Files
boocode/apps/server/src/services/__tests__/llama-args-validator.test.ts
indifferentketchup a8e475fdf4 perf(llama): unshadow cache-type + spec-decoding flags for agent opt-in
KV cache quantization (--cache-type-k q4_0) and ngram speculative decoding
(--spec-type ngram-mod) are high-value llama.cpp features that improve VRAM
usage and tokens/sec. Removing them from the shadowing lists allows agents
to enable them via llama_extra_args.
2026-06-07 22:40:23 +00:00

161 lines
5.0 KiB
TypeScript

import { describe, expect, it } from 'vitest';
import {
validateExtraArgs,
isManagedFlag,
stripShadowingFlags,
} from '../inference/llama-args-validator.js';
import { parseAgentsMd } from '../agents.js';
describe('validateExtraArgs', () => {
describe('deny list — each alias rejected', () => {
const denied = [
'-m', '--model',
'-mu', '--model-url',
'-dr', '--docker-repo',
'-hf', '-hfr', '--hf-repo',
'-hff', '--hf-file',
'-hfv', '-hfrv', '--hf-repo-v',
'-hffv', '--hf-file-v',
'-hft', '--hf-token',
'-mm', '--mmproj',
'-mmu', '--mmproj-url',
'--host', '--port', '--path', '--api-prefix', '--reuse-port',
'--api-key', '--api-key-file',
'--ssl-key-file', '--ssl-cert-file',
'--webui', '--no-webui', '--ui', '--no-ui',
'--ui-config', '--ui-config-file',
'--ui-mcp-proxy', '--no-ui-mcp-proxy',
'--models-dir', '--models-preset', '--models-max',
'--models-autoload', '--no-models-autoload',
];
for (const flag of denied) {
it(`rejects ${flag}`, () => {
expect(() => validateExtraArgs([flag])).toThrow(/managed/);
});
}
});
describe('safe flags accepted', () => {
const safe = [
'-c', '--ctx-size', '-ngl', '--gpu-layers',
'--top-k', '--cache-type-k', '--jinja', '--no-jinja',
'--spec-draft-n-max', '-fa', '--flash-attn',
'-t', '--threads', '-np', '--parallel',
];
for (const flag of safe) {
it(`accepts ${flag}`, () => {
expect(() => validateExtraArgs([flag])).not.toThrow();
expect(validateExtraArgs([flag])).toEqual([flag]);
});
}
});
it('handles --flag=value shape (denies the flag part)', () => {
expect(() => validateExtraArgs(['--model=evil.gguf'])).toThrow(/managed/);
});
it('handles --flag=value shape (accepts safe flag)', () => {
expect(validateExtraArgs(['--ctx-size=4096'])).toEqual(['--ctx-size=4096']);
});
it('returns empty array for undefined input', () => {
expect(validateExtraArgs(undefined)).toEqual([]);
});
it('returns empty array for empty input', () => {
expect(validateExtraArgs([])).toEqual([]);
});
it('treats negative numbers as values, not flags', () => {
expect(validateExtraArgs(['--seed', '-1'])).toEqual(['--seed', '-1']);
});
});
describe('isManagedFlag', () => {
it('returns true for denied flags', () => {
expect(isManagedFlag('--model')).toBe(true);
expect(isManagedFlag('-m')).toBe(true);
expect(isManagedFlag('--api-key')).toBe(true);
expect(isManagedFlag('--port')).toBe(true);
});
it('returns false for safe flags', () => {
expect(isManagedFlag('-c')).toBe(false);
expect(isManagedFlag('--ctx-size')).toBe(false);
expect(isManagedFlag('--top-k')).toBe(false);
});
});
describe('stripShadowingFlags', () => {
it('strips auto -c when user supplies -c', () => {
const result = stripShadowingFlags(['-c', '4096', '--top-k', '40']);
expect(result).toEqual(['--top-k', '40']);
});
it('retains both when no overlap', () => {
const result = stripShadowingFlags(['--top-k', '40', '--top-p', '0.95']);
expect(result).toEqual(['--top-k', '40', '--top-p', '0.95']);
});
it('strips --ctx-size=value form', () => {
const result = stripShadowingFlags(['--ctx-size=4096']);
expect(result).toEqual([]);
});
it('strips boolean --jinja flag (no value consumed)', () => {
const result = stripShadowingFlags(['--jinja', '--top-k', '40']);
expect(result).toEqual(['--top-k', '40']);
});
it('respects stripContext=false to keep context flags', () => {
const result = stripShadowingFlags(['-c', '4096'], { stripContext: false });
expect(result).toEqual(['-c', '4096']);
});
it('passes through cache flags (no longer shadowed)', () => {
const result = stripShadowingFlags(['--cache-type-k', 'q8_0']);
expect(result).toEqual(['--cache-type-k', 'q8_0']);
});
it('passes through spec flags (no longer shadowed)', () => {
const result = stripShadowingFlags(['--spec-draft-n-max', '16']);
expect(result).toEqual(['--spec-draft-n-max', '16']);
});
});
describe('AGENTS.md frontmatter validation', () => {
it('rejects agent with managed flag in llama_extra_args', () => {
const md = `## Evil Agent
---
llama_extra_args: ["--model", "evil.gguf"]
---
You are evil.`;
const { agents, errors } = parseAgentsMd(md);
expect(agents).toHaveLength(0);
expect(errors).toHaveLength(1);
expect(errors[0]!.reason).toContain('managed');
});
it('accepts agent with safe llama_extra_args', () => {
const md = `## Good Agent
---
llama_extra_args: ["--top-k", "20"]
---
You are good.`;
const { agents, errors } = parseAgentsMd(md);
expect(errors).toHaveLength(0);
expect(agents).toHaveLength(1);
expect(agents[0]!.llama_extra_args).toEqual(['--top-k', '20']);
});
it('agent without llama_extra_args has null field', () => {
const md = `## Simple Agent
---
temperature: 0.5
---
You are simple.`;
const { agents } = parseAgentsMd(md);
expect(agents[0]!.llama_extra_args).toBeNull();
});
});