Home
last modified time | relevance | path

Searched refs:group_size (Results 1 – 25 of 260) sorted by relevance

1234567891011

/aosp_15_r20/external/tensorflow/tensorflow/python/kernel_tests/
H A Dcollective_ops_test.py47 def all_reduce(t, group_size, group_key, instance_key, *args, **kwargs): argument
49 return _collective_ops.all_reduce(t, group_size, group_key, instance_key,
53 def all_gather(t, group_size, group_key, instance_key, *args, **kwargs): argument
55 return _collective_ops.all_gather(t, group_size, group_key, instance_key,
65 def all_reduce(t, group_size, group_key, instance_key, *args, **kwargs): argument
66 group_size = array_ops.identity(group_size)
69 return _collective_ops.all_reduce_v2(t, group_size, group_key, instance_key,
73 def all_gather(t, group_size, group_key, instance_key, *args, **kwargs): argument
74 group_size = array_ops.identity(group_size)
77 return _collective_ops.all_gather_v2(t, group_size, group_key, instance_key,
[all …]
H A Dcollective_ops_multi_worker_test.py111 group_size=2,
174 group_size = 2
184 group_size,
202 group_size,
212 group_size,
224 group_size,
235 group_size = 2
253 collective_ops.all_reduce(in_tensor, group_size, group_key,
259 collective_ops.all_reduce(in_tensor, group_size, group_key,
267 collective_ops.all_reduce(in_tensor, group_size, group_key, instance_key)
[all …]
/aosp_15_r20/external/tensorflow/tensorflow/python/ops/
H A Dcollective_ops.py20 group_size, argument
56 if group_size < 1:
61 group_size=group_size,
88 group_size, group_key = gen_collective_ops.collective_assign_group_v2(
92 return group_size, group_key
96 group_size, argument
147 group_size=group_size,
160 group_size, argument
186 if group_size < 1:
192 group_size=group_size,
[all …]
H A Dcollective_ops_test.py57 group_size = len(inputs)
59 reported_group_size = group_size
61 config = config_pb2.ConfigProto(device_count={device_type: group_size})
62 devices = ['/{}:{}'.format(device_type, i) for i in range(group_size)]
66 for i in range(group_size):
85 for i in range(group_size):
91 group_size = 2
94 config = config_pb2.ConfigProto(device_count={'CPU': group_size})
97 for cpu in range(group_size):
102 in_tensor, group_size, group_key, instance, 'Add', 'Div'))
[all …]
/aosp_15_r20/external/executorch/examples/models/llama/source_transformation/
H A Dquantize.py44 group_size: Optional[int] = 128,
87 groupsize=group_size,
97 if group_size is None:
102 precision=torch_dtype, groupsize=group_size
111 group_size,
155 group_size,
160 q_group_size = 256 if group_size is None else group_size
181 group_size: Optional[int] = None,
213 if group_size is None or group_size == 0:
215 elif ((x_shape_1 % group_size) == 0) or not enable_non_multiple_groups:
[all …]
H A Dpre_quantization.py25 group_size: int,
33 assert _check_linear_int4_k(child.in_features, group_size)
49 groupsize=group_size,
63 group_size: int,
72 if group_size not in [32, 64, 128, 256]:
79 group_size,
142 group_size: Optional[int] = None,
158 group_size=group_size,
172 group_size: Optional[int] = None,
178 if group_size is not None and group_size not in [0, 32, 64, 128, 256]:
[all …]
/aosp_15_r20/external/pytorch/torch/distributed/
H A D_functional_collectives.py203 group_size = c10d._get_group_size_by_name(group_name)
205 self, group_size, group_name
214 res = torch.cat(torch.chunk(res, group_size, dim=0), dim=gather_dim)
235 group_size = c10d._get_group_size_by_name(group_name)
238 self, group_size, group_name
247 res = torch.cat(torch.chunk(res, group_size, dim=0), dim=gather_dim)
274 group_size = c10d._get_group_size_by_name(group_name)
277 self.size(scatter_dim) % group_size == 0
280 tensor_list = torch.chunk(self, group_size, dim=scatter_dim)
286 group_size,
[all …]
H A D_functional_collectives_impl.py14 def _broadcast(input, src, tag, ranks, group_size): argument
23 def _all_reduce(input, reduce_op, tag, ranks, group_size): argument
32 def _all_reduce_coalesced(inputs, reduce_op, tag, ranks, group_size): argument
41 def _all_gather_into_tensor(input, tag, ranks, group_size): argument
45 group_size,
50 def _all_gather_into_tensor_coalesced(input, tag, ranks, group_size): argument
54 group_size,
64 group_size: int,
70 group_size,
80 group_size: int,
[all …]
/aosp_15_r20/external/trusty/arm-trusted-firmware/plat/xilinx/zynqmp/pm_service/
Dpm_api_pinctrl.c26 uint8_t group_size; member
42 .group_size = PINCTRL_GRP_CAN0_18 - PINCTRL_GRP_CAN0_0 + 1U,
48 .group_size = PINCTRL_GRP_CAN1_19 - PINCTRL_GRP_CAN1_0 + 1U,
54 .group_size = PINCTRL_GRP_ETHERNET0_0 - PINCTRL_GRP_ETHERNET0_0 + 1U,
60 .group_size = PINCTRL_GRP_ETHERNET1_0 - PINCTRL_GRP_ETHERNET1_0 + 1U,
66 .group_size = PINCTRL_GRP_ETHERNET2_0 - PINCTRL_GRP_ETHERNET2_0 + 1U,
72 .group_size = PINCTRL_GRP_ETHERNET3_0 - PINCTRL_GRP_ETHERNET3_0 + 1U,
78 .group_size = PINCTRL_GRP_GEMTSU0_2 - PINCTRL_GRP_GEMTSU0_0 + 1U,
84 .group_size = PINCTRL_GRP_GPIO0_77 - PINCTRL_GRP_GPIO0_0 + 1U,
90 .group_size = PINCTRL_GRP_I2C0_18 - PINCTRL_GRP_I2C0_0 + 1U,
[all …]
/aosp_15_r20/external/pytorch/torch/csrc/distributed/c10d/
H A DFunctional.cpp160 int64_t group_size) { in allocate_all_gather_output() argument
162 output_size[0] *= group_size; in allocate_all_gather_output()
170 int64_t group_size, in all_gather_into_tensor_coalesced() argument
176 outputs.push_back(allocate_all_gather_output(tensor, group_size)); in all_gather_into_tensor_coalesced()
189 int64_t group_size, in all_gather_into_tensor() argument
193 inputs, group_size, std::move(group_name))[0]; in all_gather_into_tensor()
198 int64_t group_size, in all_gather_into_tensor_out() argument
211 const int64_t group_size) { in allocate_reduce_scatter_output() argument
213 if (output_size[0] % group_size != 0) { in allocate_reduce_scatter_output()
216 << group_size << ")."; in allocate_reduce_scatter_output()
[all …]
/aosp_15_r20/external/pytorch/test/distributed/
H A Dtest_functional_api.py38 def new_subgroups(group_size: int, pg_tag=None):
43 for subgroup_id in range(world_size // group_size):
44 start_rank = subgroup_id * group_size
45 end_rank = start_rank + group_size
70 tag, rankset, group_size = ft_c._expand_group([0, 1, 2, 3])
73 self.assertEqual(4, group_size)
75 tag, rankset, group_size = ft_c._expand_group([0, 1, 2, 3], "bla")
79 tag, rankset, group_size = ft_c._expand_group([[0, 1], [2, 3]])
82 self.assertEqual(2, group_size)
84 tag, rankset, group_size = ft_c._expand_group([[0, 1], [2, 3]], "blu")
[all …]
H A Dtest_inductor_collectives.py68 def example(tensor, src, *, tag, ranks, group_size): argument
70 tensor, src, tag, ranks, group_size
100 def matmul_cat_col(a, b, c, d, e, f, *, tag, ranks, group_size): argument
104 ar = torch.ops.c10d_functional.all_reduce(z, "sum", tag, ranks, group_size)
175 def eager_func(a, b, c, d, *, tag, ranks, group_size): argument
179 ar = torch.ops.c10d_functional.all_reduce(z, "sum", tag, ranks, group_size)
214 def inductor_func(a, b, c, d, *, tag, ranks, group_size): argument
218 ar = torch.ops.c10d_functional.all_reduce(z, "sum", tag, ranks, group_size)
250 def func(a, *, tag, ranks, group_size): argument
267 def func(tensor, src_dst_pairs, *, tag, ranks, group_size): argument
[all …]
/aosp_15_r20/external/pytorch/test/distributed/pipelining/
H A Dtest_schedule.py45 self.group_size = kwargs.get("group_size", 1)
127 for num_local_stages, num_microbatches, group_size in self.test_cases:
131 group_size=group_size,
133 if num_microbatches % group_size != 0:
140 group_size,
142 num_stages = num_local_stages * group_size
144 MockPipelineStage(group_size=group_size, num_stages=num_stages)
162 for num_local_stages, num_microbatches, group_size in self.test_cases:
166 group_size=group_size,
169 num_microbatches // max(1, num_microbatches // group_size)
[all …]
/aosp_15_r20/external/tensorflow/tensorflow/core/kernels/
H A Dcollective_ops.cc114 if (col_params_->group.group_size > col_params_->group.members.size()) { in CanProceedWithCompute()
155 OP_REQUIRES_OK(c, c->GetAttr("group_size", &col_params_->group.group_size)); in CollectiveGatherOpKernel()
157 c, col_params_->group.group_size > 0, in CollectiveGatherOpKernel()
159 col_params_->group.group_size)); in CollectiveGatherOpKernel()
184 0, output_shape.dim_size(0) * col_params_->group.group_size); in ComputeAsyncImpl()
230 OP_REQUIRES_OK(c, c->GetAttr("group_size", &col_params_->group.group_size)); in CollectiveReduceOpKernel()
232 c, col_params_->group.group_size > 0, in CollectiveReduceOpKernel()
234 col_params_->group.group_size)); in CollectiveReduceOpKernel()
340 OP_REQUIRES_OK(c, c->GetAttr("group_size", &col_params_->group.group_size)); in CollectiveBcastSendOpKernel()
342 c, col_params_->group.group_size > 0, in CollectiveBcastSendOpKernel()
[all …]
H A Dcollective_nccl_reducer.cc27 Tensor group_size; in Run() local
40 Tensor(static_cast<Eigen::half>(col_params_->group.group_size)); in Run()
44 Tensor(static_cast<float>(col_params_->group.group_size)); in Run()
48 Tensor(static_cast<double>(col_params_->group.group_size)); in Run()
52 Tensor(static_cast<int32>(col_params_->group.group_size)); in Run()
56 Tensor(static_cast<int64_t>(col_params_->group.group_size)); in Run()
63 group_size = Tensor( in Run()
70 &group_size_val, col_ctx_->device, &group_size, in Run()
119 col_params_->final_op, col_ctx_->output, &group_size); in Run()
/aosp_15_r20/external/tensorflow/tensorflow/python/distribute/
H A Dcross_device_ops_test.py176 group_size = num_processes * len(devices)
179 group_size=group_size,
346 group_size = options.num_processes * (options.gpus_per_process or 1)
349 inputs = inputs_data[0:group_size]
351 if group_size == 1:
353 if group_size == 2:
355 elif group_size == 4:
394 group_size = options.num_processes * (options.gpus_per_process or 1)
406 inputs = inputs_data[0:group_size]
408 if group_size == 1:
[all …]
/aosp_15_r20/external/tensorflow/tensorflow/compiler/mlir/xla/tests/
H A Dlegalize-tf-collective.mlir6 // CHECK-SAME{LITERAL}: tf2xla.collective_info.group_size = 2
22 // CHECK-SAME{LITERAL}: tf2xla.collective_info.group_size = 2
97 // CHECK-SAME{LITERAL}: tf2xla.collective_info.group_size = 2
101 %group_size = "tf.Const"() { value = dense<2> : tensor<i32> } : () -> tensor<i32>
108 …%0 = "tf.CollectiveReduceV2"(%input, %group_size, %group_key, %instance_key) {merge_op = "Add", fi…
114 …%1 = "tf.CollectiveReduceV2"(%input, %group_size, %group_key, %instance_key) {merge_op = "Add", fi…
124 %group_size = "tf.Const"() { value = dense<2> : tensor<i32> } : () -> tensor<i32>
131 …%0 = "tf.CollectiveReduceV2"(%input, %group_size, %group_key, %instance_key) {merge_op = "Add", fi…
138 %group_size = "tf.Const"() { value = dense<2> : tensor<i32> } : () -> tensor<i32>
145 …%0 = "tf.CollectiveReduceV2"(%input, %group_size, %group_key, %instance_key) {merge_op = "Max", fi…
[all …]
/aosp_15_r20/external/tensorflow/tensorflow/compiler/xla/service/
H A Dreduce_scatter_utils.cc130 const MapIdToTableOffset& map_id, int64_t group_size, in IsPerIdOffset() argument
159 shard_size / *multiplier, map_id, group_size, ar); in IsPerIdOffset()
163 for (int64_t id = 0; id < group_size; ++id) { in IsPerIdOffset()
177 return IsPerIdOffset(offset->operand(0), shard_size, map_id, group_size, in IsPerIdOffset()
185 return IsPerIdOffset(offset->operand(0), shard_size, map_id, group_size, in IsPerIdOffset()
193 *upper_bound < (group_size - 1) * shard_size) { in IsPerIdOffset()
197 return IsPerIdOffset(offset->operand(1), shard_size, map_id, group_size, in IsPerIdOffset()
206 for (int64_t j = 0; j < group_size; ++j) { in IsPerIdOffset()
235 for (int64_t j = 0; j < group_size; ++j) { in IsPerIdOffset()
321 int64_t group_size; in MatchReduceScatter() local
[all …]
/aosp_15_r20/external/pytorch/torch/utils/data/datapipes/iter/
H A Dgrouping.py215 group_size: Optional[int] = None,
227 self.group_size = group_size
229 if group_size is not None and buffer_size is not None:
230 assert 0 < group_size <= buffer_size
231 self.guaranteed_group_size = group_size
233 assert group_size is not None and 0 < guaranteed_group_size <= group_size
274 if self.group_size is not None and self.group_size == len(
303 self.group_size,
320 self.group_size,
/aosp_15_r20/external/executorch/backends/apple/mps/test/
H A Dtest_mps_linear.py163 group_size=bl,
194 group_size=bl,
253 group_size: int = 0,
265 self.group_size = self.ic if group_size == 0 else group_size
267 if self.group_size != self.ic:
268 assert self.ic % self.group_size == 0
269 assert self.group_size % 8 == 0 # TODO make this 16
270 self.num_groups = self.ic // self.group_size
283 if self.group_size != self.ic or force_groupwise_quant:
376 self, w, n_bit=4, group_size=128, precision=torch.float32 argument
[all …]
/aosp_15_r20/external/tensorflow/tensorflow/core/common_runtime/
H A Dcollective_param_resolver_local.cc146 gr->group.group_size = group_params->group_size; in CompleteGroupLocal()
156 << " group_size=" << gr->group.group_size in CompleteGroupLocal()
203 } else if (group_params->group_size != gr->group.group_size) { in CompleteGroupLocal()
206 group_params->group_size, ", but that group has size ", in CompleteGroupLocal()
207 gr->group.group_size, " (group_key=", gr->group.group_key, ")"); in CompleteGroupLocal()
215 if (gr->group.members.size() == gr->group.group_size) { in CompleteGroupLocal()
228 group_params->user_specified_rank < gr->group.group_size)) { in CompleteGroupLocal()
243 << " group_size=" << gr->group.group_size << " (current" in CompleteGroupLocal()
246 << (gr->group.group_size - gr->group.members.size()); in CompleteGroupLocal()
263 VLOG(2) << "group_size " << gr->group.group_size << " set size " in CompleteGroupLocal()
[all …]
H A Dall_to_all.cc54 if (counter_ < 2 * col_params_->group.group_size) { in CheckCounterAndCallDone()
57 CHECK_LE(counter_, 2 * col_params_->group.group_size); // Crash ok. in CheckCounterAndCallDone()
80 if (col_ctx->input->dim_size(0) != col_ctx->col_params->group.group_size) { in InitializeCollectiveContext()
84 col_ctx->col_params->group.group_size, ")"); in InitializeCollectiveContext()
96 input_chunks_.reserve(col_params_->group.group_size); in Run()
97 output_chunks_.reserve(col_params_->group.group_size); in Run()
106 for (int i = 0; i < col_params_->group.group_size; ++i) { in Run()
113 for (int i = 0; i < col_params_->group.group_size; ++i) { in Run()
/aosp_15_r20/external/executorch/backends/vulkan/test/op_tests/
H A Dlinear_weight_int4_test.cpp87 const int group_size = 32, in test_reference_linear_int4() argument
89 assert(K % group_size == 0); in test_reference_linear_int4()
95 const int k_groups = K / group_size; in test_reference_linear_int4()
100 x, weights_4x2, group_size, scales_and_zeros, inner_k_tiles); in test_reference_linear_int4()
103 x, weights_4x2, group_size, scales_and_zeros, inner_k_tiles); in test_reference_linear_int4()
133 const int group_size = 32, in test_vulkan_linear_int4() argument
135 assert(K % group_size == 0); in test_vulkan_linear_int4()
141 const int k_groups = K / group_size; in test_vulkan_linear_int4()
146 x, weights_4x2, group_size, scales_and_zeros, inner_k_tiles); in test_vulkan_linear_int4()
177 graph.add_scalar<int64_t>(group_size), in test_vulkan_linear_int4()
/aosp_15_r20/external/executorch/exir/passes/
H A D_quant_patterns_and_replacements.py90 group_size = weight.size(1) // (
100 group_size,
137 group_size = weight.size(1) // (
147 group_size,
206 group_size = (4 * weight.size(1)) // (
224 group_size,
261 group_size = (4 * weight.size(1)) // (
279 group_size,
338 group_size = (2 * weight.size(1)) // (
354 group_size,
[all …]
/aosp_15_r20/external/executorch/backends/xnnpack/operators/
H A Dquant_params.py69 group_size: int = 0,
93 self.group_size = group_size
94 if self.group_size > 0:
102 assert group_size > 0, "Group size must be greater than 0"
103 self.is_per_channel_group = self.per_channel and self.group_size > 0
218 group_size = cast(int, quant_node_args[-2])
223 group_size = 0
243 group_size=group_size,

1234567891011