[Do Not Merge] model : LFM2.5-Audio-1.5B

common/arg.cpp

@@ -57,6 +57,7 @@ static std::initializer_list<enum llama_example> mmproj_examples = {
     LLAMA_EXAMPLE_MTMD,
     LLAMA_EXAMPLE_SERVER,
     LLAMA_EXAMPLE_CLI,
+    LLAMA_EXAMPLE_LIQUID_AUDIO,
 };
 
 static std::string read_file(const std::string & fname) {
@@ -1339,7 +1340,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         [](common_params & params, const std::string & value) {
             params.system_prompt = value;
         }
-    ).set_examples({LLAMA_EXAMPLE_COMPLETION, LLAMA_EXAMPLE_CLI, LLAMA_EXAMPLE_DIFFUSION, LLAMA_EXAMPLE_MTMD}));
+    ).set_examples({LLAMA_EXAMPLE_COMPLETION, LLAMA_EXAMPLE_CLI, LLAMA_EXAMPLE_DIFFUSION, LLAMA_EXAMPLE_MTMD, LLAMA_EXAMPLE_LIQUID_AUDIO}));
     add_opt(common_arg(
         {"--perf"},
         {"--no-perf"},
@@ -2159,7 +2160,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
                 params.image.emplace_back(item);
             }
         }
-    ).set_examples({LLAMA_EXAMPLE_MTMD, LLAMA_EXAMPLE_CLI}));
+    ).set_examples({LLAMA_EXAMPLE_MTMD, LLAMA_EXAMPLE_CLI, LLAMA_EXAMPLE_LIQUID_AUDIO}));
     add_opt(common_arg(
         {"--image-min-tokens"}, "N",
         "minimum number of tokens each image can take, only used by vision models with dynamic resolution (default: read from model)",
@@ -2639,7 +2640,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         [](common_params & params, const std::string & value) {
             params.out_file = value;
         }
-    ).set_examples({LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_CVECTOR_GENERATOR, LLAMA_EXAMPLE_EXPORT_LORA, LLAMA_EXAMPLE_TTS, LLAMA_EXAMPLE_FINETUNE}));
+    ).set_examples({LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_CVECTOR_GENERATOR, LLAMA_EXAMPLE_EXPORT_LORA, LLAMA_EXAMPLE_TTS, LLAMA_EXAMPLE_FINETUNE, LLAMA_EXAMPLE_LIQUID_AUDIO}));
     add_opt(common_arg(
         {"-ofreq", "--output-frequency"}, "N",
         string_format("output the imatrix every N iterations (default: %d)", params.n_out_freq),
@@ -2771,14 +2772,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         [](common_params & params, const std::string & value) {
             params.hostname = value;
         }
-    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_HOST"));
+    ).set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_LIQUID_AUDIO}).set_env("LLAMA_ARG_HOST"));
     add_opt(common_arg(
         {"--port"}, "PORT",
         string_format("port to listen (default: %d)", params.port),
         [](common_params & params, int value) {
             params.port = value;
         }
-    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_PORT"));
+    ).set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_LIQUID_AUDIO}).set_env("LLAMA_ARG_PORT"));
     add_opt(common_arg(
         {"--path"}, "PATH",
         string_format("path to serve static files from (default: %s)", params.public_path.c_str()),
@@ -3425,7 +3426,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         [](common_params & params, const std::string & value) {
             params.vocoder.model.path = value;
         }
-    ).set_examples({LLAMA_EXAMPLE_TTS, LLAMA_EXAMPLE_SERVER}));
+    ).set_examples({LLAMA_EXAMPLE_TTS, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_LIQUID_AUDIO}));
      add_opt(common_arg(
         {"--tts-use-guide-tokens"},
         "Use guide tokens to improve TTS word recall",
@@ -3439,7 +3440,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         [](common_params & params, const std::string & value) {
             params.vocoder.speaker_file = value;
         }
-    ).set_examples({LLAMA_EXAMPLE_TTS}));
+    ).set_examples({LLAMA_EXAMPLE_TTS, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_LIQUID_AUDIO}));
 
     add_opt(common_arg(
         {"--diffusion-steps"}, "N",

common/common.h

@@ -104,6 +104,7 @@ enum llama_example {
     LLAMA_EXAMPLE_DIFFUSION,
     LLAMA_EXAMPLE_FINETUNE,
     LLAMA_EXAMPLE_FIT_PARAMS,
+    LLAMA_EXAMPLE_LIQUID_AUDIO,
 
     LLAMA_EXAMPLE_COUNT,
 };

convert_hf_to_gguf.py

@@ -10296,7 +10296,7 @@ def _add_feed_forward_length(self):
     def set_gguf_parameters(self):
         # set num_key_value_heads only for attention layers
         self.hparams["num_key_value_heads"] = [
-            self.hparams["num_key_value_heads"] if layer_type == "full_attention" else 0
+            self.hparams["num_key_value_heads"] if layer_type != "conv" else 0
             for layer_type in self.hparams["layer_types"]
         ]
 
@@ -10345,6 +10345,25 @@ def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
         yield f"{self.dense_tensor_name}.weight", tensor.clone()
 
 
+@ModelBase.register("Lfm25AudioTokenizer")
+class LFM25AudioTokenizer(LFM2Model):
+    model_arch = gguf.MODEL_ARCH.LFM2
+
+    def set_gguf_parameters(self):
+        super().set_gguf_parameters()
+        self.gguf_writer.add_sliding_window(self.hparams["sliding_window"])
+        self.gguf_writer.add_embedding_length_out(self.hparams.get("output_size"))
+
+    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
+        if name == "istft.window" or name.startswith("emb.emb"):
+            return []
+
+        if name.startswith("lin"):
+            name = name.replace("lin", "dense_2_out")
+
+        return super().modify_tensors(data_torch, name, bid)
+
+
 @ModelBase.register("Lfm2MoeForCausalLM")
 class LFM2MoeModel(TextModel):
     model_arch = gguf.MODEL_ARCH.LFM2MOE