From 8892b6c60ba470770581f53181a2e01c302fa1ce Mon Sep 17 00:00:00 2001
From: Xuan Son Nguyen <son@huggingface.co>
Date: Wed, 29 Jul 2026 00:36:36 +0200
Subject: [PATCH] main model load ok

---
 conversion/qwen3tts.py | 2 +-
 src/llama-arch.cpp     | 1 +
 src/llama-arch.h       | 1 +
 src/llama-model.cpp    | 5 +++++
 4 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/conversion/qwen3tts.py b/conversion/qwen3tts.py
index 9e3ee84e7f..0aa2388f3a 100644
--- a/conversion/qwen3tts.py
+++ b/conversion/qwen3tts.py
@@ -70,7 +70,7 @@ class Qwen3TTSTalkerModel(TextModel):
 
         name = name[len("talker."):]
         if name == "codec_head.weight":
-            name = "lm_head.weight"
+            return None
 
         return super().filter_tensors((name, gen))
 
diff --git a/src/llama-arch.cpp b/src/llama-arch.cpp
index a0945e501d..650f5b5887 100644
--- a/src/llama-arch.cpp
+++ b/src/llama-arch.cpp
@@ -144,6 +144,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_TALKIE,           "talkie"           },
     { LLM_ARCH_MELLUM,           "mellum"           },
     { LLM_ARCH_NANBEIGE,         "nanbeige"         },
+    { LLM_ARCH_QWEN3TTS,         "qwen3tts"         },
     { LLM_ARCH_UNKNOWN,          "(unknown)"        },
 };
 
diff --git a/src/llama-arch.h b/src/llama-arch.h
index fb5eb3920b..0c6bde3e6f 100644
--- a/src/llama-arch.h
+++ b/src/llama-arch.h
@@ -149,6 +149,7 @@ enum llm_arch {
     LLM_ARCH_MINIMAX_M3,
     LLM_ARCH_DFLASH,
     LLM_ARCH_NANBEIGE,
+    LLM_ARCH_QWEN3TTS,
     LLM_ARCH_UNKNOWN,
 };
 
diff --git a/src/llama-model.cpp b/src/llama-model.cpp
index be0a0df55d..8a138e267f 100644
--- a/src/llama-model.cpp
+++ b/src/llama-model.cpp
@@ -111,6 +111,10 @@ static llama_model * llama_model_mapping(llm_arch arch, const llama_model_params
             return new llama_model_qwen3vl(params);
         case LLM_ARCH_QWEN3VLMOE:
             return new llama_model_qwen3vlmoe(params);
+        case LLM_ARCH_QWEN3TTS:
+            // Qwen3-TTS talker backbone: identical tensor layout and interleaved
+            // mrope to qwen3vl, just without vision/deepstack tensors (n_deepstack_layers is 0)
+            return new llama_model_qwen3vl(params);
         case LLM_ARCH_PHI2:
             return new llama_model_phi2(params);
         case LLM_ARCH_PHI3:
@@ -2576,6 +2580,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
         case LLM_ARCH_QWEN3VLMOE:
         case LLM_ARCH_QWEN35:
         case LLM_ARCH_QWEN35MOE:
+        case LLM_ARCH_QWEN3TTS:
             return LLAMA_ROPE_TYPE_IMROPE;
 
         case LLM_ARCH_GLM4: