| ... |
|
backward
|
|
|
activation_static_qdq.wgsl
|
620 B |
|
activation-static-qdq.d.ts
|
559 B |
|
activation-static-qdq.js
|
2.42 KB |
|
attention.d.ts
|
6.25 KB |
|
attention.js
|
73.23 KB |
|
attention.wgsl
|
9.84 KB |
|
attention_bdpa_decode_f16.wgsl
|
7.87 KB |
|
attention_decode.wgsl
|
8.19 KB |
|
attention_decode_chunked_f16.wgsl
|
5.13 KB |
|
attention_decode_chunked_f16kv.wgsl
|
6.65 KB |
|
attention_decode_contiguous_turboquant_f16kv.wgsl
|
5.75 KB |
|
attention_decode_contiguous_turboquant_prod_f16kv.wgsl
|
6.27 KB |
|
attention_decode_f16.wgsl
|
6.32 KB |
|
attention_decode_f16kv.wgsl
|
7.93 KB |
|
attention_decode_online_f16.wgsl
|
7.5 KB |
|
attention_decode_online_f16kv.wgsl
|
7.6 KB |
|
attention_decode_online_head256_f16kv.wgsl
|
7.58 KB |
|
attention_decode_online_head256_f16kv_output_gate.wgsl
|
9.19 KB |
|
attention_decode_paged_f16.wgsl
|
4.85 KB |
|
attention_decode_paged_f16kv.wgsl
|
5 KB |
|
attention_decode_subgroup.wgsl
|
7.4 KB |
|
attention_decode_tiered_f16.wgsl
|
6.42 KB |
|
attention_decode_tiered_f16kv.wgsl
|
6.58 KB |
|
attention_decode_tiered_int4_f16kv.wgsl
|
6.95 KB |
|
attention_decode_tiered_int8_f16kv.wgsl
|
6.96 KB |
|
attention_decode_tiered_turboquant_f16kv.wgsl
|
8.93 KB |
|
attention_decode_tiered_turboquant_prod_f16kv.wgsl
|
9.63 KB |
|
attention_f16.wgsl
|
10.03 KB |
|
attention_f16kv.wgsl
|
10.02 KB |
|
attention_head128_f16kv.wgsl
|
8.6 KB |
|
attention_head256_f16kv.wgsl
|
8.6 KB |
|
attention_head512_f16.wgsl
|
8.63 KB |
|
attention_head512_f16kv.wgsl
|
8.8 KB |
|
attention_prefill_flash_head256_f16kv.wgsl
|
11.89 KB |
|
attention_prefill_flash_reduce.wgsl
|
3.65 KB |
|
attention_small.wgsl
|
9.42 KB |
|
attention_small_f16.wgsl
|
8.07 KB |
|
attention_small_f16kv.wgsl
|
9.15 KB |
|
attention_streaming.wgsl
|
5.53 KB |
|
attention_streaming_f16.wgsl
|
5.75 KB |
|
attention_streaming_f16kv.wgsl
|
5.61 KB |
|
bf16_to_f16.wgsl
|
1.67 KB |
|
bf16_to_f32.wgsl
|
1.86 KB |
|
bias_add.wgsl
|
1.18 KB |
|
bias_add_f16.wgsl
|
1.4 KB |
|
cast.d.ts
|
1.37 KB |
|
cast.js
|
11.09 KB |
|
cast_f16_to_f32.wgsl
|
791 B |
|
cast_f32_to_f16.wgsl
|
903 B |
|
check-finiteness.js
|
4.64 KB |
|
check-hot-vocab-stop.js
|
3.9 KB |
|
check-stop.js
|
5.08 KB |
|
clamp.d.ts
|
467 B |
|
clamp.js
|
1.34 KB |
|
clamp.wgsl
|
517 B |
|
constants.js
|
1.58 KB |
|
conv2d.d.ts
|
818 B |
|
conv2d.js
|
3.11 KB |
|
conv2d.wgsl
|
2.08 KB |
|
conv2d_f16.wgsl
|
2.23 KB |
|
cross_entropy_loss.d.ts
|
592 B |
|
cross_entropy_loss.js
|
2.16 KB |
|
cross_entropy_loss.wgsl
|
952 B |
|
depthwise_conv1d.d.ts
|
688 B |
|
depthwise_conv1d.js
|
1.74 KB |
|
depthwise_conv1d.wgsl
|
1.27 KB |
|
depthwise_conv2d.d.ts
|
838 B |
|
depthwise_conv2d.js
|
3.27 KB |
|
depthwise_conv2d.wgsl
|
1.68 KB |
|
depthwise_conv2d_f16.wgsl
|
1.75 KB |
|
dequant.d.ts
|
2.25 KB |
|
dequant.js
|
15.44 KB |
|
dequant_f16_out.wgsl
|
4.67 KB |
|
dequant_f16_out_vec4.wgsl
|
4.81 KB |
|
dequant_f16_rowwise.wgsl
|
3.82 KB |
|
dequant_f32_rowwise.wgsl
|
3.65 KB |
|
dequant_mxfp4.wgsl
|
3.92 KB |
|
dequant_mxfp4_expert.wgsl
|
4.58 KB |
|
dequant_mxfp4_expert_f16.wgsl
|
3.13 KB |
|
dequant_q6k.wgsl
|
4.51 KB |
|
dequant_shared.wgsl
|
4.93 KB |
|
dequant_subgroup.wgsl
|
7.28 KB |
|
diffusion_gemma_sampling.wgsl
|
3.9 KB |
|
diffusion-gemma-sampling.d.ts
|
518 B |
|
diffusion-gemma-sampling.js
|
4.05 KB |
|
dispatch.js
|
3.4 KB |
|
energy.d.ts
|
2.54 KB |
|
energy.js
|
14.84 KB |
|
energy_eval.wgsl
|
713 B |
|
energy_eval_f16.wgsl
|
897 B |
|
energy_quintel_grad.wgsl
|
2.29 KB |
|
energy_quintel_grad_f16.wgsl
|
2.49 KB |
|
energy_quintel_reduce.wgsl
|
2.97 KB |
|
energy_quintel_reduce_f16.wgsl
|
3.2 KB |
|
energy_quintel_update.wgsl
|
2.33 KB |
|
energy_quintel_update_f16.wgsl
|
2.51 KB |
|
energy_update.wgsl
|
720 B |
|
energy_update_f16.wgsl
|
899 B |
|
feature-check.d.ts
|
1.18 KB |
|
feature-check.js
|
2.31 KB |
|
fused_ffn.d.ts
|
2.66 KB |
|
fused_ffn.js
|
29.66 KB |
|
fused_ffn.wgsl
|
13.38 KB |
|
fused_ffn_f16.wgsl
|
6.46 KB |
|
fused_ffn_f16w_f32a_batched.wgsl
|
3.97 KB |
|
fused_ffn_q4k.wgsl
|
13.11 KB |
|
fused_ffn_q4k_f16.wgsl
|
10.63 KB |
|
fused_ffn_q4k_metal_simd16.wgsl
|
6.45 KB |
|
fused_gate_up_gelu_gated_f16.wgsl
|
4.44 KB |
|
fused_matmul_litert_int4_f16a.wgsl
|
2.42 KB |
|
fused_matmul_litert_int4_f16a_f32out.wgsl
|
2.42 KB |
|
fused_matmul_litert_int4_f32a_f32out.wgsl
|
2.4 KB |
|
fused_matmul_q4.wgsl
|
15.37 KB |
|
fused_matmul_q4_batched.wgsl
|
6.84 KB |
|
fused_matmul_q4_batched_f16.wgsl
|
5.79 KB |
|
fused_matmul_q4_batched_f16a.wgsl
|
4.47 KB |
|
fused_matmul_q4_batched_f16acc_f16a.wgsl
|
4.86 KB |
|
fused_matmul_q4_batched_multicol_shared.wgsl
|
5.35 KB |
|
fused_matmul_q4_f16a.wgsl
|
7.19 KB |
|
fused_matmul_q4_metal_simd16.wgsl
|
4.75 KB |
|
fused_matmul_q4_multicol_f16.wgsl
|
7.58 KB |
|
fused_matmul_q4_multicol_f16a.wgsl
|
6.79 KB |
|
fused_matmul_q4_multicol_f16a_f32acc.wgsl
|
7.26 KB |
|
fused_matmul_q4_prefill_tiled_f16.wgsl
|
10.95 KB |
|
fused_matmul_q4_widetile.wgsl
|
5.49 KB |
|
fused_matmul_q4_widetile_f16.wgsl
|
7.63 KB |
|
fused_matmul_q4_widetile_f16a.wgsl
|
5.33 KB |
|
fused_matmul_q4_widetile_residual.wgsl
|
5.87 KB |
|
fused_matmul_residual.d.ts
|
1.25 KB |
|
fused_matmul_residual.js
|
4.98 KB |
|
fused_matmul_rmsnorm.d.ts
|
1.93 KB |
|
fused_matmul_rmsnorm.js
|
9.95 KB |
|
fused_matmul_rmsnorm.wgsl
|
12.08 KB |
|
fused_matmul_rmsnorm_f16.wgsl
|
11.02 KB |
|
fused_matmul_w4a16_batched_f16a.wgsl
|
3.87 KB |
|
fused_matmul_w4a16_batched_f16a_f32out.wgsl
|
3.88 KB |
|
fused_matmul_w4a16_f16a.wgsl
|
3.02 KB |
|
fused_matmul_w4a16_f16a_f32out.wgsl
|
3.02 KB |
|
fused_rmsnorm_q4_widetile.wgsl
|
10.34 KB |
|
fused_swiglu.wgsl
|
1.66 KB |
|
fused_swiglu_f16.wgsl
|
1.46 KB |
|
fused-gate-up-gelu.js
|
4.26 KB |
|
gated-short-conv.js
|
8.77 KB |
|
gather.d.ts
|
4.17 KB |
|
gather.js
|
15.96 KB |
|
gather.wgsl
|
2.24 KB |
|
gather_f16.wgsl
|
2.48 KB |
|
gather_f16_f16_out.wgsl
|
1.61 KB |
|
gather_f16_out.wgsl
|
1.63 KB |
|
gather_f16_vec4.wgsl
|
2.74 KB |
|
gather_f16_vec4_f16_out.wgsl
|
2.37 KB |
|
gather_litert_int4_vec4_f16_out.wgsl
|
2.35 KB |
|
gather_litert_int4_vec4_f32_out.wgsl
|
2.29 KB |
|
gather_split4_f16_vec4_f16_out.wgsl
|
3.69 KB |
|
gather_split8_f16_vec4_f16_out.wgsl
|
5.89 KB |
|
gather_split8_f16_vec4_f32_out.wgsl
|
6.03 KB |
|
gather_vec4.wgsl
|
2.65 KB |
|
gather_vec4_f16_out.wgsl
|
2.41 KB |
|
gelu.d.ts
|
794 B |
|
gelu.js
|
1.78 KB |
|
gelu.wgsl
|
1.84 KB |
|
gelu_f16.wgsl
|
1.94 KB |
|
gemma4_route_q4_matmul_f16a.wgsl
|
7.18 KB |
|
gemma4-route-expert.d.ts
|
845 B |
|
gemma4-route-expert.js
|
3.77 KB |
|
gptoss_mxfp4_expert_fused.wgsl
|
3.56 KB |
|
gptoss_router_topk.wgsl
|
3.8 KB |
|
grouped_pointwise_conv2d.d.ts
|
843 B |
|
grouped_pointwise_conv2d.js
|
3.23 KB |
|
grouped_pointwise_conv2d.wgsl
|
1.37 KB |
|
grouped_pointwise_conv2d_f16.wgsl
|
1.45 KB |
|
groupnorm.d.ts
|
757 B |
|
groupnorm.js
|
3.12 KB |
|
groupnorm_apply.wgsl
|
1.15 KB |
|
groupnorm_apply_f16.wgsl
|
1.38 KB |
|
groupnorm_stats.wgsl
|
2.05 KB |
|
groupnorm_stats_f16.wgsl
|
2.2 KB |
|
index.d.ts
|
9.68 KB |
|
index.js
|
7.26 KB |
|
kernel-base.js
|
711 B |
|
kernel-configs.d.ts
|
2.23 KB |
|
kernel-configs.js
|
2.05 KB |
|
kernel-prewarm.d.ts
|
106 B |
|
kernel-prewarm.js
|
2.04 KB |
|
kv_cache_write_f32_to_f16.wgsl
|
861 B |
|
kv_quantize_int4.wgsl
|
3.96 KB |
|
kv_quantize_int8.wgsl
|
3.96 KB |
|
kv_quantize_turboquant.wgsl
|
4.98 KB |
|
kv_quantize_turboquant_prod.wgsl
|
8.01 KB |
|
kv-cache-write.d.ts
|
580 B |
|
kv-cache-write.js
|
3.89 KB |
|
kv-quantize.d.ts
|
832 B |
|
kv-quantize.js
|
5.06 KB |
|
layernorm.d.ts
|
876 B |
|
layernorm.js
|
2.9 KB |
|
layernorm.wgsl
|
3.93 KB |
|
layernorm_f16.wgsl
|
3.2 KB |
|
linear_attention.d.ts
|
699 B |
|
linear_attention.js
|
3.87 KB |
|
linear_attention_apply.wgsl
|
1.35 KB |
|
linear_attention_apply_f16.wgsl
|
1.42 KB |
|
linear_attention_summary.wgsl
|
1.52 KB |
|
linear_attention_summary_f16.wgsl
|
1.54 KB |
|
linear-attention-core.js
|
38.36 KB |
|
lm_head_argmax_f16.wgsl
|
6.53 KB |
|
lm_head_argmax_q4k.wgsl
|
9.24 KB |
|
lm_head_select_logits.wgsl
|
2.02 KB |
|
lm-head-argmax.d.ts
|
775 B |
|
lm-head-argmax.js
|
12.27 KB |
|
lm-head-select-logits.d.ts
|
621 B |
|
lm-head-select-logits.js
|
6.87 KB |
|
logit-merge.d.ts
|
2.11 KB |
|
logit-merge.js
|
9.99 KB |
|
matmul.d.ts
|
2.99 KB |
|
matmul.js
|
23.83 KB |
|
matmul_f16.wgsl
|
5.78 KB |
|
matmul_f16_tiled.wgsl
|
7.69 KB |
|
matmul_f16w_f32a.wgsl
|
2.5 KB |
|
matmul_f16w_f32a_tiled.wgsl
|
6.97 KB |
|
matmul_f32.wgsl
|
3.32 KB |
|
matmul_gemv.wgsl
|
2.54 KB |
|
matmul_gemv_f16a.wgsl
|
2.6 KB |
|
matmul_gemv_residual.wgsl
|
3.67 KB |
|
matmul_gemv_residual_f16.wgsl
|
2.66 KB |
|
matmul_gemv_subgroup.wgsl
|
16.27 KB |
|
matmul_gemv_subgroup_f16a.wgsl
|
23.94 KB |
|
matmul-dispatch.js
|
6.91 KB |
|
matmul-selection.js
|
23.56 KB |
|
modulate.d.ts
|
567 B |
|
modulate.js
|
1.92 KB |
|
modulate.wgsl
|
1.12 KB |
|
modulate_f16.wgsl
|
1.33 KB |
|
moe.d.ts
|
2.63 KB |
|
moe.js
|
21.41 KB |
|
moe_gather.wgsl
|
6.25 KB |
|
moe_gather_f16.wgsl
|
2.64 KB |
|
moe_offsets.wgsl
|
1.45 KB |
|
pipeline-cache.d.ts
|
2.75 KB |
|
pipeline-cache.js
|
9.54 KB |
|
pixel_shuffle.d.ts
|
561 B |
|
pixel_shuffle.js
|
2.12 KB |
|
pixel_shuffle.wgsl
|
1.19 KB |
|
pixel_shuffle_f16.wgsl
|
1.37 KB |
|
relu.d.ts
|
471 B |
|
relu.js
|
2.12 KB |
|
relu.wgsl
|
582 B |
|
relu_f16.wgsl
|
596 B |
|
rep_penalty.wgsl
|
1.63 KB |
|
rep_penalty_f16.wgsl
|
1.37 KB |
|
rep-penalty.js
|
3.42 KB |
|
repeat_channels.d.ts
|
570 B |
|
repeat_channels.js
|
1.92 KB |
|
repeat_channels.wgsl
|
816 B |
|
repeat_channels_f16.wgsl
|
829 B |
|
residual.d.ts
|
1.44 KB |
|
residual.js
|
6.9 KB |
|
residual.wgsl
|
1.67 KB |
|
residual_f16.wgsl
|
1.22 KB |
|
residual_f16_vec4.wgsl
|
1.36 KB |
|
residual_rmsnorm_pair.wgsl
|
3.69 KB |
|
residual_vec4.wgsl
|
1.33 KB |
|
rmsnorm.d.ts
|
2.21 KB |
|
rmsnorm.js
|
10.75 KB |
|
rmsnorm.wgsl
|
15.4 KB |
|
rmsnorm_f16.wgsl
|
7.85 KB |
|
rmsnorm_pair.wgsl
|
7.04 KB |
|
rmsnorm_qk.wgsl
|
3.38 KB |
|
rmsnorm_qk_f16.wgsl
|
4.24 KB |
|
rmsnorm-pair.d.ts
|
2.05 KB |
|
rmsnorm-pair.js
|
8.08 KB |
|
rmsnorm-qk.d.ts
|
1.06 KB |
|
rmsnorm-qk.js
|
4.43 KB |
|
rmsnorm-stats.d.ts
|
684 B |
|
rmsnorm-stats.js
|
8.82 KB |
|
rope.d.ts
|
1.15 KB |
|
rope.js
|
4.02 KB |
|
rope.wgsl
|
11.96 KB |
|
rope_f16.wgsl
|
10.35 KB |
|
rope_qk.wgsl
|
3.18 KB |
|
rope_qk_f16.wgsl
|
3.22 KB |
|
rope-qk.d.ts
|
1002 B |
|
rope-qk.js
|
4.99 KB |
|
rule-matcher.js
|
1.79 KB |
|
rule-registry.js
|
2.77 KB |
|
sample.d.ts
|
1.42 KB |
|
sample.js
|
18.68 KB |
|
sample.wgsl
|
13.47 KB |
|
sample_f16.wgsl
|
11.08 KB |
|
scale.d.ts
|
1 KB |
|
scale.js
|
1.68 KB |
|
scale.wgsl
|
1.06 KB |
|
scale_f16.wgsl
|
771 B |
|
scatter_add.wgsl
|
2.92 KB |
|
scatter_add_dynamic.wgsl
|
2.23 KB |
|
scatter_add_dynamic_expert_scale.wgsl
|
1.54 KB |
|
scatter_add_dynamic_f16.wgsl
|
1.63 KB |
|
scatter_add_dynamic_f16_expert_scale.wgsl
|
1.56 KB |
|
scatter_add_dynamic_f16_weights.wgsl
|
1.47 KB |
|
scatter_add_dynamic_f16_weights_expert_scale.wgsl
|
1.57 KB |
|
scatter_add_routes_f16_weights_expert_scale.wgsl
|
1.55 KB |
|
scatter_add_vec4.wgsl
|
2.31 KB |
|
shader-cache.d.ts
|
1.85 KB |
|
shader-cache.js
|
7.03 KB |
|
silu.d.ts
|
1.77 KB |
|
silu.js
|
13 KB |
|
silu.wgsl
|
3.31 KB |
|
silu_f16.wgsl
|
3.41 KB |
|
soft_embedding_logits_f16.wgsl
|
2.7 KB |
|
soft_embedding_split_f16.wgsl
|
1.15 KB |
|
soft-embedding.d.ts
|
806 B |
|
soft-embedding.js
|
14.34 KB |
|
softmax.d.ts
|
1.25 KB |
|
softmax.js
|
4.57 KB |
|
softmax.wgsl
|
5.29 KB |
|
softmax_subgroup.wgsl
|
5.07 KB |
|
split_qg.d.ts
|
1.43 KB |
|
split_qg.js
|
1.69 KB |
|
split_qg.wgsl
|
1.95 KB |
|
split_qg_f16.wgsl
|
2.13 KB |
|
split_qkv.d.ts
|
1.26 KB |
|
split_qkv.js
|
1.99 KB |
|
split_qkv.wgsl
|
1.92 KB |
|
split_qkv_f16.wgsl
|
2.11 KB |
|
split_qkv_rmsnorm_qk.wgsl
|
4.25 KB |
|
split_qkv_rmsnorm_rope_qk.wgsl
|
6.22 KB |
|
split_qkv_rmsnorm_rope_qk_f16kv_cache.wgsl
|
6.29 KB |
|
split-qkv-rmsnorm-qk.d.ts
|
1.2 KB |
|
split-qkv-rmsnorm-qk.js
|
5.09 KB |
|
split-qkv-rmsnorm-rope-qk.d.ts
|
1.6 KB |
|
split-qkv-rmsnorm-rope-qk.js
|
8.65 KB |
|
topk.wgsl
|
7.93 KB |
|
topk_f16.wgsl
|
3.31 KB |
|
topk_f16_weights.wgsl
|
3.03 KB |
|
transpose.d.ts
|
487 B |
|
transpose.js
|
1.74 KB |
|
transpose.wgsl
|
819 B |
|
turboquant-codebook.js
|
15.5 KB |
|
types.d.ts
|
357 B |
|
uniform-utils.d.ts
|
1.96 KB |
|
uniform-utils.js
|
2.94 KB |
|
upsample2d.d.ts
|
547 B |
|
upsample2d.js
|
2.18 KB |
|
upsample2d.wgsl
|
1007 B |
|
upsample2d_f16.wgsl
|
1.17 KB |
|
utils.d.ts
|
4.29 KB |
|
utils.js
|
11.53 KB |