| /aosp_15_r20/external/pytorch/torch/nn/modules/ |
| H A D | sparse.py | 25 …padding_idx (int, optional): If specified, the entries at :attr:`padding_idx` do not contribute to…
26 … therefore, the embedding vector at :attr:`padding_idx` is not updated during training,
28 … the embedding vector at :attr:`padding_idx` will default to all zeros,
87 >>> # example with padding_idx
88 >>> embedding = nn.Embedding(10, 3, padding_idx=0)
97 >>> padding_idx = 0
98 >>> embedding = nn.Embedding(3, 3, padding_idx=padding_idx)
105 ... embedding.weight[padding_idx] = torch.ones(3)
116 "padding_idx",
125 padding_idx: Optional[int]
[all …]
|
| /aosp_15_r20/external/pytorch/torch/csrc/api/src/nn/modules/ |
| H A D | embedding.cpp | 23 if (options.padding_idx() != std::nullopt) { in reset()
24 if (*options.padding_idx() > 0) { in reset()
26 *options.padding_idx() < options.num_embeddings(), in reset()
27 "Padding_idx must be within num_embeddings"); in reset()
28 } else if (*options.padding_idx() < 0) { in reset()
30 *options.padding_idx() >= -options.num_embeddings(), in reset()
31 "Padding_idx must be within num_embedding"); in reset()
32 options.padding_idx(options.num_embeddings() + *options.padding_idx()); in reset()
53 if (options.padding_idx() != std::nullopt) { in reset_parameters()
55 weight[*options.padding_idx()].fill_(0); in reset_parameters()
[all …]
|
| /aosp_15_r20/external/pytorch/test/nn/ |
| H A D | test_embedding.py | 129 padding_idx = 2
133 embeddings[padding_idx] = padding_vec
134 embedding_nn = nn.Embedding.from_pretrained(embeddings, padding_idx=padding_idx)
135 self.assertEqual(embedding_nn.weight[padding_idx], padding_vec)
138 padding_idx = 2
141 embeddings, padding_idx=padding_idx
179 embed_old = embed_old.from_pretrained(embeddings, padding_idx=2)
181 res_F = F.embedding(a, embeddings, padding_idx=2)
206 embed_old = embed_old.from_pretrained(embeddings, padding_idx=2)
208 res_F = F.embedding_bag(a, embeddings, padding_idx=2)
[all …]
|
| /aosp_15_r20/external/pytorch/aten/src/ATen/native/ |
| H A D | EmbeddingBag.cpp | 81 bool is_fast_path_index_select(const Tensor& src, Tensor& output, index_t padding_idx) { in is_fast_path_index_select() argument
85 padding_idx < static_cast<index_t>(0); in is_fast_path_index_select()
91 …h_index_select_scale(const Tensor& src, const Tensor& scale, Tensor& output, index_t padding_idx) { in is_fast_path_index_select_scale() argument
95 scale.strides()[0] == 1 && padding_idx < static_cast<index_t>(0); in is_fast_path_index_select_scale()
99 …_path(const Tensor& src, const std::optional<Tensor>& scale, Tensor& output, index_t padding_idx) { in is_fast_path() argument
101 is_fast_path_index_select_scale(src, scale.value(), output, padding_idx) : in is_fast_path()
102 is_fast_path_index_select(src, output, padding_idx); in is_fast_path()
118 index_t padding_idx, in index_select_add() argument
138 // We can skip indices equal to padding_idx so they are not included in in index_select_add()
145 if (idx != padding_idx) { in index_select_add()
[all …]
|
| H A D | Embedding.cpp | 38 c10::SymInt padding_idx, bool scale_grad_by_freq, bool sparse) { in embedding_symint() argument
58 c10::SymInt padding_idx, bool scale_grad_by_freq, bool sparse) { in embedding_backward_symint() argument
67 padding_idx.guard_int(__FILE__, __LINE__), in embedding_backward_symint()
71 grad, indices, std::move(num_weights), padding_idx, scale_grad_by_freq); in embedding_backward_symint()
77 int64_t padding_idx, bool scale_grad_by_freq) { in embedding_sparse_backward() argument
90 if (padding_idx != -1) { in embedding_sparse_backward()
91 c10::List<std::optional<Tensor>> c({indices != padding_idx}); in embedding_sparse_backward()
100 // check if all our grad come from padding_idx in embedding_sparse_backward()
114 int64_t padding_idx, bool scale_grad_by_freq) { in embedding_dense_backward_cpu() argument
155 if (indices_data[i] != padding_idx) { in embedding_dense_backward_cpu()
|
| /aosp_15_r20/external/pytorch/torch/distributed/_shard/sharding_spec/chunk_sharding_spec_ops/ |
| H A D | _common.py | 95 padding_idx=None, argument
118 padding_idx: If specified, the entries at padding_idx do
120 vector at padding_idx is not updated during training,
122 Note that the embedding vector at padding_idx is
139 padding_idx=padding_idx,
145 padding_idx=padding_idx,
300 def _handle_row_wise_mask(gather_inp, padding_idx, weight, world_size, rank): argument
303 on the current rank. This function also adjust the ``padding_idx``
313 padding_idx: If specified, the entries at padding_idx do
315 vector at padding_idx is not updated during training,
[all …]
|
| H A D | embedding.py | 100 padding_idx = kwargs.get("padding_idx")
109 input, world_size, weight, local_shard, max_norm, norm_type, padding_idx, pg
121 padding_idx,
187 input, world_size, weight, local_shard, max_norm, norm_type, padding_idx, pg argument
203 padding_idx: If specified, the entries at padding_idx do
205 vector at padding_idx is not updated during training,
229 padding_idx=padding_idx,
235 input, world_size, weight, local_shard, max_norm, norm_type, padding_idx, rank, pg argument
251 padding_idx: If specified, the entries at padding_idx do
253 vector at padding_idx is not updated during training,
[all …]
|
| H A D | embedding_bag.py | 66 defined by the offset (excluding the row specified in ``padding_idx``).
107 padding_idx = kwargs.get("padding_idx")
127 padding_idx,
143 padding_idx,
249 padding_idx, argument
269 padding_idx: If specified, the entries at padding_idx do
271 vector at padding_idx is not updated during training,
273 Note that the embedding vector at padding_idx is
307 padding_idx=padding_idx,
322 padding_idx, argument
[all …]
|
| /aosp_15_r20/external/pytorch/torch/csrc/api/include/torch/nn/options/ |
| H A D | embedding.h | 16 /// 2).padding_idx(3).max_norm(2).norm_type(2.5).scale_grad_by_freq(true).sparse(true));
25 /// If specified, the entries at `padding_idx` do not contribute to the
26 /// gradient; therefore, the embedding vector at `padding_idx` is not updated
28 /// Embedding, the embedding vector at `padding_idx` will default to all
31 TORCH_ARG(std::optional<int64_t>, padding_idx) = std::nullopt;
55 /// If specified, the entries at `padding_idx` do not contribute to the
56 /// gradient; therefore, the embedding vector at `padding_idx` is not updated
58 TORCH_ARG(std::optional<int64_t>, padding_idx) = std::nullopt;
84 /// If specified, the entries at `padding_idx` do not contribute to the
85 /// gradient; therefore, the embedding vector at `padding_idx` is not updated
[all …]
|
| /aosp_15_r20/external/pytorch/torch/csrc/api/include/torch/nn/functional/ |
| H A D | embedding.h | 27 std::optional<int64_t> padding_idx, in embedding() argument
34 if (padding_idx != std::nullopt) { in embedding()
35 if (*padding_idx > 0) { in embedding()
37 *padding_idx < weight.size(0), in embedding()
38 "Padding_idx must be within num_embeddings"); in embedding()
39 } else if (*padding_idx < 0) { in embedding()
41 *padding_idx >= -weight.size(0), in embedding()
42 "Padding_idx must be within num_embedding"); in embedding()
43 padding_idx = weight.size(0) + *padding_idx; in embedding()
46 padding_idx = -1; in embedding()
[all …]
|
| /aosp_15_r20/external/pytorch/test/distributed/_shard/sharded_tensor/ops/ |
| H A D | test_embedding_bag.py | 42 padding_idx=None, argument
53 padding_idx=padding_idx,
63 padding_idx=padding_idx,
145 padding_idx=padding_idx,
156 padding_idx=padding_idx,
205 self._run_sharded_embedding_bag(spec, [5, 5], 17, 14, "sum", padding_idx=6)
244 self._run_sharded_embedding_bag(spec, [4, 3], 16, 14, "sum", padding_idx=12)
245 self._run_sharded_embedding_bag(spec, [4, 3], 16, 14, "mean", padding_idx=12)
246 self._run_sharded_embedding_bag(spec, [4, 3], 16, 14, "max", padding_idx=12)
255 padding_idx=10,
[all …]
|
| H A D | test_embedding.py | 39 padding_idx=None, argument
48 padding_idx=padding_idx,
56 padding_idx=padding_idx,
103 padding_idx=padding_idx,
110 padding_idx=padding_idx,
125 self._run_sharded_embedding(spec, [34], 15, 14, padding_idx=10)
126 self._run_sharded_embedding(spec, [8, 6, 5, 4], 23, 13, padding_idx=12)
167 self._run_sharded_embedding(spec, [6, 7, 6], 64, 11, padding_idx=30)
180 self._run_sharded_embedding(spec, [8, 6, 5, 4], 21, 11, padding_idx=10)
|
| /aosp_15_r20/external/pytorch/aten/src/ATen/native/cuda/ |
| H A D | EmbeddingBag.cu | 67 index_t padding_idx, int64_t numRows) { in EmbeddingBag_updateOutputKernel_max() argument
88 bool pad = (input[emb] == padding_idx); in EmbeddingBag_updateOutputKernel_max()
118 index_t padding_idx, int64_t numRows) { in EmbeddingBag_updateOutputKernel_sum_mean() argument
139 bool pad = (input[emb] == padding_idx); in EmbeddingBag_updateOutputKernel_sum_mean()
176 int64_t padding_idx) { in embedding_bag_backward_cuda_sum_avg() argument
235 count, num_weights, padding_idx, mode == EmbeddingBagMode::MEAN, offset2bag, in embedding_bag_backward_cuda_sum_avg()
243 index_t padding_idx, const index_t numel) { in EmbeddingBag_accGradParametersKernel_max() argument
258 if (word_idx >= 0 && word_idx != padding_idx) { in EmbeddingBag_accGradParametersKernel_max()
271 int64_t padding_idx) { in embedding_bag_backward_cuda_max() argument
298 padding_idx, grad_weight.numel()); in embedding_bag_backward_cuda_max()
[all …]
|
| H A D | Embedding.cu | 54 int padding_idx) in embedding_backward_feature_kernel() argument
89 if(src_row < n && f < s && dst_row != padding_idx) in embedding_backward_feature_kernel()
99 if(dst_row != padding_idx && src_row < n) // Per-warp exit condition, safe with ballot_sync in embedding_backward_feature_kernel()
137 index_t* count, int64_t numel, int64_t stride, int padding_idx) { in embedding_backward_kernel() argument
158 && input[idx] != padding_idx) { in embedding_backward_kernel()
249 int64_t num_weights, int64_t padding_idx, in embedding_dense_backward_cuda() argument
288 static_cast<int>(padding_idx)); in embedding_dense_backward_cuda()
344 sorted_indices, count, num_weights, padding_idx); in embedding_dense_backward_cuda()
|
| /aosp_15_r20/external/executorch/kernels/test/ |
| H A D | op_embedding_test.cpp | 30 int64_t padding_idx, in op_embedding_out() argument
38 padding_idx, in op_embedding_out()
65 /*padding_idx=*/0, in test_dtype()
133 /*padding_idx=*/0, in TEST_F()
175 /*padding_idx=*/0, in TEST_F()
214 /*padding_idx=*/0, in TEST_F()
250 /*padding_idx=*/0, in TEST_F()
280 /*padding_idx=*/0, in TEST_F()
310 /*padding_idx=*/0, in TEST_F()
320 /*padding_idx=*/0, in TEST_F()
[all …]
|
| /aosp_15_r20/external/pytorch/torch/ao/nn/qat/modules/ |
| H A D | embedding_ops.py | 32 padding_idx=None, argument
46 padding_idx,
66 self.padding_idx,
98 mod.padding_idx,
113 self.padding_idx,
153 padding_idx=None, argument
169 padding_idx,
192 self.padding_idx,
227 mod.padding_idx,
244 self.padding_idx,
|
| /aosp_15_r20/external/pytorch/torch/ao/nn/quantized/reference/modules/ |
| H A D | sparse.py | 26 padding_idx: Optional[int] = None,
39 padding_idx,
58 self.padding_idx,
70 mod.padding_idx,
101 padding_idx: Optional[int] = None,
116 padding_idx,
143 self.padding_idx,
158 mod.padding_idx,
|
| /aosp_15_r20/external/pytorch/torch/csrc/api/include/torch/nn/modules/ |
| H A D | embedding.h | 28 /// 2).padding_idx(3).max_norm(2).norm_type(2.5).scale_grad_by_freq(true).sparse(true));
80 .padding_idx(options.padding_idx())
104 …max_norm(2).norm_type(2.5).scale_grad_by_freq(true).sparse(true).mode(torch::kSum).padding_idx(1));
165 .padding_idx(options.padding_idx()));
|
| /aosp_15_r20/external/pytorch/torch/nn/utils/_expanded_weights/ |
| H A D | embedding_expanded_weights.py | 28 ctx.padding_idx, ctx.scale_grad_by_freq = (
29 expanded_kwargs["padding_idx"],
38 padding_idx, scale_grad_by_freq, sparse = (
39 ctx.padding_idx,
70 padding_idx,
|
| /aosp_15_r20/external/pytorch/aten/src/ATen/functorch/ |
| H A D | BatchRulesModules.cpp | 26 c10::SymInt padding_idx, bool scale_grad_by_freq, bool sparse) { in embedding_batch_rule() argument
29 …auto result = at::embedding_symint(weight, indices, std::move(padding_idx), scale_grad_by_freq, sp… in embedding_batch_rule()
35 …auto result = at::embedding_symint(weight_, indices, std::move(padding_idx), scale_grad_by_freq, s… in embedding_batch_rule()
49 …auto result = at::embedding_symint(weight_, indices_, std::move(padding_idx), scale_grad_by_freq, … in embedding_batch_rule()
57 c10::SymInt num_weights, c10::SymInt padding_idx, bool scale_grad_by_freq) { in embedding_dense_backward_batch_rule() argument
64 grad, indices, std::move(num_weights), std::move(padding_idx), scale_grad_by_freq); in embedding_dense_backward_batch_rule()
78 if (padding_idx >= 0) { in embedding_dense_backward_batch_rule()
79 result.select_symint(1, std::move(padding_idx)).fill_(0); in embedding_dense_backward_batch_rule()
|
| /aosp_15_r20/external/executorch/backends/apple/mps/runtime/operations/ |
| H A D | IndexingOps.mm | 77 int padding_idx = graphNode->padding_idx();
79 if (padding_idx != -1) {
80 MPSGraphTensor* constantTensor = [_mpsGraph constantWithScalar:padding_idx
|
| /aosp_15_r20/external/pytorch/test/distributed/_tensor/ |
| H A D | test_embedding_ops.py | 148 self._run_embedding_op_test(mesh, 1, [34], 15, 14, padding_idx=10)
149 self._run_embedding_op_test(mesh, 1, [8, 6, 5, 4], 23, 13, padding_idx=12)
159 mesh, 1, [8, 6, 5, 4], 23, 13, padding_idx=12, max_norm=2.0
168 self._run_embedding_op_test(mesh, 0, [34], 15, 14, padding_idx=10)
|
| /aosp_15_r20/external/executorch/exir/passes/ |
| H A D | _quant_patterns_and_replacements.py | 943 padding_idx,
954 out = torch.ops.aten.embedding.default(weight, indicies, padding_idx)
964 _, # padding_idx only matters for training and not when running op for inference
985 padding_idx,
999 out = torch.ops.aten.embedding.default(weight, indices, padding_idx)
1010 _, # padding_idx only matters for training and not when running op for inference
|
| /aosp_15_r20/external/pytorch/torch/nn/ |
| H A D | functional.py | 2440 padding_idx: Optional[int] = None,
2456 entries in :attr:`weight` at the row specified by :attr:`padding_idx`
2462 :attr:`padding_idx` to all zeros on construction.
2468 …padding_idx (int, optional): If specified, the entries at :attr:`padding_idx` do not contribute to…
2469 … therefore, the embedding vector at :attr:`padding_idx` is not updated during training,
2504 >>> # example with padding_idx
2509 >>> F.embedding(input, embedding_matrix, padding_idx=0)
2521 padding_idx=padding_idx,
2527 if padding_idx is not None:
2528 if padding_idx > 0:
[all …]
|
| /aosp_15_r20/external/executorch/kernels/portable/cpu/ |
| H A D | op_embedding.cpp | 71 // embedding.out(Tensor weight, Tensor indices, int padding_idx=-1, bool
77 int64_t padding_idx, in embedding_out() argument
82 (void)padding_idx; in embedding_out()
|