[update] support qwen3.5 & add model file

Author: LittleMouse

ext_components/openai-api/SConstruct

@@ -12,7 +12,7 @@ _SDK_PATH = os.path.normpath(
 
 env["GIT_REPO_LISTS"]["openai-api"] = {
     "url": "https://github.com/ZHEQIUSHUI/openai-api.cpp.git",
-    "commit": "127f3e75054a733ca1b903ff1f8ec600b90cb6b0",
+    "commit": "3d4be086e95cee55bfd8e983b20f0758b6f82182",
     "path": str(Path(_SDK_PATH) / "github_source" / "openai-api.cpp"),
 }
 

projects/llm_framework/main_llm2/models/mode_Qwen3.5-2B-ax650.json

@@ -0,0 +1,48 @@
+{
+    "mode": "Qwen3.5-2B-ax650",
+    "type": "llm",
+    "homepage": "https://huggingface.co/AXERA-TECH/Qwen3.5-2B-AX650-C128-P1152-CTX2047",
+    "capabilities": [
+        "text_generation",
+        "chat"
+    ],
+    "input_type": [
+        "llm.utf-8",
+        "llm.utf-8.stream",
+        "llm.chat_completion",
+        "llm.chat_completion.stream"
+    ],
+    "output_type": [
+        "llm.utf-8",
+        "llm.utf-8.stream"
+    ],
+    "mode_param": {
+        "system_prompt": "you are a helpful assistant.",
+        "model_name": "AXERA-TECH/Qwen3.5-2B",
+        "url_tokenizer_model": "qwen3_5_tokenizer.txt",
+        "tokenizer_type": "Qwen3_5VL",
+        "post_config_path": "post_config.json",
+        "template_filename_axmodel": "qwen3_5_text_p128_l%d_together.axmodel",
+        "axmodel_num": 24,
+        "full_attention_interval": 4,
+        "filename_post_axmodel": "qwen3_5_text_post.axmodel",
+        "filename_tokens_embed": "model.embed_tokens.weight.bfloat16.bin",
+        "tokens_embed_num": 248320,
+        "tokens_embed_size": 2048,
+        "b_use_mmap_load_embed": true,
+        "b_use_mmap_load_layer": true,
+        "vlm_type": "Qwen3VL",
+        "filename_image_encoder_axmodel": "qwen3_5_vision.axmodel",
+        "vision_patch_size": 16,
+        "vision_width": 384,
+        "vision_height": 384,
+        "vision_temporal_patch_size": 2,
+        "vision_spatial_merge_size": 2,
+        "vision_fps": 1,
+        "vision_tokens_per_second": 1,
+        "vision_cache_dir": "vision_cache",
+        "devices": [
+            0
+        ]
+    }
+}

projects/llm_framework/main_llm2/src/main.cpp

@@ -146,6 +146,7 @@ class llm_task {
             CONFIG_AUTO_SET(file_body["mode_param"], b_bos);
             CONFIG_AUTO_SET(file_body["mode_param"], b_eos);
 
+            CONFIG_AUTO_SET(file_body["mode_param"], full_attention_interval);
             CONFIG_AUTO_SET(file_body["mode_param"], filename_tokens_embed);
             CONFIG_AUTO_SET(file_body["mode_param"], tokens_embed_num);
             CONFIG_AUTO_SET(file_body["mode_param"], tokens_embed_size);
@@ -169,6 +170,16 @@ class llm_task {
             CONFIG_AUTO_SET(file_body["mode_param"], kv_cache_size);
             CONFIG_AUTO_SET(file_body["mode_param"], b_use_mmap_load_layer);
 
+            CONFIG_AUTO_SET(file_body["mode_param"], vision_cache_dir);
+
+            CONFIG_AUTO_SET(file_body["mode_param"], vision_width);
+            CONFIG_AUTO_SET(file_body["mode_param"], vision_height);
+            CONFIG_AUTO_SET(file_body["mode_param"], vision_temporal_patch_size);
+            CONFIG_AUTO_SET(file_body["mode_param"], vision_spatial_merge_size);
+            CONFIG_AUTO_SET(file_body["mode_param"], vision_patch_size);
+            CONFIG_AUTO_SET(file_body["mode_param"], vision_fps);
+            CONFIG_AUTO_SET(file_body["mode_param"], vision_tokens_per_second);
+
             mode_config_.template_filename_axmodel      = base_model + mode_config_.template_filename_axmodel;
             mode_config_.filename_post_axmodel          = base_model + mode_config_.filename_post_axmodel;
             mode_config_.filename_tokens_embed          = base_model + mode_config_.filename_tokens_embed;