feat: add speculative decoding

This commit is contained in:
jialin
2025-11-03 10:14:30 +08:00
parent 608b2e31c7
commit 3613fb3f69
15 changed files with 356 additions and 165 deletions
+7 -1
View File
@@ -218,5 +218,11 @@ export default {
'models.form.ramRatio.tips':
'Ratio of system RAM to GPU VRAM used for KV cache. For example, 2.0 means the cache in RAM can be twice as large as the GPU VRAM.',
'models.form.ramSize.tips': `Maximum size of the KV cache stored in system memory (GiB). If set, this value overrides "{content}".`,
'models.form.chunkSize.tips': 'Number of tokens per KV cache chunk.'
'models.form.chunkSize.tips': 'Number of tokens per KV cache chunk.',
'models.form.flavor.throughput.tips':
'optimized for high throughput under high request concurrency.',
'models.form.flavor.latency.tips':
'optimized for low latency under low request concurrency.',
'models.form.flavor.reference.tips':
'the most compatible option with full precision.'
};
+7 -1
View File
@@ -218,7 +218,13 @@ export default {
'models.form.ramRatio.tips':
'Ratio of system RAM to GPU VRAM used for KV cache. For example, 2.0 means the cache in RAM can be twice as large as the GPU VRAM.',
'models.form.ramSize.tips': `Maximum size of the KV cache stored in system memory (GiB). If set, this value overrides "{content}".`,
'models.form.chunkSize.tips': 'Number of tokens per KV cache chunk.'
'models.form.chunkSize.tips': 'Number of tokens per KV cache chunk.',
'models.form.flavor.throughput.tips':
'optimized for high throughput under high request concurrency.',
'models.form.flavor.latency.tips':
'optimized for low latency under low request concurrency.',
'models.form.flavor.reference.tips':
'the most compatible option with full precision.'
};
// ========== To-Do: Translate Keys (Remove After Translation) ==========