remove some commented lines:

alibaba · Oct 31, 2024 · 69e1187 · 69e1187
1 parent f8d0ddc
commit 69e1187
Show file tree

Hide file tree

Showing 8 changed files with 14 additions and 77 deletions.
diff --git a/TODO.md b/TODO.md
diff --git a/docs/transformers/llm.md b/docs/transformers/llm.md
@@ -229,6 +229,7 @@ sh package_scripts/ios/buildiOS.sh "-DMNN_ARM82=true -DMNN_LOW_MEMORY=true -DMNN
 
 #### 推理用法
 `chat_demo`的用法如下：
+pc端直接推理
 ```
 # 使用config.json
 ## 交互式聊天
@@ -243,6 +244,12 @@ sh package_scripts/ios/buildiOS.sh "-DMNN_ARM82=true -DMNN_LOW_MEMORY=true -DMNN
 ./chat_demo model_dir/llm.mnn prompt.txt
 ```
 
+手机端adb推理用法：
+```bash
+# 利用adb push将链接库push到手机上
+adb push chat_demo libllm.so libMNN_CL.so libMNN_Express.so libMNN.so tools/cv/libMNNOpenCV.so /data/local/tmp/llm
+```
+
 #### GPTQ权重加载
 - 使用脚本生成GPTQ模型权重，用法参考: [apply_gptq.py](../tools/script.html#apply-gptq-py)
 - 创建`gptq.json`配置文件

diff --git a/express/NeuralNetWorkOp.cpp b/express/NeuralNetWorkOp.cpp
@@ -476,9 +476,6 @@ VARP _Softmax(VARP logits, int axis) {
     softmax->main.AsAxis()->axis = axis;
     return (Variable::create(Expr::create(softmax.get(), {logits})));
 }
-VARP _TempratureSoftmax(VARP logits, float temperature, int axis) {
-    return _Softmax(logits * _Scalar<float>(1.0f / temperature), axis);
-}
 /*Computes softplus: log(exp(features) + 1).
 Args:
 features: A variable. Must be Halide_Type_Float.

diff --git a/include/MNN/expr/NeuralNetWorkOp.hpp b/include/MNN/expr/NeuralNetWorkOp.hpp
@@ -58,7 +58,6 @@ MNN_PUBLIC VARP _Relu(VARP x, float slope = 0.0f);
 MNN_PUBLIC VARP _Relu6(VARP x, float minValue = 0.0f, float maxValue = 6.0f);
 MNN_PUBLIC VARP _PRelu(VARP x, std::vector<float> &&slopes);
 MNN_PUBLIC VARP _Softmax(VARP logits, int axis = -1);
-MNN_PUBLIC VARP _TempratureSoftmax(VARP logits, float temperature, int axis = -1);
 MNN_PUBLIC VARP _Softplus(VARP features);
 MNN_PUBLIC VARP _Softsign(VARP features);
 MNN_PUBLIC std::vector<VARP> _Split(VARP value, INTS size_splits, int axis = 0);

diff --git a/transformers/llm/engine/android/gradle/wrapper/gradle-wrapper.properties b/transformers/llm/engine/android/gradle/wrapper/gradle-wrapper.properties
@@ -1,5 +1,5 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-7.3.3-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-8.9-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists
diff --git a/transformers/llm/engine/include/llmconfig.hpp b/transformers/llm/engine/include/llmconfig.hpp
@@ -322,22 +322,9 @@ class LlmConfig {
         return llm_config_.value("attention_fused", true);
     }
 
-    // std::string chat_template() const {
-    //     return llm_config_.value("chat_template", "");
-    // }
-
-    // std::string prompt_template() const {
-    //     return llm_config_.value("prompt_template", "");
-    // }
-
-    // std::string eos_token() const {
-    //     return llm_config_.value("eos_token", "")
-    // }
-
     std::string system_prompt_template() const {
         return llm_config_.value("system_prompt_template", "<|im_start|>system\n%s<|im_end|>\n");
     }
-
     std::string user_prompt_template() const {
         return llm_config_.value("user_prompt_template", "<|im_start|>user\n%s<|im_end|>\n");
     }

diff --git a/transformers/llm/engine/include/sampler/sampler.hpp b/transformers/llm/engine/include/sampler/sampler.hpp
@@ -24,6 +24,8 @@
 namespace MNN {
 namespace Transformer {
 
+MNN_PUBLIC VARP _TempratureSoftmax(VARP logits, float temperature, int axis = -1);
+
 class Llm;
 
 // a index and its corresponding score

diff --git a/transformers/llm/engine/src/sampler.cpp b/transformers/llm/engine/src/sampler.cpp
@@ -13,6 +13,10 @@
 namespace MNN{
 namespace Transformer{
 
+VARP _TempratureSoftmax(VARP logits, float temperature, int axis) {
+    return _Softmax(logits * _Scalar<float>(1.0f / temperature), axis);
+}
+
 /* ----------Sampler's members---------- */
 int Sampler::select(struct SubsetLogits& subset, int id) {
     if (!(subset.is_subset)) return id;