fix unused args

Author: younesbelkada

convert_hf_to_gguf.py

@@ -6647,9 +6647,6 @@ def set_gguf_parameters(self):
         self.gguf_writer.add_context_length(self.hparams.get("max_position_embeddings", 0))
         self.gguf_writer.add_feed_forward_length(self.hparams["intermediate_size"])
 
-        ## Mamba mixer params ##
-        self.gguf_writer.add_ssm_head_dim(self.d_head)
-
         ## Attention params ##
         self.gguf_writer.add_head_count(self.hparams["num_attention_heads"]) # Override value 0 from Mamba2
         self.gguf_writer.add_head_count_kv(self.hparams["num_key_value_heads"])

gguf-py/gguf/gguf_writer.py

@@ -867,9 +867,6 @@ def add_ssm_group_count(self, value: int) -> None:
     def add_ssm_dt_b_c_rms(self, value: bool) -> None:
         self.add_bool(Keys.SSM.DT_B_C_RMS.format(arch=self.arch), value)
 
-    def add_ssm_head_dim(self, value: int) -> None:
-        self.add_uint32(Keys.SSM.HEAD_DIM.format(arch=self.arch), value)
-
     def add_tokenizer_model(self, model: str) -> None:
         self.add_string(Keys.Tokenizer.MODEL, model)
 

src/llama-hparams.h

@@ -115,7 +115,6 @@ struct llama_hparams {
     uint32_t ssm_d_state = 0;
     uint32_t ssm_dt_rank = 0;
     uint32_t ssm_n_group = 0;
-    uint32_t ssm_head_dim   = 0;
 
     // for hybrid state space models
     std::array<bool, LLAMA_MAX_LAYERS> recurrent_layer_arr;