bb2a64e3d0
use buildah
2026-05-25 19:19:06 +02:00
71abff5a0d
feat: add interactive granular benchmark details to UI and update MTP summary data format
2026-05-17 11:03:23 +01:00
2a4cb50b52
chore: update benchmark execution script parameters and paths
2026-05-17 09:16:03 +01:00
451723bab8
docs: update recommended boot parameters to use amd_iommu=off for improved performance
2026-05-15 19:08:21 +01:00
a1ef189df0
feat: add MTP benchmark page with dedicated data summary and visualization components
2026-05-15 11:51:02 +01:00
adf4d1a5fa
mtp-results
2026-05-15 10:56:57 +01:00
fa27bf663c
refactor: update display name logic in benchmark script to consistently use filenames and strip multi-part suffix
2026-05-15 10:21:24 +01:00
d1e5e0b3b0
fix: enable SO_REUSEADDR on socket to prevent address binding errors in benchmark utility
2026-05-15 09:22:06 +01:00
adf72bdda7
refactor: replace immediate port kill with a 15-second wait for port release in cleanup function
2026-05-15 08:58:16 +01:00
765f635381
feat: improve container exit error reporting by adding log retrieval and manual debug guidance
2026-05-15 08:48:39 +01:00
b1d526b775
feat: add SIGINT handler for container cleanup and enable replace flag for podman execution
2026-05-15 08:43:34 +01:00
c0e33115ca
feat: implement container status monitoring and enhanced log reporting in health check loop
2026-05-15 08:38:24 +01:00
7e6c30dae6
feat: add pre-flight cleanup and port conflict resolution to benchmark runner
2026-05-15 08:36:12 +01:00
8bc5b8c5e5
feat: add automated MTP benchmark runner for llama-server via podman containers
2026-05-15 08:07:10 +01:00
dc948d766a
feat: add benchmark script for MTP acceptance and throughput testing
2026-05-14 21:02:05 +01:00
3e3f3674a8
feat: add MTP (Multi-Token Prediction) support via new ROCm 7.2.3 and Vulkan RADV toolboxes
2026-05-14 20:09:03 +01:00
7320eb3f00
updated 7.2.3 benchs
2026-05-12 12:38:43 +01:00
8648f93ad3
updated benchs
2026-05-12 12:32:07 +01:00
ed889def9d
Support Debian in toolbox/distrobox command check ( #88 )
...
* Support Debian in toolbox command check
Debian requires distrobox, like Ubuntu. This is a simple change to add support for it.
* Consolidation of logic
copilot complained, so used this as an opportunity to collapse unnecessary evals.
2026-05-12 12:25:59 +01:00
2e3dc657d2
chore: update ROCm version to 7.2.3 and remove deprecated pr21344 toolbox
2026-05-11 19:40:30 +01:00
0f9c2c85be
Update issue templates
2026-05-06 18:06:18 +01:00
07d2131d8c
added @64k benchmarks
2026-05-03 16:20:42 +01:00
1bffd6505f
feat: add longctx65536 support to standard and RPC benchmark scripts
2026-05-01 20:19:02 +01:00
d20bb42b04
updated results
2026-04-29 06:45:26 +01:00
73be068e85
feat: upgrade ROCm toolboxes to 7.2.2 and update documentation and CI configurations
2026-04-26 16:25:44 +01:00
1421e87060
feat: add Plausible analytics script to documentation index page
2026-04-21 09:47:16 +01:00
9016c0f8f8
update benchs
2026-04-15 16:54:34 +01:00
66a3314c22
refactor: update MODEL_DIR path to use absolute home directory reference
2026-04-15 11:39:35 +01:00
9707a15df7
feat: add benchmark results for rocm-7_2_1-pr21344 and update results metadata
2026-04-15 11:39:10 +01:00
c2754a810a
fix vulkan builds: add spirv-headers-devel to builder deps
2026-04-15 09:33:13 +01:00
2c2c36d3da
add rocm-7.2.1-pr21344 toolbox (gfx1151 MMQ/MMVQ tile + nwarp tuning)
...
Adds a new toolbox variant based on PR #21344 (pedapudi/llama.cpp@gfx1151-opt)
which tunes MMQ tile sizes (x_max=48, y=64) and warp counts (nwarps=4) for
RDNA3_5 gfx1151, yielding up to +100% prefill throughput at small batch sizes.
Also adds BMI2/FMA/F16C CPU SIMD flags and GGML_CUDA_FA_ALL_QUANTS=ON to match
the benchmark build used in the PR. Wire up CI (build matrix + prune), the
refresh script, and run_benchmarks.sh so results land alongside rocm-7.2.1.
2026-04-15 09:23:58 +01:00
14fae26ad0
add minimax m2.7 benchmarks
2026-04-15 08:09:12 +01:00
4b3c02a405
Update README.md
2026-04-12 21:33:46 +01:00
d74db71362
archvied old multi-node benchmarks
2026-04-11 11:20:30 +01:00
7aa6e6dea9
update benchmarks
2026-04-11 11:18:45 +01:00
a821bcb91d
chore: update rocm-7.2 benchmark configuration to version 7.2.1
2026-04-10 11:48:27 +01:00
c129a04a1c
refactor: remove hblt0 benchmark support and associated comparison scripts
2026-04-10 11:23:06 +01:00
5acf54cd67
fix: Update HuggingFace download commands ( #61 )
2026-04-10 10:56:00 +01:00
1dea385f6a
fix: remove trailing backtick causing syntax error in prune-old-toolboxes workflow
2026-04-09 19:00:44 +01:00
4ac481e7d1
chore: upgrade ROCm version from 7.2 to 7.2.1 across configuration and documentation
2026-04-09 18:33:52 +01:00
d1e49d4aa0
chore: remove llama.cpp PR 21566 patch from rocm7-nightlies Dockerfile
2026-04-07 18:33:18 +01:00
a58d133c5e
chore: update llama.cpp patch to PR 21566 for gemma-4 inference fix
2026-04-07 17:49:16 +01:00
d0281bb526
feat: apply upstream llama.cpp patch to fix Gemma-4 inference issues
2026-04-06 10:25:42 +01:00
bbd8f02014
build: remove -DGGML_CUDA_DISABLE_FUSION=1 from cmake configuration in rocm7-nightlies Dockerfile (this was for a temporary test)
2026-04-03 15:21:58 +01:00
b376d1558b
build: disable GGML CUDA fusion in ROCm build configuration (temporary test)
2026-04-03 15:16:12 +01:00
a7ace8dba7
updted benchmarks
2026-03-30 08:37:15 +01:00
614b00af3e
fixed patch (AI slop!!!)
2026-03-25 09:36:50 +00:00
ca84f4cbf3
patch: increasing MAX_REPETITION_THRESHOLD to allow complex agentic workflows
2026-03-25 09:23:19 +00:00
eb03432a50
added router mode section and example models.ini file for use with router mode ( #67 )
2026-03-04 12:09:11 +00:00
5f4698c959
build: Remove amdgpu-unroll-threshold-local CMAKE_HIP_FLAG from ROCm 7 nightlies Dockerfile.
2026-03-03 12:54:45 +00:00
Julian Wefers