reinsert cqs

Author: Nexesenex

include/llama.h

@@ -370,6 +370,14 @@ extern "C" {
         enum llama_ftype ftype;              // quantize to this llama_ftype
         enum ggml_type output_tensor_type;   // output tensor type
         enum ggml_type token_embedding_type; // token embeddings tensor type
+        enum ggml_type attn_q_type;          // attention query tensor type
+        enum ggml_type attn_k_type;          // attention key tensor type
+        enum ggml_type attn_v_type;          // attention value tensor type
+        enum ggml_type attn_qkv_type;        // attention query-key-value tensor type
+        enum ggml_type attn_output_type;     // attention output tensor type
+        enum ggml_type ffn_gate_type;        // feedforward network gate type
+        enum ggml_type ffn_down_type;        // feedforward network down type
+        enum ggml_type ffn_up_type;          // feedforward network up type
         bool allow_requantize;               // allow quantizing non-f32/f16 tensors
         bool quantize_output_tensor;         // quantize output.weight
         bool only_copy;                      // only copy tensors - ftype, allow_requantize and quantize_output_tensor are ignored

src/llama.cpp

@@ -19981,6 +19981,14 @@ struct llama_model_quantize_params llama_model_quantize_default_params() {
         /*.ftype                       =*/ LLAMA_FTYPE_MOSTLY_Q5_1,
         /*.output_tensor_type          =*/ GGML_TYPE_COUNT,
         /*.token_embedding_type        =*/ GGML_TYPE_COUNT,
+        /*.attn_q_type                 =*/ GGML_TYPE_COUNT,
+        /*.attn_k_type                 =*/ GGML_TYPE_COUNT,
+        /*.attn_v_type                 =*/ GGML_TYPE_COUNT,
+        /*.attn_qkv_type               =*/ GGML_TYPE_COUNT,
+        /*.attn_output_type            =*/ GGML_TYPE_COUNT,
+        /*.ffn_gate_type               =*/ GGML_TYPE_COUNT,
+        /*.ffn_down_type               =*/ GGML_TYPE_COUNT,
+        /*.ffn_up_type                 =*/ GGML_TYPE_COUNT,
         /*.allow_requantize            =*/ false,
         /*.quantize_output_tensor      =*/ true,
         /*.only_copy                   =*/ false,