MCPcopy Create free account
hub / github.com/ROCm/AMDMIGraphX / quantize_int8

Function quantize_int8

src/quantization.cpp:176–187  ·  view source on GitHub ↗

Source from the content-addressed store, hash-verified

174}
175
176void quantize_int8(program& prog,
177 const target& t,
178 const std::vector<parameter_map>& calibration,
179 const std::unordered_set<std::string>& ins_names)
180{
181 std::unordered_set<std::string> op_names = {"convolution", "dot"};
182 if(op_names != ins_names)
183 {
184 MIGRAPHX_THROW("QUANTIZE_INT8: only support DOT and CONVOLUTION operation");
185 }
186 quantize_8bits(prog, t, shape::int8_type, calibration, ins_names);
187}
188
189void quantize_int4_weights(program& prog)
190{

Callers 5

quantize_int8_wrapFunction · 0.50
TEST_CASEFunction · 0.50
TEST_CASEFunction · 0.50
TEST_CASEFunction · 0.50
mainFunction · 0.50

Calls 1

quantize_8bitsFunction · 0.85

Tested by 1

TEST_CASEFunction · 0.40