make qwen3-emb working

Author: foldl

models/qwen.cpp

@@ -1496,6 +1496,11 @@ namespace chatllm::qwen::v3_emb
         Tokenizer *tok = dynamic_cast<Tokenizer *>(tokenizer);
         tok->task = utils::get_opt(args, "task", tok->task);
     }
+
+    int ConditionalGeneration::get_text_embedding_dim(void) const
+    {
+        return config.hidden_size;
+    }
 }
 
 namespace chatllm::qwen::v3_ranker

models/qwen.h

@@ -605,6 +605,7 @@ namespace chatllm::qwen
                 ModelType type = ModelType::MODEL_TYPE_QWEN3_Embedding, const bool skip_lm_head = true, int extra_tensors = 0);
 
             void set_additional_args(const std::map<std::string, std::string> &args) override;
+            int get_text_embedding_dim(void) const override;
         };
     }
 

src/chat.h

@@ -1298,7 +1298,7 @@ namespace chatllm
                   cache_type(ggml::str_to_type(cache_type, ggml::type::GGML_TYPE_F16)),
                   re_quantize(ggml::str_to_type(re_quantize))
             {}
-            extra_args() : extra_args(-1, "", "", false, 1, 0, "") {}
+            extra_args() : extra_args(-1, "", false, 1, 0, "") {}
         };
 
         ModelObject(const std::string &path);

src/main.cpp

@@ -1157,7 +1157,8 @@ chatllm::BaseStreamer *get_streamer_for_log(void)
 
 static int init_vector_store(Args &args)
 {
-    chatllm::Pipeline pipeline(args.embedding_model_path);
+    DEF_ExtraArgs(pipe_args, args);
+    chatllm::Pipeline pipeline(args.embedding_model_path, pipe_args);
     args.max_length = pipeline.model->get_max_length();
 
     DEF_GenerationConfig(gen_config, args);