@qvac/inference 0.20.0 → 0.20.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (27) hide show
  1. package/NOTICE +19 -11
  2. package/dist/api/assess-model-fit.d.ts +9 -4
  3. package/dist/api/assess-model-fit.js +9 -4
  4. package/dist/handlers/assess-model-fit.d.ts +30 -6
  5. package/dist/handlers/assess-model-fit.js +68 -6
  6. package/dist/handlers/load-model/registry-download-utils.d.ts +1 -0
  7. package/dist/handlers/load-model/registry-download-utils.js +2 -1
  8. package/dist/plugins/builtin/llamacpp-completion/plugin.d.ts +61 -0
  9. package/dist/plugins/builtin/llamacpp-completion/transform.js +19 -0
  10. package/dist/resources/model-fit/assess.d.ts +10 -1
  11. package/dist/resources/model-fit/assess.js +43 -0
  12. package/dist/resources/model-fit/fit-stub/fetch-fit-stub.d.ts +83 -0
  13. package/dist/resources/model-fit/fit-stub/fetch-fit-stub.js +133 -0
  14. package/dist/resources/model-fit/fit-stub/index.d.ts +2 -0
  15. package/dist/resources/model-fit/fit-stub/index.js +2 -0
  16. package/dist/resources/model-fit/fit-stub/project-fit-from-stub.d.ts +31 -0
  17. package/dist/resources/model-fit/fit-stub/project-fit-from-stub.js +36 -0
  18. package/dist/resources/model-fit/native-probe/create-llama-fit-request.js +38 -18
  19. package/dist/schemas/assess-model-fit.d.ts +23 -0
  20. package/dist/schemas/assess-model-fit.js +16 -2
  21. package/dist/schemas/common.d.ts +67 -0
  22. package/dist/schemas/config.d.ts +138 -0
  23. package/dist/schemas/llamacpp-config.d.ts +84 -0
  24. package/dist/schemas/llamacpp-config.js +104 -4
  25. package/dist/schemas/load-model.d.ts +336 -0
  26. package/dist/schemas/model-config-schemas.d.ts +23 -0
  27. package/package.json +2 -2
@@ -65,6 +65,29 @@ export declare const loadBuiltinModelOptionsBaseSchema: z.ZodUnion<readonly [z.Z
65
65
  auto: "auto";
66
66
  }>>;
67
67
  'tensor-split': z.ZodOptional<z.ZodString>;
68
+ 'batch-size': z.ZodOptional<z.ZodNumber>;
69
+ 'ubatch-size': z.ZodOptional<z.ZodNumber>;
70
+ 'cpu-moe': z.ZodOptional<z.ZodBoolean>;
71
+ 'n-cpu-moe': z.ZodOptional<z.ZodNumber>;
72
+ 'kv-offload': z.ZodOptional<z.ZodBoolean>;
73
+ 'image-max-tokens': z.ZodOptional<z.ZodNumber>;
74
+ 'image-min-tokens': z.ZodOptional<z.ZodNumber>;
75
+ threads: z.ZodOptional<z.ZodNumber>;
76
+ 'threads-batch': z.ZodOptional<z.ZodNumber>;
77
+ 'cpu-mask': z.ZodOptional<z.ZodString>;
78
+ 'cpu-mask-batch': z.ZodOptional<z.ZodString>;
79
+ 'override-tensor': z.ZodOptional<z.ZodString>;
80
+ 'n-cpu-ffn': z.ZodOptional<z.ZodNumber>;
81
+ 'moe-cache-mib': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodLiteral<"auto">]>>;
82
+ 'prefetch-weights': z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodLiteral<"auto">]>>;
83
+ 'tensor-read-lazy': z.ZodOptional<z.ZodEnum<{
84
+ on: "on";
85
+ off: "off";
86
+ auto: "auto";
87
+ }>>;
88
+ fit: z.ZodOptional<z.ZodBoolean>;
89
+ 'fit-target': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>>;
90
+ 'fit-ctx': z.ZodOptional<z.ZodNumber>;
68
91
  openclCacheDir: z.ZodOptional<z.ZodString>;
69
92
  reasoning_budget: z.ZodOptional<z.ZodNumber>;
70
93
  projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -3043,6 +3066,29 @@ export declare const loadModelOptionsBaseSchema: z.ZodUnion<readonly [z.ZodUnion
3043
3066
  auto: "auto";
3044
3067
  }>>;
3045
3068
  'tensor-split': z.ZodOptional<z.ZodString>;
3069
+ 'batch-size': z.ZodOptional<z.ZodNumber>;
3070
+ 'ubatch-size': z.ZodOptional<z.ZodNumber>;
3071
+ 'cpu-moe': z.ZodOptional<z.ZodBoolean>;
3072
+ 'n-cpu-moe': z.ZodOptional<z.ZodNumber>;
3073
+ 'kv-offload': z.ZodOptional<z.ZodBoolean>;
3074
+ 'image-max-tokens': z.ZodOptional<z.ZodNumber>;
3075
+ 'image-min-tokens': z.ZodOptional<z.ZodNumber>;
3076
+ threads: z.ZodOptional<z.ZodNumber>;
3077
+ 'threads-batch': z.ZodOptional<z.ZodNumber>;
3078
+ 'cpu-mask': z.ZodOptional<z.ZodString>;
3079
+ 'cpu-mask-batch': z.ZodOptional<z.ZodString>;
3080
+ 'override-tensor': z.ZodOptional<z.ZodString>;
3081
+ 'n-cpu-ffn': z.ZodOptional<z.ZodNumber>;
3082
+ 'moe-cache-mib': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodLiteral<"auto">]>>;
3083
+ 'prefetch-weights': z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodLiteral<"auto">]>>;
3084
+ 'tensor-read-lazy': z.ZodOptional<z.ZodEnum<{
3085
+ on: "on";
3086
+ off: "off";
3087
+ auto: "auto";
3088
+ }>>;
3089
+ fit: z.ZodOptional<z.ZodBoolean>;
3090
+ 'fit-target': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>>;
3091
+ 'fit-ctx': z.ZodOptional<z.ZodNumber>;
3046
3092
  openclCacheDir: z.ZodOptional<z.ZodString>;
3047
3093
  reasoning_budget: z.ZodOptional<z.ZodNumber>;
3048
3094
  projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -6020,6 +6066,29 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
6020
6066
  auto: "auto";
6021
6067
  }>>;
6022
6068
  'tensor-split': z.ZodOptional<z.ZodString>;
6069
+ 'batch-size': z.ZodOptional<z.ZodNumber>;
6070
+ 'ubatch-size': z.ZodOptional<z.ZodNumber>;
6071
+ 'cpu-moe': z.ZodOptional<z.ZodBoolean>;
6072
+ 'n-cpu-moe': z.ZodOptional<z.ZodNumber>;
6073
+ 'kv-offload': z.ZodOptional<z.ZodBoolean>;
6074
+ 'image-max-tokens': z.ZodOptional<z.ZodNumber>;
6075
+ 'image-min-tokens': z.ZodOptional<z.ZodNumber>;
6076
+ threads: z.ZodOptional<z.ZodNumber>;
6077
+ 'threads-batch': z.ZodOptional<z.ZodNumber>;
6078
+ 'cpu-mask': z.ZodOptional<z.ZodString>;
6079
+ 'cpu-mask-batch': z.ZodOptional<z.ZodString>;
6080
+ 'override-tensor': z.ZodOptional<z.ZodString>;
6081
+ 'n-cpu-ffn': z.ZodOptional<z.ZodNumber>;
6082
+ 'moe-cache-mib': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodLiteral<"auto">]>>;
6083
+ 'prefetch-weights': z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodLiteral<"auto">]>>;
6084
+ 'tensor-read-lazy': z.ZodOptional<z.ZodEnum<{
6085
+ on: "on";
6086
+ off: "off";
6087
+ auto: "auto";
6088
+ }>>;
6089
+ fit: z.ZodOptional<z.ZodBoolean>;
6090
+ 'fit-target': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>>;
6091
+ 'fit-ctx': z.ZodOptional<z.ZodNumber>;
6023
6092
  openclCacheDir: z.ZodOptional<z.ZodString>;
6024
6093
  reasoning_budget: z.ZodOptional<z.ZodNumber>;
6025
6094
  projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -8985,6 +9054,25 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
8985
9054
  'split-mode'?: "none" | "layer" | "tensor" | undefined;
8986
9055
  'flash-attn'?: "on" | "off" | "auto" | undefined;
8987
9056
  'tensor-split'?: string | undefined;
9057
+ 'batch-size'?: number | undefined;
9058
+ 'ubatch-size'?: number | undefined;
9059
+ 'cpu-moe'?: boolean | undefined;
9060
+ 'n-cpu-moe'?: number | undefined;
9061
+ 'kv-offload'?: boolean | undefined;
9062
+ 'image-max-tokens'?: number | undefined;
9063
+ 'image-min-tokens'?: number | undefined;
9064
+ threads?: number | undefined;
9065
+ 'threads-batch'?: number | undefined;
9066
+ 'cpu-mask'?: string | undefined;
9067
+ 'cpu-mask-batch'?: string | undefined;
9068
+ 'override-tensor'?: string | undefined;
9069
+ 'n-cpu-ffn'?: number | undefined;
9070
+ 'moe-cache-mib'?: number | "auto" | undefined;
9071
+ 'prefetch-weights'?: boolean | "auto" | undefined;
9072
+ 'tensor-read-lazy'?: "on" | "off" | "auto" | undefined;
9073
+ fit?: boolean | undefined;
9074
+ 'fit-target'?: string | number | undefined;
9075
+ 'fit-ctx'?: number | undefined;
8988
9076
  openclCacheDir?: string | undefined;
8989
9077
  reasoning_budget?: number | undefined;
8990
9078
  projectionModelSrc?: string | {
@@ -10052,6 +10140,25 @@ export declare const loadModelOptionsSchema: z.ZodPipe<z.ZodUnion<readonly [z.Zo
10052
10140
  'split-mode'?: "none" | "layer" | "tensor" | undefined;
10053
10141
  'flash-attn'?: "on" | "off" | "auto" | undefined;
10054
10142
  'tensor-split'?: string | undefined;
10143
+ 'batch-size'?: number | undefined;
10144
+ 'ubatch-size'?: number | undefined;
10145
+ 'cpu-moe'?: boolean | undefined;
10146
+ 'n-cpu-moe'?: number | undefined;
10147
+ 'kv-offload'?: boolean | undefined;
10148
+ 'image-max-tokens'?: number | undefined;
10149
+ 'image-min-tokens'?: number | undefined;
10150
+ threads?: number | undefined;
10151
+ 'threads-batch'?: number | undefined;
10152
+ 'cpu-mask'?: string | undefined;
10153
+ 'cpu-mask-batch'?: string | undefined;
10154
+ 'override-tensor'?: string | undefined;
10155
+ 'n-cpu-ffn'?: number | undefined;
10156
+ 'moe-cache-mib'?: number | "auto" | undefined;
10157
+ 'prefetch-weights'?: boolean | "auto" | undefined;
10158
+ 'tensor-read-lazy'?: "on" | "off" | "auto" | undefined;
10159
+ fit?: boolean | undefined;
10160
+ 'fit-target'?: string | number | undefined;
10161
+ 'fit-ctx'?: number | undefined;
10055
10162
  openclCacheDir?: string | undefined;
10056
10163
  reasoning_budget?: number | undefined;
10057
10164
  projectionModelSrc?: string | {
@@ -11133,6 +11240,29 @@ export declare const loadBuiltinToRequestSchema: z.ZodDiscriminatedUnion<[z.ZodP
11133
11240
  auto: "auto";
11134
11241
  }>>;
11135
11242
  'tensor-split': z.ZodOptional<z.ZodString>;
11243
+ 'batch-size': z.ZodOptional<z.ZodNumber>;
11244
+ 'ubatch-size': z.ZodOptional<z.ZodNumber>;
11245
+ 'cpu-moe': z.ZodOptional<z.ZodBoolean>;
11246
+ 'n-cpu-moe': z.ZodOptional<z.ZodNumber>;
11247
+ 'kv-offload': z.ZodOptional<z.ZodBoolean>;
11248
+ 'image-max-tokens': z.ZodOptional<z.ZodNumber>;
11249
+ 'image-min-tokens': z.ZodOptional<z.ZodNumber>;
11250
+ threads: z.ZodOptional<z.ZodNumber>;
11251
+ 'threads-batch': z.ZodOptional<z.ZodNumber>;
11252
+ 'cpu-mask': z.ZodOptional<z.ZodString>;
11253
+ 'cpu-mask-batch': z.ZodOptional<z.ZodString>;
11254
+ 'override-tensor': z.ZodOptional<z.ZodString>;
11255
+ 'n-cpu-ffn': z.ZodOptional<z.ZodNumber>;
11256
+ 'moe-cache-mib': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodLiteral<"auto">]>>;
11257
+ 'prefetch-weights': z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodLiteral<"auto">]>>;
11258
+ 'tensor-read-lazy': z.ZodOptional<z.ZodEnum<{
11259
+ on: "on";
11260
+ off: "off";
11261
+ auto: "auto";
11262
+ }>>;
11263
+ fit: z.ZodOptional<z.ZodBoolean>;
11264
+ 'fit-target': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>>;
11265
+ 'fit-ctx': z.ZodOptional<z.ZodNumber>;
11136
11266
  openclCacheDir: z.ZodOptional<z.ZodString>;
11137
11267
  reasoning_budget: z.ZodOptional<z.ZodNumber>;
11138
11268
  projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -11284,6 +11414,25 @@ export declare const loadBuiltinToRequestSchema: z.ZodDiscriminatedUnion<[z.ZodP
11284
11414
  'split-mode'?: "none" | "layer" | "tensor" | undefined;
11285
11415
  'flash-attn'?: "on" | "off" | "auto" | undefined;
11286
11416
  'tensor-split'?: string | undefined;
11417
+ 'batch-size'?: number | undefined;
11418
+ 'ubatch-size'?: number | undefined;
11419
+ 'cpu-moe'?: boolean | undefined;
11420
+ 'n-cpu-moe'?: number | undefined;
11421
+ 'kv-offload'?: boolean | undefined;
11422
+ 'image-max-tokens'?: number | undefined;
11423
+ 'image-min-tokens'?: number | undefined;
11424
+ threads?: number | undefined;
11425
+ 'threads-batch'?: number | undefined;
11426
+ 'cpu-mask'?: string | undefined;
11427
+ 'cpu-mask-batch'?: string | undefined;
11428
+ 'override-tensor'?: string | undefined;
11429
+ 'n-cpu-ffn'?: number | undefined;
11430
+ 'moe-cache-mib'?: number | "auto" | undefined;
11431
+ 'prefetch-weights'?: boolean | "auto" | undefined;
11432
+ 'tensor-read-lazy'?: "on" | "off" | "auto" | undefined;
11433
+ fit?: boolean | undefined;
11434
+ 'fit-target'?: string | number | undefined;
11435
+ 'fit-ctx'?: number | undefined;
11287
11436
  openclCacheDir?: string | undefined;
11288
11437
  reasoning_budget?: number | undefined;
11289
11438
  projectionModelSrc?: string | {
@@ -15980,6 +16129,29 @@ export declare const loadModelOptionsToRequestSchema: z.ZodPipe<z.ZodUnknown, z.
15980
16129
  auto: "auto";
15981
16130
  }>>;
15982
16131
  'tensor-split': z.ZodOptional<z.ZodString>;
16132
+ 'batch-size': z.ZodOptional<z.ZodNumber>;
16133
+ 'ubatch-size': z.ZodOptional<z.ZodNumber>;
16134
+ 'cpu-moe': z.ZodOptional<z.ZodBoolean>;
16135
+ 'n-cpu-moe': z.ZodOptional<z.ZodNumber>;
16136
+ 'kv-offload': z.ZodOptional<z.ZodBoolean>;
16137
+ 'image-max-tokens': z.ZodOptional<z.ZodNumber>;
16138
+ 'image-min-tokens': z.ZodOptional<z.ZodNumber>;
16139
+ threads: z.ZodOptional<z.ZodNumber>;
16140
+ 'threads-batch': z.ZodOptional<z.ZodNumber>;
16141
+ 'cpu-mask': z.ZodOptional<z.ZodString>;
16142
+ 'cpu-mask-batch': z.ZodOptional<z.ZodString>;
16143
+ 'override-tensor': z.ZodOptional<z.ZodString>;
16144
+ 'n-cpu-ffn': z.ZodOptional<z.ZodNumber>;
16145
+ 'moe-cache-mib': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodLiteral<"auto">]>>;
16146
+ 'prefetch-weights': z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodLiteral<"auto">]>>;
16147
+ 'tensor-read-lazy': z.ZodOptional<z.ZodEnum<{
16148
+ on: "on";
16149
+ off: "off";
16150
+ auto: "auto";
16151
+ }>>;
16152
+ fit: z.ZodOptional<z.ZodBoolean>;
16153
+ 'fit-target': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>>;
16154
+ 'fit-ctx': z.ZodOptional<z.ZodNumber>;
15983
16155
  openclCacheDir: z.ZodOptional<z.ZodString>;
15984
16156
  reasoning_budget: z.ZodOptional<z.ZodNumber>;
15985
16157
  projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -16131,6 +16303,25 @@ export declare const loadModelOptionsToRequestSchema: z.ZodPipe<z.ZodUnknown, z.
16131
16303
  'split-mode'?: "none" | "layer" | "tensor" | undefined;
16132
16304
  'flash-attn'?: "on" | "off" | "auto" | undefined;
16133
16305
  'tensor-split'?: string | undefined;
16306
+ 'batch-size'?: number | undefined;
16307
+ 'ubatch-size'?: number | undefined;
16308
+ 'cpu-moe'?: boolean | undefined;
16309
+ 'n-cpu-moe'?: number | undefined;
16310
+ 'kv-offload'?: boolean | undefined;
16311
+ 'image-max-tokens'?: number | undefined;
16312
+ 'image-min-tokens'?: number | undefined;
16313
+ threads?: number | undefined;
16314
+ 'threads-batch'?: number | undefined;
16315
+ 'cpu-mask'?: string | undefined;
16316
+ 'cpu-mask-batch'?: string | undefined;
16317
+ 'override-tensor'?: string | undefined;
16318
+ 'n-cpu-ffn'?: number | undefined;
16319
+ 'moe-cache-mib'?: number | "auto" | undefined;
16320
+ 'prefetch-weights'?: boolean | "auto" | undefined;
16321
+ 'tensor-read-lazy'?: "on" | "off" | "auto" | undefined;
16322
+ fit?: boolean | undefined;
16323
+ 'fit-target'?: string | number | undefined;
16324
+ 'fit-ctx'?: number | undefined;
16134
16325
  openclCacheDir?: string | undefined;
16135
16326
  reasoning_budget?: number | undefined;
16136
16327
  projectionModelSrc?: string | {
@@ -20832,6 +21023,29 @@ export declare const loadLlmModelRequestSchema: z.ZodObject<{
20832
21023
  auto: "auto";
20833
21024
  }>>;
20834
21025
  'tensor-split': z.ZodOptional<z.ZodString>;
21026
+ 'batch-size': z.ZodOptional<z.ZodNumber>;
21027
+ 'ubatch-size': z.ZodOptional<z.ZodNumber>;
21028
+ 'cpu-moe': z.ZodOptional<z.ZodBoolean>;
21029
+ 'n-cpu-moe': z.ZodOptional<z.ZodNumber>;
21030
+ 'kv-offload': z.ZodOptional<z.ZodBoolean>;
21031
+ 'image-max-tokens': z.ZodOptional<z.ZodNumber>;
21032
+ 'image-min-tokens': z.ZodOptional<z.ZodNumber>;
21033
+ threads: z.ZodOptional<z.ZodNumber>;
21034
+ 'threads-batch': z.ZodOptional<z.ZodNumber>;
21035
+ 'cpu-mask': z.ZodOptional<z.ZodString>;
21036
+ 'cpu-mask-batch': z.ZodOptional<z.ZodString>;
21037
+ 'override-tensor': z.ZodOptional<z.ZodString>;
21038
+ 'n-cpu-ffn': z.ZodOptional<z.ZodNumber>;
21039
+ 'moe-cache-mib': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodLiteral<"auto">]>>;
21040
+ 'prefetch-weights': z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodLiteral<"auto">]>>;
21041
+ 'tensor-read-lazy': z.ZodOptional<z.ZodEnum<{
21042
+ on: "on";
21043
+ off: "off";
21044
+ auto: "auto";
21045
+ }>>;
21046
+ fit: z.ZodOptional<z.ZodBoolean>;
21047
+ 'fit-target': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>>;
21048
+ 'fit-ctx': z.ZodOptional<z.ZodNumber>;
20835
21049
  openclCacheDir: z.ZodOptional<z.ZodString>;
20836
21050
  reasoning_budget: z.ZodOptional<z.ZodNumber>;
20837
21051
  projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -23343,6 +23557,29 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
23343
23557
  auto: "auto";
23344
23558
  }>>;
23345
23559
  'tensor-split': z.ZodOptional<z.ZodString>;
23560
+ 'batch-size': z.ZodOptional<z.ZodNumber>;
23561
+ 'ubatch-size': z.ZodOptional<z.ZodNumber>;
23562
+ 'cpu-moe': z.ZodOptional<z.ZodBoolean>;
23563
+ 'n-cpu-moe': z.ZodOptional<z.ZodNumber>;
23564
+ 'kv-offload': z.ZodOptional<z.ZodBoolean>;
23565
+ 'image-max-tokens': z.ZodOptional<z.ZodNumber>;
23566
+ 'image-min-tokens': z.ZodOptional<z.ZodNumber>;
23567
+ threads: z.ZodOptional<z.ZodNumber>;
23568
+ 'threads-batch': z.ZodOptional<z.ZodNumber>;
23569
+ 'cpu-mask': z.ZodOptional<z.ZodString>;
23570
+ 'cpu-mask-batch': z.ZodOptional<z.ZodString>;
23571
+ 'override-tensor': z.ZodOptional<z.ZodString>;
23572
+ 'n-cpu-ffn': z.ZodOptional<z.ZodNumber>;
23573
+ 'moe-cache-mib': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodLiteral<"auto">]>>;
23574
+ 'prefetch-weights': z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodLiteral<"auto">]>>;
23575
+ 'tensor-read-lazy': z.ZodOptional<z.ZodEnum<{
23576
+ on: "on";
23577
+ off: "off";
23578
+ auto: "auto";
23579
+ }>>;
23580
+ fit: z.ZodOptional<z.ZodBoolean>;
23581
+ 'fit-target': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>>;
23582
+ 'fit-ctx': z.ZodOptional<z.ZodNumber>;
23346
23583
  openclCacheDir: z.ZodOptional<z.ZodString>;
23347
23584
  reasoning_budget: z.ZodOptional<z.ZodNumber>;
23348
23585
  projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -25813,6 +26050,25 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
25813
26050
  'split-mode'?: "none" | "layer" | "tensor" | undefined;
25814
26051
  'flash-attn'?: "on" | "off" | "auto" | undefined;
25815
26052
  'tensor-split'?: string | undefined;
26053
+ 'batch-size'?: number | undefined;
26054
+ 'ubatch-size'?: number | undefined;
26055
+ 'cpu-moe'?: boolean | undefined;
26056
+ 'n-cpu-moe'?: number | undefined;
26057
+ 'kv-offload'?: boolean | undefined;
26058
+ 'image-max-tokens'?: number | undefined;
26059
+ 'image-min-tokens'?: number | undefined;
26060
+ threads?: number | undefined;
26061
+ 'threads-batch'?: number | undefined;
26062
+ 'cpu-mask'?: string | undefined;
26063
+ 'cpu-mask-batch'?: string | undefined;
26064
+ 'override-tensor'?: string | undefined;
26065
+ 'n-cpu-ffn'?: number | undefined;
26066
+ 'moe-cache-mib'?: number | "auto" | undefined;
26067
+ 'prefetch-weights'?: boolean | "auto" | undefined;
26068
+ 'tensor-read-lazy'?: "on" | "off" | "auto" | undefined;
26069
+ fit?: boolean | undefined;
26070
+ 'fit-target'?: string | number | undefined;
26071
+ 'fit-ctx'?: number | undefined;
25816
26072
  openclCacheDir?: string | undefined;
25817
26073
  reasoning_budget?: number | undefined;
25818
26074
  projectionModelSrc?: string | {
@@ -26776,6 +27032,25 @@ export declare const loadModelSrcRequestSchema: z.ZodPipe<z.ZodUnion<readonly [z
26776
27032
  'split-mode'?: "none" | "layer" | "tensor" | undefined;
26777
27033
  'flash-attn'?: "on" | "off" | "auto" | undefined;
26778
27034
  'tensor-split'?: string | undefined;
27035
+ 'batch-size'?: number | undefined;
27036
+ 'ubatch-size'?: number | undefined;
27037
+ 'cpu-moe'?: boolean | undefined;
27038
+ 'n-cpu-moe'?: number | undefined;
27039
+ 'kv-offload'?: boolean | undefined;
27040
+ 'image-max-tokens'?: number | undefined;
27041
+ 'image-min-tokens'?: number | undefined;
27042
+ threads?: number | undefined;
27043
+ 'threads-batch'?: number | undefined;
27044
+ 'cpu-mask'?: string | undefined;
27045
+ 'cpu-mask-batch'?: string | undefined;
27046
+ 'override-tensor'?: string | undefined;
27047
+ 'n-cpu-ffn'?: number | undefined;
27048
+ 'moe-cache-mib'?: number | "auto" | undefined;
27049
+ 'prefetch-weights'?: boolean | "auto" | undefined;
27050
+ 'tensor-read-lazy'?: "on" | "off" | "auto" | undefined;
27051
+ fit?: boolean | undefined;
27052
+ 'fit-target'?: string | number | undefined;
27053
+ 'fit-ctx'?: number | undefined;
26779
27054
  openclCacheDir?: string | undefined;
26780
27055
  reasoning_budget?: number | undefined;
26781
27056
  projectionModelSrc?: string | {
@@ -27770,6 +28045,29 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
27770
28045
  auto: "auto";
27771
28046
  }>>;
27772
28047
  'tensor-split': z.ZodOptional<z.ZodString>;
28048
+ 'batch-size': z.ZodOptional<z.ZodNumber>;
28049
+ 'ubatch-size': z.ZodOptional<z.ZodNumber>;
28050
+ 'cpu-moe': z.ZodOptional<z.ZodBoolean>;
28051
+ 'n-cpu-moe': z.ZodOptional<z.ZodNumber>;
28052
+ 'kv-offload': z.ZodOptional<z.ZodBoolean>;
28053
+ 'image-max-tokens': z.ZodOptional<z.ZodNumber>;
28054
+ 'image-min-tokens': z.ZodOptional<z.ZodNumber>;
28055
+ threads: z.ZodOptional<z.ZodNumber>;
28056
+ 'threads-batch': z.ZodOptional<z.ZodNumber>;
28057
+ 'cpu-mask': z.ZodOptional<z.ZodString>;
28058
+ 'cpu-mask-batch': z.ZodOptional<z.ZodString>;
28059
+ 'override-tensor': z.ZodOptional<z.ZodString>;
28060
+ 'n-cpu-ffn': z.ZodOptional<z.ZodNumber>;
28061
+ 'moe-cache-mib': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodLiteral<"auto">]>>;
28062
+ 'prefetch-weights': z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodLiteral<"auto">]>>;
28063
+ 'tensor-read-lazy': z.ZodOptional<z.ZodEnum<{
28064
+ on: "on";
28065
+ off: "off";
28066
+ auto: "auto";
28067
+ }>>;
28068
+ fit: z.ZodOptional<z.ZodBoolean>;
28069
+ 'fit-target': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>>;
28070
+ 'fit-ctx': z.ZodOptional<z.ZodNumber>;
27773
28071
  openclCacheDir: z.ZodOptional<z.ZodString>;
27774
28072
  reasoning_budget: z.ZodOptional<z.ZodNumber>;
27775
28073
  projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
@@ -30240,6 +30538,25 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
30240
30538
  'split-mode'?: "none" | "layer" | "tensor" | undefined;
30241
30539
  'flash-attn'?: "on" | "off" | "auto" | undefined;
30242
30540
  'tensor-split'?: string | undefined;
30541
+ 'batch-size'?: number | undefined;
30542
+ 'ubatch-size'?: number | undefined;
30543
+ 'cpu-moe'?: boolean | undefined;
30544
+ 'n-cpu-moe'?: number | undefined;
30545
+ 'kv-offload'?: boolean | undefined;
30546
+ 'image-max-tokens'?: number | undefined;
30547
+ 'image-min-tokens'?: number | undefined;
30548
+ threads?: number | undefined;
30549
+ 'threads-batch'?: number | undefined;
30550
+ 'cpu-mask'?: string | undefined;
30551
+ 'cpu-mask-batch'?: string | undefined;
30552
+ 'override-tensor'?: string | undefined;
30553
+ 'n-cpu-ffn'?: number | undefined;
30554
+ 'moe-cache-mib'?: number | "auto" | undefined;
30555
+ 'prefetch-weights'?: boolean | "auto" | undefined;
30556
+ 'tensor-read-lazy'?: "on" | "off" | "auto" | undefined;
30557
+ fit?: boolean | undefined;
30558
+ 'fit-target'?: string | number | undefined;
30559
+ 'fit-ctx'?: number | undefined;
30243
30560
  openclCacheDir?: string | undefined;
30244
30561
  reasoning_budget?: number | undefined;
30245
30562
  projectionModelSrc?: string | {
@@ -31203,6 +31520,25 @@ export declare const loadModelRequestSchema: z.ZodUnion<readonly [z.ZodPipe<z.Zo
31203
31520
  'split-mode'?: "none" | "layer" | "tensor" | undefined;
31204
31521
  'flash-attn'?: "on" | "off" | "auto" | undefined;
31205
31522
  'tensor-split'?: string | undefined;
31523
+ 'batch-size'?: number | undefined;
31524
+ 'ubatch-size'?: number | undefined;
31525
+ 'cpu-moe'?: boolean | undefined;
31526
+ 'n-cpu-moe'?: number | undefined;
31527
+ 'kv-offload'?: boolean | undefined;
31528
+ 'image-max-tokens'?: number | undefined;
31529
+ 'image-min-tokens'?: number | undefined;
31530
+ threads?: number | undefined;
31531
+ 'threads-batch'?: number | undefined;
31532
+ 'cpu-mask'?: string | undefined;
31533
+ 'cpu-mask-batch'?: string | undefined;
31534
+ 'override-tensor'?: string | undefined;
31535
+ 'n-cpu-ffn'?: number | undefined;
31536
+ 'moe-cache-mib'?: number | "auto" | undefined;
31537
+ 'prefetch-weights'?: boolean | "auto" | undefined;
31538
+ 'tensor-read-lazy'?: "on" | "off" | "auto" | undefined;
31539
+ fit?: boolean | undefined;
31540
+ 'fit-target'?: string | number | undefined;
31541
+ 'fit-ctx'?: number | undefined;
31206
31542
  openclCacheDir?: string | undefined;
31207
31543
  reasoning_budget?: number | undefined;
31208
31544
  projectionModelSrc?: string | {
@@ -47,6 +47,29 @@ export declare const MODEL_CONFIG_SCHEMA_BY_TYPE: {
47
47
  auto: "auto";
48
48
  }>>;
49
49
  'tensor-split': z.ZodOptional<z.ZodString>;
50
+ 'batch-size': z.ZodOptional<z.ZodNumber>;
51
+ 'ubatch-size': z.ZodOptional<z.ZodNumber>;
52
+ 'cpu-moe': z.ZodOptional<z.ZodBoolean>;
53
+ 'n-cpu-moe': z.ZodOptional<z.ZodNumber>;
54
+ 'kv-offload': z.ZodOptional<z.ZodBoolean>;
55
+ 'image-max-tokens': z.ZodOptional<z.ZodNumber>;
56
+ 'image-min-tokens': z.ZodOptional<z.ZodNumber>;
57
+ threads: z.ZodOptional<z.ZodNumber>;
58
+ 'threads-batch': z.ZodOptional<z.ZodNumber>;
59
+ 'cpu-mask': z.ZodOptional<z.ZodString>;
60
+ 'cpu-mask-batch': z.ZodOptional<z.ZodString>;
61
+ 'override-tensor': z.ZodOptional<z.ZodString>;
62
+ 'n-cpu-ffn': z.ZodOptional<z.ZodNumber>;
63
+ 'moe-cache-mib': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodLiteral<"auto">]>>;
64
+ 'prefetch-weights': z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodLiteral<"auto">]>>;
65
+ 'tensor-read-lazy': z.ZodOptional<z.ZodEnum<{
66
+ on: "on";
67
+ off: "off";
68
+ auto: "auto";
69
+ }>>;
70
+ fit: z.ZodOptional<z.ZodBoolean>;
71
+ 'fit-target': z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodString]>>;
72
+ 'fit-ctx': z.ZodOptional<z.ZodNumber>;
50
73
  openclCacheDir: z.ZodOptional<z.ZodString>;
51
74
  reasoning_budget: z.ZodOptional<z.ZodNumber>;
52
75
  projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@qvac/inference",
3
- "version": "0.20.0",
3
+ "version": "0.20.1",
4
4
  "description": "Bare-only in-process core of the QVAC SDK. Runs inference directly on the Bare runtime with explicit plugin assembly and no RPC, worker, or subprocess layer.",
5
5
  "license": "Apache-2.0",
6
6
  "repository": {
@@ -165,7 +165,7 @@
165
165
  "@qvac/logging": "^0.1.1",
166
166
  "@qvac/model-fit": "^0.12.0",
167
167
  "@qvac/rag": "^0.8.1",
168
- "@qvac/registry-client": "^0.6.1",
168
+ "@qvac/registry-client": "^0.7.0",
169
169
  "bare-abort-controller": "^1.1.2",
170
170
  "bare-buffer": "^3.6.2",
171
171
  "bare-cpu-info": "0.1.1",