| backward(const Tensor &grad_output) override | pulsatrix::MultiHeadAttentionModule | virtual |
| clear_key_padding_mask() | pulsatrix::MultiHeadAttentionModule | |
| compute_device() const override | pulsatrix::MultiHeadAttentionModule | inlinevirtual |
| config() const | pulsatrix::MultiHeadAttentionModule | inline |
| d_model() const | pulsatrix::MultiHeadAttentionModule | inline |
| forward(const Tensor &input) | pulsatrix::Module | inline |
| forward_cached(const Tensor &input, KVCache &cache) | pulsatrix::MultiHeadAttentionModule | |
| forward_impl(const Tensor &input) override | pulsatrix::MultiHeadAttentionModule | protectedvirtual |
| forward_traced(const Tensor &input, NodeId input_node, ComputationGraph &graph, Autograd &autograd) | pulsatrix::Module | inline |
| has_key_padding_mask() const | pulsatrix::MultiHeadAttentionModule | inline |
| head_dim() const | pulsatrix::MultiHeadAttentionModule | inline |
| is_causal() const | pulsatrix::MultiHeadAttentionModule | inline |
| is_training() const | pulsatrix::Module | inline |
| k_norm() | pulsatrix::MultiHeadAttentionModule | inline |
| k_proj() | pulsatrix::MultiHeadAttentionModule | inline |
| key_relevance() const | pulsatrix::MultiHeadAttentionModule | |
| last_attention_weights() const | pulsatrix::MultiHeadAttentionModule | inline |
| MakeKVCache(int64_t batch, int64_t max_length) const | pulsatrix::MultiHeadAttentionModule | |
| MultiHeadAttentionModule(int64_t d_model, int64_t num_heads, DeviceBackend *backend, bool use_rope=true, bool use_qk_norm=false) | pulsatrix::MultiHeadAttentionModule | |
| MultiHeadAttentionModule(const AttentionConfig &config, DeviceBackend *backend) | pulsatrix::MultiHeadAttentionModule | |
| named_buffers() override | pulsatrix::MultiHeadAttentionModule | virtual |
| named_children() override | pulsatrix::MultiHeadAttentionModule | virtual |
| named_parameters() override | pulsatrix::MultiHeadAttentionModule | virtual |
| num_heads() const | pulsatrix::MultiHeadAttentionModule | inline |
| num_kv_heads() const | pulsatrix::MultiHeadAttentionModule | inline |
| op_type() const override | pulsatrix::MultiHeadAttentionModule | inlinevirtual |
| out_proj() | pulsatrix::MultiHeadAttentionModule | inline |
| parameters() | pulsatrix::Module | inlinevirtual |
| position_offset() const | pulsatrix::MultiHeadAttentionModule | inline |
| propagate_relevance(const Tensor &relevance_out, const LRPRuleConfig &config) override | pulsatrix::MultiHeadAttentionModule | virtual |
| q_norm() | pulsatrix::MultiHeadAttentionModule | inline |
| q_proj() | pulsatrix::MultiHeadAttentionModule | inline |
| release_activations() override | pulsatrix::MultiHeadAttentionModule | virtual |
| set_key_padding_mask(const Tensor &key_keep) | pulsatrix::MultiHeadAttentionModule | |
| set_position_offset(int64_t offset) | pulsatrix::MultiHeadAttentionModule | |
| set_requires_grad(bool requires_grad, const std::string &prefix="") | pulsatrix::Module | inline |
| set_training(bool training) override | pulsatrix::MultiHeadAttentionModule | virtual |
| supports_lrp_rule(LRPRule rule) const | pulsatrix::Module | inlinevirtual |
| uses_qk_norm() const | pulsatrix::MultiHeadAttentionModule | inline |
| uses_rope() const | pulsatrix::MultiHeadAttentionModule | inline |
| v_proj() | pulsatrix::MultiHeadAttentionModule | inline |
| value_relevance() const | pulsatrix::MultiHeadAttentionModule | |
| ~Module()=default | pulsatrix::Module | virtual |