Frank
57264f7faa
- frank-default-large: GGML_CUDA_FA_ALL_QUANTS=ON is mandatory in the FastMTP build; without it q8_0 KV + FA falls back to a slow path and prefill collapses ~12x (95 -> 1312 t/s at 8K, measured) - frank-default-small: FLASH_ATTENTION_TRITON_AMD_ENABLE=TRUE + pure-python flash_attn (triton_amd) + stub; without it the vision encoder warmup OOMs via torch SDPA
63 lines
2.1 KiB
Bash
63 lines
2.1 KiB
Bash
# Tunables for frank-default-large (llama.cpp, dankserver, RTX 3090).
|
|
#
|
|
# Runtime: patched llama-server (HauhauCS FastMTP build) serving
|
|
# HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF IQ4_XS with
|
|
# the FastMTP 32K draft sidecar and the ngram-mod fallback. Single machine,
|
|
# no pipeline split. Alias frank-default-large, port 7770.
|
|
#
|
|
# WARNING: do not name any variable TEMP, HOST, or GPU_MAX_HEAP_SIZE. The ROCm
|
|
# HIP runtime reads those bare env vars (see the 2026-08-13 TEMP incident).
|
|
# This box is NVIDIA/CUDA, but the rule still applies to keep the env files
|
|
# portable.
|
|
|
|
# Model files (downloaded to this directory from the HF repo).
|
|
MODEL_DIR=/home/dank/models/hauhaucs-qwen38
|
|
MODEL_FILE=Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-IQ4_XS.gguf
|
|
DRAFT_FILE=Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-FastMTP-32K.gguf
|
|
# Patched llama.cpp binary. The FastMTP sidecar needs the patch; see AGENTS.md.
|
|
# REQUIRED build flags: -DGGML_CUDA_FA=ON -DGGML_CUDA_FA_ALL_QUANTS=ON.
|
|
# Without FA_ALL_QUANTS the q8_0 KV + flash attention fall back to a slow
|
|
# path and prefill collapses ~12x (95 t/s vs 1300+ t/s at 8K prompt,
|
|
# measured 2026-08-17). Do not rebuild without it.
|
|
LLAMA_BIN=/home/dank/llama.cpp-fastmtp/build/bin/llama-server
|
|
|
|
ALIAS=frank-default-large
|
|
BIND_ADDR=0.0.0.0
|
|
LISTEN_PORT=7770
|
|
|
|
# Context, single slot. 131072 (128K) leaves ~1.4 GiB VRAM headroom on the
|
|
# 3090 with IQ4_XS + FastMTP + K q8_0 / V f16 KV. Lower to 122880 (120K) for
|
|
# ~1.8 GiB headroom. KV per token = 16 attention layers x 4 KV heads x 256
|
|
# dim: K q8_0 (1088 B/token) + V f16 (2048 B/token) = 3136 B/token/layer.
|
|
CTX_SIZE=131072
|
|
PARALLEL=1
|
|
|
|
# Speculative decoding. FastMTP (MTP depth 2) with the ngram-mod fallback.
|
|
SPEC_DEFAULT=1
|
|
MTP=1
|
|
SPEC_DRAFT_N_MAX=2
|
|
SPEC_DRAFT_NGL=all
|
|
SPEC_DRAFT_P_MIN=0
|
|
SPEC_NGRAM_MOD_N_MIN=48
|
|
SPEC_NGRAM_MOD_N_MAX=64
|
|
SPEC_NGRAM_MOD_N_MATCH=24
|
|
|
|
# KV quantization on GPU: K q8_0 + V f16 (user request: K=8, V=16).
|
|
CACHE_TYPE_K=q8_0
|
|
CACHE_TYPE_V=f16
|
|
|
|
# Batch sizing (user request).
|
|
BATCH_SIZE=2048
|
|
UBATCH_SIZE=1024
|
|
|
|
# Sampling - Qwen3.8 thinking-mode defaults.
|
|
TEMPERATURE=1.0
|
|
TOP_K=20
|
|
TOP_P=0.95
|
|
MIN_P=0.0
|
|
|
|
THREADS=8
|
|
THREADS_HTTP=1
|
|
|
|
VERBOSITY=4
|