Merge pull request #618 from chenghuaWang/wch-main

feat(core): Introduce kBool data type for Qnn ElewiseEqual Op

Author: chenghuaWang

examples/qwen3_qnn_aot/aot_run.cpp

@@ -29,7 +29,7 @@ MLLM_MAIN({
 
   RunnerConfig config;
   config.num_layers = qwen3_cfg.num_hidden_layers;
-  config.num_heads = qwen3_cfg.num_attention_heads;
+  config.num_heads = qwen3_cfg.num_key_value_heads;
   config.head_dim = qwen3_cfg.head_dim;
   config.vocab_size = qwen3_cfg.vocab_size;
   config.context_len = 1024;

examples/qwen3_qnn_aot/modeling_qwen_qnn_aot_sha.hpp

@@ -764,8 +764,7 @@ inline void prepareParametersForSHA(const ParameterFile::ptr_t& params, const Qw
           std::string h_str = std::to_string(h);
           // Weight: slice per head
           std::string new_weight_name = layer_prefix + "q_norm." + h_str + ".weight";  // NOLINT
-          auto sliced = orig_weight.slice({{h * head_dim, (h + 1) * head_dim}}, false);
-          params->push(new_weight_name, sliced.contiguous().setMemType(kParamsNormal).setName(new_weight_name));
+          params->push(new_weight_name, orig_weight.contiguous().setMemType(kParamsNormal).setName(new_weight_name));
 
           // Scale: copy (scalar)
           if (has_scale) {
@@ -801,8 +800,7 @@ inline void prepareParametersForSHA(const ParameterFile::ptr_t& params, const Qw
           std::string h_str = std::to_string(h);
           // Weight: slice per head
           std::string new_weight_name = layer_prefix + "k_norm." + h_str + ".weight";  // NOLINT
-          auto sliced = orig_weight.slice({{h * head_dim, (h + 1) * head_dim}}, false);
-          params->push(new_weight_name, sliced.contiguous().setMemType(kParamsNormal).setName(new_weight_name));
+          params->push(new_weight_name, orig_weight.contiguous().setMemType(kParamsNormal).setName(new_weight_name));
 
           // Scale: copy (scalar)
           if (has_scale) {