mirror of
https://github.com/diegosouzapw/OmniRoute.git
synced 2026-09-17 20:32:25 +03:00
Landed with the design call resolved per the owner's pick — **option 1**: the synced store is now endpoint-agnostic (persistDiscoveredModels and managedModelImport no longer drop non-chat models at write time), and chat selectability moved to read time (auto-pool expansion in autoStrategy applies filterChatSelectableModels; the models-route projection already had its chatOnly filter). Your discovery test now passes end-to-end (3/3): /api/show capabilities persist per connection and image/embedding requests route through the advertising host. Reconciliation notes: conflicted areas merged onto the current tip (adobe discovery import, requestedModel preflight signature, resolvedProvider fast-path coexists with the synced-route override — explicit resolution wins); carried base-red drains (#10055 memoization, #11071 test variants) dropped as already-landed; the managed-model-import exclusion test was propagated to the new contract (image/video models persist; the read filter still hides them from chat pickers — pinned by a new assertion). Full battery: 205/206 focused (the one red is a confirmed periodic-timer timing flake on the loaded devbox — 20/20 isolated), autoCombo vitest 30/30, combo suites 46/46, gates + typecheck clean. Thank you @yourspraveen — the capability probe + routing design was right; it just needed the store contract opened up. Fixes #11087.
100 lines
3.9 KiB
TypeScript
100 lines
3.9 KiB
TypeScript
import { test } from "node:test";
|
|
import assert from "node:assert";
|
|
import { estimateCompressionTokens } from "../../../open-sse/services/compression/stats.ts";
|
|
import { transformAnthropicMessages, transformOpenAIChatCompletions } from "omniglyph";
|
|
|
|
const CHARS_PER_TOKEN = 4;
|
|
|
|
// Corpo Claude denso o bastante para o gate de rentabilidade do omniglyph converter
|
|
// (mesmo fixture usado em omniglyph-adapter.test.ts / omniglyph-plumbing.test.ts).
|
|
const DENSE =
|
|
"X".repeat(500) +
|
|
"\n" +
|
|
Array.from(
|
|
{ length: 400 },
|
|
(_, i) => `const row_${i} = compute(${i * 17}, "${"v".repeat(80)}");`
|
|
).join("\n");
|
|
|
|
const DENSE_CLAUDE_BODY = {
|
|
model: "claude-fable-5",
|
|
max_tokens: 128,
|
|
system: DENSE,
|
|
messages: [{ role: "user", content: [{ type: "text", text: "oi" }] }],
|
|
};
|
|
|
|
test("estimateCompressionTokens é image-aware: encolhe de verdade numa página omniglyph real", async () => {
|
|
const encoded = new TextEncoder().encode(JSON.stringify(DENSE_CLAUDE_BODY));
|
|
const result = await transformAnthropicMessages({ body: encoded, model: "claude-fable-5" });
|
|
assert.ok(
|
|
result.applied,
|
|
`omniglyph deveria ter convertido o corpo denso (reason=${result.reason})`
|
|
);
|
|
const outBody = JSON.parse(new TextDecoder().decode(result.body)) as Record<string, unknown>;
|
|
assert.ok(
|
|
JSON.stringify(outBody).includes('"type":"image"'),
|
|
"saída deveria conter bloco de imagem"
|
|
);
|
|
|
|
const naiveCharEstimate = Math.ceil(JSON.stringify(outBody).length / CHARS_PER_TOKEN);
|
|
const imageAwareEstimate = estimateCompressionTokens(outBody);
|
|
const originalTextEstimate = estimateCompressionTokens(DENSE_CLAUDE_BODY);
|
|
|
|
console.log(
|
|
`naive-char=${naiveCharEstimate} image-aware=${imageAwareEstimate} original-text=${originalTextEstimate}`
|
|
);
|
|
|
|
// O estimador ciente de imagem deve ser MUITO menor que o char-count ingênuo do
|
|
// próprio base64 (prova que o base64 não é contado por char).
|
|
assert.ok(
|
|
imageAwareEstimate < naiveCharEstimate,
|
|
`esperado image-aware (${imageAwareEstimate}) < naive-char (${naiveCharEstimate})`
|
|
);
|
|
// E deve representar encolhimento real frente ao texto original (não só frente ao
|
|
// próprio char-count do PNG).
|
|
assert.ok(
|
|
imageAwareEstimate < originalTextEstimate,
|
|
`esperado image-aware (${imageAwareEstimate}) < original-text (${originalTextEstimate})`
|
|
);
|
|
});
|
|
|
|
test("regressão: corpo sem imagem estima o MESMO valor de antes (char-count puro)", () => {
|
|
const plainBody = {
|
|
model: "claude-sonnet-5",
|
|
max_tokens: 128,
|
|
system: "prompt curto",
|
|
messages: [{ role: "user", content: [{ type: "text", text: "olá, tudo bem?" }] }],
|
|
};
|
|
const expected = Math.ceil(JSON.stringify(plainBody).length / CHARS_PER_TOKEN);
|
|
assert.equal(estimateCompressionTokens(plainBody), expected);
|
|
});
|
|
|
|
test("estimateCompressionTokens contabiliza image_url PNG no wire OpenAI", async () => {
|
|
const originalBody = {
|
|
model: "gpt-5.6",
|
|
messages: [
|
|
{ role: "system", content: DENSE },
|
|
{ role: "user", content: "oi" },
|
|
],
|
|
};
|
|
const encoded = new TextEncoder().encode(JSON.stringify(originalBody));
|
|
const result = await transformOpenAIChatCompletions(encoded);
|
|
assert.equal(result.info.compressed, true);
|
|
const compressedBody = JSON.parse(new TextDecoder().decode(result.body)) as Record<
|
|
string,
|
|
unknown
|
|
>;
|
|
assert.ok(JSON.stringify(compressedBody).includes('"type":"image_url"'));
|
|
|
|
const originalEstimate = estimateCompressionTokens(originalBody);
|
|
const compressedEstimate = estimateCompressionTokens(compressedBody);
|
|
const naiveEstimate = Math.ceil(JSON.stringify(compressedBody).length / CHARS_PER_TOKEN);
|
|
assert.ok(
|
|
compressedEstimate < naiveEstimate,
|
|
`expected image-aware (${compressedEstimate}) < base64 char estimate (${naiveEstimate})`
|
|
);
|
|
assert.ok(
|
|
compressedEstimate < originalEstimate,
|
|
`expected OpenAI image compression (${compressedEstimate}) < original (${originalEstimate})`
|
|
);
|
|
});
|