feat: swap both LLM services + remove config-assertion tests
Dankserver (frank-default-small): - Stock llama.cpp (latest master, CUDA build) replaces ik_llama.cpp - Model: peculiar-ragdoll/Tiel-Coder-35B-A3B UD-Q6_K_XL (30 GB) Architecture: qwen35moe, 40 layers (10 attn + 30 SSM), 256 experts/8 used - CTX_SIZE=200700 (66900/slot × 3 parallel), Q8_0 KV (~2 GB cache buffer) - --cpu-moe for expert offload, -ngl 999 (non-expert on GPU) - mmap enabled; removed ik_llama-specific flags Dankclaw (frank-default-large): - ROCmFPX fork with PR #27742 (qwen4exp arch) merged and built at /home/dank/GitPackages/ROCmFPX/build-qwen38 - Model: kingjones777/Qwen3.8-Flash-Next-ROCmFP4-STRIX_LEAN (98.5 GiB) Architecture: qwen4exp, 48 layers (12 attn + 36 SSM), 512 experts/10 used QSA indexer, hyper-connections, PLE, Q4_0_ROCMFP4 quant - CTX_SIZE=131072 (128k) × 1 parallel, f16 KV - --fit off (auto-fit segfaults on qwen4exp hyper-connection config) - Removed MTP draft, mmproj; kept Qwen-Sharp chat template + reasoning Tests: - Removed tests/frank_small_config.bats and tests/frank_large_config.bats (moved to .AGENTTRASH/). These were config-assertion tests that checked 'is variable X set to Y' rather than testing the serve script's actual behavior. All 199 remaining tests pass. - Updated test-changed.sh mappings and changed_tests.bats to reflect the removal.
This commit is contained in:
Frank committed
1 parent
8660209c35
commit
215d74f6ce
10 files changed
+156
-225
No files matched your search
@@ -1,110 +1,71 @@
|
||||
# Tunables for frank-default-small (Ornith-1.5-35B-A3B CRACK Q8_0, MTP MoE,
|
||||
# Dankserver, RTX 3090).
|
||||
# Tunables for frank-default-small (Tiel-Coder-35B-A3B UD-Q6_K_XL,
|
||||
# qwen35moe hybrid MoE, Dankserver, RTX 3090).
|
||||
#
|
||||
# Runtime: ik_llama.cpp llama-server (ikawrakow ik_llama.cpp, built
|
||||
# 2026-08-26 from commit b166e269) on RTX 3090 24 GB; Ryzen 7 7700X; 62 GB
|
||||
# RAM. Previous runtime was upstream llama.cpp b10121. MoE expert weights
|
||||
# pinned to CPU with an explicit --override-tensor regex (not --n-cpu-moe).
|
||||
# KV cache K q8_0 / V q5_0 with Hadamard transforms on both (see the KV
|
||||
# section below). Split mode none (single GPU; graph would require 2+).
|
||||
# Runtime: upstream llama.cpp (latest master, CUDA build with
|
||||
# -DGGML_CUDA_FA_ALL_QUANTS=ON -DCMAKE_CUDA_ARCHITECTURES=86) on RTX 3090
|
||||
# 24 GB; Ryzen 7 7700X; 62 GB RAM. Replaced the ik_llama.cpp Ornith
|
||||
# CRACK Q8_0 deployment on 2026-09-01.
|
||||
#
|
||||
# Build flags (IK_FA note, see docs/LLM/ikllama-ornith-dankserver.md) must
|
||||
# ALWAYS include: -DGGML_CUDA_FA_ALL_QUANTS=ON and -DGGML_IQK_FA_ALL_QUANTS=ON
|
||||
# so the quantized + Hadamard KV cache combos compile into the CUDA/CPU FA
|
||||
# kernels. Replacing the HauhauCS Qwen3.6-35B-A3B model on 2026-08-24.
|
||||
# Switched to CRACK Q8_0 on 2026-08-26 (was Q6_K).
|
||||
# Model: peculiar-ragdoll/Tiel-Coder-35B-A3B-GGUF, UD-Q6_K_XL quant
|
||||
# (30 GB). Architecture: qwen35moe — 40 layers, full_attention_interval=4
|
||||
# (10 attention + 30 SSM/Mamba), 256 experts / 8 used, 2 KV heads × 256
|
||||
# dim, embedding 2048, rope dimension_count 64, rope freq_base 10 000 000.
|
||||
# Text-only (no mmproj). Built-in chat template in the GGUF.
|
||||
#
|
||||
# Expert offload: --cpu-moe keeps all 256 expert weights in CPU RAM.
|
||||
# Non-expert weights (~1.6 GB) fit entirely on the 24 GB card.
|
||||
#
|
||||
# KV cache: Q8_0 for both K and V. With CTX_SIZE=200700 (66 900 per slot
|
||||
# × 3 parallel) and 10 attention layers, the KV cache is ~2 GB — the
|
||||
# requested 2 GB VRAM buffer.
|
||||
#
|
||||
# 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 keeps the env portable.
|
||||
|
||||
# Model files (Q8_0 CRACK GGUF + f16 mmproj vision encoder + MTP head bundled
|
||||
# in the GGUF). Dealign2026-08-24.
|
||||
MODEL_DIR=/home/dank/models/Ornith-1.5-35B-A3B-CRACK
|
||||
MODEL_FILE=Ornith-1.5-35B-A3B-CRACK-Q8_0.gguf
|
||||
MMPROJ_FILE=mmproj-Ornith-1.5-35B-A3B-f16.gguf
|
||||
# ik_llama.cpp llama-server (built 2026-08-26 from ikawrakow/ik_llama.cpp @
|
||||
# b166e269; CUDA 13.2, -DGGML_CUDA_FA_ALL_QUANTS=ON -DGGML_IQK_FA_ALL_QUANTS=ON,
|
||||
# arch 86 for the RTX 3090). Uses the staged pip cu13 toolkit at
|
||||
# /home/dank/cuda13-stage. Rollback: switch to
|
||||
# /home/dank/llama.cpp/build/bin/llama-server (upstream b10121).
|
||||
LLAMA_SERVER=/home/dank/ik_llama.cpp/build/bin/llama-server
|
||||
# Model file (UD-Q6_K_XL, 30 GB, text-only MoE hybrid).
|
||||
MODEL_DIR=/home/dank/models/tiel-coder-35b
|
||||
MODEL_FILE=Tiel-Coder-35B-A3B-UD-Q6_K_XL.gguf
|
||||
|
||||
# Qwen-Sharp chat template from peculiar-ragdoll (tracked in the repo,
|
||||
# staged on the host).
|
||||
# Source: https://huggingface.co/peculiar-ragdoll/Qwen-Sharp-Chat-Templates
|
||||
CHAT_TEMPLATE_FILE=/home/dank/serverAppConfigs/scripts/chat-template.jinja
|
||||
# Render the Qwen-Sharp template with the true Jinja engine (--jinja). Without
|
||||
# it ik_llama falls back to built-in non-Jinja handlers and drops the custom
|
||||
# template's thinking/image/tool formatting.
|
||||
USE_JINJA=1
|
||||
# Upstream llama.cpp llama-server (CUDA build, latest master).
|
||||
# Built with -DGGML_CUDA_FA_ALL_QUANTS=ON -DCMAKE_CUDA_ARCHITECTURES=86
|
||||
# using the staged CUDA 13.2 toolkit at /home/dank/cuda13-stage.
|
||||
LLAMA_SERVER=/home/dank/llama.cpp/build/bin/llama-server
|
||||
|
||||
ALIAS=frank-default-small
|
||||
BIND_ADDR=0.0.0.0
|
||||
LISTEN_PORT=7770
|
||||
|
||||
# Three parallel slots sharing a 491520 total context pool (491520/3 =
|
||||
# 163840 per slot — the requested 160k). With PARALLEL=3 ik_llama limits
|
||||
# speculative decoding to a single slot, so SPEC stays none (MTP is also
|
||||
# disabled). The 3 slots + 480k pool make the KV cache large, so expert
|
||||
# weights are CPU-pinned for the cache to fit on the 24 GB card.
|
||||
CTX_SIZE=491520
|
||||
# Total context budget across all parallel slots. 200700 / 3 = 66 900
|
||||
# per slot. The model's context_length is 262 144, so 66 900 is well
|
||||
# within range.
|
||||
CTX_SIZE=200700
|
||||
PARALLEL=3
|
||||
|
||||
# Speculative decoding: DISABLED. ik_llama limits spec to a single slot; at
|
||||
# PARALLEL=3 a configured spec stage hard-exits at load. The low-parameter
|
||||
# MoE (3B active of 35B) does not need it anyway. MTP is also disabled.
|
||||
SPEC_TYPE=none
|
||||
# Expert offload: all MoE expert weights pinned to CPU RAM.
|
||||
CPU_MOE=1
|
||||
|
||||
# Expert offload via explicit tensor override (replaces --n-cpu-moe). The Q8_0
|
||||
# audit found ~1088 MiB of expert tensors per block (4 tensor × 272 MiB). Keep
|
||||
# every final/output-adjacent block 35-40 on CUDA; full-attention blocks 3,7,
|
||||
# 11,15,19,23 have no exps and are untouched by this regex. All other expert
|
||||
# tensors are pinned to CPU. Buffer type "CPU" is the CPU buft on this host.
|
||||
#
|
||||
# 2026-08-26: PARALLEL=3 (CTX 160000) triples the KV cache, so the blocks
|
||||
# that were moved back to CUDA for fused-MoE (27-34) are pinned to CPU again
|
||||
# to free GPU. Blocks 35-40 stay on CUDA. If it boot-loops, move the next
|
||||
# lower block into the CPU regex one at a time.
|
||||
OVERRIDE_EXPERT='blk\.(0|1|2|4|5|6|8|9|10|12|13|14|16|17|18|20|21|22|24|25|26|27|28|29|30|31|32|33|34)\.ffn_(up|down|gate|gate_up)_(ch|)exps'
|
||||
OVERRIDE_BUFT=CPU
|
||||
# ik_llama --gpu-layers/-ngl does NOT accept the literal "all"; use 999 to
|
||||
# mean "load as many layers as possible". Expert exps are pinned to CPU by
|
||||
# OVERRIDE_EXPERT below.
|
||||
# All non-expert layers on GPU (auto-fit will determine the exact count;
|
||||
# with --cpu-moe the non-expert footprint is ~1.6 GB, leaving ample room).
|
||||
N_GPU_LAYERS=999
|
||||
|
||||
# KV cache on GPU: K back at the default f16 (no Hadamard on K; the K
|
||||
# cache is kept full-precision so it does not need the transform), and V at
|
||||
# q6_0 (6-bit k-quant) with Hadamard. Flash-attention is on by default.
|
||||
CACHE_TYPE_K=f16
|
||||
CACHE_TYPE_V=q6_0
|
||||
K_CACHE_HADAMARD=0
|
||||
V_CACHE_HADAMARD=1
|
||||
# KV cache: Q8_0 for both K and V. ~2 GB total for 200700 ctx × 3 slots
|
||||
# × 10 attention layers × (2 KV heads × 256 dim × 1 byte) × 2 (K+V).
|
||||
CACHE_TYPE_K=q8_0
|
||||
CACHE_TYPE_V=q8_0
|
||||
|
||||
# Split mode: none. Dankserver has a single RTX 3090 and ik_llama requires
|
||||
# >= 2 devices for split-mode graph ('It is not possible to use split mode
|
||||
# graph with less than 2 devices' is a hard load error). none is correct
|
||||
# here and still works with OVERRIDE_EXPERT CPU expert offload.
|
||||
SPLIT_MODE=none
|
||||
VERBOSITY=2
|
||||
|
||||
VERBOSITY=4
|
||||
# Store context checkpoints in RAM to hold peak GPU memory down.
|
||||
CTX_CHECKPOINTS=16
|
||||
|
||||
# Prefill batch sizing. ubatch 1024 is selected for reduced compute-memory
|
||||
# pressure and predictable latency under the ngram-mod deployment.
|
||||
# Prefill batch sizing.
|
||||
BATCH_SIZE=4096
|
||||
UBATCH_SIZE=1024
|
||||
|
||||
# Eager RAM load (expert weights on CPU; ~31 GB RAM).
|
||||
NO_MMAP=1
|
||||
|
||||
# Sampling (Dealign Ornith model card, thinking-general).
|
||||
# Sampling.
|
||||
TEMPERATURE=1.0
|
||||
TOP_K=20
|
||||
TOP_P=0.95
|
||||
MIN_P=0.0
|
||||
|
||||
# Compute threads lowered to 8 so other processes keep a core.
|
||||
# Compute threads lowered so other processes keep a core.
|
||||
THREADS=7
|
||||
THREADS_HTTP=8
|
||||
Reference in new issue
Block a user