From ce885a33c1d75d3adb817d58a079244e68e9be91 Mon Sep 17 00:00:00 2001
From: "ziyang.wang" <ziyang.wang@sophgo.com>
Date: Mon, 11 Nov 2024 17:54:42 +0800
Subject: [PATCH] [llama3.2] fix wrong vocab size

---
 models/Llama3_2-Vision/compile/export_onnx.py     | 2 +-
 models/Llama3_2-Vision/python_demo/CMakeLists.txt | 4 +++-
 models/Llama3_2-Vision/python_demo/pipeline.py    | 3 +--
 models/Qwen2_5/python_demo/CMakeLists.txt         | 6 ++++--
 4 files changed, 9 insertions(+), 6 deletions(-)

diff --git a/models/Llama3_2-Vision/compile/export_onnx.py b/models/Llama3_2-Vision/compile/export_onnx.py
index 5a56c42..6c0453b 100755
--- a/models/Llama3_2-Vision/compile/export_onnx.py
+++ b/models/Llama3_2-Vision/compile/export_onnx.py
@@ -60,7 +60,7 @@
 NUM_ATTENTION_HEADS = config.text_config.num_attention_heads
 NUM_KEY_VALUE_HEADS = config.text_config.num_key_value_heads
 HEAD_DIM = HIDDEN_SIZE // NUM_ATTENTION_HEADS
-VOCAB_SIZE = config.text_config.hidden_size
+VOCAB_SIZE = config.text_config.vocab_size
 CROSS_ATTN_LAYERS = config.text_config.cross_attention_layers
 print(f'\nLLM config:\n\
     Layers: {NUM_LAYERS}\n\
diff --git a/models/Llama3_2-Vision/python_demo/CMakeLists.txt b/models/Llama3_2-Vision/python_demo/CMakeLists.txt
index a439063..73c5cc0 100755
--- a/models/Llama3_2-Vision/python_demo/CMakeLists.txt
+++ b/models/Llama3_2-Vision/python_demo/CMakeLists.txt
@@ -5,6 +5,8 @@ if (NOT DEFINED TARGET_ARCH)
     set(TARGET_ARCH pcie)
 endif()
 
+set(CMAKE_CXX_STANDARD 17)
+
 include_directories(${PROJECT_SOURCE_DIR}/../../../support/include)
 
 if (${CMAKE_HOST_SYSTEM_PROCESSOR} STREQUAL "aarch64")
@@ -17,7 +19,7 @@ elseif (${TARGET_ARCH} STREQUAL "pcie")
     message("PCIE mode, starting......")
 endif()
 
-add_definitions(-DDEBUG --std=c++17 -fPIC -Wall -Werror)
+add_definitions(-DDEBUG -fPIC -Wall -Werror)
 set(CMAKE_BUILD_TYPE "Debug")
 
 find_package(pybind11 REQUIRED CONFIG)
diff --git a/models/Llama3_2-Vision/python_demo/pipeline.py b/models/Llama3_2-Vision/python_demo/pipeline.py
index 0e63d43..c9bca65 100755
--- a/models/Llama3_2-Vision/python_demo/pipeline.py
+++ b/models/Llama3_2-Vision/python_demo/pipeline.py
@@ -137,8 +137,7 @@ def stream_answer(self, inputs):
 
             self.answer_token += full_word_tokens
             print(word, flush=True, end="")
-            if tok_num > self.model.max_new_tokens and word == '.':
-                break
+
             token = self.model.forward_next()
             tok_num += 1
             full_word_tokens = []
diff --git a/models/Qwen2_5/python_demo/CMakeLists.txt b/models/Qwen2_5/python_demo/CMakeLists.txt
index c32d1d2..87e068e 100755
--- a/models/Qwen2_5/python_demo/CMakeLists.txt
+++ b/models/Qwen2_5/python_demo/CMakeLists.txt
@@ -1,10 +1,12 @@
 cmake_minimum_required(VERSION 3.10)
-project(qwen2)
+project(qwen2_5)
 
 if (NOT DEFINED TARGET_ARCH)
     set(TARGET_ARCH pcie)
 endif()
 
+set(CMAKE_CXX_STANDARD 17)
+
 include_directories(${PROJECT_SOURCE_DIR}/../support/include)
 
 if (${CMAKE_HOST_SYSTEM_PROCESSOR} STREQUAL "aarch64")
@@ -17,7 +19,7 @@ elseif (${TARGET_ARCH} STREQUAL "pcie")
 	message("PCIE mode, starting......")
 endif()
 
-add_definitions(-DDEBUG --std=c++17 -fPIC -Wall -Werror)
+add_definitions(-DDEBUG -fPIC -Wall -Werror)
 set(CMAKE_BUILD_TYPE "Debug")
 
 find_package(pybind11 REQUIRED CONFIG)