ComfyUI Node
Llama.cpp N-gram Speculative Config
Produces the shared Compact speculative input using model-free prompt-history N-gram drafting.
Llama.cpp N-gram Speculative Config
- speculative
◄speculative_modeoff►
◄ngram_size3►
◄num_pred_tokens10►
◄ngram_modek►
◄ngram_min_hits2►
◄ngram_max_entries_per_key8►
◄ngram_sync_check_tokens16►
CategoryOllama/llama_cpp/compact
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| speculative_mode | COMBO | off | 2 options: off, ngram |
| ngram_size | INT | 31–8 | — |
| num_pred_tokens | INT | 101–32 | — |
| ngram_mode | COMBO | k | 2 options: k, k4v |
| ngram_min_hits | INT | 21–16 | — |
| ngram_max_entries_per_key | INT | 80–1024 | — |
| ngram_sync_check_tokens | INT | 161–256 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| speculative | OLLAMA_IMAGE_LIST_LLAMA_CPP_SPECULATIVE_CONFIG | — |