From e61c6980dcea3ae67fd2e3cf81bdee9a09a80e63 Mon Sep 17 00:00:00 2001
From: City <125218114+city96@users.noreply.github.com>
Date: Wed, 27 Aug 2025 12:13:38 +0200
Subject: [PATCH] Shape fix

---
 tools/lcpp.patch | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/tools/lcpp.patch b/tools/lcpp.patch
index 56e2f8d..3e449ac 100644
--- a/tools/lcpp.patch
+++ b/tools/lcpp.patch
@@ -39,7 +39,7 @@ index b16c462f..6d1568f1 100644
      const int idx = gguf_find_tensor(ctx, name);
      if (idx < 0) {
 diff --git a/src/llama.cpp b/src/llama.cpp
-index 24e1f1f0..569778fd 100644
+index 24e1f1f0..8a1e9ef8 100644
 --- a/src/llama.cpp
 +++ b/src/llama.cpp
 @@ -205,6 +205,18 @@ enum llm_arch {
@@ -296,7 +296,7 @@ index 24e1f1f0..569778fd 100644
      }
  
      size_t total_size_org = 0;
-@@ -18547,6 +18757,51 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
+@@ -18547,6 +18757,57 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
              ctx_outs[i_split] = gguf_init_empty();
          }
          gguf_add_tensor(ctx_outs[i_split], tensor);
@@ -344,11 +344,17 @@ index 24e1f1f0..569778fd 100644
 +                gguf_set_tensor_ndim(ctx_outs[i_split], tensor->name, n_dim);
 +                LLAMA_LOG_INFO("\n%s: Correcting shape for Wan FLF2V: [key:%s]\n", __func__, tensor->name);
 +            }
++            // S2V model only
++            if (name == "casual_audio_encoder.weights" || name == "casual_audio_encoder.encoder.padding_tokens") {
++                const int n_dim = 4;
++                gguf_set_tensor_ndim(ctx_outs[i_split], tensor->name, n_dim);
++                LLAMA_LOG_INFO("\n%s: Correcting shape for Wan S2V: [key:%s]\n", __func__, tensor->name);
++            }
 +        }
      }
  
      // Set split info if needed
-@@ -18647,6 +18902,124 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
+@@ -18647,6 +18908,124 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
          // do not quantize relative position bias (T5)
          quantize &= name.find("attn_rel_b.weight") == std::string::npos;
  
@@ -473,7 +479,7 @@ index 24e1f1f0..569778fd 100644
          enum ggml_type new_type;
          void * new_data;
          size_t new_size;
-@@ -18655,6 +19028,9 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
+@@ -18655,6 +19034,9 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
              new_type = default_type;
  
              // get more optimal quantization type based on the tensor shape, layer, etc.
@@ -483,7 +489,7 @@ index 24e1f1f0..569778fd 100644
              if (!params->pure && ggml_is_quantized(default_type)) {
                  new_type = llama_tensor_get_type(qs, new_type, tensor, ftype);
              }
-@@ -18664,6 +19040,7 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
+@@ -18664,6 +19046,7 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
              if (params->output_tensor_type < GGML_TYPE_COUNT && strcmp(tensor->name, "output.weight") == 0) {
                  new_type = params->output_tensor_type;
              }