[cuda backend] int4/8 matvec: vectorized activation load #13969

Run time

Learn about OS pricing on GitHub Actions

Job	Run time
Get changed files / get-changed-files	3s
CI run decision / decide	32s
unittest-cuda / linux-job	52m 48s
test-models-cuda / linux-job	46m 37s
test-executorch-cuda-build-12.6 / linux-job	18m 56s
test-executorch-cuda-build-13.0 / linux-job	16m 33s
export-model-cuda-artifact (openai, whisper-small, non-quantized) / linux-job	18m 30s
export-model-cuda-artifact (facebook, dinov2-small-imagenet1k-1-layer, non-quantized) / linux-job	14m 32s
export-model-cuda-artifact (SocialLocalMobile, gemma-4-31B-it-HQQ-INT4, quantized-int4-tile-packed) / linux-job	53m 56s
export-model-cuda-artifact (google, gemma-3-4b-it, non-quantized) / linux-job	29m 37s
export-model-cuda-artifact (nvidia, diar_streaming_sortformer_4spk-v2, non-quantized) / linux-job	31m 2s
export-model-cuda-artifact (mistralai, Voxtral-Mini-4B-Realtime-2602, quantized-int4-tile-packed) / linux-job	23m 40s
export-model-cuda-artifact (google, gemma-3-4b-it, quantized-int4-tile-packed) / linux-job	28m 31s
export-model-cuda-artifact (mistralai, Voxtral-Mini-3B-2507, quantized-int4-weight-only) / linux-job	26m 22s
export-model-cuda-artifact (SocialLocalMobile, Qwen3.5-35B-A3B-HQQ-INT4, quantized-int4-tile-packed) / linux-job	46m 1s
export-model-cuda-artifact (Qwen, Qwen3-0.6B, quantized-int4-tile-packed) / linux-job	18m 49s
export-model-cuda-artifact (openai, whisper-large-v3-turbo, quantized-int4-tile-packed) / linux-job	20m 11s
export-model-cuda-artifact (openai, whisper-large-v3-turbo, quantized-int4-weight-only) / linux-job	18m 23s
export-model-cuda-artifact (Qwen, Qwen3-0.6B, non-quantized) / linux-job	17m 30s
export-model-cuda-artifact (mistralai, Voxtral-Mini-3B-2507, non-quantized) / linux-job	28m 4s
export-model-cuda-artifact (nvidia, parakeet-tdt, quantized-int4-tile-packed) / linux-job	25m 48s
export-model-cuda-artifact (nvidia, parakeet-tdt, non-quantized) / linux-job	23m 53s
export-model-cuda-artifact (mistralai, Voxtral-Mini-3B-2507, quantized-int4-tile-packed) / linux-job	26m 15s
check-all-cuda-builds	4s
test-cuda-pybind (gemma3-4b, --quantize, google-gemma-3-4b-it-cuda-quantized-int4-tile-packed) / linux-job	21m 10s
test-cuda-pybind (qwen3-0.6b, Qwen-Qwen3-0.6B-cuda-non-quantized) / linux-job	20m 30s
test-cuda-pybind (qwen3-0.6b, --quantize, Qwen-Qwen3-0.6B-cuda-quantized-int4-tile-packed) / linux-job	19m 22s
test-model-cuda-e2e (mistralai, Voxtral-Mini-3B-2507, quantized-int4-weight-only) / linux-job	10m 0s
test-model-cuda-e2e (google, gemma-3-4b-it, non-quantized) / linux-job	13m 18s
test-model-cuda-e2e (mistralai, Voxtral-Mini-3B-2507, quantized-int4-tile-packed) / linux-job	10m 38s
test-model-cuda-e2e (nvidia, parakeet-tdt, non-quantized) / linux-job	11m 36s
test-model-cuda-e2e (google, gemma-3-4b-it, quantized-int4-tile-packed) / linux-job	10m 5s
test-model-cuda-e2e (mistralai, Voxtral-Mini-3B-2507, non-quantized) / linux-job	13m 24s
test-model-cuda-e2e (SocialLocalMobile, Qwen3.5-35B-A3B-HQQ-INT4, quantized-int4-tile-packed) / linux-job	14m 1s
test-model-cuda-e2e (nvidia, parakeet-tdt, quantized-int4-tile-packed) / linux-job	10m 4s
test-model-cuda-e2e (nvidia, diar_streaming_sortformer_4spk-v2, non-quantized) / linux-job	8m 18s
test-model-cuda-e2e (openai, whisper-large-v3-turbo, quantized-int4-tile-packed) / linux-job	23m 48s
test-model-cuda-e2e (mistralai, Voxtral-Mini-4B-Realtime-2602, quantized-int4-tile-packed) / linux-job	25m 57s
test-model-cuda-e2e (openai, whisper-large-v3-turbo, quantized-int4-weight-only) / linux-job	23m 58s
test-model-cuda-e2e (facebook, dinov2-small-imagenet1k-1-layer, non-quantized) / linux-job	10m 20s
test-model-cuda-e2e (openai, whisper-small, non-quantized) / linux-job	23m 53s
test-model-cuda-e2e (SocialLocalMobile, gemma-4-31B-it-HQQ-INT4, quantized-int4-tile-packed) / linux-job	16m 17s
test-executorch-cuda-build-12.6 / linux-job	19m 18s
CI run decision / decide	32s
test-executorch-cuda-build-13.0 / linux-job	16m 33s
unittest-cuda / linux-job	52m 48s
export-model-cuda-artifact (openai, whisper-large-v3-turbo, quantized-int4-weight-only) / linux-job	18m 23s
test-models-cuda / linux-job	46m 37s
export-model-cuda-artifact (nvidia, parakeet-tdt, non-quantized) / linux-job	23m 53s
export-model-cuda-artifact (Qwen, Qwen3-0.6B, non-quantized) / linux-job	17m 30s
export-model-cuda-artifact (mistralai, Voxtral-Mini-3B-2507, non-quantized) / linux-job	28m 4s
export-model-cuda-artifact (mistralai, Voxtral-Mini-4B-Realtime-2602, quantized-int4-tile-packed) / linux-job	23m 40s
export-model-cuda-artifact (nvidia, diar_streaming_sortformer_4spk-v2, non-quantized) / linux-job	31m 2s
export-model-cuda-artifact (openai, whisper-large-v3-turbo, quantized-int4-tile-packed) / linux-job	20m 11s
export-model-cuda-artifact (SocialLocalMobile, Qwen3.5-35B-A3B-HQQ-INT4, quantized-int4-tile-packed) / linux-job	46m 1s
export-model-cuda-artifact (openai, whisper-small, non-quantized) / linux-job	18m 30s
export-model-cuda-artifact (mistralai, Voxtral-Mini-3B-2507, quantized-int4-weight-only) / linux-job	26m 22s
export-model-cuda-artifact (google, gemma-3-4b-it, quantized-int4-tile-packed) / linux-job	28m 31s
export-model-cuda-artifact (facebook, dinov2-small-imagenet1k-1-layer, non-quantized) / linux-job	14m 32s
export-model-cuda-artifact (google, gemma-3-4b-it, non-quantized) / linux-job	29m 37s
export-model-cuda-artifact (SocialLocalMobile, gemma-4-31B-it-HQQ-INT4, quantized-int4-tile-packed) / linux-job	53m 56s
export-model-cuda-artifact (Qwen, Qwen3-0.6B, quantized-int4-tile-packed) / linux-job	18m 49s
export-model-cuda-artifact (nvidia, parakeet-tdt, quantized-int4-tile-packed) / linux-job	25m 48s
export-model-cuda-artifact (mistralai, Voxtral-Mini-3B-2507, quantized-int4-tile-packed) / linux-job	26m 15s
test-cuda-pybind (qwen3-0.6b, --quantize, Qwen-Qwen3-0.6B-cuda-quantized-int4-tile-packed) / linux-job	19m 22s
test-model-cuda-e2e (nvidia, parakeet-tdt, non-quantized) / linux-job	11m 36s
test-model-cuda-e2e (mistralai, Voxtral-Mini-3B-2507, quantized-int4-tile-packed) / linux-job	10m 38s
test-model-cuda-e2e (facebook, dinov2-small-imagenet1k-1-layer, non-quantized) / linux-job	10m 20s
test-model-cuda-e2e (mistralai, Voxtral-Mini-4B-Realtime-2602, quantized-int4-tile-packed) / linux-job	25m 57s
test-model-cuda-e2e (SocialLocalMobile, gemma-4-31B-it-HQQ-INT4, quantized-int4-tile-packed) / linux-job	16m 17s
test-model-cuda-e2e (google, gemma-3-4b-it, non-quantized) / linux-job	13m 18s
test-cuda-pybind (qwen3-0.6b, Qwen-Qwen3-0.6B-cuda-non-quantized) / linux-job	20m 30s
test-cuda-pybind (gemma3-4b, --quantize, google-gemma-3-4b-it-cuda-quantized-int4-tile-packed) / linux-job	21m 10s
test-model-cuda-e2e (mistralai, Voxtral-Mini-3B-2507, non-quantized) / linux-job	13m 24s
test-model-cuda-e2e (nvidia, diar_streaming_sortformer_4spk-v2, non-quantized) / linux-job	8m 18s
test-model-cuda-e2e (mistralai, Voxtral-Mini-3B-2507, quantized-int4-weight-only) / linux-job	10m 0s
test-model-cuda-e2e (openai, whisper-large-v3-turbo, quantized-int4-tile-packed) / linux-job	23m 48s
test-model-cuda-e2e (openai, whisper-large-v3-turbo, quantized-int4-weight-only) / linux-job	23m 58s
test-model-cuda-e2e (SocialLocalMobile, Qwen3.5-35B-A3B-HQQ-INT4, quantized-int4-tile-packed) / linux-job	14m 1s
test-model-cuda-e2e (openai, whisper-small, non-quantized) / linux-job	23m 53s
test-model-cuda-e2e (google, gemma-3-4b-it, quantized-int4-tile-packed) / linux-job	10m 5s
test-model-cuda-e2e (nvidia, parakeet-tdt, quantized-int4-tile-packed) / linux-job	10m 4s
Get changed files / get-changed-files	3s
check-all-cuda-builds	3s
	1d 5h 6m 53s

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[cuda backend] int4/8 matvec: vectorized activation load #13969

Usage

[cuda backend] int4/8 matvec: vectorized activation load #13969

Uh oh!

Run time