From 3b51b5a94a15a09173a3e96cdf1cf92d483f7db4 Mon Sep 17 00:00:00 2001 From: Frank Date: Mon, 17 Aug 2026 21:12:35 -0700 Subject: [PATCH] dankclaw: replace vLLM with CIRU ROCmFPX llama.cpp runtime Retire the vLLM service (frank-default-small, port 52625) on Dankclaw and serve jcbtc/Qwen3.8-27B-CIRU-ActiveFPX-PromptForge through the CIRU ROCmFPX llama.cpp runtime instead. - Build llama-server from ciru-ai/ROCmFPX tag qwen3.8-activefpx-promptforge-v2.3 (commit 103af8b41) against the private TheRock ROCm 7.15 SDK in the retired ling3-ciru venv. The model card requires ROCm 7.15 for the advertised Strix Halo performance; the system ROCm 7.2.4 selects slower kernels. CK pinned at fdf4bb7fcc984811cef48ce817d89aac064b984a; generated ck/config.h + ck/version.h copied into the CK source tree so the ggml-hip build can include them. LLAMA_BUILD_WEBUI=OFF (HF bucket llama-ui returns 404; server is consumed via the OpenAI API). - New scripts/serve-frank-small.sh + frank-default-small.env: 64K ctx, one slot, f16 KV, context checkpoints, prompt cache, native depth-4 MTP on ROCm0, PromptForge DualView sidecars (FFN + GDN), temp 0.7 top-p 0.95, reasoning off. LD_LIBRARY_PATH puts the 7.15 SDK libs before any system ROCm. - Update the systemd unit description (same service name and port). - deploy.sh: update the dankclaw target comment. - Docs: new docs/ciru-activefpx-promptforge.md with the build runbook; update docs/index.md and AGENTS.md topology. - The vLLM venvs (vllm-venv, vllm-vk-venv) were moved to /home/dank/.AGENTTRASH/ on the target; the vllm-vulkan plugin and docs were removed from the repo in an earlier commit. --- AGENTS.md | 60 +++++----- docs/ciru-activefpx-promptforge.md | 167 ++++++++++++++++++++++++++++ docs/index.md | 26 +++-- scripts/deploy.sh | 7 +- scripts/frank-default-small.env | 115 +++++++++---------- scripts/frank-default-small.service | 3 +- scripts/serve-frank-small.sh | 138 +++++++++++++++-------- 7 files changed, 359 insertions(+), 157 deletions(-) create mode 100644 docs/ciru-activefpx-promptforge.md diff --git a/AGENTS.md b/AGENTS.md index d4bfb50..72f97dc 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -3,16 +3,18 @@ ## Overview This project manages deployment scripts for the LLM inference fleet on two -local machines. Topology (2026-08-17 redo): +local machines. Topology (2026-08-18): | Target | Host | Runtime | Service | Model | Port | |--------|------|---------|---------|-------|------| | Dankserver | dank@dankserver | llama.cpp (patched FastMTP build) | frank-default-large | HauhauCS Qwen3.8-27B-Uncensored-Aggressive IQ4_XS + FastMTP 32K draft + ngram-mod | 7770 | -| Dankclaw | dank@dankclaw | vLLM (ROCm, /home/dank/vllm-venv) | frank-default-small | nameistoken Qwen3.6-35B-A3B Quark W8A8 INT8, ngram spec, tool calling | 52625 | +| Dankclaw | dank@dankclaw | CIRU ROCmFPX llama.cpp (ROCm 7.15 SDK in the ling3-ciru venv) | frank-default-small | jcbtc Qwen3.8-27B-CIRU-ActiveFPX-PromptForge v2, native MTP, PromptForge DualView | 52625 | The 2026-08-17 redo retired the pipeline-parallel split and the FastFlowLM -NPU endpoint. The RTX 3090 now runs the whole 27B model standalone -(no RPC worker). Dankclaw runs vLLM on the Strix Halo iGPU (gfx1151). +NPU endpoint. The RTX 3090 runs the whole 27B model standalone +(no RPC worker). The 2026-08-18 swap retired the vLLM service on Dankclaw; +frank-default-small now runs the CIRU ROCmFPX llama.cpp runtime on the +Strix Halo iGPU (gfx1151). See docs/ciru-activefpx-promptforge.md. The old scripts live in `old/scripts/` for reference; they describe the retired PP topology. @@ -60,7 +62,7 @@ Use `scripts/deploy.sh` to deploy committed changes to a target. It pushes `main ``` FRANK_SUDO_PW=... ./scripts/deploy.sh dankserver # llama.cpp FastMTP, port 7770 -FRANK_SUDO_PW=... ./scripts/deploy.sh dankclaw # vLLM, port 52625 +FRANK_SUDO_PW=... ./scripts/deploy.sh dankclaw # CIRU llama.cpp, port 52625 ./scripts/deploy.sh dankserver --dry-run # preview without changes ``` @@ -96,37 +98,27 @@ back to a slow path and prefill collapses ~12x (measured 2026-08-17: 95 t/s vs 1300+ t/s at an 8K prompt on the 3090). See the HauhauCS HF repo README for the base recipe. -### vLLM ROCm install (Dankclaw) +### CIRU ROCmFPX llama.cpp (Dankclaw) -vLLM runs from `/home/dank/vllm-venv` (Python 3.12). Install recipe: +The CIRU llama-server serves `jcbtc/Qwen3.8-27B-CIRU-ActiveFPX-PromptForge` +from `/home/dank/ciru-qwen38/ROCmFPX/build-promptforge/bin/llama-server`. +The runtime is the ciru-ai/ROCmFPX tag `qwen3.8-activefpx-promptforge-v2.3` +(commit `103af8b41`) built against the private TheRock ROCm 7.15 SDK kept +in the retired ling3-ciru venv +(`.venv/lib/python3.12/site-packages/_rocm_sdk_*`). The model card requires +ROCm 7.15 for the advertised Strix Halo performance; the system ROCm 7.2.4 +selects slower kernels. The serve script exports the 7.15 SDK libraries +first in `LD_LIBRARY_PATH`. -``` -uv venv --python /usr/bin/python3.12 /home/dank/vllm-venv -uv pip install --python /home/dank/vllm-venv vllm --torch-backend=auto -``` +The model files (v2 GGUF + FFN.pfs + GDN.pfs, ~34.9 GiB) live in +`/home/dank/models/CIRU-Qwen3.8-27B-ActiveFPX/`. The full build runbook is +in docs/ciru-activefpx-promptforge.md. The retired vLLM venvs +(`/home/dank/vllm-venv`, `/home/dank/vllm-vk-venv`) were moved to +`/home/dank/.AGENTTRASH/` on 2026-08-18. -`--torch-backend=auto` resolves the AMD ROCm wheel set (torch + -amd-torch-device-gfx1151 + the vllm rocm variant) from the AMD index. -The model cache is the default HF cache -(`/home/dank/.cache/huggingface`); the HF token lives at -`~/.cache/huggingface/token`. The retired CIRU package -(`/home/dank/ling3-ciru`) documents a source-build fallback for gfx1151. - -**ROCm flash-attention requirement:** the vision encoder needs the Triton -flash-attention path. Set `FLASH_ATTENTION_TRITON_AMD_ENABLE=TRUE` (in -`frank-default-small.env`) and keep the pure-Python `flash_attn` package -(copied from the CIRU venv) plus a `flash_attn_2_cuda` stub in the venv so -`flash_attn.flash_attn_triton_amd` is findable. Without it the mm encoder -falls back to torch SDPA and engine init OOMs (256 GiB score matrix). - -**Tool calling (frank-default-small):** the server runs -`--enable-auto-tool-choice --tool-call-parser qwen3_xml` with a custom -chat template (`/home/dank/models/qwen36-chat_template-tools.jinja`). -The template adds a concrete worked tool-call example and defaults tool -requests to non-thinking (Qwen3 tool calls only work reliably with -thinking off; normal requests keep thinking). The Avesed INT8 quant -emitted malformed Qwen3 XML tool calls; the Quark W8A8 model follows the -format cleanly. +**Env var naming rule still applies:** the HIP runtime reads `TEMP`, `HOST`, +and `GPU_MAX_HEAP_SIZE`. The serve script passes `--host`/`--port`/`--temp` +as CLI flags only. ### Env var naming rule — HIP runtime collision @@ -171,7 +163,7 @@ graph TD - `sudo` on both boxes needs a tty. Remote sudo steps fail non-interactively. Use `echo '' | sudo -S `. - Never run `pkill -f` with a pattern that appears in your own command line. It kills your ssh session. Use `pkill -x `. -- Model load is slow: Dankserver ~5 minutes cold (15.7 GB IQ4_XS through RAM, `--no-mmap`); Dankclaw vLLM ~5-10 minutes (38 GB checkpoint). Health returns 503 while loading. Do not kill the process. `deploy.sh` waits up to 10 minutes on Dankserver and 20 minutes on Dankclaw (see `HEALTH_WAIT`). +- Model load is slow: Dankserver ~5 minutes cold (15.7 GB IQ4_XS through RAM, `--no-mmap`); Dankclaw CIRU ~5-10 minutes (34.9 GiB ActiveFPX payload). Health returns 503 while loading. Do not kill the process. `deploy.sh` waits up to 10 minutes on Dankserver and 20 minutes on Dankclaw (see `HEALTH_WAIT`). - A client abort and a server release look different in the journal. A client abort logs `W srv stop: cancel task, id_task = N`. A natural stop (EOS or limit) logs only `slot release` with `truncated = 0`. The server never preempts a running generation; when all slots are busy, new tasks defer into a queue. See the 2026-08-13 incident in [docs/deployment.md](docs/deployment.md). - Retired hardware history: the Dankserver 35B cuBLAS crash and the DFlash premature-EOS bug are historical; see [docs/dankserver-vram-growth.md](docs/dankserver-vram-growth.md) and [docs/qwen3.8-spec-matrix.md](docs/qwen3.8-spec-matrix.md). The FastFlowLM NPU endpoint is retired; see [docs/FastFlowLM NPU Inference.md](docs/FastFlowLM%20NPU%20Inference.md). diff --git a/docs/ciru-activefpx-promptforge.md b/docs/ciru-activefpx-promptforge.md new file mode 100644 index 0000000..1e82eb7 --- /dev/null +++ b/docs/ciru-activefpx-promptforge.md @@ -0,0 +1,167 @@ +--- +type: reference +title: CIRU ActiveFPX PromptForge on Dankclaw +description: jcbtc/Qwen3.8-27B-CIRU-ActiveFPX-PromptForge served by the CIRU ROCmFPX llama.cpp runtime on the Strix Halo iGPU. Install runbook, ROCm 7.15 toolchain, and the frank-default-small service. +tags: + - llama.cpp + - rocm + - dankclaw + - gfx1151 + - ciru + - promptforge +timestamp: 2026-08-18 +--- + +# CIRU ActiveFPX PromptForge on Dankclaw + +`frank-default-small` (port 52625) now serves +`jcbtc/Qwen3.8-27B-CIRU-ActiveFPX-PromptForge` through the CIRU ROCmFPX +llama.cpp runtime. This replaced the retired vLLM service that previously +occupied the slot. + +## What This Is + +The model is a **GGUF**, not a vLLM checkpoint. It is the quality-and-speed +release of Qwen3.8-27B for AMD ROCm. CIRU DualView keeps a compact ActiveFPX +GGUF for generation and load optional PromptForge companion views to +accelerate large prompts. The model carries native depth-4 multi-token +prediction (MTP). + +| Artifact | Size | +|----------|------| +| `Qwen3.8-27B-CIRU-ActiveFPX-PromptForge-v2.gguf` | 15.19 GiB | +| `Qwen3.8-27B-CIRU-ActiveFPX-PromptForge-FFN.pfs` | 15.95 GiB | +| `Qwen3.8-27B-CIRU-ActiveFPX-PromptForge-GDN.pfs` | 3.75 GiB | +| Three-file payload | ~34.89 GiB | + +The two `.pfs` companions are optional. Omit them to run the GGUF alone and +save ~19.7 GiB of memory. The GGUF remains independently usable. + +## Requirements + +- **ROCm 7.15 is required** for the advertised Strix Halo performance. + The system ROCm (`/opt/rocm-7.2.4`) selects slower or fallback kernels. +- The pinned runtime is `ciru-ai/ROCmFPX` tag + `qwen3.8-activefpx-promptforge-v2.3` (commit `103af8b41`). +- Composable Kernel at `fdf4bb7fcc984811cef48ce817d89aac064b984a`. +- Validated GPU: AMD Radeon 8060S / gfx1151. + +## ROCm 7.15 Toolchain + +The retired `ling3-ciru` package left a complete TheRock ROCm 7.15 SDK +(`7.15.0a20260728`) inside its venv: + +``` +/home/dank/ling3-ciru/Ling-3.0-Flash-CIRU-int4-Strix-native/.venv/lib/python3.12/site-packages/ + _rocm_sdk_devel/ # hipcc, amdclang 23, cmake, headers + _rocm_sdk_libraries/ # hipblas, hipblaslt, rocroller + _rocm_sdk_core/ # libamdhip64, libhsa-runtime64 + rocm-sdk-device-gfx1151 +``` + +The compiler is amdclang 23. `rocm-sdk version` reports `7.15.0a20260728`. +The binary is built against and run with these 7.15 libraries. + +## Build Runbook (2026-08-18) + +Performed on Dankclaw: + +1. Install host build tools (ninja-build was missing): `sudo apt install ninja-build`. +2. Clone the runtime and CK at their pinned revisions: + ``` + git clone --depth 1 --branch qwen3.8-activefpx-promptforge-v2.3 \ + https://github.com/ciru-ai/ROCmFPX.git + git clone --depth 1 https://github.com/ROCm/composable_kernel.git + git -C composable_kernel checkout fdf4bb7fcc984811cef48ce817d89aac064b984a + ``` +3. CK headers are generated at CK configure time (`ck/config.h`, + `ck/version.h`). The llama.cpp build includes the CK source tree, so run + CK's configure once and copy the headers into `include/ck/`: + ``` + cmake -S composable_kernel -B build-ck -DAMDGPU_TARGETS=gfx1151 -DCK_ENABLE_WMMA=ON + cp build-ck/include/ck/config.h build-ck/include/ck/version.h composable_kernel/include/ck/ + ``` +4. Configure the llama.cpp build against the venv ROCm 7.15 SDK: + ``` + export ROCM_PATH=$SDK/_rocm_sdk_devel + cmake -S ROCmFPX -B build-promptforge -G Ninja \ + -DCMAKE_BUILD_TYPE=Release \ + -DCMAKE_C_COMPILER=$ROCM_PATH/llvm/bin/clang \ + -DCMAKE_CXX_COMPILER=$ROCM_PATH/llvm/bin/clang++ \ + -DCMAKE_HIP_COMPILER=$ROCM_PATH/llvm/bin/clang++ \ + -DCMAKE_PREFIX_PATH=$ROCM_PATH \ + -DGGML_HIP=ON -DGGML_CUDA=OFF -DGGML_VULKAN=OFF \ + -DGGML_HIP_FORCE_MMQ=ON -DGGML_HIP_GRAPHS=ON \ + -DGGML_HIP_MMQ_MFMA=ON -DGGML_HIP_NO_VMM=ON \ + -DGGML_HIP_ROCWMMA_FATTN=OFF -DGGML_NATIVE=ON \ + -DAMDGPU_TARGETS=gfx1151 -DGPU_BUILD_TARGETS=gfx1151 \ + -DPROMPTFORGE_CK_ROOT=$CK_ROOT \ + -DLLAMA_BUILD_SERVER=ON \ + -DLLAMA_BUILD_WEBUI=OFF + cmake --build build-promptforge --target llama-server -j$(nproc) + ``` + `LLAMA_BUILD_WEBUI=OFF` is required. The embedded browser UI downloads + assets from an HF bucket (`llama-ui`) that returns 404; the service is + consumed through the OpenAI-compatible API, so the UI is not needed. + +Build output: `build-promptforge/bin/llama-server`. + +```mermaid +graph TD + A[Clone ROCmFPX v2.3 + CK pinned] --> B[Configure CK, copy generated headers] + B --> C[Configure llama-server vs ROCm 7.15 SDK] + C --> D[Build llama-server] + D --> E[Verify --list-devices shows gfx1151] + E --> F[Download v2 GGUF + FFN + GDN] + F --> G[Deploy frank-default-small] + G --> H[Health check + chat completion] +``` + +## Model Download + +Downloaded with the HF token to +`/home/dank/models/CIRU-Qwen3.8-27B-ActiveFPX/`. + +## Service + +- Unit: `frank-default-small.service`, port 52625 +- Script: `scripts/serve-frank-small.sh`, env `scripts/frank-default-small.env` +- The script exports the ROCm 7.15 SDK libraries in `LD_LIBRARY_PATH` before + the system ROCm. `ROCM_PATH` points at the devel SDK. + +### Launch profile (from the model card) + +- 64K context, one slot, f16 KV, context checkpoints (32), prompt cache +- Native MTP draft, depth 4, device ROCm0 +- `--temp 0.7 --top-p 0.95`, reasoning off +- PromptForge env: `PROMPTFORGE_MODE=m2048_fused_tail1476`, + `GGML_CUDA_GRAPH_OPT=0`, `LLAMA_MTP_CPU_ARGMAX_FASTPATH=1` + +### Required environment + +``` +export PROMPTFORGE_SIDECAR=...FFN.pfs +export PROMPTFORGE_GDN_SIDECAR=...GDN.pfs +export PROMPTFORGE_MODE=m2048_fused_tail1476 +export GGML_CUDA_GRAPH_OPT=0 +export LLAMA_MTP_CPU_ARGMAX_FASTPATH=1 +``` + +The `p_min=0` MTP profile uses the CPU argmax fast path. This avoids ROCm's +unsupported TOP_K operator. + +## Operational Notes + +- Health: `curl http://127.0.0.1:52625/health`. Returns 503 while loading. +- Model load is slow (34.9 GiB payload); do not kill the process during + warmup. +- `deploy.sh dankclaw` waits 20 minutes for health. +- The systemd unit must not set `HOST` or `TEMP` env vars (HIP collision + rule). `--host`/`--port`/`--temp` are CLI flags only. + +## See Also + +- [[frank-redo-2026-08-17]] — the fleet topologies this slot has served. +- [[Shared Memory Architecture Dankclaw]] — unified memory sizing. +- [[qwen3.8-strix-optimizations]] — llama.cpp ROCm0 baselines on this iGPU. +- [[VLLM Overview]] — vLLM reference (retired for this slot). diff --git a/docs/index.md b/docs/index.md index 93cead5..c29d8e3 100644 --- a/docs/index.md +++ b/docs/index.md @@ -1,8 +1,8 @@ # vLLM Scripts — Knowledge Base Documentation for the LLM inference fleet on two local machines -(2026-08-17 redo): llama.cpp FastMTP on Dankserver and vLLM ROCm on -Dankclaw. +(2026-08-18): llama.cpp FastMTP on Dankserver and CIRU ROCmFPX llama.cpp +on Dankclaw. ## Deployment Topology @@ -19,10 +19,10 @@ FastMTP 32K draft + ngram-mod, 128K ctx] subgraph Dankclaw[AMD Strix Halo - Ubuntu 26.04] D[gfx1151 iGPU unified memory] - E[vLLM ROCm + E[CIRU ROCmFPX llama.cpp frank-default-small, port 52625] - F[nameistoken Qwen3.6-35B-A3B Quark W8A8 INT8 -full 262144 ctx, ngram spec + tools, 4 seqs] + F[jcbtc Qwen3.8-27B-CIRU-ActiveFPX-PromptForge +64K ctx, native MTP, PromptForge DualView] D --> E --> F end @@ -40,17 +40,23 @@ full 262144 ctx, ngram spec + tools, 4 seqs] ## Active Deployment - [[deployment]] — deploy and update process for both servers -- [[frank-redo-2026-08-17]] — the full redo: topology change, model - choices, context math, and the two install runbooks +- [[frank-redo-2026-08-17]] — the 2026-08-17 topology change +- [[ciru-activefpx-promptforge]] — the CIRU runtime on Dankclaw + (frank-default-small) - [[Dankserver]] — the llama.cpp FastMTP box (frank-default-large) -- [[Dankclaw]] — the vLLM ROCm box (frank-default-small) +- [[Dankclaw]] — the CIRU ROCmFPX box (frank-default-small) -## vLLM Reference (Dankclaw) +## CIRU Reference (Dankclaw) + +- [[ciru-activefpx-promptforge]] — model, ROCm 7.15 toolchain, build, + and the frank-default-small service + +## vLLM Reference (historical) - [[VLLM Overview]] — installation, CLI, API, architecture, quantization, parallelism - [[VLLM Index]] — quick reference with commands, endpoints, and parameter table - [[Running Qwen3.6 35B A3B FP8 on Dankserver]] — Qwen3.6-35B-A3B hardware - notes and sizing (the model family served by frank-default-small) + notes and sizing (the model family formerly served by frank-default-small) ## Memory Management diff --git a/scripts/deploy.sh b/scripts/deploy.sh index 9b2aba9..d7993eb 100755 --- a/scripts/deploy.sh +++ b/scripts/deploy.sh @@ -16,9 +16,9 @@ # ./deploy.sh [--dry-run] # FRANK_SUDO_PW=... ./deploy.sh # -# Targets (2026-08-17 topology): +# Targets (2026-08-18 topology): # dankserver - llama.cpp FastMTP (frank-default-large.service, port 7770) -# dankclaw - vLLM ROCm (frank-default-small.service, port 52625) +# dankclaw - CIRU ROCmFPX llama.cpp (frank-default-small.service, port 52625) set -euo pipefail # --------------------------------------------------------------------------- @@ -41,7 +41,8 @@ REMOTE_DIR[dankclaw]="/home/dank/vllmScripts" SERVICE[dankclaw]="frank-default-small" SERVICE_HEALTH[dankclaw]="http://127.0.0.1:52625/health" DROP_CACHES[dankclaw]=0 -# vLLM loads a 38 GB checkpoint; allow 20 minutes. +# CIRU llama.cpp (ROCm 7.15) loads the 34.9 GiB ActiveFPX payload + MTP +# draft and initializes PromptForge views; allow 20 minutes. HEALTH_WAIT[dankclaw]=600 TARGET="${1:-}" diff --git a/scripts/frank-default-small.env b/scripts/frank-default-small.env index b18f6cb..0039554 100644 --- a/scripts/frank-default-small.env +++ b/scripts/frank-default-small.env @@ -1,72 +1,65 @@ -# Tunables for frank-default-small (vLLM, dankclaw, Strix Halo gfx1151). +# Tunables for frank-default-small (CIRU llama.cpp, dankclaw, Strix Halo +# gfx1151). # -# Runtime: vLLM (ROCm) in /home/dank/vllm-venv serving -# Avesed/Qwen3.6-35B-A3B-INT8-W8A8 (compressed-tensors INT8 W8A8, MoE 35B -# total / 3B active, Qwen3.5-MoE hybrid: 30 of 40 layers are GatedDeltaNet -# linear attention). Alias frank-default-small, port 52625. +# Runtime: the PromptForge llama-server built from ciru-ai/ROCmFPX tag +# qwen3.8-activefpx-promptforge-v2.3 serving +# jcbtc/Qwen3.8-27B-CIRU-ActiveFPX-PromptForge (v2). ActiveFPX compact +# GGUF with PromptForge companions for CIRU DualView prompt acceleration and +# native depth-4 MTP. Alias frank-default-small, port 52625. # -# The ROCm vLLM install was created with: -# uv venv --python /usr/bin/python3.12 /home/dank/vllm-venv -# uv pip install --python /home/dank/vllm-venv vllm --torch-backend=auto -# The --torch-backend=auto resolver pulls the AMD ROCm wheels (torch + -# amd-torch-device-gfx1151 + vllm rocm variant) from the AMD index. +# This runtime replaced the retired vLLM service that previously occupied +# this slot. See docs/ciru-activefpx-promptforge.md. +# +# 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). +# llama.cpp reads --host/--port/--temp as CLI flags, never as env vars. + +# CIRU llama-server binary (ROCmFPX v2.3 build, see docs/ciru runtime page). +# Built against the private TheRock ROCm 7.15 SDK kept in the retired +# ling3-ciru venv. ROCm 7.15 is required for the advertised Strix Halo +# performance; the system ROCm 7.2.4 selects slower/fallback kernels. +LLAMA_BIN=/home/dank/ciru-qwen38/ROCmFPX/build-promptforge/bin/llama-server + +# Model files. The three-file payload (v2 GGUF + FFN + GDN) enables CIRU +# DualView prompt acceleration; omit the two .pfs files to run the GGUF alone +# and save ~19.7 GiB of memory at the cost of prompt throughput. +MODEL_DIR=/home/dank/models/CIRU-Qwen3.8-27B-ActiveFPX +MODEL_FILE=Qwen3.8-27B-CIRU-ActiveFPX-PromptForge-v2.gguf +FFN_PFS=Qwen3.8-27B-CIRU-ActiveFPX-PromptForge-FFN.pfs +GDN_PFS=Qwen3.8-27B-CIRU-ActiveFPX-PromptForge-GDN.pfs + +# ROCm 7.15 SDK paths (inside the retired ling3-ciru venv). The serve script +# exports ROCM_PATH and LD_LIBRARY_PATH from these so the 7.15 libraries +# precede any system ROCm 7.2.4. +ROCM_DEVEL=/home/dank/ling3-ciru/Ling-3.0-Flash-CIRU-int4-Strix-native/.venv/lib/python3.12/site-packages/_rocm_sdk_devel +ROCM_LIB=/home/dank/ling3-ciru/Ling-3.0-Flash-CIRU-int4-Strix-native/.venv/lib/python3.12/site-packages/_rocm_sdk_libraries +ROCM_CORE=/home/dank/ling3-ciru/Ling-3.0-Flash-CIRU-int4-Strix-native/.venv/lib/python3.12/site-packages/_rocm_sdk_core -VLLM_BIN=/home/dank/vllm-venv/bin/vllm -# 2026-08-17 (evening): swapped from Avesed/Qwen3.6-35B-A3B-INT8-W8A8 to -# nameistoken/Qwen3.6-35B-A3B-Quark-W8A8-INT8 (AMD Quark W8A8, same -# architecture, vLLM-native expert layout). The Avesed quant's tool-call -# XML fidelity was poor (malformed tags); the Quark model -# follows the Qwen3 XML tool format cleanly. -MODEL=nameistoken/Qwen3.6-35B-A3B-Quark-W8A8-INT8 ALIAS=frank-default-small BIND_ADDR=0.0.0.0 LISTEN_PORT=52625 -# Full native context. The KV cache stays unquantized (auto = model dtype, -# bf16). Worst-case KV at 262144 x 4 sequences is ~10.7 GiB; weights are -# ~35 GiB; total fits the Strix Halo unified pool with room to spare. -MAX_MODEL_LEN=262144 -KV_CACHE_DTYPE=auto +# Serving defaults from the model card: 64K context, single slot, f16 KV, +# context checkpoints, prompt cache. Native MTP depth-4. +CTX_SIZE=65536 +PARALLEL=1 +BATCH_SIZE=2048 +UBATCH_SIZE=2048 +CACHE_TYPE_K=f16 +CACHE_TYPE_V=f16 +CTX_CHECKPOINTS=32 +CACHE_RAM=8192 -# Concurrency and batching (user request). max-num-batched-tokens lowered -# to 8192 on 2026-08-17 (evening). vLLM has no separate ubatch knob. -MAX_NUM_SEQS=4 -MAX_NUM_BATCHED_TOKENS=8192 +# Native MTP draft (depth-4, greedy, ROCm-safe CPU argmax fast path). +SPEC_DRAFT_N_MAX=4 +SPEC_DRAFT_N_MIN=0 +SPEC_DRAFT_P_MIN=0.0 +SPEC_DRAFT_P_SPLIT=0.10 -# vLLM has no separate ubatch knob. The scheduler issues one batch per step -# with up to max-num-batched-tokens tokens; the KV cache is preallocated at -# max-model-len. Set gpu-memory-utilization below 1.0 so the unified pool -# keeps headroom for the host. -GPU_MEMORY_UTILIZATION=0.92 +# Sampling (model card defaults). +TEMPERATURE=0.7 +TOP_P=0.95 -# Prefix caching. This model is hybrid (GatedDeltaNet + full attention); vLLM -# keeps prefix caching OPT-IN by default for hybrid models ("while the feature -# matures"), so it is off unless we ask for it. Turning it on reuses the KV of -# the shared prompt prefix (pi re-sends the same system prompt + tool schema -# every turn), so follow-up requests skip re-prefilling thousands of tokens. -# Big TTFT win for the agent loop. Set 0 to disable. -ENABLE_PREFIX_CACHING=1 +REASONING=off -# ngram speculative decoding (vLLM ngram proposer). -# prompt_lookup_max = ngram window; num_speculative_tokens = draft depth. -NGRAM_LOOKUP_MAX=5 -NGRAM_LOOKUP_MIN=1 -NUM_SPEC_TOKENS=5 - -# Tool calling. Clients (pi, OpenWebUI) send tool_choice=auto; without -# --enable-auto-tool-choice vLLM rejects it with HTTP 400. The Avesed -# chat template renders tools in the Qwen3 XML format, so the qwen3_xml -# parser is the matching one. CHAT_TEMPLATE must point at the repo's -# chat_template.jinja (the tokenizer_config.json ships none; vLLM's -# built-in template does not render the XML tool instructions). -ENABLE_AUTO_TOOL_CHOICE=1 -TOOL_CALL_PARSER=qwen3_xml -CHAT_TEMPLATE=/home/dank/models/qwen36-chat_template-tools.jinja - -# RDNA flash-attention for the vision encoder (and general FA paths). -# Without this, the mm encoder falls back to torch SDPA and the warmup -# OOMs (256 GiB score matrix at max_model_len; measured 2026-08-17). -# Requires the flash_attn package with flash_attn_triton_amd findable -# (the pure-Python copy from the CIRU package plus a flash_attn_2_cuda -# stub; see docs/frank-redo-2026-08-17.md). -FLASH_ATTENTION_TRITON_AMD_ENABLE=TRUE +VERBOSITY=4 diff --git a/scripts/frank-default-small.service b/scripts/frank-default-small.service index 779715c..4946b1f 100644 --- a/scripts/frank-default-small.service +++ b/scripts/frank-default-small.service @@ -1,5 +1,5 @@ [Unit] -Description=vLLM Qwen3.6-35B-A3B INT8 W8A8 ngram spec (dankclaw, frank-default-small) +Description=CIRU llama.cpp Qwen3.8-27B ActiveFPX PromptForge ngram MTP (dankclaw, frank-default-small) After=network-online.target Wants=network-online.target @@ -8,7 +8,6 @@ Type=simple User=dank Group=dank Environment="FRANK_SMALL_ENV_FILE=/home/dank/vllmScripts/scripts/frank-default-small.env" -Environment="HF_HOME=/home/dank/.cache/huggingface" ExecStart=/home/dank/vllmScripts/scripts/serve-frank-small.sh Restart=on-failure RestartSec=10 diff --git a/scripts/serve-frank-small.sh b/scripts/serve-frank-small.sh index 1ec83be..78ecd54 100755 --- a/scripts/serve-frank-small.sh +++ b/scripts/serve-frank-small.sh @@ -1,10 +1,16 @@ #!/usr/bin/env bash -# Serve Avesed/Qwen3.6-35B-A3B-INT8-W8A8 via vLLM on Dankclaw (Strix Halo). +# Serve jcbtc/Qwen3.8-27B-CIRU-ActiveFPX-PromptForge via the CIRU ROCmFPX +# llama.cpp runtime on Dankclaw (Strix Halo gfx1151). # -# Runs the ROCm vLLM build in /home/dank/vllm-venv. The model is a -# compressed-tensors INT8 W8A8 quant (MoE, 35B total / 3B active). Full -# 262144 context, unquantized KV cache, 4 concurrent sequences, ngram -# speculative decoding. +# Runtime: the PromptForge llama-server built from ciru-ai/ROCmFPX tag +# qwen3.8-activefpx-promptforge-v2.3 (commit 103af8b41) against the private +# TheRock ROCm 7.15 SDK (kept in the retired ling3-ciru venv). The model is +# a GGUF with optional PromptForge companion views (FFN/GDN .pfs) that enable +# CIRU DualView prompt acceleration. Native depth-4 MTP speculative decoding. +# +# The model card requires ROCm 7.15 (not the system ROCm 7.2.4) for the +# advertised Strix Halo performance. LD_LIBRARY_PATH therefore points at the +# 7.15 SDK libraries first. An older system ROCm must not precede them. # # Tunables come from frank-default-small.env in this directory unless # FRANK_SMALL_ENV_FILE overrides that path. @@ -20,55 +26,93 @@ if [[ -f "$ENV_FILE" ]]; then fi # --- defaults (used when the env file does not set them) ------------------- -: "${VLLM_BIN:=/home/dank/vllm-venv/bin/vllm}" -: "${MODEL:=nameistoken/Qwen3.6-35B-A3B-Quark-W8A8-INT8}" +: "${LLAMA_BIN:=/home/dank/ciru-qwen38/ROCmFPX/build-promptforge/bin/llama-server}" +: "${MODEL_DIR:=/home/dank/models/CIRU-Qwen3.8-27B-ActiveFPX}" +: "${MODEL_FILE:=Qwen3.8-27B-CIRU-ActiveFPX-PromptForge-v2.gguf}" +: "${FFN_PFS:=Qwen3.8-27B-CIRU-ActiveFPX-PromptForge-FFN.pfs}" +: "${GDN_PFS:=Qwen3.8-27B-CIRU-ActiveFPX-PromptForge-GDN.pfs}" +: "${ROCM_DEVEL:=/home/dank/ling3-ciru/Ling-3.0-Flash-CIRU-int4-Strix-native/.venv/lib/python3.12/site-packages/_rocm_sdk_devel}" +: "${ROCM_LIB:=/home/dank/ling3-ciru/Ling-3.0-Flash-CIRU-int4-Strix-native/.venv/lib/python3.12/site-packages/_rocm_sdk_libraries}" +: "${ROCM_CORE:=/home/dank/ling3-ciru/Ling-3.0-Flash-CIRU-int4-Strix-native/.venv/lib/python3.12/site-packages/_rocm_sdk_core}" : "${ALIAS:=frank-default-small}" : "${BIND_ADDR:=0.0.0.0}" : "${LISTEN_PORT:=52625}" -: "${MAX_MODEL_LEN:=262144}" -: "${KV_CACHE_DTYPE:=auto}" -: "${MAX_NUM_SEQS:=4}" -: "${MAX_NUM_BATCHED_TOKENS:=8192}" -: "${GPU_MEMORY_UTILIZATION:=0.92}" -: "${NGRAM_LOOKUP_MAX:=5}" -: "${NGRAM_LOOKUP_MIN:=1}" -: "${NUM_SPEC_TOKENS:=5}" -: "${ENABLE_AUTO_TOOL_CHOICE:=1}" -: "${TOOL_CALL_PARSER:=qwen3_xml}" -: "${ENABLE_PREFIX_CACHING:=1}" -# The Avesed tokenizer_config.json has no chat_template; vLLM would use its -# built-in template which does not render the model's XML tool format. Use -# the repo's chat_template.jinja (ships the instructions). -: "${CHAT_TEMPLATE:=/home/dank/models/qwen36-chat_template-tools.jinja}" +: "${CTX_SIZE:=65536}" +: "${PARALLEL:=1}" +: "${BATCH_SIZE:=2048}" +: "${UBATCH_SIZE:=2048}" +: "${CACHE_TYPE_K:=f16}" +: "${CACHE_TYPE_V:=f16}" +: "${CTX_CHECKPOINTS:=32}" +: "${CACHE_RAM:=8192}" +: "${SPEC_DRAFT_N_MAX:=4}" +: "${SPEC_DRAFT_N_MIN:=0}" +: "${SPEC_DRAFT_P_MIN:=0.0}" +: "${SPEC_DRAFT_P_SPLIT:=0.10}" +: "${TEMPERATURE:=0.7}" +: "${TOP_P:=0.95}" +: "${REASONING:=off}" +: "${VERBOSITY:=4}" -if [[ ! -x "$VLLM_BIN" ]]; then - echo "missing vllm binary: $VLLM_BIN" >&2 +MODEL="$MODEL_DIR/$MODEL_FILE" +export PROMPTFORGE_SIDECAR="$MODEL_DIR/$FFN_PFS" +export PROMPTFORGE_GDN_SIDECAR="$MODEL_DIR/$GDN_PFS" +export PROMPTFORGE_MODE="m2048_fused_tail1476" +export GGML_CUDA_GRAPH_OPT=0 +export LLAMA_MTP_CPU_ARGMAX_FASTPATH=1 + +if [[ ! -f "$MODEL" ]]; then + echo "missing required file: $MODEL" >&2 + exit 1 +fi +if [[ ! -f "$PROMPTFORGE_SIDECAR" ]]; then + echo "missing PromptForge FFN companion: $PROMPTFORGE_SIDECAR" >&2 + exit 1 +fi +if [[ ! -f "$PROMPTFORGE_GDN_SIDECAR" ]]; then + echo "missing PromptForge GDN companion: $PROMPTFORGE_GDN_SIDECAR" >&2 + exit 1 +fi +if [[ ! -x "$LLAMA_BIN" ]]; then + echo "missing llama-server binary: $LLAMA_BIN" >&2 exit 1 fi -TOOL_ARGS=() -if [[ "${ENABLE_AUTO_TOOL_CHOICE}" == "1" ]]; then - TOOL_ARGS+=(--enable-auto-tool-choice) - TOOL_ARGS+=(--tool-call-parser "${TOOL_CALL_PARSER}") -fi +# ROCm 7.15 SDK libraries must precede any system ROCm in LD_LIBRARY_PATH. +export ROCM_PATH="$ROCM_DEVEL" +export LD_LIBRARY_PATH="${ROCM_LIB}/lib:${ROCM_CORE}/lib:${ROCM_DEVEL}/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" -PC_ARGS=() -if [[ "${ENABLE_PREFIX_CACHING}" == "1" ]]; then - PC_ARGS+=(--enable-prefix-caching) -fi - -exec "$VLLM_BIN" serve "$MODEL" \ +exec "$LLAMA_BIN" \ + --model "$MODEL" \ + --alias "$ALIAS" \ --host "$BIND_ADDR" \ --port "$LISTEN_PORT" \ - --served-model-name "$ALIAS" \ - --max-model-len "$MAX_MODEL_LEN" \ - --kv-cache-dtype "$KV_CACHE_DTYPE" \ - --max-num-seqs "$MAX_NUM_SEQS" \ - --max-num-batched-tokens "$MAX_NUM_BATCHED_TOKENS" \ - --gpu-memory-utilization "$GPU_MEMORY_UTILIZATION" \ - --speculative-config "{\"method\":\"ngram\",\"prompt_lookup_max\":$NGRAM_LOOKUP_MAX,\"prompt_lookup_min\":$NGRAM_LOOKUP_MIN,\"num_speculative_tokens\":$NUM_SPEC_TOKENS}" \ - --reasoning-parser qwen3 \ - --trust-remote-code \ - --chat-template "$CHAT_TEMPLATE" \ - "${PC_ARGS[@]}" \ - "${TOOL_ARGS[@]}" + -dev ROCm0 \ + -ngl 999 \ + -c "$CTX_SIZE" \ + -b "$BATCH_SIZE" \ + -ub "$UBATCH_SIZE" \ + -fa on \ + -ctk "$CACHE_TYPE_K" \ + -ctv "$CACHE_TYPE_V" \ + -np "$PARALLEL" \ + -ctxcp "$CTX_CHECKPOINTS" \ + --cache-ram "$CACHE_RAM" \ + --cache-prompt \ + --metrics \ + --spec-type draft-mtp \ + --spec-draft-device ROCm0 \ + --spec-draft-ngl 999 \ + --spec-draft-type-k f16 \ + --spec-draft-type-v f16 \ + --spec-draft-n-max "$SPEC_DRAFT_N_MAX" \ + --spec-draft-n-min "$SPEC_DRAFT_N_MIN" \ + --spec-draft-p-min "$SPEC_DRAFT_P_MIN" \ + --spec-draft-p-split "$SPEC_DRAFT_P_SPLIT" \ + --spec-draft-backend-sampling \ + --temp "$TEMPERATURE" \ + --top-p "$TOP_P" \ + --reasoning "$REASONING" \ + --reasoning-format none \ + --reasoning-budget -1 \ + -lv "$VERBOSITY"