DeepSeek V4 (#24162)

* convert: add dsv4 conversion

* add basic setup

* add llm_graph_input_dsv4

* add save-load state

* add sinkhorn eps - correction by @fairydreaming

* add rope fix

* cleanup dead code

* fix bugs

* support pro model: added by @fairydreaming

* remove redundant V cache

* Chat template

* remove debugging leftovers

* Add mechanism for inlining templates based on architecture

* s/deepseek-v4-flash/deepseek4/g

* s/deepseek-v4-flash/deepseek4/g continued

* enable graph reuse

* enable FA

* fix test llama archs

* rename

* compatibility with antirez ds4 GGUFs

* simplified set_gguf_parameters() by calling super class method, replaced moe.score_func with expert_gating_func.

* reserve worst-case kv-cache

* revert max split inputs

* address review comments

* add padding to enable FA

* pad only the final value of plan.n_kv to 256

* remove built-in cpp chat template

* cont: remove cpp built-in template

* rm outdated test

* replace ggml_view_3d() with ggml_reshape_3d()

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* only support n_seq=1 for now

* remove unused var

* cont: remove unused var

* use scale bias

* use correct ptr for can_reuse

* remove gen-chat-inline-templates.py

* simplify graph reuse

* cont: cleanup

* remove unused inputs

* enable partial checkpointing

* add correct shape for kq_mask + set llama_model_n_swa to 0 for dsv4

* precompute source_idx + add comment about dummy write

* support multi-seq

* remove restored_trim_pos

* use split_equal when possible

* fix indent

* address review comments

* use LLM_KV

* fix ci

---------

Co-authored-by: Piotr Wilkin <piotr.wilkin@syndatis.com>
Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com>
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
Co-authored-by: fairydreaming <166155368+fairydreaming@users.noreply.github.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

src/llama-kv-cache.cpp

@@ -211,10 +211,12 @@ llama_kv_cache::llama_kv_cache(
             n_embd_head_k_all = -1;
         }
 
-        if (n_embd_head_v_all == 0) {
-            n_embd_head_v_all = (int32_t) hparams.n_embd_head_v(il);
-        } else if (n_embd_head_v_all > 0 && n_embd_head_v_all != (int32_t) hparams.n_embd_head_v(il)) {
-            n_embd_head_v_all = -1;
+        if (!is_mla) {
+            if (n_embd_head_v_all == 0) {
+                n_embd_head_v_all = (int32_t) hparams.n_embd_head_v(il);
+            } else if (n_embd_head_v_all > 0 && n_embd_head_v_all != (int32_t) hparams.n_embd_head_v(il)) {
+                n_embd_head_v_all = -1;
+            }
         }
 
         // [TAG_V_CACHE_VARIABLE]
@@ -336,8 +338,9 @@ llama_kv_cache::llama_kv_cache(
             ggml_is_quantized(type_k) &&
             hparams.n_embd_head_k() % 64 == 0;
 
-        // always create Hadamard rotation tensors for DeepSeek V3.2 DSA lightning indexer
-        if (model.arch == LLM_ARCH_DEEPSEEK32 && hparams.n_embd_head_k_full == hparams.indexer_head_size) {
+        // always create Hadamard rotation tensors for DeepSeek lightning indexers
+        if ((model.arch == LLM_ARCH_DEEPSEEK32 || model.arch == LLM_ARCH_DEEPSEEK4) &&
+                hparams.n_embd_head_k_full == hparams.indexer_head_size) {
             attn_rot_k = true;
         }
 
@@ -1220,6 +1223,23 @@ ggml_type llama_kv_cache::type_v() const {
     return layers[0].v->type;
 }
 
+std::vector<uint32_t> llama_kv_cache::get_layer_ids() const {
+    std::vector<uint32_t> res;
+    res.reserve(layers.size());
+
+    for (const auto & layer : layers) {
+        res.push_back(layer.il);
+    }
+
+    return res;
+}
+
+ggml_tensor * llama_kv_cache::get_k_storage(int32_t il) const {
+    const int32_t ikv = map_layer_ids.at(il);
+
+    return layers[ikv].k;
+}
+
 uint32_t llama_kv_cache::get_n_kv(const slot_info & sinfo) const {
     uint32_t result = 0;