File size: 8,444 Bytes
31c7d49
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
import { compareFloat32, type TensorComparison } from './models/triposplat/tensorMath'
import { OrtWorkerClient } from './runtime/OrtWorkerClient'
import { createTensorPayload, type TensorPayload } from './runtime/tensors'

const SESSION_ID = 'triposplat/gaussian-decoder-parity'
const ATOL = 0.02
const RTOL = 0.01
const MINIMUM_COSINE = 0.9999
const SHAPES = {
  points: [1, 8192, 3],
  cond: [1, 8192, 16],
  features: [1, 8192, 480],
} as const

interface OutputGate extends TensorComparison {
  fractionWithinTolerance: number
  maxErrorIndex: number
  referenceAtMaxError: number
  candidateAtMaxError: number
  passed: boolean
}

interface GaussianLabResult {
  passed: boolean
  executionProvider: string
  modelLoadMs: number
  modelTransferBytes?: number
  inferenceMs: number
  readbackMs: number
  comparison: OutputGate
  tolerance: { absolute: number; relative: number; minimumCosineSimilarity: number }
  environment: { userAgent: string; crossOriginIsolated: boolean; webgpu: boolean }
}

declare global {
  interface Window {
    __TRIPOSPLAT_GAUSSIAN_RESULT__?: GaussianLabResult
  }
}

const modelInput = document.querySelector<HTMLInputElement>('#model')!
const fixtureInput = document.querySelector<HTMLInputElement>('#fixture')!
const runButton = document.querySelector<HTMLButtonElement>('#run')!
const statusElement = document.querySelector<HTMLElement>('[data-testid="gaussian-status"]')!
const errorElement = document.querySelector<HTMLPreElement>('[data-testid="gaussian-error"]')!
const resultElement = document.querySelector<HTMLPreElement>('[data-testid="gaussian-result"]')!
let activeClient: OrtWorkerClient | undefined
let busy = false

function elementCount(shape: readonly number[]): number {
  return shape.reduce((product, value) => product * value, 1)
}

async function fetchFloat32(url: string, expectedElements: number): Promise<Float32Array> {
  const response = await fetch(url)
  if (!response.ok) throw new Error(`Could not fetch ${url}: HTTP ${response.status}`)
  const buffer = await response.arrayBuffer()
  if (buffer.byteLength !== expectedElements * 4) {
    throw new Error(`${url} has ${buffer.byteLength} bytes; expected ${expectedElements * 4}.`)
  }
  return new Float32Array(buffer)
}

async function contentLength(url: string): Promise<number | undefined> {
  try {
    const response = await fetch(url, { method: 'HEAD' })
    if (!response.ok) return undefined
    const value = Number(response.headers.get('content-length'))
    return Number.isFinite(value) && value >= 0 ? value : undefined
  } catch {
    return undefined
  }
}

function payloadFloat32(payload: TensorPayload | undefined): Float32Array {
  if (!payload || payload.type !== 'float32') {
    throw new Error("Output 'features' is missing or is not float32.")
  }
  return new Float32Array(payload.data)
}

function gate(reference: Float32Array, candidate: Float32Array): OutputGate {
  const comparison = compareFloat32(reference, candidate)
  let within = 0
  let maxError = -1
  let maxErrorIndex = 0
  for (let index = 0; index < reference.length; index += 1) {
    const error = Math.abs(reference[index] - candidate[index])
    if (error <= ATOL + RTOL * Math.abs(reference[index])) within += 1
    if (error > maxError) {
      maxError = error
      maxErrorIndex = index
    }
  }
  const fractionWithinTolerance = within / reference.length
  return {
    ...comparison,
    fractionWithinTolerance,
    maxErrorIndex,
    referenceAtMaxError: reference[maxErrorIndex],
    candidateAtMaxError: candidate[maxErrorIndex],
    passed: comparison.finite
      && fractionWithinTolerance === 1
      && comparison.cosineSimilarity >= MINIMUM_COSINE,
  }
}

async function run(): Promise<void> {
  if (busy) return
  busy = true
  runButton.disabled = true
  runButton.textContent = 'Running…'
  errorElement.hidden = true
  resultElement.hidden = true
  delete window.__TRIPOSPLAT_GAUSSIAN_RESULT__
  try {
    if (activeClient) await activeClient.dispose()
    const modelUrl = modelInput.value
    const fixtureUrl = fixtureInput.value
    const client = new OrtWorkerClient({ onStatus: ({ message }) => { statusElement.textContent = message } })
    activeClient = client
    statusElement.textContent = 'Fetching the official fp32 Gaussian fixture…'
    const [points, cond, reference] = await Promise.all([
      fetchFloat32(`${fixtureUrl}/points.f32`, elementCount(SHAPES.points)),
      fetchFloat32(`${fixtureUrl}/cond.f32`, elementCount(SHAPES.cond)),
      fetchFloat32(`${fixtureUrl}/features.f32`, elementCount(SHAPES.features)),
    ])
    const sidecarUrl = `${modelUrl}.data`
    const graphName = new URL(modelUrl, document.baseURI).pathname.split('/').at(-1)
    if (!graphName) throw new Error(`Could not derive graph name from ${modelUrl}.`)
    const transferParts = await Promise.all([contentLength(modelUrl), contentLength(sidecarUrl)])
    const modelTransferBytes = transferParts.every((value) => value !== undefined)
      ? transferParts.reduce<number>((sum, value) => sum + (value ?? 0), 0)
      : undefined
    const loaded = await client.loadSession({
      sessionId: SESSION_ID,
      manifest: {
        graphUrl: modelUrl,
        externalData: [{ path: `${decodeURIComponent(graphName)}.data`, url: sidecarUrl }],
      },
      options: { allowWasmFallback: false, graphOptimizationLevel: 'disabled' },
    })
    if (loaded.executionProvider !== 'webgpu') {
      throw new Error(`Expected WebGPU, loaded ${loaded.executionProvider}.`)
    }
    const response = await client.runSession({
      sessionId: SESSION_ID,
      inputs: {
        points: createTensorPayload('float32', points, SHAPES.points),
        cond: createTensorPayload('float32', cond, SHAPES.cond),
      },
      outputs: ['features'],
      tag: 'gaussian-decoder-parity',
    })
    const comparison = gate(reference, payloadFloat32(response.outputs.features))
    const result: GaussianLabResult = {
      passed: comparison.passed,
      executionProvider: loaded.executionProvider,
      modelLoadMs: loaded.loadMs,
      modelTransferBytes,
      inferenceMs: response.timings.inferenceMs,
      readbackMs: response.timings.readbackMs,
      comparison,
      tolerance: { absolute: ATOL, relative: RTOL, minimumCosineSimilarity: MINIMUM_COSINE },
      environment: {
        userAgent: navigator.userAgent,
        crossOriginIsolated: self.crossOriginIsolated,
        webgpu: 'gpu' in navigator,
      },
    }
    await client.dispose()
    if (activeClient === client) activeClient = undefined
    window.__TRIPOSPLAT_GAUSSIAN_RESULT__ = result
    resultElement.textContent = JSON.stringify(result, null, 2)
    resultElement.hidden = false
    statusElement.textContent = result.passed
      ? 'PASS: WebGPU Gaussian features match official fp32 PyTorch.'
      : 'FAIL: WebGPU Gaussian features exceed tolerance.'
  } catch (caught) {
    const message = caught instanceof Error ? caught.message : String(caught)
    errorElement.textContent = message
    errorElement.hidden = false
    statusElement.textContent = 'Gaussian decoder validation failed.'
    if (activeClient) await activeClient.dispose().catch(() => undefined)
    activeClient = undefined
  } finally {
    busy = false
    runButton.disabled = false
    runButton.textContent = 'Run Gaussian decoder parity gate'
  }
}

runButton.addEventListener('click', () => { void run() })
addEventListener('beforeunload', () => { if (activeClient) void activeClient.dispose() })
if (new URLSearchParams(location.search).get('autorun') === '1') queueMicrotask(() => { void run() })

const style = document.createElement('style')
style.textContent = `
  :root { color: #ececf3; background: #101014; font: 15px/1.5 ui-monospace, SFMono-Regular, Menlo, monospace; }
  body { margin: 0; } main { max-width: 920px; margin: 0 auto; padding: 48px 24px; }
  h1 { font: 600 28px/1.2 system-ui, sans-serif; } label { display: grid; gap: 6px; margin: 18px 0; }
  input { box-sizing: border-box; width: 100%; padding: 10px; color: inherit; background: #1b1b22; border: 1px solid #3a3a48; border-radius: 6px; }
  button { padding: 10px 16px; color: #08080a; background: #f8cf00; border: 0; border-radius: 6px; font-weight: 700; cursor: pointer; }
  button:disabled { opacity: .55; cursor: wait; } pre { overflow: auto; padding: 16px; background: #18181f; border-radius: 8px; }
  .error { color: #ff9b9b; }
`
document.head.append(style)