haylanandClaude Bot 8927e440aa feat(llm): make spec decoding and KV cache type .env toggles ...

Move all speculative-decoding flags into LLAMA_SPEC_ARGS (blank = off)
and the KV cache quant into LLAMA_KV_CACHE_TYPE, so token-corruption
A/B tests are a single .env edit instead of a compose edit.
Replaces LLAMA_DRAFT_MODEL_FILE and LLAMA_SPEC_N_MAX.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

2026-10-31 18:22:44 +01:00

..

agents

docs(agents): add writing-docs guide and new-term procedure

2026-09-24 21:52:19 +02:00

coding-cli-setup

docs(opencode): document v2 config, mark v1 deprecated

2026-10-05 21:32:58 +02:00

research

feat(llm): raise prompt cache RAM budget via --cache-ram

2026-10-20 21:05:27 +02:00

architecture.md

feat(llm): make spec decoding and KV cache type .env toggles

2026-10-31 18:22:44 +01:00

gateway.md

feat(llm): switch to DFlash2 speculative decoding, drop qwen-classifier

2026-10-05 21:28:55 +02:00

network-access.md

feat(llm): switch to DFlash2 speculative decoding, drop qwen-classifier

2026-10-05 21:28:55 +02:00

operations.md

refactor(models): merge downloaders into one, driven by models.yml

2026-10-31 16:48:58 +01:00

server-planing.md

Add build plan documentation for 4x Radeon AI PRO R9700 rackmount LLM server

2026-08-25 07:20:58 +02:00