wip wip wip

llama.cpp

@@ -3438,6 +3438,75 @@ static struct ggml_tensor * llm_build_kqv(
     return cur;
 }
 
+struct llm_build_context {
+    llm_build_context(
+        llama_context  & lctx,
+    const llama_batch  & batch,
+    const llm_build_cb & cb,
+                  bool   worst_case) :
+        model        (lctx.model),
+        hparams      (model.hparams),
+        cparams      (lctx.cparams),
+        kv_self      (lctx.kv_self),
+        n_embd       (hparams.n_embd),
+        n_layer      (hparams.n_layer),
+        n_ctx        (cparams.n_ctx),
+        n_head       (hparams.n_head),
+        n_head_kv    (hparams.n_head_kv),
+        n_embd_head  (hparams.n_embd_head()),
+        freq_base    (cparams.rope_freq_base),
+        freq_scale   (cparams.rope_freq_scale),
+        norm_rms_eps (hparams.f_norm_rms_eps),
+        n_tokens     (batch.n_tokens),
+        n_kv         (worst_case ? n_ctx            : kv_self.n),
+        kv_head      (worst_case ? n_ctx - n_tokens : kv_self.head),
+        do_rope_shift(worst_case || kv_self.has_shift),
+        cb           (cb),
+        buf_compute  (lctx.buf_compute) {
+            GGML_ASSERT(!!kv_self.ctx);
+    struct ggml_init_params params = {
+        /*.mem_size   =*/ buf_compute.size,
+        /*.mem_buffer =*/ buf_compute.data,
+        /*.no_alloc   =*/ true,
+    };
+
+    struct ggml_context * ctx0 = ggml_init(params);
+
+        }
+
+    const llama_model & model;
+    const llama_hparams & hparams;
+    const llama_cparams & cparams;
+
+    const llama_kv_cache & kv_self;
+
+    const int64_t n_embd;
+    const int64_t n_layer;
+    const int64_t n_ctx;
+    const int64_t n_head;
+    const int64_t n_head_kv;
+    const int64_t n_embd_head;
+
+    const float freq_base;
+    const float freq_scale;
+    const float norm_rms_eps;
+
+    const int32_t n_tokens;
+    const int32_t n_kv;
+    const int32_t kv_head;
+
+    const bool do_rope_shift;
+
+    const llm_build_cb & cb;
+
+    llama_buffer & buf_compute;
+
+    struct ggml_cgraph * llama() {
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
+    }
+
+};
+
 static struct ggml_cgraph * llm_build_llama(
         llama_context  & lctx,
     const llama_batch  & batch,