spec : refactor params

[ggerganov]

Overview

Refactor the speculative decoding parameters.

Additional information

Important

Some old CLI arguments are now replaced. See the table below

Old Parameter(s)	New Parameter(s)
--draft, --draft-n, --draft-max	--spec-draft-n-max or --spec-ngram-mod-n-max
--draft-min, --draft-n-min	--spec-draft-n-min or --spec-ngram-mod-n-min
--spec-ngram-size-n	--spec-ngram-*-size-n or --spec-ngram-mod-n-match
--spec-ngram-size-m	--spec-ngram-*-size-m
--spec-ngram-min-hits	--spec-ngram-*-min-hits

Important

Some old llama-server request parameters are removed

Parameter	Description
speculative.ngram_size_n	N-gram size for lookup
speculative.ngram_size_m	M-gram size for speculative tokens
speculative.ngram_m_hits	Minimum hits at n-gram/m-gram lookup

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: NO

[ggerganov]

With this change we can now combine ngram-based speculative decoding with draft-based spec. dec.:

llama-server -hf ggml-org/Qwen3.6-27B-GGUF:Q8_0 \
  --spec-draft-hf ggml-org/Qwen3.5-0.8B-GGUF:Q4_0 \
  --spec-type ngram-mod \
  --spec-ngram-mod-n-match 24 \
  --spec-ngram-mod-n-min 48 \
  --spec-ngram-mod-n-max 64