From 2a783325f0223dff0c96ea4ba68f19c7c65dd6a0 Mon Sep 17 00:00:00 2001 From: Donato Capitella Date: Sat, 6 Dec 2025 18:13:25 +0000 Subject: [PATCH] added ministral-3-8B --- ...6__rocm-7-nightly-rocwmma__fa1__single.log | 10 + ...-7-nightly-rocwmma__hblt0__fa1__single.log | 10 + ...2512-BF16__rocm-7-nightly__fa1__single.log | 10 + ...16__rocm-7-nightly__hblt0__fa1__single.log | 10 + ...12-BF16__rocm-7.9-rocwmma__fa1__single.log | 10 + ...__rocm-7.9-rocwmma__hblt0__fa1__single.log | 10 + ...truct-2512-BF16__rocm-7.9__fa1__single.log | 10 + ...512-BF16__rocm-7.9__hblt0__fa1__single.log | 10 + ...2-BF16__rocm6_4_4-rocwmma__fa1__single.log | 10 + ..._rocm6_4_4-rocwmma__hblt0__fa1__single.log | 10 + ...ruct-2512-BF16__rocm6_4_4__fa1__single.log | 10 + ...12-BF16__rocm6_4_4__hblt0__fa1__single.log | 10 + ...512-BF16__rocm7.1-rocwmma__fa1__single.log | 10 + ...6__rocm7.1-rocwmma__hblt0__fa1__single.log | 10 + ...struct-2512-BF16__rocm7.1__fa1__single.log | 10 + ...2512-BF16__rocm7.1__hblt0__fa1__single.log | 10 + ...-2512-BF16__vulkan_amdvlk__fa1__single.log | 12 + ...ct-2512-BF16__vulkan_radv__fa1__single.log | 12 + docs/results.json | 1046 ++++++++++++++++- 19 files changed, 1229 insertions(+), 1 deletion(-) create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly-rocwmma__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly-rocwmma__hblt0__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly__hblt0__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9-rocwmma__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9-rocwmma__hblt0__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9__hblt0__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4-rocwmma__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4-rocwmma__hblt0__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4__hblt0__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1-rocwmma__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1-rocwmma__hblt0__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1__hblt0__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__vulkan_amdvlk__fa1__single.log create mode 100644 benchmark/results/Ministral-3-8B-Instruct-2512-BF16__vulkan_radv__fa1__single.log diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly-rocwmma__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly-rocwmma__fa1__single.log new file mode 100644 index 0000000..0c755c6 --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly-rocwmma__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 4566.69 ± 9.35 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.17 ± 0.00 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly-rocwmma__hblt0__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly-rocwmma__hblt0__fa1__single.log new file mode 100644 index 0000000..27d1bb8 --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly-rocwmma__hblt0__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 624.69 ± 0.18 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.15 ± 0.01 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly__fa1__single.log new file mode 100644 index 0000000..f07206f --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 4367.96 ± 32.60 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.22 ± 0.00 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly__hblt0__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly__hblt0__fa1__single.log new file mode 100644 index 0000000..400f262 --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly__hblt0__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 618.44 ± 0.31 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.20 ± 0.01 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9-rocwmma__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9-rocwmma__fa1__single.log new file mode 100644 index 0000000..c868db3 --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9-rocwmma__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 4254.15 ± 13.82 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.42 ± 0.01 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9-rocwmma__hblt0__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9-rocwmma__hblt0__fa1__single.log new file mode 100644 index 0000000..531ec13 --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9-rocwmma__hblt0__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 599.01 ± 0.38 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.40 ± 0.02 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9__fa1__single.log new file mode 100644 index 0000000..69d9b90 --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 4356.51 ± 12.09 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.44 ± 0.01 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9__hblt0__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9__hblt0__fa1__single.log new file mode 100644 index 0000000..2dd67ad --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9__hblt0__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 598.84 ± 0.40 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.41 ± 0.01 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4-rocwmma__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4-rocwmma__fa1__single.log new file mode 100644 index 0000000..5d8ac31 --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4-rocwmma__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 3963.40 ± 14.39 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.44 ± 0.00 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4-rocwmma__hblt0__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4-rocwmma__hblt0__fa1__single.log new file mode 100644 index 0000000..5080769 --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4-rocwmma__hblt0__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 612.14 ± 0.41 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.43 ± 0.01 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4__fa1__single.log new file mode 100644 index 0000000..f8e0f9d --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 4172.76 ± 21.37 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.48 ± 0.01 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4__hblt0__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4__hblt0__fa1__single.log new file mode 100644 index 0000000..7699da2 --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4__hblt0__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 617.68 ± 0.32 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.49 ± 0.01 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1-rocwmma__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1-rocwmma__fa1__single.log new file mode 100644 index 0000000..e2b75fd --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1-rocwmma__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 4243.27 ± 9.54 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.40 ± 0.01 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1-rocwmma__hblt0__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1-rocwmma__hblt0__fa1__single.log new file mode 100644 index 0000000..2deb530 --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1-rocwmma__hblt0__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 596.53 ± 0.31 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.39 ± 0.01 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1__fa1__single.log new file mode 100644 index 0000000..fff9513 --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 4358.63 ± 10.50 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.46 ± 0.01 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1__hblt0__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1__hblt0__fa1__single.log new file mode 100644 index 0000000..30f7a88 --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1__hblt0__fa1__single.log @@ -0,0 +1,10 @@ +ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no +ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no +ggml_cuda_init: found 1 ROCm devices: + Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32 +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | pp512 | 598.95 ± 0.48 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | ROCm | 99 | 1 | tg128 | 31.45 ± 0.01 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__vulkan_amdvlk__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__vulkan_amdvlk__fa1__single.log new file mode 100644 index 0000000..bd4c86e --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__vulkan_amdvlk__fa1__single.log @@ -0,0 +1,12 @@ +WARNING: radv is not a conformant Vulkan implementation, testing use only. +WARNING: radv is not a conformant Vulkan implementation, testing use only. +ggml_vulkan: Found 3 Vulkan devices: +ggml_vulkan: 0 = AMD Radeon AI PRO R9700 (AMD open-source driver) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 64 | shared memory: 32768 | int dot: 1 | matrix cores: KHR_coopmat +ggml_vulkan: 1 = AMD Radeon AI PRO R9700 (AMD open-source driver) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 64 | shared memory: 32768 | int dot: 1 | matrix cores: KHR_coopmat +ggml_vulkan: 2 = AMD Ryzen 9 9900X3D 12-Core Processor (AMD open-source driver) | uma: 1 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 32768 | int dot: 1 | matrix cores: none +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | Vulkan | 99 | 1 | pp512 | 874.94 ± 7.12 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | Vulkan | 99 | 1 | tg128 | 29.20 ± 0.09 | + +build: 6016d0bd4 (7285) diff --git a/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__vulkan_radv__fa1__single.log b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__vulkan_radv__fa1__single.log new file mode 100644 index 0000000..a910346 --- /dev/null +++ b/benchmark/results/Ministral-3-8B-Instruct-2512-BF16__vulkan_radv__fa1__single.log @@ -0,0 +1,12 @@ +WARNING: radv is not a conformant Vulkan implementation, testing use only. +WARNING: radv is not a conformant Vulkan implementation, testing use only. +ggml_vulkan: Found 3 Vulkan devices: +ggml_vulkan: 0 = AMD Ryzen 9 9900X3D 12-Core Processor (RADV RAPHAEL_MENDOCINO) (radv) | uma: 1 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 65536 | int dot: 1 | matrix cores: none +ggml_vulkan: 1 = AMD Radeon AI PRO R9700 (RADV GFX1201) (radv) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 64 | shared memory: 65536 | int dot: 1 | matrix cores: KHR_coopmat +ggml_vulkan: 2 = AMD Radeon AI PRO R9700 (RADV GFX1201) (radv) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 64 | shared memory: 65536 | int dot: 1 | matrix cores: KHR_coopmat +| model | size | params | backend | ngl | fa | test | t/s | +| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | Vulkan | 99 | 1 | pp512 | 1578.59 ± 104.98 | +| mistral3 8B BF16 | 15.81 GiB | 8.49 B | Vulkan | 99 | 1 | tg128 | 27.74 ± 0.07 | + +build: 6016d0bd4 (7285) diff --git a/docs/results.json b/docs/results.json index 7f59bdc..f74561d 100644 --- a/docs/results.json +++ b/docs/results.json @@ -1,6 +1,6 @@ { "meta": { - "generated_at": "2025-12-06T11:02:47Z", + "generated_at": "2025-12-06T18:13:04Z", "os_kernel": "Fedora 42 \u2014 Linux 6.15.9-201.fc42.x86_64 (Sat Aug 2 11:37:34 UTC 2025)", "llamacpp_builds": [ { @@ -1043,6 +1043,1050 @@ "number": "7285" } }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7-nightly-rocwmma", + "env_base": "rocm", + "env_variant": "7-nightly-rocwmma", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 4566.69, + "tps_std": 9.35, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly-rocwmma__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7-nightly-rocwmma", + "env_base": "rocm", + "env_variant": "7-nightly-rocwmma", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.17, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly-rocwmma__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7-nightly-rocwmma-hblt0", + "env_base": "rocm", + "env_variant": "7-nightly-rocwmma-hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 624.69, + "tps_std": 0.18, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly-rocwmma__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7-nightly-rocwmma-hblt0", + "env_base": "rocm", + "env_variant": "7-nightly-rocwmma-hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.15, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly-rocwmma__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7-nightly", + "env_base": "rocm", + "env_variant": "7-nightly", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 4367.96, + "tps_std": 32.6, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7-nightly", + "env_base": "rocm", + "env_variant": "7-nightly", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.22, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7-nightly-hblt0", + "env_base": "rocm", + "env_variant": "7-nightly-hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 618.44, + "tps_std": 0.31, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7-nightly-hblt0", + "env_base": "rocm", + "env_variant": "7-nightly-hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.2, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7-nightly__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7.9-rocwmma", + "env_base": "rocm", + "env_variant": "7.9-rocwmma", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 4254.15, + "tps_std": 13.82, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9-rocwmma__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7.9-rocwmma", + "env_base": "rocm", + "env_variant": "7.9-rocwmma", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.42, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9-rocwmma__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7.9-rocwmma-hblt0", + "env_base": "rocm", + "env_variant": "7.9-rocwmma-hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 599.01, + "tps_std": 0.38, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9-rocwmma__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7.9-rocwmma-hblt0", + "env_base": "rocm", + "env_variant": "7.9-rocwmma-hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.4, + "tps_std": 0.02, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9-rocwmma__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7.9", + "env_base": "rocm", + "env_variant": "7.9", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 4356.51, + "tps_std": 12.09, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7.9", + "env_base": "rocm", + "env_variant": "7.9", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.44, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7.9-hblt0", + "env_base": "rocm", + "env_variant": "7.9-hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 598.84, + "tps_std": 0.4, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm-7.9-hblt0", + "env_base": "rocm", + "env_variant": "7.9-hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.41, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm-7.9__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm6_4_4-rocwmma", + "env_base": "rocm6_4_4", + "env_variant": "rocwmma", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 3963.4, + "tps_std": 14.39, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4-rocwmma__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm6_4_4-rocwmma", + "env_base": "rocm6_4_4", + "env_variant": "rocwmma", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.44, + "tps_std": 0.0, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4-rocwmma__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm6_4_4-rocwmma-hblt0", + "env_base": "rocm6_4_4", + "env_variant": "rocwmma-hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 612.14, + "tps_std": 0.41, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4-rocwmma__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm6_4_4-rocwmma-hblt0", + "env_base": "rocm6_4_4", + "env_variant": "rocwmma-hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.43, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4-rocwmma__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm6_4_4", + "env_base": "rocm6_4_4", + "env_variant": null, + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 4172.76, + "tps_std": 21.37, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm6_4_4", + "env_base": "rocm6_4_4", + "env_variant": null, + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.48, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm6_4_4-hblt0", + "env_base": "rocm6_4_4", + "env_variant": "hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 617.68, + "tps_std": 0.32, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm6_4_4-hblt0", + "env_base": "rocm6_4_4", + "env_variant": "hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.49, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm6_4_4__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm7.1-rocwmma", + "env_base": "rocm7.1", + "env_variant": "rocwmma", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 4243.27, + "tps_std": 9.54, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1-rocwmma__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm7.1-rocwmma", + "env_base": "rocm7.1", + "env_variant": "rocwmma", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.4, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1-rocwmma__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm7.1-rocwmma-hblt0", + "env_base": "rocm7.1", + "env_variant": "rocwmma-hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 596.53, + "tps_std": 0.31, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1-rocwmma__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm7.1-rocwmma-hblt0", + "env_base": "rocm7.1", + "env_variant": "rocwmma-hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.39, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1-rocwmma__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm7.1", + "env_base": "rocm7.1", + "env_variant": null, + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 4358.63, + "tps_std": 10.5, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm7.1", + "env_base": "rocm7.1", + "env_variant": null, + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.46, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm7.1-hblt0", + "env_base": "rocm7.1", + "env_variant": "hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 598.95, + "tps_std": 0.48, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "rocm7.1-hblt0", + "env_base": "rocm7.1", + "env_variant": "hblt0", + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 31.45, + "tps_std": 0.01, + "error": false, + "error_type": null, + "backend": "ROCm", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__rocm7.1__hblt0__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "vulkan_amdvlk", + "env_base": "vulkan_amdvlk", + "env_variant": null, + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 874.94, + "tps_std": 7.12, + "error": false, + "error_type": null, + "backend": "Vulkan", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__vulkan_amdvlk__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "vulkan_amdvlk", + "env_base": "vulkan_amdvlk", + "env_variant": null, + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 29.2, + "tps_std": 0.09, + "error": false, + "error_type": null, + "backend": "Vulkan", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__vulkan_amdvlk__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "vulkan_radv", + "env_base": "vulkan_radv", + "env_variant": null, + "fa": true, + "context": "default", + "context_tokens": null, + "test": "pp512", + "tps_mean": 1578.59, + "tps_std": 104.98, + "error": false, + "error_type": null, + "backend": "Vulkan", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__vulkan_radv__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, + { + "model": "Ministral-3-8B-Instruct-2512-BF16", + "model_clean": "Ministral-3-8B-Instruct-2512-BF16", + "env": "vulkan_radv", + "env_base": "vulkan_radv", + "env_variant": null, + "fa": true, + "context": "default", + "context_tokens": null, + "test": "tg128", + "tps_mean": 27.74, + "tps_std": 0.07, + "error": false, + "error_type": null, + "backend": "Vulkan", + "ngl": 99, + "mmap": null, + "params_b": 8.49, + "file_size_gib": 15.81, + "name_params_b": 8.49, + "quant": "BF16", + "log": "results/Ministral-3-8B-Instruct-2512-BF16__vulkan_radv__fa1__single.log", + "rpc": false, + "gpu_config": "single", + "build": { + "hash": "6016d0bd4", + "number": "7285" + } + }, { "model": "Qwen3-Coder-30B-A3B-Instruct-BF16-00001-of-00002", "model_clean": "Qwen3-Coder-30B-A3B-Instruct-BF16",