|
| 1 | +#!/usr/bin/env node |
| 2 | + |
| 3 | +import { GGMLQuantizationType, gguf } from "."; |
| 4 | + |
| 5 | +interface PrintColumnHeader { |
| 6 | + name: string; |
| 7 | + maxWidth?: number; |
| 8 | + alignRight?: boolean; |
| 9 | +} |
| 10 | + |
| 11 | +const mapDtypeToName = Object.fromEntries(Object.entries(GGMLQuantizationType).map(([name, value]) => [value, name])); |
| 12 | + |
| 13 | +async function main() { |
| 14 | + const ggufPath = process.argv[2]; |
| 15 | + const { metadata, tensorInfos } = await gguf(ggufPath, { |
| 16 | + allowLocalFile: true, |
| 17 | + }); |
| 18 | + |
| 19 | + // TODO: print info about endianess |
| 20 | + console.log(`* Dumping ${Object.keys(metadata).length} key/value pair(s)`); |
| 21 | + printTable( |
| 22 | + [ |
| 23 | + { name: "Idx", alignRight: true }, |
| 24 | + // { name: 'Type' }, // TODO: support this |
| 25 | + { name: "Count", alignRight: true }, |
| 26 | + { name: "Value" }, |
| 27 | + ], |
| 28 | + Object.entries(metadata).map(([key, value], i) => { |
| 29 | + const MAX_LEN = 50; |
| 30 | + let strVal = ""; |
| 31 | + let count = 1; |
| 32 | + if (Array.isArray(value)) { |
| 33 | + strVal = JSON.stringify(value); |
| 34 | + count = value.length; |
| 35 | + } else if (value instanceof String || typeof value === "string") { |
| 36 | + strVal = JSON.stringify(value); |
| 37 | + } else { |
| 38 | + strVal = value.toString(); |
| 39 | + } |
| 40 | + strVal = strVal.length > MAX_LEN ? strVal.slice(0, MAX_LEN) + "..." : strVal; |
| 41 | + return [(i + 1).toString(), count.toString(), `${key} = ${strVal}`]; |
| 42 | + }) |
| 43 | + ); |
| 44 | + |
| 45 | + console.log(); |
| 46 | + console.log(`* Dumping ${tensorInfos.length} tensor(s)`); |
| 47 | + printTable( |
| 48 | + [ |
| 49 | + { name: "Idx", alignRight: true }, |
| 50 | + { name: "Num Elements", alignRight: true }, |
| 51 | + { name: "Shape" }, |
| 52 | + { name: "Data Type" }, |
| 53 | + { name: "Name" }, |
| 54 | + ], |
| 55 | + tensorInfos.map((tensorInfo, i) => { |
| 56 | + const shape = [1n, 1n, 1n, 1n]; |
| 57 | + tensorInfo.shape.forEach((dim, i) => { |
| 58 | + shape[i] = dim; |
| 59 | + }); |
| 60 | + return [ |
| 61 | + (i + 1).toString(), |
| 62 | + shape.reduce((acc, n) => acc * n, 1n).toString(), |
| 63 | + shape.map((n) => n.toString().padStart(6)).join(", "), |
| 64 | + mapDtypeToName[tensorInfo.dtype], |
| 65 | + tensorInfo.name, |
| 66 | + ]; |
| 67 | + }) |
| 68 | + ); |
| 69 | +} |
| 70 | + |
| 71 | +function printTable(header: PrintColumnHeader[], rows: string[][], leftPad = 2) { |
| 72 | + const leftPadStr = " ".repeat(leftPad); |
| 73 | + |
| 74 | + // Calculate column widths |
| 75 | + const columnWidths = header.map((h, i) => { |
| 76 | + const maxContentWidth = Math.max(h.name.length, ...rows.map((row) => (row[i] || "").length)); |
| 77 | + return h.maxWidth ? Math.min(maxContentWidth, h.maxWidth) : maxContentWidth; |
| 78 | + }); |
| 79 | + |
| 80 | + // Print header |
| 81 | + const headerLine = header |
| 82 | + .map((h, i) => { |
| 83 | + return h.name.padEnd(columnWidths[i]); |
| 84 | + }) |
| 85 | + .join(" | "); |
| 86 | + console.log(leftPadStr + headerLine); |
| 87 | + |
| 88 | + // Print separator |
| 89 | + console.log(leftPadStr + columnWidths.map((w) => "-".repeat(w)).join("-|-")); |
| 90 | + |
| 91 | + // Print rows |
| 92 | + for (const row of rows) { |
| 93 | + const line = header |
| 94 | + .map((h, i) => { |
| 95 | + return h.alignRight ? (row[i] || "").padStart(columnWidths[i]) : (row[i] || "").padEnd(columnWidths[i]); |
| 96 | + }) |
| 97 | + .join(" | "); |
| 98 | + console.log(leftPadStr + line); |
| 99 | + } |
| 100 | +} |
| 101 | + |
| 102 | +main(); |
0 commit comments