,UAAU,IAAI,IAAI,IAAI,UAAU,CAAC,WAAW,IAAI,IAAI,EAAE;QACtD,MAAM,WAAW,GAAG,YAAY,CAAC,mCAAmC,CAAC;YACjE,WAAW,EAAE,UAAU,CAAC,WAAW;YACnC,SAAS,EAAE,0BAA0B,CAAC,EAAC,WAAW,EAAE,UAAU,CAAC,WAAW,EAAE,SAAS,EAAE,CAAC,EAAC,CAAC;YAC1F,cAAc,EAAE,SAAS;YACzB,SAAS,EAAE,CAAC;YACZ,kBAAkB,EAAE,UAAU,CAAC,gBAAgB,IAAI,KAAK;SAC3D,CAAC,CAAC,OAAO,CAAC;QAEX,MAAM,SAAS,GAAG,SAAS,GAAG,WAAW,CAAC;QAC1C,IAAI,SAAS,GAAG,WAAW;YACvB,OAAO,IAAI,CAAC;QAEhB,OAAO;YACH,WAAW,EAAE,UAAU,CAAC,WAAW;YACnC,WAAW;YACX,SAAS;SACZ,CAAC;KACL;IAED,MAAM,UAAU,GAAG,iCAAiC,CAAC;QACjD,SAAS;QACT,YAAY;QACZ,IAAI,EAAE,WAAW,GAAG,SAAS;QAC7B,kBAAkB,EAAE,UAAU,EAAE,gBAAgB,IAAI,KAAK;KAC5D,CAAC,CAAC;IAEH,IAAI,UAAU,IAAI,IAAI,IAAI,SAAS,GAAG,UAAU,CAAC,IAAI,GAAG,WAAW;QAC/D,OAAO,IAAI,CAAC;IAEhB,OAAO;QACH,WAAW,EAAE,UAAU,CAAC,WAAW;QACnC,WAAW,EAAE,UAAU,CAAC,IAAI;QAC5B,SAAS,EAAE,SAAS,GAAG,UAAU,CAAC,IAAI;KACzC,CAAC;AACN,CAAC;AAED,SAAS,iCAAiC,CAAC,EAAC,SAAS,EAAE,YAAY,EAAE,IAAI,EAAE,kBAAkB,EAE5F;IACG,MAAM,cAAc,GAAG,0BAA0B,CAAC,EAAC,gBAAgB,EAAE,YAAY,CAAC,gBAAgB,EAAC,CAAC,CAAC;IAErG,KAAK,IAAI,WAAW,GAAG,cAAc,EAAE,WAAW,IAAI,mCAAmC,EAAE,WAAW,EAAE,EAAE;QACtG,MAAM,WAAW,GAAG,YAAY,CAAC,mCAAmC,CAAC;YACjE,WAAW;YACX,SAAS,EAAE,0BAA0B,CAAC,EAAC,WAAW,EAAE,SAAS,EAAE,CAAC,EAAC,CAAC;YAClE,cAAc,EAAE,SAAS;YACzB,SAAS,EAAE,CAAC;YACZ,kBAAkB;SACrB,CAAC,CAAC,OAAO,CAAC;QAEX,IAAI,WAAW,IAAI,IAAI;YACnB,OAAO;gBACH,WAAW;gBACX,IAAI,EAAE,WAAW;aACpB,CAAC;KACT;IAED,OAAO,IAAI,CAAC;AAChB,CAAC"}

package/dist/gguf/insights/utils/scoreLevels.d.ts

@@ -0,0 +1,5 @@
+export declare function scoreLevels(num: number, levels: {
+    start: number;
+    end?: number;
+    points: number;
+}[]): number;

package/dist/gguf/insights/utils/scoreLevels.js

@@ -0,0 +1,16 @@
+export function scoreLevels(num, levels) {
+    let res = 0;
+    for (let i = 0; i < levels.length; i++) {
+        const level = levels[i];
+        const start = level.start;
+        const end = level.end ?? levels[i + 1]?.start ?? Math.max(start, num);
+        if (num < start)
+            break;
+        else if (num >= end)
+            res += level.points;
+        else
+            res += level.points * ((num - start) / (end - start));
+    }
+    return res;
+}
+//# sourceMappingURL=scoreLevels.js.map

package/dist/gguf/insights/utils/scoreLevels.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"scoreLevels.js","sourceRoot":"","sources":["../../../../src/gguf/insights/utils/scoreLevels.ts"],"names":[],"mappings":"AAAA,MAAM,UAAU,WAAW,CAAC,GAAW,EAAE,MAAyD;IAC9F,IAAI,GAAG,GAAG,CAAC,CAAC;IAEZ,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE;QACpC,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;QACxB,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC;QAC1B,MAAM,GAAG,GAAG,KAAK,CAAC,GAAG,IAAI,MAAM,CAAC,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,IAAI,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;QAEtE,IAAI,GAAG,GAAG,KAAK;YACX,MAAM;aACL,IAAI,GAAG,IAAI,GAAG;YACf,GAAG,IAAI,KAAK,CAAC,MAAM,CAAC;;YAEpB,GAAG,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,GAAG,GAAG,KAAK,CAAC,GAAG,CAAC,GAAG,GAAG,KAAK,CAAC,CAAC,CAAC;KAC7D;IAED,OAAO,GAAG,CAAC;AACf,CAAC"}

package/dist/gguf/readGgufFileInfo.js

@@ -1,3 +1,4 @@
+import { isUrl } from "../utils/isUrl.js";
 import { parseGguf } from "./parser/parseGguf.js";
 import { GgufNetworkFetchFileReader } from "./fileReaders/GgufNetworkFetchFileReader.js";
 import { GgufFsFileReader } from "./fileReaders/GgufFsFileReader.js";
@@ -9,7 +10,7 @@ import { normalizeGgufDownloadUrl } from "./utils/normalizeGgufDownloadUrl.js";
  */
 export async function readGgufFileInfo(pathOrUrl, { readTensorInfo = true, sourceType, ignoreKeys = [], logWarnings = true, fetchRetryOptions = ggufDefaultFetchRetryOptions, fetchHeaders = {}, signal } = {}) {
     function createFileReader() {
-        if (sourceType === "network" || (sourceType == null && (pathOrUrl.startsWith("http://") || pathOrUrl.startsWith("https://")))) {
+        if (sourceType === "network" || (sourceType == null && isUrl(pathOrUrl))) {
             return new GgufNetworkFetchFileReader({
                 url: normalizeGgufDownloadUrl(pathOrUrl),
                 retryOptions: fetchRetryOptions,

package/dist/gguf/readGgufFileInfo.js.map

@@ -1 +1 @@
-{"version":3,"file":"readGgufFileInfo.js","sourceRoot":"","sources":["../../src/gguf/readGgufFileInfo.ts"],"names":[],"mappings":"AACA,OAAO,EAAC,SAAS,EAAC,MAAM,uBAAuB,CAAC;AAChD,OAAO,EAAC,0BAA0B,EAAC,MAAM,6CAA6C,CAAC;AACvF,OAAO,EAAC,gBAAgB,EAAC,MAAM,mCAAmC,CAAC;AACnE,OAAO,EAAC,4BAA4B,EAAC,MAAM,aAAa,CAAC;AACzD,OAAO,EAAC,wBAAwB,EAAC,MAAM,qCAAqC,CAAC;AAG7E;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,SAAiB,EAAE,EACtD,cAAc,GAAG,IAAI,EACrB,UAAU,EACV,UAAU,GAAG,EAAE,EACf,WAAW,GAAG,IAAI,EAClB,iBAAiB,GAAG,4BAA4B,EAChD,YAAY,GAAG,EAAE,EACjB,MAAM,KA8BN,EAAE;IACF,SAAS,gBAAgB;QACrB,IAAI,UAAU,KAAK,SAAS,IAAI,CAAC,UAAU,IAAI,IAAI,IAAI,CAAC,SAAS,CAAC,UAAU,CAAC,SAAS,CAAC,IAAI,SAAS,CAAC,UAAU,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE;YAC3H,OAAO,IAAI,0BAA0B,CAAC;gBAClC,GAAG,EAAE,wBAAwB,CAAC,SAAS,CAAC;gBACxC,YAAY,EAAE,iBAAiB;gBAC/B,OAAO,EAAE,YAAY;gBACrB,MAAM;aACT,CAAC,CAAC;SACN;aAAM,IAAI,UAAU,KAAK,YAAY,IAAI,UAAU,IAAI,IAAI,EAAE;YAC1D,OAAO,IAAI,gBAAgB,CAAC;gBACxB,QAAQ,EAAE,SAAS;gBACnB,MAAM;aACT,CAAC,CAAC;SACN;QAED,KAAK,CAAC,UAA0B,CAAC,CAAC;QAClC,MAAM,IAAI,KAAK,CAAC,2BAA2B,UAAU,EAAE,CAAC,CAAC;IAC7D,CAAC;IAED,MAAM,UAAU,GAAG,gBAAgB,EAAE,CAAC;IACtC,OAAO,MAAM,SAAS,CAAC;QACnB,UAAU;QACV,UAAU;QACV,cAAc;QACd,WAAW;KACd,CAAC,CAAC;AACP,CAAC"}
+{"version":3,"file":"readGgufFileInfo.js","sourceRoot":"","sources":["../../src/gguf/readGgufFileInfo.ts"],"names":[],"mappings":"AACA,OAAO,EAAC,KAAK,EAAC,MAAM,mBAAmB,CAAC;AACxC,OAAO,EAAC,SAAS,EAAC,MAAM,uBAAuB,CAAC;AAChD,OAAO,EAAC,0BAA0B,EAAC,MAAM,6CAA6C,CAAC;AACvF,OAAO,EAAC,gBAAgB,EAAC,MAAM,mCAAmC,CAAC;AACnE,OAAO,EAAC,4BAA4B,EAAC,MAAM,aAAa,CAAC;AACzD,OAAO,EAAC,wBAAwB,EAAC,MAAM,qCAAqC,CAAC;AAG7E;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,SAAiB,EAAE,EACtD,cAAc,GAAG,IAAI,EACrB,UAAU,EACV,UAAU,GAAG,EAAE,EACf,WAAW,GAAG,IAAI,EAClB,iBAAiB,GAAG,4BAA4B,EAChD,YAAY,GAAG,EAAE,EACjB,MAAM,KA8BN,EAAE;IACF,SAAS,gBAAgB;QACrB,IAAI,UAAU,KAAK,SAAS,IAAI,CAAC,UAAU,IAAI,IAAI,IAAI,KAAK,CAAC,SAAS,CAAC,CAAC,EAAE;YACtE,OAAO,IAAI,0BAA0B,CAAC;gBAClC,GAAG,EAAE,wBAAwB,CAAC,SAAS,CAAC;gBACxC,YAAY,EAAE,iBAAiB;gBAC/B,OAAO,EAAE,YAAY;gBACrB,MAAM;aACT,CAAC,CAAC;SACN;aAAM,IAAI,UAAU,KAAK,YAAY,IAAI,UAAU,IAAI,IAAI,EAAE;YAC1D,OAAO,IAAI,gBAAgB,CAAC;gBACxB,QAAQ,EAAE,SAAS;gBACnB,MAAM;aACT,CAAC,CAAC;SACN;QAED,KAAK,CAAC,UAA0B,CAAC,CAAC;QAClC,MAAM,IAAI,KAAK,CAAC,2BAA2B,UAAU,EAAE,CAAC,CAAC;IAC7D,CAAC;IAED,MAAM,UAAU,GAAG,gBAAgB,EAAE,CAAC;IACtC,OAAO,MAAM,SAAS,CAAC;QACnB,UAAU;QACV,UAAU;QACV,cAAc;QACd,WAAW;KACd,CAAC,CAAC;AACP,CAAC"}

package/dist/index.d.ts

@@ -11,6 +11,7 @@ import { LlamaGrammarEvaluationState, LlamaGrammarEvaluationStateOptions } from
 import { LlamaContext, LlamaContextSequence } from "./evaluator/LlamaContext/LlamaContext.js";
 import { LlamaEmbeddingContext, type LlamaEmbeddingContextOptions, LlamaEmbedding, type LlamaEmbeddingJSON } from "./evaluator/LlamaEmbeddingContext.js";
 import { type LlamaContextOptions, type BatchingOptions, type LlamaContextSequenceRepeatPenalty, type CustomBatchingDispatchSchedule, type CustomBatchingPrioritizationStrategy, type BatchItem, type PrioritizedBatchItem, type ContextShiftOptions, type ContextTokensDeleteRange, type EvaluationPriority } from "./evaluator/LlamaContext/types.js";
+import { TokenBias } from "./evaluator/TokenBias.js";
 import { LlamaChatSession, type LlamaChatSessionOptions, type LlamaChatSessionContextShiftOptions, type LLamaChatPromptOptions, type LlamaChatSessionRepeatPenalty } from "./evaluator/LlamaChatSession/LlamaChatSession.js";
 import { defineChatSessionFunction } from "./evaluator/LlamaChatSession/utils/defineChatSessionFunction.js";
 import { LlamaChat, type LlamaChatOptions, type LLamaChatGenerateResponseOptions, type LLamaChatContextShiftOptions, type LlamaChatResponse, type LlamaChatResponseFunctionCall } from "./evaluator/LlamaChat/LlamaChat.js";
@@ -39,4 +40,4 @@ import { type GbnfJsonArraySchema, type GbnfJsonBasicSchema, type GbnfJsonConstS
 import { type GgufFileInfo } from "./gguf/types/GgufFileInfoTypes.js";
 import { type GgufMetadata, type GgufMetadataLlmToType, GgufArchitectureType, GgufFileType, GgufMetadataTokenizerTokenType, GgufMetadataArchitecturePoolingType, type GgufMetadataGeneral, type GgufMetadataTokenizer, type GgufMetadataDefaultArchitectureType, type GgufMetadataLlmLLaMA, type GgufMetadataMPT, type GgufMetadataGPTNeoX, type GgufMetadataGPTJ, type GgufMetadataGPT2, type GgufMetadataBloom, type GgufMetadataFalcon, type GgufMetadataMamba, type GgufMetadataRWKV, isGgufMetadataOfArchitectureType } from "./gguf/types/GgufMetadataTypes.js";
 import { GgmlType, type GgufTensorInfo } from "./gguf/types/GgufTensorInfoTypes.js";
-export { Llama, getLlama, type LlamaOptions, LlamaLogLevel, NoBinaryFoundError, LlamaModel, LlamaModelTokens, LlamaModelInfillTokens, type LlamaModelOptions, LlamaGrammar, type LlamaGrammarOptions, LlamaJsonSchemaGrammar, LlamaJsonSchemaValidationError, LlamaGrammarEvaluationState, type LlamaGrammarEvaluationStateOptions, LlamaContext, LlamaContextSequence, type LlamaContextOptions, type BatchingOptions, type CustomBatchingDispatchSchedule, type CustomBatchingPrioritizationStrategy, type BatchItem, type PrioritizedBatchItem, type ContextShiftOptions, type ContextTokensDeleteRange, type EvaluationPriority, type LlamaContextSequenceRepeatPenalty, LlamaEmbeddingContext, type LlamaEmbeddingContextOptions, LlamaEmbedding, type LlamaEmbeddingJSON, LlamaChatSession, defineChatSessionFunction, type LlamaChatSessionOptions, type LlamaChatSessionContextShiftOptions, type LLamaChatPromptOptions, type LlamaChatSessionRepeatPenalty, LlamaChat, type LlamaChatOptions, type LLamaChatGenerateResponseOptions, type LLamaChatContextShiftOptions, type LLamaContextualRepeatPenalty, type LlamaChatResponse, type LlamaChatResponseFunctionCall, LlamaCompletion, type LlamaCompletionOptions, type LlamaCompletionGenerationOptions, type LlamaInfillGenerationOptions, TokenMeter, type TokenMeterState, UnsupportedError, InsufficientMemoryError, DisposedError, ChatWrapper, type ChatWrapperSettings, EmptyChatWrapper, LlamaChatWrapper, GeneralChatWrapper, ChatMLChatWrapper, FalconChatWrapper, AlpacaChatWrapper, FunctionaryChatWrapper, GemmaChatWrapper, TemplateChatWrapper, type TemplateChatWrapperOptions, JinjaTemplateChatWrapper, type JinjaTemplateChatWrapperOptions, resolveChatWrapper, type ResolveChatWrapperOptions, resolvableChatWrapperTypeNames, type ResolvableChatWrapperTypeName, specializedChatWrapperTypeNames, type SpecializedChatWrapperTypeName, templateChatWrapperTypeNames, type TemplateChatWrapperTypeName, LlamaText, SpecialTokensText, SpecialToken, isLlamaText, tokenizeText, type LlamaTextJSON, type LlamaTextJSONValue, type LlamaTextSpecialTokensTextJSON, type LlamaTextSpecialTokenJSON, appendUserMessageToChatHistory, getModuleVersion, type ChatHistoryItem, type ChatModelFunctionCall, type ChatModelFunctions, type ChatModelResponse, type ChatSessionModelFunction, type ChatSessionModelFunctions, type ChatSystemMessage, type ChatUserMessage, type Token, isChatModelResponseFunctionCall, type GbnfJsonSchema, type GbnfJsonSchemaToType, type GbnfJsonSchemaImmutableType, type GbnfJsonBasicSchema, type GbnfJsonConstSchema, type GbnfJsonEnumSchema, type GbnfJsonOneOfSchema, type GbnfJsonObjectSchema, type GbnfJsonArraySchema, LlamaVocabularyType, LlamaLogLevelGreaterThan, LlamaLogLevelGreaterThanOrEqual, readGgufFileInfo, type GgufFileInfo, type GgufMetadata, type GgufTensorInfo, type GgufMetadataLlmToType, GgufArchitectureType, GgufFileType, GgufMetadataTokenizerTokenType, GgufMetadataArchitecturePoolingType, type GgufMetadataGeneral, type GgufMetadataTokenizer, type GgufMetadataDefaultArchitectureType, type GgufMetadataLlmLLaMA, type GgufMetadataMPT, type GgufMetadataGPTNeoX, type GgufMetadataGPTJ, type GgufMetadataGPT2, type GgufMetadataBloom, type GgufMetadataFalcon, type GgufMetadataMamba, type GgufMetadataRWKV, GgmlType, isGgufMetadataOfArchitectureType };
+export { Llama, getLlama, type LlamaOptions, LlamaLogLevel, NoBinaryFoundError, LlamaModel, LlamaModelTokens, LlamaModelInfillTokens, type LlamaModelOptions, LlamaGrammar, type LlamaGrammarOptions, LlamaJsonSchemaGrammar, LlamaJsonSchemaValidationError, LlamaGrammarEvaluationState, type LlamaGrammarEvaluationStateOptions, LlamaContext, LlamaContextSequence, type LlamaContextOptions, type BatchingOptions, type CustomBatchingDispatchSchedule, type CustomBatchingPrioritizationStrategy, type BatchItem, type PrioritizedBatchItem, type ContextShiftOptions, type ContextTokensDeleteRange, type EvaluationPriority, type LlamaContextSequenceRepeatPenalty, TokenBias, LlamaEmbeddingContext, type LlamaEmbeddingContextOptions, LlamaEmbedding, type LlamaEmbeddingJSON, LlamaChatSession, defineChatSessionFunction, type LlamaChatSessionOptions, type LlamaChatSessionContextShiftOptions, type LLamaChatPromptOptions, type LlamaChatSessionRepeatPenalty, LlamaChat, type LlamaChatOptions, type LLamaChatGenerateResponseOptions, type LLamaChatContextShiftOptions, type LLamaContextualRepeatPenalty, type LlamaChatResponse, type LlamaChatResponseFunctionCall, LlamaCompletion, type LlamaCompletionOptions, type LlamaCompletionGenerationOptions, type LlamaInfillGenerationOptions, TokenMeter, type TokenMeterState, UnsupportedError, InsufficientMemoryError, DisposedError, ChatWrapper, type ChatWrapperSettings, EmptyChatWrapper, LlamaChatWrapper, GeneralChatWrapper, ChatMLChatWrapper, FalconChatWrapper, AlpacaChatWrapper, FunctionaryChatWrapper, GemmaChatWrapper, TemplateChatWrapper, type TemplateChatWrapperOptions, JinjaTemplateChatWrapper, type JinjaTemplateChatWrapperOptions, resolveChatWrapper, type ResolveChatWrapperOptions, resolvableChatWrapperTypeNames, type ResolvableChatWrapperTypeName, specializedChatWrapperTypeNames, type SpecializedChatWrapperTypeName, templateChatWrapperTypeNames, type TemplateChatWrapperTypeName, LlamaText, SpecialTokensText, SpecialToken, isLlamaText, tokenizeText, type LlamaTextJSON, type LlamaTextJSONValue, type LlamaTextSpecialTokensTextJSON, type LlamaTextSpecialTokenJSON, appendUserMessageToChatHistory, getModuleVersion, type ChatHistoryItem, type ChatModelFunctionCall, type ChatModelFunctions, type ChatModelResponse, type ChatSessionModelFunction, type ChatSessionModelFunctions, type ChatSystemMessage, type ChatUserMessage, type Token, isChatModelResponseFunctionCall, type GbnfJsonSchema, type GbnfJsonSchemaToType, type GbnfJsonSchemaImmutableType, type GbnfJsonBasicSchema, type GbnfJsonConstSchema, type GbnfJsonEnumSchema, type GbnfJsonOneOfSchema, type GbnfJsonObjectSchema, type GbnfJsonArraySchema, LlamaVocabularyType, LlamaLogLevelGreaterThan, LlamaLogLevelGreaterThanOrEqual, readGgufFileInfo, type GgufFileInfo, type GgufMetadata, type GgufTensorInfo, type GgufMetadataLlmToType, GgufArchitectureType, GgufFileType, GgufMetadataTokenizerTokenType, GgufMetadataArchitecturePoolingType, type GgufMetadataGeneral, type GgufMetadataTokenizer, type GgufMetadataDefaultArchitectureType, type GgufMetadataLlmLLaMA, type GgufMetadataMPT, type GgufMetadataGPTNeoX, type GgufMetadataGPTJ, type GgufMetadataGPT2, type GgufMetadataBloom, type GgufMetadataFalcon, type GgufMetadataMamba, type GgufMetadataRWKV, GgmlType, isGgufMetadataOfArchitectureType };

package/dist/index.js

@@ -10,6 +10,7 @@ import { LlamaJsonSchemaValidationError } from "./utils/gbnfJson/utils/validateO
 import { LlamaGrammarEvaluationState } from "./evaluator/LlamaGrammarEvaluationState.js";
 import { LlamaContext, LlamaContextSequence } from "./evaluator/LlamaContext/LlamaContext.js";
 import { LlamaEmbeddingContext, LlamaEmbedding } from "./evaluator/LlamaEmbeddingContext.js";
+import { TokenBias } from "./evaluator/TokenBias.js";
 import { LlamaChatSession } from "./evaluator/LlamaChatSession/LlamaChatSession.js";
 import { defineChatSessionFunction } from "./evaluator/LlamaChatSession/utils/defineChatSessionFunction.js";
 import { LlamaChat } from "./evaluator/LlamaChat/LlamaChat.js";
@@ -36,5 +37,5 @@ import { readGgufFileInfo } from "./gguf/readGgufFileInfo.js";
 import { isChatModelResponseFunctionCall } from "./types.js";
 import { GgufArchitectureType, GgufFileType, GgufMetadataTokenizerTokenType, GgufMetadataArchitecturePoolingType, isGgufMetadataOfArchitectureType } from "./gguf/types/GgufMetadataTypes.js";
 import { GgmlType } from "./gguf/types/GgufTensorInfoTypes.js";
-export { Llama, getLlama, LlamaLogLevel, NoBinaryFoundError, LlamaModel, LlamaModelTokens, LlamaModelInfillTokens, LlamaGrammar, LlamaJsonSchemaGrammar, LlamaJsonSchemaValidationError, LlamaGrammarEvaluationState, LlamaContext, LlamaContextSequence, LlamaEmbeddingContext, LlamaEmbedding, LlamaChatSession, defineChatSessionFunction, LlamaChat, LlamaCompletion, TokenMeter, UnsupportedError, InsufficientMemoryError, DisposedError, ChatWrapper, EmptyChatWrapper, LlamaChatWrapper, GeneralChatWrapper, ChatMLChatWrapper, FalconChatWrapper, AlpacaChatWrapper, FunctionaryChatWrapper, GemmaChatWrapper, TemplateChatWrapper, JinjaTemplateChatWrapper, resolveChatWrapper, resolvableChatWrapperTypeNames, specializedChatWrapperTypeNames, templateChatWrapperTypeNames, LlamaText, SpecialTokensText, SpecialToken, isLlamaText, tokenizeText, appendUserMessageToChatHistory, getModuleVersion, isChatModelResponseFunctionCall, LlamaVocabularyType, LlamaLogLevelGreaterThan, LlamaLogLevelGreaterThanOrEqual, readGgufFileInfo, GgufArchitectureType, GgufFileType, GgufMetadataTokenizerTokenType, GgufMetadataArchitecturePoolingType, GgmlType, isGgufMetadataOfArchitectureType };
+export { Llama, getLlama, LlamaLogLevel, NoBinaryFoundError, LlamaModel, LlamaModelTokens, LlamaModelInfillTokens, LlamaGrammar, LlamaJsonSchemaGrammar, LlamaJsonSchemaValidationError, LlamaGrammarEvaluationState, LlamaContext, LlamaContextSequence, TokenBias, LlamaEmbeddingContext, LlamaEmbedding, LlamaChatSession, defineChatSessionFunction, LlamaChat, LlamaCompletion, TokenMeter, UnsupportedError, InsufficientMemoryError, DisposedError, ChatWrapper, EmptyChatWrapper, LlamaChatWrapper, GeneralChatWrapper, ChatMLChatWrapper, FalconChatWrapper, AlpacaChatWrapper, FunctionaryChatWrapper, GemmaChatWrapper, TemplateChatWrapper, JinjaTemplateChatWrapper, resolveChatWrapper, resolvableChatWrapperTypeNames, specializedChatWrapperTypeNames, templateChatWrapperTypeNames, LlamaText, SpecialTokensText, SpecialToken, isLlamaText, tokenizeText, appendUserMessageToChatHistory, getModuleVersion, isChatModelResponseFunctionCall, LlamaVocabularyType, LlamaLogLevelGreaterThan, LlamaLogLevelGreaterThanOrEqual, readGgufFileInfo, GgufArchitectureType, GgufFileType, GgufMetadataTokenizerTokenType, GgufMetadataArchitecturePoolingType, GgmlType, isGgufMetadataOfArchitectureType };
 //# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,aAAa,EAAC,MAAM,iBAAiB,CAAC;AAC9C,OAAO,EAAC,KAAK,EAAC,MAAM,qBAAqB,CAAC;AAC1C,OAAO,EAAC,QAAQ,EAAe,MAAM,wBAAwB,CAAC;AAC9D,OAAO,EAAC,kBAAkB,EAAC,MAAM,wCAAwC,CAAC;AAC1E,OAAO,EAAC,aAAa,EAAE,wBAAwB,EAAE,+BAA+B,EAAE,mBAAmB,EAAC,MAAM,qBAAqB,CAAC;AAClI,OAAO,EAAC,UAAU,EAAE,sBAAsB,EAA0B,gBAAgB,EAAC,MAAM,2BAA2B,CAAC;AACvH,OAAO,EAAC,YAAY,EAA2B,MAAM,6BAA6B,CAAC;AACnF,OAAO,EAAC,sBAAsB,EAAC,MAAM,uCAAuC,CAAC;AAC7E,OAAO,EAAC,8BAA8B,EAAC,MAAM,2DAA2D,CAAC;AACzG,OAAO,EAAC,2BAA2B,EAAqC,MAAM,4CAA4C,CAAC;AAC3H,OAAO,EAAC,YAAY,EAAE,oBAAoB,EAAC,MAAM,0CAA0C,CAAC;AAC5F,OAAO,EACH,qBAAqB,EAAqC,cAAc,EAC3E,MAAM,sCAAsC,CAAC;AAM9C,OAAO,EACH,gBAAgB,EAEnB,MAAM,kDAAkD,CAAC;AAC1D,OAAO,EAAC,yBAAyB,EAAC,MAAM,iEAAiE,CAAC;AAC1G,OAAO,EACH,SAAS,EAEZ,MAAM,oCAAoC,CAAC;AAC5C,OAAO,EACH,eAAe,EAClB,MAAM,gCAAgC,CAAC;AACxC,OAAO,EAAC,UAAU,EAAuB,MAAM,2BAA2B,CAAC;AAC3E,OAAO,EAAC,gBAAgB,EAAC,MAAM,6BAA6B,CAAC;AAC7D,OAAO,EAAC,uBAAuB,EAAC,MAAM,oCAAoC,CAAC;AAC3E,OAAO,EAAC,WAAW,EAA2B,MAAM,kBAAkB,CAAC;AACvE,OAAO,EAAC,gBAAgB,EAAC,MAAM,oCAAoC,CAAC;AACpE,OAAO,EAAC,gBAAgB,EAAC,MAAM,oCAAoC,CAAC;AACpE,OAAO,EAAC,kBAAkB,EAAC,MAAM,sCAAsC,CAAC;AACxE,OAAO,EAAC,iBAAiB,EAAC,MAAM,qCAAqC,CAAC;AACtE,OAAO,EAAC,iBAAiB,EAAC,MAAM,qCAAqC,CAAC;AACtE,OAAO,EAAC,iBAAiB,EAAC,MAAM,qCAAqC,CAAC;AACtE,OAAO,EAAC,sBAAsB,EAAC,MAAM,0CAA0C,CAAC;AAChF,OAAO,EAAC,gBAAgB,EAAC,MAAM,oCAAoC,CAAC;AACpE,OAAO,EAAC,mBAAmB,EAAkC,MAAM,+CAA+C,CAAC;AACnH,OAAO,EAAC,wBAAwB,EAAuC,MAAM,oDAAoD,CAAC;AAClI,OAAO,EACH,8BAA8B,EAAsC,+BAA+B,EAC9D,4BAA4B,EAAoC,kBAAkB,EAE1H,MAAM,4CAA4C,CAAC;AACpD,OAAO,EACH,SAAS,EAAE,iBAAiB,EAAE,YAAY,EAAE,WAAW,EAAE,YAAY,EAExE,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAAC,8BAA8B,EAAC,MAAM,2CAA2C,CAAC;AACzF,OAAO,EAAC,gBAAgB,EAAC,MAAM,6BAA6B,CAAC;AAC7D,OAAO,EAAC,gBAAgB,EAAC,MAAM,4BAA4B,CAAC;AAE5D,OAAO,EAGS,+BAA+B,EAC9C,MAAM,YAAY,CAAC;AAMpB,OAAO,EAC4C,oBAAoB,EAAE,YAAY,EAAE,8BAA8B,EACjH,mCAAmC,EAE6D,gCAAgC,EACnI,MAAM,mCAAmC,CAAC;AAC3C,OAAO,EAAC,QAAQ,EAAsB,MAAM,qCAAqC,CAAC;AAGlF,OAAO,EACH,KAAK,EACL,QAAQ,EAER,aAAa,EACb,kBAAkB,EAClB,UAAU,EACV,gBAAgB,EAChB,sBAAsB,EAEtB,YAAY,EAEZ,sBAAsB,EACtB,8BAA8B,EAC9B,2BAA2B,EAE3B,YAAY,EACZ,oBAAoB,EAWpB,qBAAqB,EAErB,cAAc,EAEd,gBAAgB,EAChB,yBAAyB,EAKzB,SAAS,EAOT,eAAe,EAIf,UAAU,EAEV,gBAAgB,EAChB,uBAAuB,EACvB,aAAa,EACb,WAAW,EAEX,gBAAgB,EAChB,gBAAgB,EAChB,kBAAkB,EAClB,iBAAiB,EACjB,iBAAiB,EACjB,iBAAiB,EACjB,sBAAsB,EACtB,gBAAgB,EAChB,mBAAmB,EAEnB,wBAAwB,EAExB,kBAAkB,EAElB,8BAA8B,EAE9B,+BAA+B,EAE/B,4BAA4B,EAE5B,SAAS,EACT,iBAAiB,EACjB,YAAY,EACZ,WAAW,EACX,YAAY,EAKZ,8BAA8B,EAC9B,gBAAgB,EAUhB,+BAA+B,EAU/B,mBAAmB,EACnB,wBAAwB,EACxB,+BAA+B,EAC/B,gBAAgB,EAKhB,oBAAoB,EACpB,YAAY,EACZ,8BAA8B,EAC9B,mCAAmC,EAanC,QAAQ,EACR,gCAAgC,EACnC,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,aAAa,EAAC,MAAM,iBAAiB,CAAC;AAC9C,OAAO,EAAC,KAAK,EAAC,MAAM,qBAAqB,CAAC;AAC1C,OAAO,EAAC,QAAQ,EAAe,MAAM,wBAAwB,CAAC;AAC9D,OAAO,EAAC,kBAAkB,EAAC,MAAM,wCAAwC,CAAC;AAC1E,OAAO,EAAC,aAAa,EAAE,wBAAwB,EAAE,+BAA+B,EAAE,mBAAmB,EAAC,MAAM,qBAAqB,CAAC;AAClI,OAAO,EAAC,UAAU,EAAE,sBAAsB,EAA0B,gBAAgB,EAAC,MAAM,2BAA2B,CAAC;AACvH,OAAO,EAAC,YAAY,EAA2B,MAAM,6BAA6B,CAAC;AACnF,OAAO,EAAC,sBAAsB,EAAC,MAAM,uCAAuC,CAAC;AAC7E,OAAO,EAAC,8BAA8B,EAAC,MAAM,2DAA2D,CAAC;AACzG,OAAO,EAAC,2BAA2B,EAAqC,MAAM,4CAA4C,CAAC;AAC3H,OAAO,EAAC,YAAY,EAAE,oBAAoB,EAAC,MAAM,0CAA0C,CAAC;AAC5F,OAAO,EACH,qBAAqB,EAAqC,cAAc,EAC3E,MAAM,sCAAsC,CAAC;AAM9C,OAAO,EAAC,SAAS,EAAC,MAAM,0BAA0B,CAAC;AACnD,OAAO,EACH,gBAAgB,EAEnB,MAAM,kDAAkD,CAAC;AAC1D,OAAO,EAAC,yBAAyB,EAAC,MAAM,iEAAiE,CAAC;AAC1G,OAAO,EACH,SAAS,EAEZ,MAAM,oCAAoC,CAAC;AAC5C,OAAO,EACH,eAAe,EAClB,MAAM,gCAAgC,CAAC;AACxC,OAAO,EAAC,UAAU,EAAuB,MAAM,2BAA2B,CAAC;AAC3E,OAAO,EAAC,gBAAgB,EAAC,MAAM,6BAA6B,CAAC;AAC7D,OAAO,EAAC,uBAAuB,EAAC,MAAM,oCAAoC,CAAC;AAC3E,OAAO,EAAC,WAAW,EAA2B,MAAM,kBAAkB,CAAC;AACvE,OAAO,EAAC,gBAAgB,EAAC,MAAM,oCAAoC,CAAC;AACpE,OAAO,EAAC,gBAAgB,EAAC,MAAM,oCAAoC,CAAC;AACpE,OAAO,EAAC,kBAAkB,EAAC,MAAM,sCAAsC,CAAC;AACxE,OAAO,EAAC,iBAAiB,EAAC,MAAM,qCAAqC,CAAC;AACtE,OAAO,EAAC,iBAAiB,EAAC,MAAM,qCAAqC,CAAC;AACtE,OAAO,EAAC,iBAAiB,EAAC,MAAM,qCAAqC,CAAC;AACtE,OAAO,EAAC,sBAAsB,EAAC,MAAM,0CAA0C,CAAC;AAChF,OAAO,EAAC,gBAAgB,EAAC,MAAM,oCAAoC,CAAC;AACpE,OAAO,EAAC,mBAAmB,EAAkC,MAAM,+CAA+C,CAAC;AACnH,OAAO,EAAC,wBAAwB,EAAuC,MAAM,oDAAoD,CAAC;AAClI,OAAO,EACH,8BAA8B,EAAsC,+BAA+B,EAC9D,4BAA4B,EAAoC,kBAAkB,EAE1H,MAAM,4CAA4C,CAAC;AACpD,OAAO,EACH,SAAS,EAAE,iBAAiB,EAAE,YAAY,EAAE,WAAW,EAAE,YAAY,EAExE,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAAC,8BAA8B,EAAC,MAAM,2CAA2C,CAAC;AACzF,OAAO,EAAC,gBAAgB,EAAC,MAAM,6BAA6B,CAAC;AAC7D,OAAO,EAAC,gBAAgB,EAAC,MAAM,4BAA4B,CAAC;AAE5D,OAAO,EAGS,+BAA+B,EAC9C,MAAM,YAAY,CAAC;AAMpB,OAAO,EAC4C,oBAAoB,EAAE,YAAY,EAAE,8BAA8B,EACjH,mCAAmC,EAE6D,gCAAgC,EACnI,MAAM,mCAAmC,CAAC;AAC3C,OAAO,EAAC,QAAQ,EAAsB,MAAM,qCAAqC,CAAC;AAGlF,OAAO,EACH,KAAK,EACL,QAAQ,EAER,aAAa,EACb,kBAAkB,EAClB,UAAU,EACV,gBAAgB,EAChB,sBAAsB,EAEtB,YAAY,EAEZ,sBAAsB,EACtB,8BAA8B,EAC9B,2BAA2B,EAE3B,YAAY,EACZ,oBAAoB,EAWpB,SAAS,EACT,qBAAqB,EAErB,cAAc,EAEd,gBAAgB,EAChB,yBAAyB,EAKzB,SAAS,EAOT,eAAe,EAIf,UAAU,EAEV,gBAAgB,EAChB,uBAAuB,EACvB,aAAa,EACb,WAAW,EAEX,gBAAgB,EAChB,gBAAgB,EAChB,kBAAkB,EAClB,iBAAiB,EACjB,iBAAiB,EACjB,iBAAiB,EACjB,sBAAsB,EACtB,gBAAgB,EAChB,mBAAmB,EAEnB,wBAAwB,EAExB,kBAAkB,EAElB,8BAA8B,EAE9B,+BAA+B,EAE/B,4BAA4B,EAE5B,SAAS,EACT,iBAAiB,EACjB,YAAY,EACZ,WAAW,EACX,YAAY,EAKZ,8BAA8B,EAC9B,gBAAgB,EAUhB,+BAA+B,EAU/B,mBAAmB,EACnB,wBAAwB,EACxB,+BAA+B,EAC/B,gBAAgB,EAKhB,oBAAoB,EACpB,YAAY,EACZ,8BAA8B,EAC9B,mCAAmC,EAanC,QAAQ,EACR,gCAAgC,EACnC,CAAC"}

package/dist/utils/getConsoleLogPrefix.js

@@ -4,7 +4,7 @@ export function getConsoleLogPrefix(forcePrefix = false, padEnd = true) {
     const isInCLI = getIsRunningFromCLI();
     const forceShowLogPrefix = getForceShowConsoleLogPrefix();
     if (!isInCLI || forceShowLogPrefix || forcePrefix)
-        return chalk.grey("[node-llama-cpp]") + (padEnd ? " " : "");
+        return chalk.gray("[node-llama-cpp]") + (padEnd ? " " : "");
     return "";
 }
 //# sourceMappingURL=getConsoleLogPrefix.js.map

package/dist/utils/getReadableContextSize.d.ts

@@ -0,0 +1 @@
+export declare function getReadableContextSize(contextSize: number): string;

package/dist/utils/getReadableContextSize.js

@@ -0,0 +1,7 @@
+export function getReadableContextSize(contextSize) {
+    return contextSize.toLocaleString("en-US", {
+        notation: "compact",
+        compactDisplay: "short"
+    });
+}
+//# sourceMappingURL=getReadableContextSize.js.map

package/dist/utils/getReadableContextSize.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"getReadableContextSize.js","sourceRoot":"","sources":["../../src/utils/getReadableContextSize.ts"],"names":[],"mappings":"AAAA,MAAM,UAAU,sBAAsB,CAAC,WAAmB;IACtD,OAAO,WAAW,CAAC,cAAc,CAAC,OAAO,EAAE;QACvC,QAAQ,EAAE,SAAS;QACnB,cAAc,EAAE,OAAO;KAC1B,CAAC,CAAC;AACP,CAAC"}

package/dist/utils/isToken.d.ts

@@ -0,0 +1,2 @@
+import { Token } from "../types.js";
+export declare function isToken(token: any): token is Token;

package/dist/utils/isToken.js

@@ -0,0 +1,4 @@
+export function isToken(token) {
+    return typeof token === "number";
+}
+//# sourceMappingURL=isToken.js.map

package/dist/utils/isToken.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"isToken.js","sourceRoot":"","sources":["../../src/utils/isToken.ts"],"names":[],"mappings":"AAEA,MAAM,UAAU,OAAO,CAAC,KAAU;IAC9B,OAAO,OAAO,KAAK,KAAK,QAAQ,CAAC;AACrC,CAAC"}

package/dist/utils/isUrl.d.ts

@@ -0,0 +1 @@
+export declare function isUrl(text: string, throwOnInvalidUrl?: boolean): boolean;

package/dist/utils/isUrl.js

@@ -0,0 +1,15 @@
+export function isUrl(text, throwOnInvalidUrl = true) {
+    if (text.startsWith("http://") || text.startsWith("https://")) {
+        try {
+            new URL(text);
+            return true;
+        }
+        catch {
+            if (throwOnInvalidUrl)
+                throw new Error(`Invalid URL: ${text}`);
+            return false;
+        }
+    }
+    return false;
+}
+//# sourceMappingURL=isUrl.js.map

package/dist/utils/isUrl.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"isUrl.js","sourceRoot":"","sources":["../../src/utils/isUrl.ts"],"names":[],"mappings":"AAAA,MAAM,UAAU,KAAK,CAAC,IAAY,EAAE,oBAA6B,IAAI;IACjE,IAAI,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC,IAAI,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC,EAAE;QAC3D,IAAI;YACA,IAAI,GAAG,CAAC,IAAI,CAAC,CAAC;YACd,OAAO,IAAI,CAAC;SACf;QAAC,MAAM;YACJ,IAAI,iBAAiB;gBACjB,MAAM,IAAI,KAAK,CAAC,gBAAgB,IAAI,EAAE,CAAC,CAAC;YAE5C,OAAO,KAAK,CAAC;SAChB;KACJ;IAED,OAAO,KAAK,CAAC;AACjB,CAAC"}

package/dist/utils/tokenizeInput.d.ts

@@ -1,3 +1,3 @@
 import { Token, Tokenizer } from "../types.js";
 import { LlamaText } from "./LlamaText.js";
-export declare function tokenizeInput(input: Token[] | string | LlamaText, tokenizer: Tokenizer, options?: "trimLeadingSpace"): Token[];
+export declare function tokenizeInput(input: Token | Token[] | string | LlamaText, tokenizer: Tokenizer, options?: "trimLeadingSpace"): Token[];

package/dist/utils/tokenizeInput.js

@@ -1,9 +1,12 @@
 import { isLlamaText } from "./LlamaText.js";
+import { isToken } from "./isToken.js";
 export function tokenizeInput(input, tokenizer, options) {
     if (typeof input === "string")
         return tokenizer(input, false, options);
     else if (isLlamaText(input))
         return input.tokenize(tokenizer, options);
+    else if (isToken(input))
+        return [input];
     return input;
 }
 //# sourceMappingURL=tokenizeInput.js.map

package/dist/utils/tokenizeInput.js.map

@@ -1 +1 @@
-{"version":3,"file":"tokenizeInput.js","sourceRoot":"","sources":["../../src/utils/tokenizeInput.ts"],"names":[],"mappings":"AACA,OAAO,EAAC,WAAW,EAAY,MAAM,gBAAgB,CAAC;AAEtD,MAAM,UAAU,aAAa,CAAC,KAAmC,EAAE,SAAoB,EAAE,OAA4B;IACjH,IAAI,OAAO,KAAK,KAAK,QAAQ;QACzB,OAAO,SAAS,CAAC,KAAK,EAAE,KAAK,EAAE,OAAO,CAAC,CAAC;SACvC,IAAI,WAAW,CAAC,KAAK,CAAC;QACvB,OAAO,KAAK,CAAC,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;IAE9C,OAAO,KAAK,CAAC;AACjB,CAAC"}
+{"version":3,"file":"tokenizeInput.js","sourceRoot":"","sources":["../../src/utils/tokenizeInput.ts"],"names":[],"mappings":"AACA,OAAO,EAAC,WAAW,EAAY,MAAM,gBAAgB,CAAC;AACtD,OAAO,EAAC,OAAO,EAAC,MAAM,cAAc,CAAC;AAErC,MAAM,UAAU,aAAa,CAAC,KAA2C,EAAE,SAAoB,EAAE,OAA4B;IACzH,IAAI,OAAO,KAAK,KAAK,QAAQ;QACzB,OAAO,SAAS,CAAC,KAAK,EAAE,KAAK,EAAE,OAAO,CAAC,CAAC;SACvC,IAAI,WAAW,CAAC,KAAK,CAAC;QACvB,OAAO,KAAK,CAAC,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;SACzC,IAAI,OAAO,CAAC,KAAK,CAAC;QACnB,OAAO,CAAC,KAAK,CAAC,CAAC;IAEnB,OAAO,KAAK,CAAC;AACjB,CAAC"}

package/dist/utils/withOra.d.ts

@@ -3,4 +3,5 @@ export default function withOra<T>(message: string | {
     success?: string;
     fail?: string;
     useStatusLogs?: boolean;
+    noSuccessLiveStatus?: boolean;
 }, callback: () => Promise<T>): Promise<T>;

package/dist/utils/withOra.js

@@ -1,9 +1,9 @@
 import ora from "ora";
-import { isRunningInsideGoogleColab } from "../config.js";
+import { useCiLogs } from "../config.js";
 import { getConsoleLogPrefix } from "./getConsoleLogPrefix.js";
 import withStatusLogs from "./withStatusLogs.js";
 export default async function withOra(message, callback) {
-    if (isRunningInsideGoogleColab || (typeof message !== "string" && message.useStatusLogs))
+    if (useCiLogs || (typeof message !== "string" && message.useStatusLogs))
         return withStatusLogs(message, callback);
     const spinner = ora({
         prefixText: getConsoleLogPrefix(),
@@ -11,20 +11,26 @@ export default async function withOra(message, callback) {
             ? { text: message }
             : { loading: message.loading, success: message.success, fail: message.fail })
     });
-    spinner.start();
+    spinner.start(typeof message === "string"
+        ? message
+        : message.loading);
     try {
         const res = await callback();
-        if (typeof message !== "string")
-            spinner.succeed(message.success);
+        if (typeof message !== "string") {
+            if (message.noSuccessLiveStatus)
+                spinner.clear();
+            else
+                spinner.succeed(message.success);
+        }
         else
-            spinner.succeed();
+            spinner.succeed(message);
         return res;
     }
     catch (er) {
         if (typeof message !== "string")
             spinner.fail(message.fail);
         else
-            spinner.fail();
+            spinner.fail(message);
         throw er;
     }
 }

package/dist/utils/withOra.js.map

@@ -1 +1 @@
-{"version":3,"file":"withOra.js","sourceRoot":"","sources":["../../src/utils/withOra.ts"],"names":[],"mappings":"AAAA,OAAO,GAAG,MAAM,KAAK,CAAC;AACtB,OAAO,EAAC,0BAA0B,EAAC,MAAM,cAAc,CAAC;AACxD,OAAO,EAAC,mBAAmB,EAAC,MAAM,0BAA0B,CAAC;AAC7D,OAAO,cAAc,MAAM,qBAAqB,CAAC;AAEjD,MAAM,CAAC,OAAO,CAAC,KAAK,UAAU,OAAO,CACjC,OAKC,EACD,QAA0B;IAE1B,IAAI,0BAA0B,IAAI,CAAC,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,CAAC,aAAa,CAAC;QACpF,OAAO,cAAc,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;IAE7C,MAAM,OAAO,GAAG,GAAG,CAAC;QAChB,UAAU,EAAE,mBAAmB,EAAE;QACjC,GAAG,CACC,OAAO,OAAO,KAAK,QAAQ;YACvB,CAAC,CAAC,EAAC,IAAI,EAAE,OAAO,EAAqC;YACrD,CAAC,CAAC,EAAC,OAAO,EAAE,OAAO,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,CAAC,OAAO,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAC,CACjF;KACJ,CAAC,CAAC;IAEH,OAAO,CAAC,KAAK,EAAE,CAAC;IAEhB,IAAI;QACA,MAAM,GAAG,GAAG,MAAM,QAAQ,EAAE,CAAC;QAE7B,IAAI,OAAO,OAAO,KAAK,QAAQ;YAC3B,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;;YAEjC,OAAO,CAAC,OAAO,EAAE,CAAC;QAEtB,OAAO,GAAG,CAAC;KACd;IAAC,OAAO,EAAE,EAAE;QACT,IAAI,OAAO,OAAO,KAAK,QAAQ;YAC3B,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;;YAE3B,OAAO,CAAC,IAAI,EAAE,CAAC;QAEnB,MAAM,EAAE,CAAC;KACZ;AACL,CAAC"}
+{"version":3,"file":"withOra.js","sourceRoot":"","sources":["../../src/utils/withOra.ts"],"names":[],"mappings":"AAAA,OAAO,GAAG,MAAM,KAAK,CAAC;AACtB,OAAO,EAAC,SAAS,EAAC,MAAM,cAAc,CAAC;AACvC,OAAO,EAAC,mBAAmB,EAAC,MAAM,0BAA0B,CAAC;AAC7D,OAAO,cAAc,MAAM,qBAAqB,CAAC;AAEjD,MAAM,CAAC,OAAO,CAAC,KAAK,UAAU,OAAO,CACjC,OAMC,EACD,QAA0B;IAE1B,IAAI,SAAS,IAAI,CAAC,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,CAAC,aAAa,CAAC;QACnE,OAAO,cAAc,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;IAE7C,MAAM,OAAO,GAAG,GAAG,CAAC;QAChB,UAAU,EAAE,mBAAmB,EAAE;QACjC,GAAG,CACC,OAAO,OAAO,KAAK,QAAQ;YACvB,CAAC,CAAC,EAAC,IAAI,EAAE,OAAO,EAAqC;YACrD,CAAC,CAAC,EAAC,OAAO,EAAE,OAAO,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,CAAC,OAAO,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAC,CACjF;KACJ,CAAC,CAAC;IAEH,OAAO,CAAC,KAAK,CACT,OAAO,OAAO,KAAK,QAAQ;QACvB,CAAC,CAAC,OAAO;QACT,CAAC,CAAC,OAAO,CAAC,OAAO,CACxB,CAAC;IAEF,IAAI;QACA,MAAM,GAAG,GAAG,MAAM,QAAQ,EAAE,CAAC;QAE7B,IAAI,OAAO,OAAO,KAAK,QAAQ,EAAE;YAC7B,IAAI,OAAO,CAAC,mBAAmB;gBAC3B,OAAO,CAAC,KAAK,EAAE,CAAC;;gBAEhB,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;SACxC;;YACG,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;QAE7B,OAAO,GAAG,CAAC;KACd;IAAC,OAAO,EAAE,EAAE;QACT,IAAI,OAAO,OAAO,KAAK,QAAQ;YAC3B,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;;YAE3B,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAE1B,MAAM,EAAE,CAAC;KACZ;AACL,CAAC"}

package/dist/utils/withProgressLog.d.ts

@@ -0,0 +1,23 @@
+/// <reference types="node" />
+export declare function withProgressLog<T>({ loadingText, successText, failText, liveUpdates, statusIcons, initialPercentage, initialProgressBarText, progressBarLength, minPercentageChangeForNonLiveUpdates, eta, etaUpdateInterval, noProgress, progressFractionDigits, noSuccessLiveStatus, liveCtrlCSendsAbortSignal }: {
+    loadingText: string;
+    successText: string;
+    failText: string;
+    liveUpdates?: boolean;
+    statusIcons?: boolean;
+    initialPercentage?: number;
+    initialProgressBarText?: string;
+    progressBarLength?: number;
+    minPercentageChangeForNonLiveUpdates?: number;
+    eta?: boolean;
+    etaUpdateInterval?: number;
+    noProgress?: boolean;
+    progressFractionDigits?: boolean;
+    noSuccessLiveStatus?: boolean;
+    liveCtrlCSendsAbortSignal?: boolean;
+}, callback: (progressUpdater: ProgressUpdater) => Promise<T>): Promise<T>;
+type ProgressUpdater = {
+    setProgress(percentage: number, progressText?: string): ProgressUpdater;
+    abortSignal?: AbortSignal;
+};
+export {};