[ExecuTorch][WebGPU] Add linear_dq8ca_q4gsw + choose_qparams_affine (8da4w) - #21229
Merged
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21229
Note: Links to docs will display an error until the docs builds have been completed. ⏳ No Failures, 52 PendingAs of commit 7643e17 with merge base 4a26c64 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Jul 22, 2026
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
JCNTH
added a commit
that referenced
this pull request
Jul 24, 2026
…8da4w) Pull Request resolved: #21229 Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers `et_vk.linear_dq8ca_q4gsw` + `torchao.choose_qparams_affine`; WebGPU delegated neither, so any `Int8DynamicActivationIntxWeightConfig` model fell back to CPU. Solution: Author the reachable pair. `torchao.choose_qparams_affine` computes per-row asymmetric int8 activation scale/zp; `et_vk.linear_dq8ca_q4gsw` folds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (`ChooseQParams.cpp`, `QuantizedLinear.cpp:760`); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring. Impl: `choose_qparams_affine.wgsl` does a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) then `calculate_scale_and_zero_point` mirroring Vulkan. `linear_dq8ca_q4gsw.wgsl` is the register-tiled q4gsw GEMM with `out[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w)`, reading the packed-int8 zp. int8 zp (`elem_size` 1) is bound word-aligned; `num_rows` must be <=4 or a multiple of 4 (int8 buffers alloc `max(nbytes,4)`) — arbitrary-M prefill is a documented follow-up. ghstack-source-id: 406388760 @exported-using-ghexport Differential Revision: [D112257680](https://our.internmc.facebook.com/intern/diff/D112257680/)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stack from ghstack (oldest at bottom):
Problem: WebGPU lacked the dynamic-8bit-activation x 4-bit-group-weight linear (the 8da4w path real LLMs use for dynamic quant). Vulkan registers
et_vk.linear_dq8ca_q4gsw+torchao.choose_qparams_affine; WebGPU delegated neither, so anyInt8DynamicActivationIntxWeightConfigmodel fell back to CPU.Solution: Author the reachable pair.
torchao.choose_qparams_affinecomputes per-row asymmetric int8 activation scale/zp;et_vk.linear_dq8ca_q4gswfolds that dynamic activation quant into the existing q4gsw 4-bit-group GEMM. Both mirror the Vulkan reference (ChooseQParams.cpp,QuantizedLinear.cpp:760); the activation-quant + weight-dequant math was CPU-de-risked exact against torchao before authoring.Impl:
choose_qparams_affine.wgsldoes a cooperative per-row min/max reduction (one workgroup per block of 4 rows so the 4 int8 zps pack into one u32 with no write race) thencalculate_scale_and_zero_pointmirroring Vulkan.linear_dq8ca_q4gsw.wgslis the register-tiled q4gsw GEMM without[m,n] = s[m] * sum_k (xq[m,k]-z[m]) * dequant(w), reading the packed-int8 zp. int8 zp (elem_size1) is bound word-aligned;num_rowsmust be <=4 or a multiple of 4 (int8 buffers allocmax(nbytes,4)) — arbitrary-M prefill is a documented follow-up.@exported-using-ghexport
Differential Revision: D112257680
Differential Revision: D112257680