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* rm outdated test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>

gguf-py/gguf/gguf_writer.py

@@ -715,6 +715,9 @@ class GGUFWriter:
     def add_full_attention_interval(self, interval: int) -> None:
         self.add_uint32(Keys.LLM.FULL_ATTENTION_INTERVAL.format(arch=self.arch), interval)
 
+    def add_hash_layer_count(self, count: int) -> None:
+        self.add_uint32(Keys.LLM.HASH_LAYER_COUNT.format(arch=self.arch), count)
+
     def add_feed_forward_length(self, length: int | Sequence[int]) -> None:
         if isinstance(length, int):
             self.add_uint32(Keys.LLM.FEED_FORWARD_LENGTH.format(arch=self.arch), length)
@@ -952,6 +955,27 @@ class GGUFWriter:
     def add_norm_before_residual(self, value: bool) -> None:
         self.add_bool(Keys.LLM.NORM_BEFORE_RESIDUAL.format(arch=self.arch), value)
 
+    def add_attention_output_group_count(self, count: int) -> None:
+        self.add_uint32(Keys.Attention.OUTPUT_GROUP_COUNT.format(arch=self.arch), count)
+
+    def add_attention_output_lora_rank(self, length: int) -> None:
+        self.add_uint32(Keys.Attention.OUTPUT_LORA_RANK.format(arch=self.arch), length)
+
+    def add_attention_compress_ratios(self, values: Sequence[int]) -> None:
+        self.add_array(Keys.Attention.COMPRESS_RATIOS.format(arch=self.arch), values)
+
+    def add_attention_compress_rope_freq_base(self, value: float) -> None:
+        self.add_float32(Keys.Attention.COMPRESS_ROPE_FREQ_BASE.format(arch=self.arch), value)
+
+    def add_hyper_connection_count(self, count: int) -> None:
+        self.add_uint32(Keys.HyperConnection.COUNT.format(arch=self.arch), count)
+
+    def add_hyper_connection_sinkhorn_iterations(self, count: int) -> None:
+        self.add_uint32(Keys.HyperConnection.SINKHORN_ITERATIONS.format(arch=self.arch), count)
+
+    def add_hyper_connection_epsilon(self, value: float) -> None:
+        self.add_float32(Keys.HyperConnection.EPSILON.format(arch=self.arch), value)
+
     def add_attention_scale(self, value: float) -> None:
         self.add_float32(Keys.Attention.SCALE.format(arch=self.arch), value)