@qvac/sdk 0.14.1 → 0.15.0

package/dist/schemas/load-model.d.ts

@@ -1,17 +1,17 @@
-import { z } from "zod";
-import type { Logger } from "../logging/index";
-import { llmConfigBaseSchema, embedConfigBaseSchema, type LlmConfig, type EmbedConfig } from "./llamacpp-config";
-import { whisperConfigSchema } from "./transcription-config";
-import type { parakeetConfigSchema } from "./transcription-config";
-import { bciConfigSchema } from "./bci-config";
-import { delegateSchema } from "./delegate";
-import { nmtConfigSchema } from "./translation-config";
-import { ttsConfigSchema } from "./text-to-speech";
-import { ocrConfigSchema } from "./ocr";
-import { type ModelDescriptor } from "./model-src-utils";
-import { ModelType, type CanonicalModelType, type ModelTypeInput } from "./model-types";
-import { sdcppConfigSchema } from "./sdcpp-config";
-import { vlaConfigSchema } from "./vla";
+import { z } from 'zod';
+import type { Logger } from '../logging/index';
+import { llmConfigBaseSchema, embedConfigBaseSchema, type LlmConfig, type EmbedConfig } from './llamacpp-config';
+import { whisperConfigSchema } from './transcription-config';
+import type { parakeetConfigSchema } from './transcription-config';
+import { bciConfigSchema } from './bci-config';
+import { delegateSchema } from './delegate';
+import { nmtConfigSchema } from './translation-config';
+import { ttsConfigSchema } from './text-to-speech';
+import { ocrConfigSchema } from './ocr';
+import { type ModelDescriptor } from './model-src-utils';
+import { ModelType, type CanonicalModelType, type ModelTypeInput } from './model-types';
+import { sdcppConfigSchema } from './sdcpp-config';
+import { vlaConfigSchema } from './vla';
 export declare function isBuiltInModelType(modelType: unknown): boolean;
 export declare const loadBuiltinModelOptionsBaseSchema: z.ZodUnion<readonly [z.ZodObject<{
     modelType: z.ZodEnum<{
@@ -36,23 +36,24 @@ export declare const loadBuiltinModelOptionsBaseSchema: z.ZodUnion<readonly [z.Z
         repeat_penalty: z.ZodOptional<z.ZodNumber>;
         stop_sequences: z.ZodOptional<z.ZodArray<z.ZodString>>;
         n_discarded: z.ZodOptional<z.ZodNumber>;
+        parallel: z.ZodOptional<z.ZodNumber>;
         tools: z.ZodOptional<z.ZodBoolean>;
         toolsMode: z.ZodOptional<z.ZodEnum<{
             static: "static";
             dynamic: "dynamic";
         }>>;
-        "cache-type-k": z.ZodOptional<z.ZodString>;
-        "cache-type-v": z.ZodOptional<z.ZodString>;
-        "main-gpu": z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
+        'cache-type-k': z.ZodOptional<z.ZodString>;
+        'cache-type-v': z.ZodOptional<z.ZodString>;
+        'main-gpu': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
             integrated: "integrated";
             dedicated: "dedicated";
         }>]>>;
-        "split-mode": z.ZodOptional<z.ZodEnum<{
+        'split-mode': z.ZodOptional<z.ZodEnum<{
             none: "none";
             layer: "layer";
             row: "row";
         }>>;
-        "tensor-split": z.ZodOptional<z.ZodString>;
+        'tensor-split': z.ZodOptional<z.ZodString>;
         openclCacheDir: z.ZodOptional<z.ZodString>;
         reasoning_budget: z.ZodOptional<z.ZodNumber>;
         projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -97,6 +98,7 @@ export declare const loadBuiltinModelOptionsBaseSchema: z.ZodUnion<readonly [z.Z
             batched: "batched";
             sequential: "sequential";
         }>>;
+        'mmproj-use-gpu': z.ZodOptional<z.ZodBoolean>;
     }, z.core.$strict>>;
     modelSrc: z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
         src: z.ZodString;
@@ -472,8 +474,8 @@ export declare const loadBuiltinModelOptionsBaseSchema: z.ZodUnion<readonly [z.Z
     modelConfig: z.ZodOptional<z.ZodObject<{
         gpuLayers: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         batchSize: z.ZodOptional<z.ZodNumber>;
         pooling: z.ZodOptional<z.ZodEnum<{
@@ -1380,8 +1382,8 @@ export declare const loadBuiltinModelOptionsBaseSchema: z.ZodUnion<readonly [z.Z
         }>>;
         threads: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         prediction: z.ZodOptional<z.ZodEnum<{
             auto: "auto";
@@ -1963,23 +1965,24 @@ export declare const loadModelOptionsBaseSchema: z.ZodUnion<readonly [z.ZodUnion
         repeat_penalty: z.ZodOptional<z.ZodNumber>;
         stop_sequences: z.ZodOptional<z.ZodArray<z.ZodString>>;
         n_discarded: z.ZodOptional<z.ZodNumber>;
+        parallel: z.ZodOptional<z.ZodNumber>;
         tools: z.ZodOptional<z.ZodBoolean>;
         toolsMode: z.ZodOptional<z.ZodEnum<{
             static: "static";
             dynamic: "dynamic";
         }>>;
-        "cache-type-k": z.ZodOptional<z.ZodString>;
-        "cache-type-v": z.ZodOptional<z.ZodString>;
-        "main-gpu": z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
+        'cache-type-k': z.ZodOptional<z.ZodString>;
+        'cache-type-v': z.ZodOptional<z.ZodString>;
+        'main-gpu': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
             integrated: "integrated";
             dedicated: "dedicated";
         }>]>>;
-        "split-mode": z.ZodOptional<z.ZodEnum<{
+        'split-mode': z.ZodOptional<z.ZodEnum<{
             none: "none";
             layer: "layer";
             row: "row";
         }>>;
-        "tensor-split": z.ZodOptional<z.ZodString>;
+        'tensor-split': z.ZodOptional<z.ZodString>;
         openclCacheDir: z.ZodOptional<z.ZodString>;
         reasoning_budget: z.ZodOptional<z.ZodNumber>;
         projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -2024,6 +2027,7 @@ export declare const loadModelOptionsBaseSchema: z.ZodUnion<readonly [z.ZodUnion
             batched: "batched";
             sequential: "sequential";
         }>>;
+        'mmproj-use-gpu': z.ZodOptional<z.ZodBoolean>;
     }, z.core.$strict>>;
     modelSrc: z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
         src: z.ZodString;
@@ -2399,8 +2403,8 @@ export declare const loadModelOptionsBaseSchema: z.ZodUnion<readonly [z.ZodUnion
     modelConfig: z.ZodOptional<z.ZodObject<{
         gpuLayers: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         batchSize: z.ZodOptional<z.ZodNumber>;
         pooling: z.ZodOptional<z.ZodEnum<{
@@ -3307,8 +3311,8 @@ export declare const loadModelOptionsBaseSchema: z.ZodUnion<readonly [z.ZodUnion
         }>>;
         threads: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         prediction: z.ZodOptional<z.ZodEnum<{
             auto: "auto";
@@ -3889,23 +3893,24 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
         repeat_penalty: z.ZodOptional<z.ZodNumber>;
         stop_sequences: z.ZodOptional<z.ZodArray<z.ZodString>>;
         n_discarded: z.ZodOptional<z.ZodNumber>;
+        parallel: z.ZodOptional<z.ZodNumber>;
         tools: z.ZodOptional<z.ZodBoolean>;
         toolsMode: z.ZodOptional<z.ZodEnum<{
             static: "static";
             dynamic: "dynamic";
         }>>;
-        "cache-type-k": z.ZodOptional<z.ZodString>;
-        "cache-type-v": z.ZodOptional<z.ZodString>;
-        "main-gpu": z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
+        'cache-type-k': z.ZodOptional<z.ZodString>;
+        'cache-type-v': z.ZodOptional<z.ZodString>;
+        'main-gpu': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
             integrated: "integrated";
             dedicated: "dedicated";
         }>]>>;
-        "split-mode": z.ZodOptional<z.ZodEnum<{
+        'split-mode': z.ZodOptional<z.ZodEnum<{
             none: "none";
             layer: "layer";
             row: "row";
         }>>;
-        "tensor-split": z.ZodOptional<z.ZodString>;
+        'tensor-split': z.ZodOptional<z.ZodString>;
         openclCacheDir: z.ZodOptional<z.ZodString>;
         reasoning_budget: z.ZodOptional<z.ZodNumber>;
         projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -3950,6 +3955,7 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
             batched: "batched";
             sequential: "sequential";
         }>>;
+        'mmproj-use-gpu': z.ZodOptional<z.ZodBoolean>;
     }, z.core.$strict>>;
     modelSrc: z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
         src: z.ZodString;
@@ -4325,8 +4331,8 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
     modelConfig: z.ZodOptional<z.ZodObject<{
         gpuLayers: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         batchSize: z.ZodOptional<z.ZodNumber>;
         pooling: z.ZodOptional<z.ZodEnum<{
@@ -5233,8 +5239,8 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
         }>>;
         threads: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         prediction: z.ZodOptional<z.ZodEnum<{
             auto: "auto";
@@ -5825,13 +5831,14 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
         repeat_penalty?: number | undefined;
         stop_sequences?: string[] | undefined;
         n_discarded?: number | undefined;
+        parallel?: number | undefined;
         tools?: boolean | undefined;
         toolsMode?: "static" | "dynamic" | undefined;
-        "cache-type-k"?: string | undefined;
-        "cache-type-v"?: string | undefined;
-        "main-gpu"?: number | "integrated" | "dedicated" | undefined;
-        "split-mode"?: "none" | "layer" | "row" | undefined;
-        "tensor-split"?: string | undefined;
+        'cache-type-k'?: string | undefined;
+        'cache-type-v'?: string | undefined;
+        'main-gpu'?: number | "integrated" | "dedicated" | undefined;
+        'split-mode'?: "none" | "layer" | "row" | undefined;
+        'tensor-split'?: string | undefined;
         openclCacheDir?: string | undefined;
         reasoning_budget?: number | undefined;
         projectionModelSrc?: string | {
@@ -5848,6 +5855,7 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
             addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
         } | undefined;
         image_tile_mode?: "disabled" | "batched" | "sequential" | undefined;
+        'mmproj-use-gpu'?: boolean | undefined;
     } | undefined;
     delegate?: {
         providerPublicKey: string;
@@ -6058,7 +6066,7 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
     };
     modelConfig?: {
         gpuLayers?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         batchSize?: number | undefined;
         pooling?: "none" | "mean" | "cls" | "last" | "rank" | undefined;
         attention?: "causal" | "non-causal" | undefined;
@@ -6297,7 +6305,7 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
     modelConfig?: {
         mode: "diffusion" | "upscale" | "video";
         threads?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         prediction?: "auto" | "eps" | "v" | "edm_v" | "flow" | "flux2_flow" | undefined;
         type?: "auto" | "f32" | "f16" | "bf16" | "q2_k" | "q3_k" | "q4_0" | "q4_1" | "q4_k" | "q5_0" | "q5_1" | "q5_k" | "q6_k" | "q8_0" | undefined;
         rng?: "cpu" | "cuda" | "std_default" | undefined;
@@ -6541,13 +6549,14 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
         repeat_penalty?: number | undefined;
         stop_sequences?: string[] | undefined;
         n_discarded?: number | undefined;
+        parallel?: number | undefined;
         tools?: boolean | undefined;
         toolsMode?: "static" | "dynamic" | undefined;
-        "cache-type-k"?: string | undefined;
-        "cache-type-v"?: string | undefined;
-        "main-gpu"?: number | "integrated" | "dedicated" | undefined;
-        "split-mode"?: "none" | "layer" | "row" | undefined;
-        "tensor-split"?: string | undefined;
+        'cache-type-k'?: string | undefined;
+        'cache-type-v'?: string | undefined;
+        'main-gpu'?: number | "integrated" | "dedicated" | undefined;
+        'split-mode'?: "none" | "layer" | "row" | undefined;
+        'tensor-split'?: string | undefined;
         openclCacheDir?: string | undefined;
         reasoning_budget?: number | undefined;
         projectionModelSrc?: string | {
@@ -6564,6 +6573,7 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
             addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
         } | undefined;
         image_tile_mode?: "disabled" | "batched" | "sequential" | undefined;
+        'mmproj-use-gpu'?: boolean | undefined;
     } | undefined;
     seed?: boolean | undefined;
     delegate?: {
@@ -6774,7 +6784,7 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
     };
     modelConfig?: {
         gpuLayers?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         batchSize?: number | undefined;
         pooling?: "none" | "mean" | "cls" | "last" | "rank" | undefined;
         attention?: "causal" | "non-causal" | undefined;
@@ -7013,7 +7023,7 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
     modelConfig?: {
         mode: "diffusion" | "upscale" | "video";
         threads?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         prediction?: "auto" | "eps" | "v" | "edm_v" | "flow" | "flux2_flow" | undefined;
         type?: "auto" | "f32" | "f16" | "bf16" | "q2_k" | "q3_k" | "q4_0" | "q4_1" | "q4_k" | "q5_0" | "q5_1" | "q5_k" | "q6_k" | "q8_0" | undefined;
         rng?: "cpu" | "cuda" | "std_default" | undefined;
@@ -7249,23 +7259,24 @@ export declare const loadBuiltinToRequestSchema: z.ZodDiscriminatedUnion<[z.ZodP
         repeat_penalty: z.ZodOptional<z.ZodNumber>;
         stop_sequences: z.ZodOptional<z.ZodArray<z.ZodString>>;
         n_discarded: z.ZodOptional<z.ZodNumber>;
+        parallel: z.ZodOptional<z.ZodNumber>;
         tools: z.ZodOptional<z.ZodBoolean>;
         toolsMode: z.ZodOptional<z.ZodEnum<{
             static: "static";
             dynamic: "dynamic";
         }>>;
-        "cache-type-k": z.ZodOptional<z.ZodString>;
-        "cache-type-v": z.ZodOptional<z.ZodString>;
-        "main-gpu": z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
+        'cache-type-k': z.ZodOptional<z.ZodString>;
+        'cache-type-v': z.ZodOptional<z.ZodString>;
+        'main-gpu': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
             integrated: "integrated";
             dedicated: "dedicated";
         }>]>>;
-        "split-mode": z.ZodOptional<z.ZodEnum<{
+        'split-mode': z.ZodOptional<z.ZodEnum<{
             none: "none";
             layer: "layer";
             row: "row";
         }>>;
-        "tensor-split": z.ZodOptional<z.ZodString>;
+        'tensor-split': z.ZodOptional<z.ZodString>;
         openclCacheDir: z.ZodOptional<z.ZodString>;
         reasoning_budget: z.ZodOptional<z.ZodNumber>;
         projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -7310,6 +7321,7 @@ export declare const loadBuiltinToRequestSchema: z.ZodDiscriminatedUnion<[z.ZodP
             batched: "batched";
             sequential: "sequential";
         }>>;
+        'mmproj-use-gpu': z.ZodOptional<z.ZodBoolean>;
     }, z.core.$strict>>;
     onProgress: z.ZodOptional<z.ZodUnknown>;
     logger: z.ZodOptional<z.ZodUnknown>;
@@ -7409,13 +7421,14 @@ export declare const loadBuiltinToRequestSchema: z.ZodDiscriminatedUnion<[z.ZodP
         repeat_penalty?: number | undefined;
         stop_sequences?: string[] | undefined;
         n_discarded?: number | undefined;
+        parallel?: number | undefined;
         tools?: boolean | undefined;
         toolsMode?: "static" | "dynamic" | undefined;
-        "cache-type-k"?: string | undefined;
-        "cache-type-v"?: string | undefined;
-        "main-gpu"?: number | "integrated" | "dedicated" | undefined;
-        "split-mode"?: "none" | "layer" | "row" | undefined;
-        "tensor-split"?: string | undefined;
+        'cache-type-k'?: string | undefined;
+        'cache-type-v'?: string | undefined;
+        'main-gpu'?: number | "integrated" | "dedicated" | undefined;
+        'split-mode'?: "none" | "layer" | "row" | undefined;
+        'tensor-split'?: string | undefined;
         openclCacheDir?: string | undefined;
         reasoning_budget?: number | undefined;
         projectionModelSrc?: string | {
@@ -7432,6 +7445,7 @@ export declare const loadBuiltinToRequestSchema: z.ZodDiscriminatedUnion<[z.ZodP
             addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
         } | undefined;
         image_tile_mode?: "disabled" | "batched" | "sequential" | undefined;
+        'mmproj-use-gpu'?: boolean | undefined;
     } | undefined;
     onProgress?: unknown;
     logger?: unknown;
@@ -8142,8 +8156,8 @@ export declare const loadBuiltinToRequestSchema: z.ZodDiscriminatedUnion<[z.ZodP
     modelConfig: z.ZodOptional<z.ZodObject<{
         gpuLayers: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         batchSize: z.ZodOptional<z.ZodNumber>;
         pooling: z.ZodOptional<z.ZodEnum<{
@@ -8258,7 +8272,7 @@ export declare const loadBuiltinToRequestSchema: z.ZodDiscriminatedUnion<[z.ZodP
     };
     modelConfig?: {
         gpuLayers?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         batchSize?: number | undefined;
         pooling?: "none" | "mean" | "cls" | "last" | "rank" | undefined;
         attention?: "causal" | "non-causal" | undefined;
@@ -9513,8 +9527,8 @@ export declare const loadBuiltinToRequestSchema: z.ZodDiscriminatedUnion<[z.ZodP
         }>>;
         threads: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         prediction: z.ZodOptional<z.ZodEnum<{
             auto: "auto";
@@ -9949,7 +9963,7 @@ export declare const loadBuiltinToRequestSchema: z.ZodDiscriminatedUnion<[z.ZodP
     modelConfig?: {
         mode: "diffusion" | "upscale" | "video";
         threads?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         prediction?: "auto" | "eps" | "v" | "edm_v" | "flow" | "flux2_flow" | undefined;
         type?: "auto" | "f32" | "f16" | "bf16" | "q2_k" | "q3_k" | "q4_0" | "q4_1" | "q4_k" | "q5_0" | "q5_1" | "q5_k" | "q6_k" | "q8_0" | undefined;
         rng?: "cpu" | "cuda" | "std_default" | undefined;
@@ -10429,23 +10443,24 @@ export declare const loadModelOptionsToRequestSchema: z.ZodPipe<z.ZodUnknown, z.
         repeat_penalty: z.ZodOptional<z.ZodNumber>;
         stop_sequences: z.ZodOptional<z.ZodArray<z.ZodString>>;
         n_discarded: z.ZodOptional<z.ZodNumber>;
+        parallel: z.ZodOptional<z.ZodNumber>;
         tools: z.ZodOptional<z.ZodBoolean>;
         toolsMode: z.ZodOptional<z.ZodEnum<{
             static: "static";
             dynamic: "dynamic";
         }>>;
-        "cache-type-k": z.ZodOptional<z.ZodString>;
-        "cache-type-v": z.ZodOptional<z.ZodString>;
-        "main-gpu": z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
+        'cache-type-k': z.ZodOptional<z.ZodString>;
+        'cache-type-v': z.ZodOptional<z.ZodString>;
+        'main-gpu': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
             integrated: "integrated";
             dedicated: "dedicated";
         }>]>>;
-        "split-mode": z.ZodOptional<z.ZodEnum<{
+        'split-mode': z.ZodOptional<z.ZodEnum<{
             none: "none";
             layer: "layer";
             row: "row";
         }>>;
-        "tensor-split": z.ZodOptional<z.ZodString>;
+        'tensor-split': z.ZodOptional<z.ZodString>;
         openclCacheDir: z.ZodOptional<z.ZodString>;
         reasoning_budget: z.ZodOptional<z.ZodNumber>;
         projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -10490,6 +10505,7 @@ export declare const loadModelOptionsToRequestSchema: z.ZodPipe<z.ZodUnknown, z.
             batched: "batched";
             sequential: "sequential";
         }>>;
+        'mmproj-use-gpu': z.ZodOptional<z.ZodBoolean>;
     }, z.core.$strict>>;
     onProgress: z.ZodOptional<z.ZodUnknown>;
     logger: z.ZodOptional<z.ZodUnknown>;
@@ -10589,13 +10605,14 @@ export declare const loadModelOptionsToRequestSchema: z.ZodPipe<z.ZodUnknown, z.
         repeat_penalty?: number | undefined;
         stop_sequences?: string[] | undefined;
         n_discarded?: number | undefined;
+        parallel?: number | undefined;
         tools?: boolean | undefined;
         toolsMode?: "static" | "dynamic" | undefined;
-        "cache-type-k"?: string | undefined;
-        "cache-type-v"?: string | undefined;
-        "main-gpu"?: number | "integrated" | "dedicated" | undefined;
-        "split-mode"?: "none" | "layer" | "row" | undefined;
-        "tensor-split"?: string | undefined;
+        'cache-type-k'?: string | undefined;
+        'cache-type-v'?: string | undefined;
+        'main-gpu'?: number | "integrated" | "dedicated" | undefined;
+        'split-mode'?: "none" | "layer" | "row" | undefined;
+        'tensor-split'?: string | undefined;
         openclCacheDir?: string | undefined;
         reasoning_budget?: number | undefined;
         projectionModelSrc?: string | {
@@ -10612,6 +10629,7 @@ export declare const loadModelOptionsToRequestSchema: z.ZodPipe<z.ZodUnknown, z.
             addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
         } | undefined;
         image_tile_mode?: "disabled" | "batched" | "sequential" | undefined;
+        'mmproj-use-gpu'?: boolean | undefined;
     } | undefined;
     onProgress?: unknown;
     logger?: unknown;
@@ -11322,8 +11340,8 @@ export declare const loadModelOptionsToRequestSchema: z.ZodPipe<z.ZodUnknown, z.
     modelConfig: z.ZodOptional<z.ZodObject<{
         gpuLayers: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         batchSize: z.ZodOptional<z.ZodNumber>;
         pooling: z.ZodOptional<z.ZodEnum<{
@@ -11438,7 +11456,7 @@ export declare const loadModelOptionsToRequestSchema: z.ZodPipe<z.ZodUnknown, z.
     };
     modelConfig?: {
         gpuLayers?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         batchSize?: number | undefined;
         pooling?: "none" | "mean" | "cls" | "last" | "rank" | undefined;
         attention?: "causal" | "non-causal" | undefined;
@@ -12693,8 +12711,8 @@ export declare const loadModelOptionsToRequestSchema: z.ZodPipe<z.ZodUnknown, z.
         }>>;
         threads: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         prediction: z.ZodOptional<z.ZodEnum<{
             auto: "auto";
@@ -13129,7 +13147,7 @@ export declare const loadModelOptionsToRequestSchema: z.ZodPipe<z.ZodUnknown, z.
     modelConfig?: {
         mode: "diffusion" | "upscale" | "video";
         threads?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         prediction?: "auto" | "eps" | "v" | "edm_v" | "flow" | "flux2_flow" | undefined;
         type?: "auto" | "f32" | "f16" | "bf16" | "q2_k" | "q3_k" | "q4_0" | "q4_1" | "q4_k" | "q5_0" | "q5_1" | "q5_k" | "q6_k" | "q8_0" | undefined;
         rng?: "cpu" | "cuda" | "std_default" | undefined;
@@ -13618,23 +13636,24 @@ export declare const loadLlmModelRequestSchema: z.ZodObject<{
         repeat_penalty: z.ZodOptional<z.ZodNumber>;
         stop_sequences: z.ZodOptional<z.ZodArray<z.ZodString>>;
         n_discarded: z.ZodOptional<z.ZodNumber>;
+        parallel: z.ZodOptional<z.ZodNumber>;
         tools: z.ZodOptional<z.ZodBoolean>;
         toolsMode: z.ZodOptional<z.ZodEnum<{
             static: "static";
             dynamic: "dynamic";
         }>>;
-        "cache-type-k": z.ZodOptional<z.ZodString>;
-        "cache-type-v": z.ZodOptional<z.ZodString>;
-        "main-gpu": z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
+        'cache-type-k': z.ZodOptional<z.ZodString>;
+        'cache-type-v': z.ZodOptional<z.ZodString>;
+        'main-gpu': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
             integrated: "integrated";
             dedicated: "dedicated";
         }>]>>;
-        "split-mode": z.ZodOptional<z.ZodEnum<{
+        'split-mode': z.ZodOptional<z.ZodEnum<{
             none: "none";
             layer: "layer";
             row: "row";
         }>>;
-        "tensor-split": z.ZodOptional<z.ZodString>;
+        'tensor-split': z.ZodOptional<z.ZodString>;
         openclCacheDir: z.ZodOptional<z.ZodString>;
         reasoning_budget: z.ZodOptional<z.ZodNumber>;
         projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -13679,6 +13698,7 @@ export declare const loadLlmModelRequestSchema: z.ZodObject<{
             batched: "batched";
             sequential: "sequential";
         }>>;
+        'mmproj-use-gpu': z.ZodOptional<z.ZodBoolean>;
     }, z.core.$strip>;
 }, z.core.$strict>;
 export declare const loadWhisperModelRequestSchema: z.ZodObject<{
@@ -13918,8 +13938,8 @@ export declare const loadEmbeddingsModelRequestSchema: z.ZodObject<{
     modelConfig: z.ZodObject<{
         gpuLayers: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         batchSize: z.ZodOptional<z.ZodNumber>;
         pooling: z.ZodOptional<z.ZodEnum<{
@@ -14690,8 +14710,8 @@ export declare const loadDiffusionModelRequestSchema: z.ZodObject<{
         }>>;
         threads: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         prediction: z.ZodOptional<z.ZodEnum<{
             auto: "auto";
@@ -15138,23 +15158,24 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
         repeat_penalty: z.ZodOptional<z.ZodNumber>;
         stop_sequences: z.ZodOptional<z.ZodArray<z.ZodString>>;
         n_discarded: z.ZodOptional<z.ZodNumber>;
+        parallel: z.ZodOptional<z.ZodNumber>;
         tools: z.ZodOptional<z.ZodBoolean>;
         toolsMode: z.ZodOptional<z.ZodEnum<{
             static: "static";
             dynamic: "dynamic";
         }>>;
-        "cache-type-k": z.ZodOptional<z.ZodString>;
-        "cache-type-v": z.ZodOptional<z.ZodString>;
-        "main-gpu": z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
+        'cache-type-k': z.ZodOptional<z.ZodString>;
+        'cache-type-v': z.ZodOptional<z.ZodString>;
+        'main-gpu': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
             integrated: "integrated";
             dedicated: "dedicated";
         }>]>>;
-        "split-mode": z.ZodOptional<z.ZodEnum<{
+        'split-mode': z.ZodOptional<z.ZodEnum<{
             none: "none";
             layer: "layer";
             row: "row";
         }>>;
-        "tensor-split": z.ZodOptional<z.ZodString>;
+        'tensor-split': z.ZodOptional<z.ZodString>;
         openclCacheDir: z.ZodOptional<z.ZodString>;
         reasoning_budget: z.ZodOptional<z.ZodNumber>;
         projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -15199,6 +15220,7 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
             batched: "batched";
             sequential: "sequential";
         }>>;
+        'mmproj-use-gpu': z.ZodOptional<z.ZodBoolean>;
     }, z.core.$strip>;
 }, z.core.$strict>, z.ZodObject<{
     type: z.ZodLiteral<"loadModel">;
@@ -15434,8 +15456,8 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
     modelConfig: z.ZodObject<{
         gpuLayers: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         batchSize: z.ZodOptional<z.ZodNumber>;
         pooling: z.ZodOptional<z.ZodEnum<{
@@ -16202,8 +16224,8 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
         }>>;
         threads: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         prediction: z.ZodOptional<z.ZodEnum<{
             auto: "auto";
@@ -16636,13 +16658,14 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
         repeat_penalty?: number | undefined;
         stop_sequences?: string[] | undefined;
         n_discarded?: number | undefined;
+        parallel?: number | undefined;
         tools?: boolean | undefined;
         toolsMode?: "static" | "dynamic" | undefined;
-        "cache-type-k"?: string | undefined;
-        "cache-type-v"?: string | undefined;
-        "main-gpu"?: number | "integrated" | "dedicated" | undefined;
-        "split-mode"?: "none" | "layer" | "row" | undefined;
-        "tensor-split"?: string | undefined;
+        'cache-type-k'?: string | undefined;
+        'cache-type-v'?: string | undefined;
+        'main-gpu'?: number | "integrated" | "dedicated" | undefined;
+        'split-mode'?: "none" | "layer" | "row" | undefined;
+        'tensor-split'?: string | undefined;
         openclCacheDir?: string | undefined;
         reasoning_budget?: number | undefined;
         projectionModelSrc?: string | {
@@ -16659,6 +16682,7 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
             addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
         } | undefined;
         image_tile_mode?: "disabled" | "batched" | "sequential" | undefined;
+        'mmproj-use-gpu'?: boolean | undefined;
     };
     modelName?: string | undefined;
     withProgress?: boolean | undefined;
@@ -16837,7 +16861,7 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
     modelType: "llamacpp-embedding";
     modelConfig: {
         gpuLayers?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         batchSize?: number | undefined;
         pooling?: "none" | "mean" | "cls" | "last" | "rank" | undefined;
         attention?: "causal" | "non-causal" | undefined;
@@ -17054,7 +17078,7 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
     modelConfig?: {
         mode: "diffusion" | "upscale" | "video";
         threads?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         prediction?: "auto" | "eps" | "v" | "edm_v" | "flow" | "flux2_flow" | undefined;
         type?: "auto" | "f32" | "f16" | "bf16" | "q2_k" | "q3_k" | "q4_0" | "q4_1" | "q4_k" | "q5_0" | "q5_1" | "q5_k" | "q6_k" | "q8_0" | undefined;
         rng?: "cpu" | "cuda" | "std_default" | undefined;
@@ -17256,13 +17280,14 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
         repeat_penalty?: number | undefined;
         stop_sequences?: string[] | undefined;
         n_discarded?: number | undefined;
+        parallel?: number | undefined;
         tools?: boolean | undefined;
         toolsMode?: "static" | "dynamic" | undefined;
-        "cache-type-k"?: string | undefined;
-        "cache-type-v"?: string | undefined;
-        "main-gpu"?: number | "integrated" | "dedicated" | undefined;
-        "split-mode"?: "none" | "layer" | "row" | undefined;
-        "tensor-split"?: string | undefined;
+        'cache-type-k'?: string | undefined;
+        'cache-type-v'?: string | undefined;
+        'main-gpu'?: number | "integrated" | "dedicated" | undefined;
+        'split-mode'?: "none" | "layer" | "row" | undefined;
+        'tensor-split'?: string | undefined;
         openclCacheDir?: string | undefined;
         reasoning_budget?: number | undefined;
         projectionModelSrc?: string | {
@@ -17279,6 +17304,7 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
             addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
         } | undefined;
         image_tile_mode?: "disabled" | "batched" | "sequential" | undefined;
+        'mmproj-use-gpu'?: boolean | undefined;
     };
     modelName?: string | undefined;
     withProgress?: boolean | undefined;
@@ -17457,7 +17483,7 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
     modelType: "llamacpp-embedding";
     modelConfig: {
         gpuLayers?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         batchSize?: number | undefined;
         pooling?: "none" | "mean" | "cls" | "last" | "rank" | undefined;
         attention?: "causal" | "non-causal" | undefined;
@@ -17675,7 +17701,7 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
     modelConfig?: {
         mode: "diffusion" | "upscale" | "video";
         threads?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         prediction?: "auto" | "eps" | "v" | "edm_v" | "flow" | "flux2_flow" | undefined;
         type?: "auto" | "f32" | "f16" | "bf16" | "q2_k" | "q3_k" | "q4_0" | "q4_1" | "q4_k" | "q5_0" | "q5_1" | "q5_k" | "q6_k" | "q8_0" | undefined;
         rng?: "cpu" | "cuda" | "std_default" | undefined;
@@ -17889,23 +17915,24 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
         repeat_penalty: z.ZodOptional<z.ZodNumber>;
         stop_sequences: z.ZodOptional<z.ZodArray<z.ZodString>>;
         n_discarded: z.ZodOptional<z.ZodNumber>;
+        parallel: z.ZodOptional<z.ZodNumber>;
         tools: z.ZodOptional<z.ZodBoolean>;
         toolsMode: z.ZodOptional<z.ZodEnum<{
             static: "static";
             dynamic: "dynamic";
         }>>;
-        "cache-type-k": z.ZodOptional<z.ZodString>;
-        "cache-type-v": z.ZodOptional<z.ZodString>;
-        "main-gpu": z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
+        'cache-type-k': z.ZodOptional<z.ZodString>;
+        'cache-type-v': z.ZodOptional<z.ZodString>;
+        'main-gpu': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
             integrated: "integrated";
             dedicated: "dedicated";
         }>]>>;
-        "split-mode": z.ZodOptional<z.ZodEnum<{
+        'split-mode': z.ZodOptional<z.ZodEnum<{
             none: "none";
             layer: "layer";
             row: "row";
         }>>;
-        "tensor-split": z.ZodOptional<z.ZodString>;
+        'tensor-split': z.ZodOptional<z.ZodString>;
         openclCacheDir: z.ZodOptional<z.ZodString>;
         reasoning_budget: z.ZodOptional<z.ZodNumber>;
         projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -17950,6 +17977,7 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
             batched: "batched";
             sequential: "sequential";
         }>>;
+        'mmproj-use-gpu': z.ZodOptional<z.ZodBoolean>;
     }, z.core.$strip>;
 }, z.core.$strict>, z.ZodObject<{
     type: z.ZodLiteral<"loadModel">;
@@ -18185,8 +18213,8 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
     modelConfig: z.ZodObject<{
         gpuLayers: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         batchSize: z.ZodOptional<z.ZodNumber>;
         pooling: z.ZodOptional<z.ZodEnum<{
@@ -18953,8 +18981,8 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
         }>>;
         threads: z.ZodOptional<z.ZodNumber>;
         device: z.ZodOptional<z.ZodEnum<{
-            cpu: "cpu";
             gpu: "gpu";
+            cpu: "cpu";
         }>>;
         prediction: z.ZodOptional<z.ZodEnum<{
             auto: "auto";
@@ -19387,13 +19415,14 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
         repeat_penalty?: number | undefined;
         stop_sequences?: string[] | undefined;
         n_discarded?: number | undefined;
+        parallel?: number | undefined;
         tools?: boolean | undefined;
         toolsMode?: "static" | "dynamic" | undefined;
-        "cache-type-k"?: string | undefined;
-        "cache-type-v"?: string | undefined;
-        "main-gpu"?: number | "integrated" | "dedicated" | undefined;
-        "split-mode"?: "none" | "layer" | "row" | undefined;
-        "tensor-split"?: string | undefined;
+        'cache-type-k'?: string | undefined;
+        'cache-type-v'?: string | undefined;
+        'main-gpu'?: number | "integrated" | "dedicated" | undefined;
+        'split-mode'?: "none" | "layer" | "row" | undefined;
+        'tensor-split'?: string | undefined;
         openclCacheDir?: string | undefined;
         reasoning_budget?: number | undefined;
         projectionModelSrc?: string | {
@@ -19410,6 +19439,7 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
             addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
         } | undefined;
         image_tile_mode?: "disabled" | "batched" | "sequential" | undefined;
+        'mmproj-use-gpu'?: boolean | undefined;
     };
     modelName?: string | undefined;
     withProgress?: boolean | undefined;
@@ -19588,7 +19618,7 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
     modelType: "llamacpp-embedding";
     modelConfig: {
         gpuLayers?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         batchSize?: number | undefined;
         pooling?: "none" | "mean" | "cls" | "last" | "rank" | undefined;
         attention?: "causal" | "non-causal" | undefined;
@@ -19805,7 +19835,7 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
     modelConfig?: {
         mode: "diffusion" | "upscale" | "video";
         threads?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         prediction?: "auto" | "eps" | "v" | "edm_v" | "flow" | "flux2_flow" | undefined;
         type?: "auto" | "f32" | "f16" | "bf16" | "q2_k" | "q3_k" | "q4_0" | "q4_1" | "q4_k" | "q5_0" | "q5_1" | "q5_k" | "q6_k" | "q8_0" | undefined;
         rng?: "cpu" | "cuda" | "std_default" | undefined;
@@ -20007,13 +20037,14 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
         repeat_penalty?: number | undefined;
         stop_sequences?: string[] | undefined;
         n_discarded?: number | undefined;
+        parallel?: number | undefined;
         tools?: boolean | undefined;
         toolsMode?: "static" | "dynamic" | undefined;
-        "cache-type-k"?: string | undefined;
-        "cache-type-v"?: string | undefined;
-        "main-gpu"?: number | "integrated" | "dedicated" | undefined;
-        "split-mode"?: "none" | "layer" | "row" | undefined;
-        "tensor-split"?: string | undefined;
+        'cache-type-k'?: string | undefined;
+        'cache-type-v'?: string | undefined;
+        'main-gpu'?: number | "integrated" | "dedicated" | undefined;
+        'split-mode'?: "none" | "layer" | "row" | undefined;
+        'tensor-split'?: string | undefined;
         openclCacheDir?: string | undefined;
         reasoning_budget?: number | undefined;
         projectionModelSrc?: string | {
@@ -20030,6 +20061,7 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
             addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
         } | undefined;
         image_tile_mode?: "disabled" | "batched" | "sequential" | undefined;
+        'mmproj-use-gpu'?: boolean | undefined;
     };
     modelName?: string | undefined;
     withProgress?: boolean | undefined;
@@ -20208,7 +20240,7 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
     modelType: "llamacpp-embedding";
     modelConfig: {
         gpuLayers?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         batchSize?: number | undefined;
         pooling?: "none" | "mean" | "cls" | "last" | "rank" | undefined;
         attention?: "causal" | "non-causal" | undefined;
@@ -20426,7 +20458,7 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
     modelConfig?: {
         mode: "diffusion" | "upscale" | "video";
         threads?: number | undefined;
-        device?: "cpu" | "gpu" | undefined;
+        device?: "gpu" | "cpu" | undefined;
         prediction?: "auto" | "eps" | "v" | "edm_v" | "flow" | "flux2_flow" | undefined;
         type?: "auto" | "f32" | "f16" | "bf16" | "q2_k" | "q3_k" | "q4_0" | "q4_1" | "q4_k" | "q5_0" | "q5_1" | "q5_k" | "q6_k" | "q8_0" | undefined;
         rng?: "cpu" | "cuda" | "std_default" | undefined;
@@ -20799,7 +20831,7 @@ export type LoadModelOptions = z.input<typeof loadBuiltinModelOptionsBaseSchema>
  * `loadModel` options for custom plugin model types (any non-built-in string).
  * Built-in model types use {@link LoadModelOptions}.
  */
-export type LoadCustomPluginModelOptions<T extends string> = Omit<z.input<typeof loadCustomPluginModelOptionsBaseSchema>, "modelType"> & {
+export type LoadCustomPluginModelOptions<T extends string> = Omit<z.input<typeof loadCustomPluginModelOptionsBaseSchema>, 'modelType'> & {
     modelType: T extends ModelTypeInput ? never : T;
     onProgress?: (progress: ModelProgressUpdate) => void;
     logger?: Logger;
@@ -20863,7 +20895,7 @@ export type LoadModelDescriptorInferredOptions<S extends ModelDescriptor> = {
  */
 export type LoadModelDescriptorParam<S extends ModelDescriptor> = S extends {
     engine: CanonicalModelType;
-} ? LoadModelDescriptorInferredOptions<S> : Omit<LoadModelDescriptorOnlyOptions, "modelSrc"> & {
+} ? LoadModelDescriptorInferredOptions<S> : Omit<LoadModelDescriptorOnlyOptions, 'modelSrc'> & {
     modelSrc: S;
 };
 //# sourceMappingURL=load-model.d.ts.map