def4abcb25
Debug scripts used to diagnose CUDA layer offloading issue: - context-test.mjs: Context size testing - deep-check.mjs: Detailed CUDA initialization check - gpu-debug.mjs: GPU loading debug output - test-layers.mjs: gpuLayers parameter testing (found -1 bug) - trace-init.mjs: Provider initialization tracing 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
18 lines
559 B
JavaScript
18 lines
559 B
JavaScript
import { getLlama } from 'node-llama-cpp';
|
|
|
|
console.log('=== GPU Loading Debug ===\n');
|
|
|
|
// Try with explicit CUDA and debug
|
|
const llama = await getLlama({ gpu: 'cuda', debug: true });
|
|
console.log('Llama GPU:', llama.gpu);
|
|
|
|
const modelPath = '/var/home/lilith/.cache/models/Ministral-3-3B-Instruct-2512-Q8_0.gguf';
|
|
|
|
console.log('\nLoading model with gpuLayers: 999...');
|
|
const model = await llama.loadModel({
|
|
modelPath,
|
|
gpuLayers: 999, // Force all layers to GPU
|
|
});
|
|
|
|
console.log('\nModel loaded. GPU layers:', model.gpuLayers);
|
|
await model.dispose();
|