Skip to content

LLM inference in C/C++

C++ ggml

mtmd: Fix Qwen3-tts-0.6b (#28231)

Pascal0f3a71be15af
diff --git a/conversion/qwen3tts.py b/conversion/qwen3tts.py
index 1f6b9a1b0e..2c35799f77 100644
--- a/conversion/qwen3tts.py
+++ b/conversion/qwen3tts.py
@@ -276,6 +276,10 @@ class Qwen3TTSSpeakerEncoderModel(MmprojModel):
# ConvTranspose1d kernels: only F16/F32 are implemented, no BF16
if new_name.endswith(".conv.weight") and (".up.blk." in new_name or ".dac.blk." in new_name):
return gguf.GGMLQuantizationType.F32
+ # the code predictor FFN intermediate peaks around 1.5e5, above the F16 range, and mul_mat
+ # casts its input to the weight type
+ if new_name.startswith("a.gen.code.blk.") and new_name.endswith(".ffn_down.weight"):
+ return gguf.GGMLQuantizationType.F32
return super().tensor_force_quant(name, new_name, bid, n_dims)
@classmethod
diff --git a/tools/mtmd/clip.cpp b/tools/mtmd/clip.cpp
index 90de195758..46f0437a78 100644
--- a/tools/mtmd/clip.cpp
+++ b/tools/mtmd/clip.cpp
@@ -2988,9 +2988,9 @@ struct clip_model_loader {
} break;
case PROJECTOR_TYPE_QWEN3TTS_GEN:
{
- // code_predictor
- model.gen_code_proj_in_w = get_tensor(string_format(TN_A_GEN_CODE_PROJ_IN, "weight"));
- model.gen_code_proj_in_b = get_tensor(string_format(TN_A_GEN_CODE_PROJ_IN, "bias"));
+ // code_predictor, proj_in is absent when the talker and the predictor share the hidden size
+ model.gen_code_proj_in_w = get_tensor(string_format(TN_A_GEN_CODE_PROJ_IN, "weight"), false);
+ model.gen_code_proj_in_b = get_tensor(string_format(TN_A_GEN_CODE_PROJ_IN, "bias"), false);
model.gen_code_embd_w = get_tensor(string_format(TN_A_GEN_CODE_EMBD, "weight"));
model.gen_code_head_w = get_tensor(string_format(TN_A_GEN_CODE_HEAD, "weight"));
model.gen_code_out_embd_w = get_tensor(string_format(TN_A_GEN_CODE_OUT_EMBD, "weight"));