fix: Fix logic for initializing inputs and attn layers for hybrid caches

Branch: GraniteFour

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

Author: gabe-l-hart

src/llama-graph.cpp

@@ -413,13 +413,6 @@ void llm_graph_input_attn_cross::set_input(const llama_ubatch * ubatch) {
     }
 }
 
-llm_graph_input_attn_kv_hybrid_recurrent::llm_graph_input_attn_kv_hybrid_recurrent(
-        const llama_hparams & hparams,
-        const llama_cparams & cparams,
-        const llama_kv_cache_hybrid_recurrent_state * kv_state) :
-    llm_graph_input_attn_kv_unified(hparams, cparams, kv_state->get_state_attn()) {
-}
-
 //
 // llm_graph_context
 //
@@ -1280,7 +1273,9 @@ ggml_tensor * llm_graph_context::build_attn(
     ggml_build_forward_expand(gf, k_cur);
     ggml_build_forward_expand(gf, v_cur);
 
-    const auto * kv_state = static_cast<const llama_kv_cache_unified_state *>(mstate);
+    // NOTE: For hybrid caches, this may be a child of mstate, so we use the one
+    //  encapsulated in inp
+    const auto * kv_state = inp->kv_state;
 
     // store to KV cache
     {
@@ -1312,10 +1307,10 @@ ggml_tensor * llm_graph_context::build_attn(
     return cur;
 }
 
-llm_graph_input_attn_kv_hybrid_recurrent * llm_graph_context::build_attn_inp_kv_hybrid_recurrent() const {
+llm_graph_input_attn_kv_unified * llm_graph_context::build_attn_inp_kv_hybrid_recurrent() const {
     const auto * kv_state = static_cast<const llama_kv_cache_hybrid_recurrent_state *>(mstate);
 
-    auto inp = std::make_unique<llm_graph_input_attn_kv_hybrid_recurrent>(hparams, cparams, kv_state);
+    auto inp = std::make_unique<llm_graph_input_attn_kv_unified>(hparams, cparams, kv_state->get_state_attn());
 
     {
         GGML_ASSERT(hparams.swa_type == LLAMA_SWA_TYPE_NONE && "Hybrid recurrent is not supported with SWA attention layers");
@@ -1329,25 +1324,7 @@ llm_graph_input_attn_kv_hybrid_recurrent * llm_graph_context::build_attn_inp_kv_
         inp->self_kq_mask_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->self_kq_mask, GGML_TYPE_F16) : inp->self_kq_mask;
     }
 
-    return (llm_graph_input_attn_kv_hybrid_recurrent *) res->add_input(std::move(inp));
-}
-
-ggml_tensor * llm_graph_context::build_attn(
-        llm_graph_input_attn_kv_hybrid_recurrent * inp,
-        ggml_cgraph * gf,
-        ggml_tensor * wo,
-        ggml_tensor * wo_b,
-        ggml_tensor * q_cur,
-        ggml_tensor * k_cur,
-        ggml_tensor * v_cur,
-        ggml_tensor * kq_b,
-        ggml_tensor * v_mla,
-            float     kq_scale,
-            int       il) const {
-    return build_attn(
-        static_cast<llm_graph_input_attn_kv_unified *>(inp),
-        gf, wo, wo_b, q_cur, k_cur, v_cur, kq_b, v_mla, kq_scale, il
-    );
+    return (llm_graph_input_attn_kv_unified *) res->add_input(std::move(inp));
 }
 
 llm_graph_input_attn_kv_unified_iswa * llm_graph_context::build_attn_inp_kv_unified_iswa() const {
@@ -1490,13 +1467,17 @@ ggml_tensor * llm_graph_context::build_attn(
 }
 
 ggml_tensor * llm_graph_context::build_copy_mask_state(
-         ggml_cgraph * gf,
-         ggml_tensor * s,
-         ggml_tensor * state_copy,
-         ggml_tensor * state_mask,
-             int32_t   n_state,
-             int32_t   n_seqs) const {
-    const auto * kv_state = static_cast<const llama_kv_cache_recurrent_state *>(mstate);
+                             ggml_cgraph * gf,
+                             ggml_tensor * s,
+                             ggml_tensor * state_copy,
+                             ggml_tensor * state_mask,
+                                 int32_t   n_state,
+                                 int32_t   n_seqs,
+    const llama_kv_cache_recurrent_state * kv_state) const {
+
+    if (kv_state == nullptr) {
+        kv_state = static_cast<const llama_kv_cache_recurrent_state *>(mstate);
+    }
 
     const auto n_kv    = kv_state->get_n_kv();
     const auto kv_head = kv_state->get_head();

src/llama-graph.h

@@ -297,16 +297,6 @@ class llm_graph_input_attn_kv_unified_iswa : public llm_graph_input_i {
     const llama_kv_cache_unified_iswa_state * kv_state;
 };
 
-class llm_graph_input_attn_kv_hybrid_recurrent : public llm_graph_input_attn_kv_unified {
-public:
-    llm_graph_input_attn_kv_hybrid_recurrent(
-            const llama_hparams & hparams,
-            const llama_cparams & cparams,
-            const llama_kv_cache_hybrid_recurrent_state * kv_state);
-
-    virtual ~llm_graph_input_attn_kv_hybrid_recurrent() = default;
-};
-
 class llm_graph_input_attn_cross : public llm_graph_input_i {
 public:
     llm_graph_input_attn_cross(const llama_cross * cross) : cross(cross) {}
@@ -597,20 +587,7 @@ struct llm_graph_context {
                   float   kq_scale,
                     int   il) const;
 
-    llm_graph_input_attn_kv_hybrid_recurrent * build_attn_inp_kv_hybrid_recurrent() const;
-
-    ggml_tensor * build_attn(
-            llm_graph_input_attn_kv_hybrid_recurrent * inp,
-            ggml_cgraph * gf,
-            ggml_tensor * wo,
-            ggml_tensor * wo_b,
-            ggml_tensor * q_cur, // [n_embd_head_q, n_head_q, n_tokens]
-            ggml_tensor * k_cur, // [n_embd_head_k, n_head_k, n_tokens]
-            ggml_tensor * v_cur, // [n_embd_head_v, n_head_v, n_tokens]
-            ggml_tensor * kq_b,
-            ggml_tensor * v_mla, // [n_embd_head_v_mla, n_embd_head_v, n_head_v]
-                  float   kq_scale,
-                    int   il) const;
+    llm_graph_input_attn_kv_unified * build_attn_inp_kv_hybrid_recurrent() const;
 
     llm_graph_input_attn_cross * build_attn_inp_cross() const;
 
@@ -632,12 +609,13 @@ struct llm_graph_context {
     //
 
     ggml_tensor * build_copy_mask_state(
-             ggml_cgraph * gf,
-             ggml_tensor * s,
-             ggml_tensor * state_copy,
-             ggml_tensor * state_mask,
-                 int32_t   n_state,
-                 int32_t   n_seqs) const;
+                                 ggml_cgraph * gf,
+                                 ggml_tensor * s,
+                                 ggml_tensor * state_copy,
+                                 ggml_tensor * state_mask,
+                                     int32_t   n_state,
+                                     int32_t   n_seqs,
+        const llama_kv_cache_recurrent_state * kv_state = nullptr) const;
 
     ggml_tensor * build_rwkv_token_shift_load(
              ggml_cgraph * gf,