doppler-gpu 0.4.10 → 0.4.14
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.
- package/CHANGELOG.md +95 -0
- package/README.md +40 -1
- package/models/catalog.json +637 -32
- package/package.json +24 -3
- package/src/client/doppler-registry.js +6 -0
- package/src/client/doppler-registry.json +135 -13
- package/src/client/runtime/lora.d.ts +10 -0
- package/src/client/runtime/lora.js +69 -0
- package/src/client/runtime/model-manager.js +6 -167
- package/src/client/runtime/model-session.d.ts +3 -1
- package/src/client/runtime/model-session.js +6 -0
- package/src/client/wrap-pipeline-handle.js +7 -0
- package/src/config/conversion/amplify/amplify-120m-f16-af32.json +516 -0
- package/src/config/conversion/esm/esm2-t12-35m-ur50d-f32-af32.json +571 -0
- package/src/config/conversion/esmc/esmc-300m-f32-af32.json +342 -0
- package/src/config/conversion/nucleotide-transformer/nucleotide-transformer-v2-50m-f32-af32.json +388 -0
- package/src/config/conversion/qwen3/qwen-3-5-0-8b-q4k-ehaf16.json +3 -2
- package/src/config/kernels/kernel-ref-digests.js +2 -1
- package/src/config/kernels/registry.json +73 -11
- package/src/config/merge.js +24 -0
- package/src/config/required-inference-fields-contract-check.js +2 -0
- package/src/config/schema/adapter.schema.d.ts +6 -1
- package/src/config/schema/adapter.schema.js +5 -0
- package/src/config/schema/converter.schema.d.ts +7 -0
- package/src/config/schema/converter.schema.js +15 -0
- package/src/config/schema/inference.schema.d.ts +8 -0
- package/src/config/schema/manifest.schema.d.ts +32 -1
- package/src/config/schema/manifest.schema.js +26 -0
- package/src/converter/conversion-plan.js +15 -1
- package/src/converter/core.js +5 -4
- package/src/converter/source-tensor-rules.d.ts +50 -0
- package/src/converter/source-tensor-rules.js +180 -0
- package/src/converter/source-tokenizer.d.ts +45 -0
- package/src/converter/source-tokenizer.js +155 -0
- package/src/experimental/adapters/artifact-contract.d.ts +83 -0
- package/src/experimental/adapters/artifact-contract.js +254 -0
- package/src/experimental/adapters/index.d.ts +25 -0
- package/src/experimental/adapters/index.js +12 -0
- package/src/experimental/adapters/lora-loader.js +24 -4
- package/src/experimental/bridge/trainer-artifact-bridge.d.ts +2 -1
- package/src/experimental/bridge/trainer-artifact-bridge.js +35 -13
- package/src/experimental/training/artifacts.d.ts +160 -0
- package/src/experimental/training/attention-backward.d.ts +64 -0
- package/src/experimental/training/attention-forward.d.ts +22 -0
- package/src/experimental/training/attention-forward.js +82 -0
- package/src/experimental/training/autograd.d.ts +76 -0
- package/src/experimental/training/checkpoint.d.ts +36 -0
- package/src/experimental/training/clip.d.ts +9 -0
- package/src/experimental/training/dataloader.d.ts +8 -0
- package/src/experimental/training/distillation/student-fixture.js +1 -0
- package/src/experimental/training/export.d.ts +41 -0
- package/src/experimental/training/lora-pipeline.d.ts +186 -0
- package/src/experimental/training/lora-pipeline.js +5 -0
- package/src/experimental/training/lora.d.ts +20 -0
- package/src/experimental/training/lora.js +4 -3
- package/src/experimental/training/loss-scaling.d.ts +21 -0
- package/src/experimental/training/loss.d.ts +10 -0
- package/src/experimental/training/native-qwen-lora.d.ts +106 -0
- package/src/experimental/training/native-qwen-lora.js +449 -0
- package/src/experimental/training/objectives/base.d.ts +58 -0
- package/src/experimental/training/optimizer.d.ts +22 -0
- package/src/experimental/training/optimizer.js +1 -0
- package/src/experimental/training/runner.d.ts +249 -0
- package/src/experimental/training/trainer.d.ts +89 -0
- package/src/experimental/training/workloads.d.ts +205 -0
- package/src/gpu/kernels/backward/adam.d.ts +1 -0
- package/src/gpu/kernels/backward/adam.js +6 -2
- package/src/gpu/kernels/backward/adam.wgsl +3 -2
- package/src/gpu/kernels/backward/matmul_backward_q4k.wgsl +90 -0
- package/src/gpu/kernels/backward/utils.js +18 -4
- package/src/inference/pipelines/text/attention/attn-config.js +2 -0
- package/src/inference/pipelines/text/attention/projections.js +161 -24
- package/src/inference/pipelines/text/attention/record.js +67 -9
- package/src/inference/pipelines/text/attention/run.js +60 -7
- package/src/inference/pipelines/text/attention/types.js +18 -9
- package/src/inference/pipelines/text/config.d.ts +5 -0
- package/src/inference/pipelines/text/config.js +91 -0
- package/src/inference/pipelines/text/ffn/dense.js +205 -2
- package/src/inference/pipelines/text/ffn/standard.js +50 -3
- package/src/inference/pipelines/text/generator-helpers.js +11 -1
- package/src/inference/pipelines/text/generator-runtime.js +63 -15
- package/src/inference/pipelines/text/generator-steps.js +3 -1
- package/src/inference/pipelines/text/generator.d.ts +5 -1
- package/src/inference/pipelines/text/generator.js +196 -3
- package/src/inference/pipelines/text/init-qkv-fusion.js +85 -1
- package/src/inference/pipelines/text/init.d.ts +2 -0
- package/src/inference/pipelines/text/init.js +2 -0
- package/src/inference/pipelines/text/layer.js +6 -0
- package/src/inference/pipelines/text/linear-attention.d.ts +2 -0
- package/src/inference/pipelines/text/linear-attention.js +77 -6
- package/src/inference/pipelines/text/logits/cpu.d.ts +6 -0
- package/src/inference/pipelines/text/logits/cpu.js +22 -0
- package/src/inference/pipelines/text/logits/index.d.ts +1 -1
- package/src/inference/pipelines/text/logits/index.js +85 -11
- package/src/inference/pipelines/text/logits/types.d.ts +4 -0
- package/src/inference/pipelines/text/lora-types.d.ts +6 -1
- package/src/inference/pipelines/text/lora-types.js +5 -0
- package/src/inference/pipelines/text/ops.js +33 -0
- package/src/inference/pipelines/text/types.d.ts +66 -0
- package/src/inference/pipelines/text/weights.d.ts +10 -0
- package/src/inference/pipelines/text/weights.js +33 -1
- package/src/inference/pipelines/text.d.ts +9 -2
- package/src/inference/pipelines/text.js +116 -0
- package/src/inference/tokenizers/behavior-flags.js +5 -1
- package/src/inference/tokenizers/bundled.js +99 -8
- package/src/loader/doppler-loader.d.ts +2 -0
- package/src/loader/doppler-loader.js +9 -0
- package/src/loader/final-weights-loader.d.ts +8 -0
- package/src/loader/final-weights-loader.js +42 -0
- package/src/loader/layer-loader.d.ts +6 -0
- package/src/loader/layer-loader.js +102 -3
- package/src/loader/loader-types.d.ts +15 -2
- package/src/models/diffusiongemma.js +1 -1
- package/src/models/embeddinggemma.js +1 -1
- package/src/models/gemma3.js +1 -1
- package/src/models/gemma4.js +1 -1
- package/src/models/qwen3.js +1 -1
- package/src/tooling/hf-registry-utils.js +1 -1
- package/src/tooling/node-browser-command-runner.d.ts +16 -0
- package/src/tooling/node-browser-command-runner.js +1 -1
- package/src/tooling/node-converter.js +33 -1
- package/src/tooling/node-webgpu.js +85 -2
- package/src/training-browser.d.ts +19 -0
- package/src/training-browser.js +13 -0
- package/src/training.d.ts +137 -0
- package/src/training.js +219 -0
- package/src/version.js +1 -1
- package/tests/kernels/browser/kernel-suite.js +9 -1
- package/tests/kernels/browser/test-page.js +4 -2
- package/tests/kernels/reference/adam.js +4 -2
- package/NOTICE +0 -5
|
@@ -0,0 +1,186 @@
|
|
|
1
|
+
import type { LoadedTrainingWorkload } from './workloads.js';
|
|
2
|
+
|
|
3
|
+
export declare const LORA_RUNNER_SUPPORT_CONTRACT: Readonly<{
|
|
4
|
+
supportedBaseModelId: string;
|
|
5
|
+
supportedDatasetFormat: string;
|
|
6
|
+
registeredBaseModelIds: readonly string[];
|
|
7
|
+
registeredDatasetFormats: readonly string[];
|
|
8
|
+
implementedRunnerKeys: readonly string[];
|
|
9
|
+
}>;
|
|
10
|
+
|
|
11
|
+
export declare const LORA_RUNNER_BASE_MODEL_REGISTRY: Readonly<Record<string, Readonly<{
|
|
12
|
+
baseModelId: string;
|
|
13
|
+
modelRef?: string;
|
|
14
|
+
family: string;
|
|
15
|
+
runnerKind: string;
|
|
16
|
+
requiresExternalTrainer?: boolean;
|
|
17
|
+
nativeLoraTargets?: readonly Readonly<{ module: 'down_proj'; layer: 'last' }>[];
|
|
18
|
+
}>>>;
|
|
19
|
+
|
|
20
|
+
export declare const LORA_RUNNER_DATASET_FORMAT_REGISTRY: Readonly<Record<string, Readonly<{
|
|
21
|
+
datasetFormat: string;
|
|
22
|
+
datasetKind: string;
|
|
23
|
+
}>>>;
|
|
24
|
+
|
|
25
|
+
export declare function getLoraRunnerCompatibility(workload: unknown): {
|
|
26
|
+
schemaVersion: 1;
|
|
27
|
+
supported: boolean;
|
|
28
|
+
runnerContract: typeof LORA_RUNNER_SUPPORT_CONTRACT;
|
|
29
|
+
observed: {
|
|
30
|
+
baseModelId: string;
|
|
31
|
+
datasetFormat: string;
|
|
32
|
+
taskType: string;
|
|
33
|
+
runnerKey: string;
|
|
34
|
+
baseModelFamily: string | null;
|
|
35
|
+
baseModelRunnerKind: string | null;
|
|
36
|
+
requiresExternalTrainer: boolean;
|
|
37
|
+
nativeLoraTargets: readonly Readonly<{ module: 'down_proj'; layer: 'last' }>[];
|
|
38
|
+
datasetKind: string | null;
|
|
39
|
+
registeredBaseModel: boolean;
|
|
40
|
+
registeredDatasetFormat: boolean;
|
|
41
|
+
};
|
|
42
|
+
blockedReasons: string[];
|
|
43
|
+
};
|
|
44
|
+
|
|
45
|
+
export declare function assertLoraRunnerCompatibility(workload: unknown): {
|
|
46
|
+
schemaVersion: 1;
|
|
47
|
+
supported: true;
|
|
48
|
+
runnerContract: typeof LORA_RUNNER_SUPPORT_CONTRACT;
|
|
49
|
+
observed: {
|
|
50
|
+
baseModelId: string;
|
|
51
|
+
datasetFormat: string;
|
|
52
|
+
taskType: string;
|
|
53
|
+
runnerKey: string;
|
|
54
|
+
baseModelFamily: string | null;
|
|
55
|
+
baseModelRunnerKind: string | null;
|
|
56
|
+
requiresExternalTrainer: boolean;
|
|
57
|
+
nativeLoraTargets: readonly Readonly<{ module: 'down_proj'; layer: 'last' }>[];
|
|
58
|
+
datasetKind: string | null;
|
|
59
|
+
registeredBaseModel: boolean;
|
|
60
|
+
registeredDatasetFormat: boolean;
|
|
61
|
+
};
|
|
62
|
+
blockedReasons: [];
|
|
63
|
+
};
|
|
64
|
+
|
|
65
|
+
export declare function preflightCausalLmLoraWorkload(
|
|
66
|
+
workload: unknown,
|
|
67
|
+
options?: {
|
|
68
|
+
datasetPath?: string;
|
|
69
|
+
fetch?: (url: string) => Promise<string>;
|
|
70
|
+
readFile?: (path: string) => Promise<string>;
|
|
71
|
+
}
|
|
72
|
+
): Promise<{
|
|
73
|
+
schemaVersion: 1;
|
|
74
|
+
supported: boolean;
|
|
75
|
+
runnerKey: string;
|
|
76
|
+
baseModelId: string;
|
|
77
|
+
baseModelFamily: string | null;
|
|
78
|
+
datasetPath: string;
|
|
79
|
+
datasetFormat: string;
|
|
80
|
+
taskType: string;
|
|
81
|
+
rowCount: number;
|
|
82
|
+
firstRowId: string | null;
|
|
83
|
+
lastRowId: string | null;
|
|
84
|
+
textPairFields: {
|
|
85
|
+
prompt: string | null;
|
|
86
|
+
completion: string | null;
|
|
87
|
+
};
|
|
88
|
+
textPairLengths: {
|
|
89
|
+
minPromptChars: number;
|
|
90
|
+
maxPromptChars: number;
|
|
91
|
+
minCompletionChars: number;
|
|
92
|
+
maxCompletionChars: number;
|
|
93
|
+
};
|
|
94
|
+
adapter: {
|
|
95
|
+
rank: number;
|
|
96
|
+
alpha: number;
|
|
97
|
+
targetModules: string[];
|
|
98
|
+
};
|
|
99
|
+
blockedReasons: string[];
|
|
100
|
+
}>;
|
|
101
|
+
|
|
102
|
+
export interface CausalLmLoraTrainerTensor {
|
|
103
|
+
name: string;
|
|
104
|
+
shape: [number, number];
|
|
105
|
+
tensor?: Float32Array | GPUBuffer | { buffer?: GPUBuffer; shape?: [number, number]; dtype?: string };
|
|
106
|
+
data?: Float32Array | number[];
|
|
107
|
+
values?: Float32Array | number[];
|
|
108
|
+
dtype?: 'f16' | 'f32';
|
|
109
|
+
}
|
|
110
|
+
|
|
111
|
+
export interface CausalLmLoraTrainerResult {
|
|
112
|
+
checkpointId?: string | null;
|
|
113
|
+
checkpointStep?: number | null;
|
|
114
|
+
adapterId?: string | null;
|
|
115
|
+
adapterName?: string | null;
|
|
116
|
+
trainerId?: string | null;
|
|
117
|
+
runnerId?: string | null;
|
|
118
|
+
metrics?: Record<string, unknown>;
|
|
119
|
+
receipts?: unknown[];
|
|
120
|
+
evalReports?: Array<Record<string, unknown>>;
|
|
121
|
+
tensors?: CausalLmLoraTrainerTensor[];
|
|
122
|
+
weights?: CausalLmLoraTrainerTensor[];
|
|
123
|
+
}
|
|
124
|
+
|
|
125
|
+
export interface CausalLmLoraTrainerInput {
|
|
126
|
+
schemaVersion: 1;
|
|
127
|
+
runnerKind: 'causal_lm_lora';
|
|
128
|
+
workload: LoadedTrainingWorkload['workload'];
|
|
129
|
+
loadedWorkload: LoadedTrainingWorkload;
|
|
130
|
+
compatibility: ReturnType<typeof getLoraRunnerCompatibility>;
|
|
131
|
+
preflight: Awaited<ReturnType<typeof preflightCausalLmLoraWorkload>>;
|
|
132
|
+
dataset: {
|
|
133
|
+
absolutePath: string;
|
|
134
|
+
rowCount: number;
|
|
135
|
+
rows: Array<Record<string, unknown>>;
|
|
136
|
+
datasetHash: string;
|
|
137
|
+
};
|
|
138
|
+
adapter: Record<string, unknown>;
|
|
139
|
+
training: Record<string, unknown>;
|
|
140
|
+
export: Record<string, unknown> | null;
|
|
141
|
+
layout: Record<string, string>;
|
|
142
|
+
}
|
|
143
|
+
|
|
144
|
+
export type CausalLmLoraTrainer = (input: CausalLmLoraTrainerInput) => Promise<CausalLmLoraTrainerResult> | CausalLmLoraTrainerResult;
|
|
145
|
+
|
|
146
|
+
export declare function runLoraPipeline(options: {
|
|
147
|
+
loadedWorkload: LoadedTrainingWorkload;
|
|
148
|
+
runRoot?: string | null;
|
|
149
|
+
timestamp?: string | Date | null;
|
|
150
|
+
causalLmTrainer?: CausalLmLoraTrainer;
|
|
151
|
+
fetch?: (url: string) => Promise<string>;
|
|
152
|
+
readFile?: (path: string) => Promise<string>;
|
|
153
|
+
}): Promise<Record<string, unknown>>;
|
|
154
|
+
|
|
155
|
+
export declare function evaluateLoraCheckpoint(options: {
|
|
156
|
+
loadedWorkload: LoadedTrainingWorkload;
|
|
157
|
+
checkpointPath: string;
|
|
158
|
+
checkpointId?: string | null;
|
|
159
|
+
checkpointStep?: number | null;
|
|
160
|
+
layout?: Record<string, string> | null;
|
|
161
|
+
}): Promise<Record<string, unknown>[]>;
|
|
162
|
+
|
|
163
|
+
export declare function exportLoraCheckpoint(options: {
|
|
164
|
+
loadedWorkload: LoadedTrainingWorkload;
|
|
165
|
+
checkpointPath: string;
|
|
166
|
+
checkpointId?: string | null;
|
|
167
|
+
checkpointStep?: number | null;
|
|
168
|
+
layout?: Record<string, string> | null;
|
|
169
|
+
exportsDir?: string | null;
|
|
170
|
+
datasetHash?: string | null;
|
|
171
|
+
}): Promise<Record<string, unknown>>;
|
|
172
|
+
|
|
173
|
+
export declare function watchLoraCheckpoints(options: {
|
|
174
|
+
loadedWorkload: LoadedTrainingWorkload;
|
|
175
|
+
runRoot: string;
|
|
176
|
+
pollIntervalMs?: number | null;
|
|
177
|
+
stopWhenIdle?: boolean;
|
|
178
|
+
}): Promise<{ ok: true; processedCount: number; manifestPath: string }>;
|
|
179
|
+
|
|
180
|
+
export declare function compareLoraRun(options: {
|
|
181
|
+
runRoot: string;
|
|
182
|
+
}): Promise<Record<string, unknown>>;
|
|
183
|
+
|
|
184
|
+
export declare function qualityGateLoraRun(options: {
|
|
185
|
+
runRoot: string;
|
|
186
|
+
}): Promise<Record<string, unknown>>;
|
|
@@ -122,6 +122,9 @@ export const LORA_RUNNER_BASE_MODEL_REGISTRY = Object.freeze({
|
|
|
122
122
|
family: 'qwen3',
|
|
123
123
|
runnerKind: 'causal_lm_text_generation',
|
|
124
124
|
requiresExternalTrainer: true,
|
|
125
|
+
nativeLoraTargets: Object.freeze([
|
|
126
|
+
Object.freeze({ module: 'down_proj', layer: 'last' }),
|
|
127
|
+
]),
|
|
125
128
|
}),
|
|
126
129
|
'qwen-3-5-2b-q4k-ehaf16': Object.freeze({
|
|
127
130
|
baseModelId: 'qwen-3-5-2b-q4k-ehaf16',
|
|
@@ -238,6 +241,7 @@ export function getLoraRunnerCompatibility(workload) {
|
|
|
238
241
|
baseModelFamily: baseModel?.family || null,
|
|
239
242
|
baseModelRunnerKind: baseModel?.runnerKind || null,
|
|
240
243
|
requiresExternalTrainer: baseModel?.requiresExternalTrainer === true,
|
|
244
|
+
nativeLoraTargets: baseModel?.nativeLoraTargets || Object.freeze([]),
|
|
241
245
|
datasetKind: dataset?.datasetKind || null,
|
|
242
246
|
registeredBaseModel: Boolean(baseModel),
|
|
243
247
|
registeredDatasetFormat: Boolean(dataset),
|
|
@@ -401,6 +405,7 @@ function createToyLoraModel(workload) {
|
|
|
401
405
|
outDim: 2,
|
|
402
406
|
rank: workload.pipeline.adapter.rank,
|
|
403
407
|
alpha: workload.pipeline.adapter.alpha,
|
|
408
|
+
dtype: 'f32',
|
|
404
409
|
});
|
|
405
410
|
const model = {
|
|
406
411
|
adapter,
|
|
@@ -0,0 +1,20 @@
|
|
|
1
|
+
import type { Tensor } from '../../gpu/tensor.js';
|
|
2
|
+
import type { AutogradTape } from './autograd.js';
|
|
3
|
+
|
|
4
|
+
export interface LoraAdapterConfig {
|
|
5
|
+
inDim: number;
|
|
6
|
+
outDim: number;
|
|
7
|
+
rank: number;
|
|
8
|
+
alpha: number;
|
|
9
|
+
dtype: 'f16' | 'f32';
|
|
10
|
+
}
|
|
11
|
+
|
|
12
|
+
export declare class LoraAdapter {
|
|
13
|
+
constructor(config: LoraAdapterConfig);
|
|
14
|
+
A: Tensor;
|
|
15
|
+
B: Tensor;
|
|
16
|
+
alpha: number;
|
|
17
|
+
rank: number;
|
|
18
|
+
forward(input: Tensor, tape: AutogradTape): Promise<Tensor>;
|
|
19
|
+
dispose(): void;
|
|
20
|
+
}
|
|
@@ -1,6 +1,5 @@
|
|
|
1
1
|
import { acquireBuffer, releaseBuffer, uploadData, BufferUsage } from '../../memory/buffer-pool.js';
|
|
2
2
|
import { createTensor, tensorBytes } from '../../gpu/tensor.js';
|
|
3
|
-
import { getTrainingConfig } from '../../config/training-defaults.js';
|
|
4
3
|
import { runMatmul, runScale } from '../../gpu/kernels/index.js';
|
|
5
4
|
import { OpType } from './autograd.js';
|
|
6
5
|
import { f32ToF16Array } from '../../inference/kv-cache/types.js';
|
|
@@ -25,8 +24,10 @@ function uploadLoraInit(buffer, dtype, values) {
|
|
|
25
24
|
|
|
26
25
|
export class LoraAdapter {
|
|
27
26
|
constructor(config) {
|
|
28
|
-
const { inDim, outDim, rank, alpha } = config;
|
|
29
|
-
|
|
27
|
+
const { inDim, outDim, rank, alpha, dtype } = config;
|
|
28
|
+
if (dtype !== 'f16' && dtype !== 'f32') {
|
|
29
|
+
throw new Error(`LoraAdapter dtype must be f16 or f32; got ${String(dtype)}.`);
|
|
30
|
+
}
|
|
30
31
|
|
|
31
32
|
const aBytes = tensorBytes([inDim, rank], dtype);
|
|
32
33
|
const bBytes = tensorBytes([rank, outDim], dtype);
|
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
import type { TrainingLossScalingConfigSchema } from '../../config/schema/training.schema.js';
|
|
2
|
+
import type { Tensor } from '../../gpu/tensor.js';
|
|
3
|
+
|
|
4
|
+
export declare function detectOverflow(grads: Map<Tensor, Tensor>): Promise<boolean>;
|
|
5
|
+
|
|
6
|
+
export declare class DynamicLossScaler {
|
|
7
|
+
enabled: boolean;
|
|
8
|
+
scale: number;
|
|
9
|
+
minScale: number;
|
|
10
|
+
maxScale: number;
|
|
11
|
+
scaleFactor: number;
|
|
12
|
+
backoffFactor: number;
|
|
13
|
+
growthInterval: number;
|
|
14
|
+
overflowCheck: boolean;
|
|
15
|
+
|
|
16
|
+
constructor(config?: TrainingLossScalingConfigSchema);
|
|
17
|
+
shouldScale(): boolean;
|
|
18
|
+
scaleLoss(loss: Tensor): Promise<Tensor>;
|
|
19
|
+
unscaleGradients(grads: Map<Tensor, Tensor>): Promise<Map<Tensor, Tensor>>;
|
|
20
|
+
update(hasOverflow: boolean): void;
|
|
21
|
+
}
|
|
@@ -0,0 +1,10 @@
|
|
|
1
|
+
import type { TrainingConfigSchema } from '../../config/training-defaults.d.ts';
|
|
2
|
+
import type { Tensor } from '../../gpu/tensor.js';
|
|
3
|
+
import type { AutogradTape } from './autograd.js';
|
|
4
|
+
|
|
5
|
+
export declare function crossEntropyLoss(
|
|
6
|
+
logits: Tensor,
|
|
7
|
+
targets: Tensor,
|
|
8
|
+
config: TrainingConfigSchema,
|
|
9
|
+
tape: AutogradTape
|
|
10
|
+
): Promise<Tensor>;
|
|
@@ -0,0 +1,106 @@
|
|
|
1
|
+
import type { InferencePipeline } from '../../inference/pipelines/text.js';
|
|
2
|
+
import type { LoRAExportResult } from './export.js';
|
|
3
|
+
import type { Tensor } from '../../gpu/tensor.js';
|
|
4
|
+
|
|
5
|
+
export interface NativeQwenLoRATrainerOptions {
|
|
6
|
+
pipeline: InferencePipeline;
|
|
7
|
+
baseModelId: string;
|
|
8
|
+
layerIdx: number;
|
|
9
|
+
module: 'down_proj';
|
|
10
|
+
rank: number;
|
|
11
|
+
alpha: number;
|
|
12
|
+
optimizer: {
|
|
13
|
+
type?: string;
|
|
14
|
+
lr: number;
|
|
15
|
+
beta1?: number;
|
|
16
|
+
beta2?: number;
|
|
17
|
+
eps?: number;
|
|
18
|
+
weightDecay?: number;
|
|
19
|
+
};
|
|
20
|
+
gradient?: { maxNorm?: number };
|
|
21
|
+
precision?: { activations?: 'f16' | 'f32'; gradients?: 'f32'; loraParams?: 'f16' | 'f32' };
|
|
22
|
+
}
|
|
23
|
+
|
|
24
|
+
export interface NativeQwenLoRAStepResult {
|
|
25
|
+
loss: number;
|
|
26
|
+
gradientNorm: number | null;
|
|
27
|
+
optimizer: Record<string, unknown> | null;
|
|
28
|
+
}
|
|
29
|
+
|
|
30
|
+
export interface NativeQwenLoRACheckpointTensor {
|
|
31
|
+
dtype: 'f16' | 'f32';
|
|
32
|
+
shape: number[];
|
|
33
|
+
bytes: string;
|
|
34
|
+
}
|
|
35
|
+
|
|
36
|
+
export interface NativeQwenLoRACheckpoint {
|
|
37
|
+
schemaVersion: 1;
|
|
38
|
+
backend: 'webgpu_native';
|
|
39
|
+
baseModelId: string;
|
|
40
|
+
target: { layerIdx: number; module: 'down_proj' };
|
|
41
|
+
rank: number;
|
|
42
|
+
alpha: number;
|
|
43
|
+
stepCount: number;
|
|
44
|
+
adapter: { A: NativeQwenLoRACheckpointTensor; B: NativeQwenLoRACheckpointTensor };
|
|
45
|
+
optimizer: Array<{ m: NativeQwenLoRACheckpointTensor; v: NativeQwenLoRACheckpointTensor }>;
|
|
46
|
+
}
|
|
47
|
+
|
|
48
|
+
export declare class NativeQwenLoRATrainer {
|
|
49
|
+
constructor(options: NativeQwenLoRATrainerOptions);
|
|
50
|
+
readonly pipeline: InferencePipeline;
|
|
51
|
+
readonly layerIdx: number;
|
|
52
|
+
readonly module: 'down_proj';
|
|
53
|
+
readonly baseModelId: string;
|
|
54
|
+
readonly adapter: { A: Tensor; B: Tensor; rank: number; alpha: number };
|
|
55
|
+
paramGroups(): { base: Tensor[]; lora: Tensor[] };
|
|
56
|
+
trainStep(
|
|
57
|
+
inputIds: readonly number[],
|
|
58
|
+
targetIds: readonly number[],
|
|
59
|
+
supervisedTokenCount: number
|
|
60
|
+
): Promise<NativeQwenLoRAStepResult>;
|
|
61
|
+
exportAdapter(options: {
|
|
62
|
+
id: string;
|
|
63
|
+
name: string;
|
|
64
|
+
weightsPath: string;
|
|
65
|
+
}): Promise<LoRAExportResult>;
|
|
66
|
+
createCheckpoint(): Promise<NativeQwenLoRACheckpoint>;
|
|
67
|
+
restoreCheckpoint(checkpoint: NativeQwenLoRACheckpoint): void;
|
|
68
|
+
dispose(): void;
|
|
69
|
+
}
|
|
70
|
+
|
|
71
|
+
export declare function createNativeQwenLoRATrainer(
|
|
72
|
+
options: NativeQwenLoRATrainerOptions
|
|
73
|
+
): NativeQwenLoRATrainer;
|
|
74
|
+
|
|
75
|
+
export declare function loadNativeQwenTrainingPipeline(
|
|
76
|
+
modelUrl: string,
|
|
77
|
+
options?: {
|
|
78
|
+
log?: (message: string) => void;
|
|
79
|
+
onProgress?: (stage: string, progress: number, message: string) => void;
|
|
80
|
+
runtime?: Record<string, unknown>;
|
|
81
|
+
}
|
|
82
|
+
): Promise<{
|
|
83
|
+
pipeline: InferencePipeline;
|
|
84
|
+
manifest: Record<string, unknown> & { modelId: 'qwen-3-5-0-8b-q4k-ehaf16' };
|
|
85
|
+
capabilities: Record<string, unknown>;
|
|
86
|
+
configSnapshot: Record<string, unknown>;
|
|
87
|
+
}>;
|
|
88
|
+
|
|
89
|
+
export interface NativeQwenSftLoRAOptions extends NativeQwenLoRATrainerOptions {
|
|
90
|
+
samples: Array<{
|
|
91
|
+
inputIds: readonly number[];
|
|
92
|
+
targetIds: readonly number[];
|
|
93
|
+
supervisedTokenCount: number;
|
|
94
|
+
}>;
|
|
95
|
+
maxSteps?: number;
|
|
96
|
+
export?: { id: string; name: string; weightsPath: string } | null;
|
|
97
|
+
}
|
|
98
|
+
|
|
99
|
+
export declare function trainNativeQwenSftLoRA(options: NativeQwenSftLoRAOptions): Promise<{
|
|
100
|
+
backend: 'webgpu_native';
|
|
101
|
+
surfaces: ['browser', 'node', 'bun'];
|
|
102
|
+
baseModelId: string;
|
|
103
|
+
target: { layerIdx: number; module: 'down_proj' };
|
|
104
|
+
metrics: NativeQwenLoRAStepResult[];
|
|
105
|
+
adapter: LoRAExportResult | null;
|
|
106
|
+
}>;
|