Press n or j to go to the next uncovered block, b, p or k for the previous block.
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 | 1x 1x 1x 6x 6x 6x 6x 6x 3x 3x 3x 3x 3x 3x 3x 6x 3x 3x 6x 6x 6x 3x 3x 3x 3x 2x | /**
* Model configuration — frozen singleton derived from model-descriptor.json.
*
* The descriptor is the single source of truth for all model constants.
* This module resolves it and provides the validated, frozen ModelConfig.
*/
import type { ModelDescriptor, ModelConfig } from './types';
import * as fs from 'node:fs';
import * as path from 'node:path';
/**
* Default config for nomic-embed-code-v1 (BERT-base, 137M parameters).
* Used as fallback when no model-descriptor.json is found.
*/
export const EMBED_CODE_V1_CONFIG: Readonly<ModelConfig> = Object.freeze({
embeddingDim: 768,
maxTokens: 8192,
poolingStrategy: 'mean' as const,
normalize: true,
inputIdsName: 'input_ids',
attentionMaskName: 'attention_mask',
outputName: 'last_hidden_state',
taskPrefixes: {
query: 'search_query: ',
document: 'search_document: ',
},
});
/**
* Default config for nomic-embed-text-v1.5 (137M BERT-based).
*/
export const EMBED_TEXT_V15_CONFIG: Readonly<ModelConfig> = Object.freeze({
embeddingDim: 768,
maxTokens: 8192,
poolingStrategy: 'mean' as const,
normalize: true,
inputIdsName: 'input_ids',
attentionMaskName: 'attention_mask',
outputName: 'last_hidden_state',
taskPrefixes: {
query: 'search_query: ',
document: 'search_document: ',
},
});
/** Fallback config to use when descriptor is absent */
const FALLBACK_CONFIG = EMBED_CODE_V1_CONFIG;
/**
* Resolve model config from the model-descriptor.json co-located with the weights file.
*
* This is the single source of truth pattern — the descriptor that ships alongside
* the weights defines the architecture, input/output names, and runtime config.
* Falls back to EMBED_CODE_V1_CONFIG when no descriptor is present.
*
* @param modelPath Path to the weights file (weights.int8.bin)
* @param fallbackConfig Config to use when no descriptor is found
* @returns { config, descriptor } — validated ModelConfig and the raw descriptor (or null)
*/
export function resolveModelConfig(
modelPath: string,
fallbackConfig: Readonly<ModelConfig> = FALLBACK_CONFIG,
): { config: Readonly<ModelConfig>; descriptor: ModelDescriptor | null } {
const modelDir = path.dirname(modelPath);
const descriptorPath = path.join(modelDir, 'model-descriptor.json');
let descriptor: ModelDescriptor;
try {
const raw = fs.readFileSync(descriptorPath, 'utf-8');
descriptor = JSON.parse(raw);
} catch {
// Descriptor not found or unparseable — use fallback
console.warn(
'[embed-code-core] Model descriptor not found or invalid, using fallback configuration.',
);
return { config: fallbackConfig, descriptor: null };
}
// Validate critical descriptor values
const arch = descriptor.architecture;
Iif (typeof arch.embedding_dim !== 'number' || arch.embedding_dim <= 0) {
throw new Error(
`[embed-code-core] Invalid descriptor: embedding_dim=${arch.embedding_dim}, must be > 0`,
);
}
Iif (typeof arch.max_position_embeddings !== 'number' || arch.max_position_embeddings <= 0) {
throw new Error(
`[embed-code-core] Invalid descriptor: max_position_embeddings=${arch.max_position_embeddings}, must be > 0`,
);
}
const validStrategies = ['mean', 'cls', 'last_token'];
const strategy = descriptor.pooling?.strategy;
Iif (strategy !== undefined && !validStrategies.includes(strategy)) {
throw new Error(
`[embed-code-core] Invalid descriptor: pooling.strategy="${strategy}", must be one of: ${validStrategies.join(', ')}`,
);
}
// Build ModelConfig from descriptor — try `weights`, fall back to `onnx` for compat
const d = descriptor;
const w = d.weights ?? d.onnx;
const config: ModelConfig = {
embeddingDim: d.architecture.embedding_dim,
maxTokens: d.architecture.max_position_embeddings,
poolingStrategy: d.pooling.strategy,
normalize: d.pooling.normalize,
inputIdsName: w?.input_ids_name ?? 'input_ids',
attentionMaskName: w?.attention_mask_name ?? 'attention_mask',
outputName: w?.output_name ?? 'last_hidden_state',
taskPrefixes: d.task_prefixes,
};
const frozenConfig: ModelConfig = {
...config,
taskPrefixes: Object.freeze({ ...config.taskPrefixes }),
};
return { config: Object.freeze(frozenConfig), descriptor };
}
/**
* Read model descriptor from a directory.
* Returns null if not found.
*/
export function readModelDescriptor(dirPath: string): ModelDescriptor | null {
try {
const descriptorPath = path.join(dirPath, 'model-descriptor.json');
const raw = fs.readFileSync(descriptorPath, 'utf-8');
return JSON.parse(raw) as ModelDescriptor;
} catch {
return null;
}
}
|