-
```
What steps will reproduce the problem?
1. update to latest depot_tools
build/gyp_chromium --depth=. -fninja -G msvs_version=2012
talk/libjingle_all.gyp
ninja -C out\Debug …
-
```
What steps will reproduce the problem?
1. update to latest depot_tools
build/gyp_chromium --depth=. -fninja -G msvs_version=2012
talk/libjingle_all.gyp
ninja -C out\Debug …
-
```
What steps will reproduce the problem?
1. update to latest depot_tools
build/gyp_chromium --depth=. -fninja -G msvs_version=2012
talk/libjingle_all.gyp
ninja -C out\Debug …
-
## Checklist:
1. 查找[历史相关issue](https://github.com/PaddlePaddle/PaddleX/issues)寻求解答
2. 翻阅[FAQ常见问题汇总和答疑](https://github.com/PaddlePaddle/PaddleX/blob/develop/docs/FAQ/FAQ.md)
3. 确认bug是否在新版本里还未修复
4…
-
已经尝试过chatglm和chatglm2均成功加速,但是在尝试moss模型时(moss-moon-003-sft-plugin-int4),成功转化模型,但无法推理。同时stopping_criteria_list也不知道应该放哪里。
代码如下
```
tokenizer = AutoTokenizer.from_pretrained("../pre_model/moss-moon-0…
-
Ideas, anyone?
It would also be helpful to have a flowchart regarding which prompts lead to which pages...
First prompt with more detailed custom compute:
execute
|-benchmark
|-stresser
|-custom…
-
Replace `new` and `delete` to allocate and de-allocate, respectively, through a allocator that re-cycles CPU heaps.
-
Hey @ffulin, thanks for the work on FastBuild!
I've recently seen lots of improvements mentioned in FB version 1.11 and 1.12 so I was eager to try them with my FB setup.
However, to my surprise, I…
-
some functions, like TransAddTranslation, take a vec3_t as a parameter, and do SSE operations with it. From a quick read, it seems that SSE operations expect 16 bytes (octets) operand size, and vec3_t…
-
**Describe the bug**
```shell
1. I'm build rknpu backend with cross compiling on linux;
2. build arm linux without rknpu is ok;
3. it also met the incompatibility issue between vector and vector ,…