djl
djl copied to clipboard
An Engine-Agnostic Deep Learning Framework in Java
Allows -1 in shape string presentation ## Description ## Brief description of what this PR is about - If this change is a backward incompatible change, why must this change...
## Description ## Brief description of what this PR is about - If this change is a backward incompatible change, why must this change be made? - Interesting edge cases...
## Description (A clear and concise description of what the feature is.) Generating Face matching 128 dimension Will this change the current api? How? As of now embedding returned by...
## Description ## Brief description of what this PR is about - If this change is a backward incompatible change, why must this change be made? - Interesting edge cases...
## Description I have set `interOpNumThreads` and `intraOpNumThreads` option,but the cpu usage is 3200%, it's to high. so how to reduce the cpu usage ,such as 400%. ## Code ```scala...
## Description ## This PR convert Gradle Groovy scripts to Kotlin, which is also [recommended by Gradle](https://docs.gradle.org/current/userguide/implementing_gradle_plugins.html), being a statically typed language this change is backward compatible Interesting edge cases...
## Description **Using Llama for inference, some Chinese characters may become garbled.**  **Additionally, I would like to ask if DJL can fine tune the llama model?**
I found that the inference speed in TorchScript(converted by [model_zoo_importer.py](https://github.com/deepjavalibrary/djl/blob/5c691c0979d92702f7bfe1aba2f989d8c9b1c568/extensions/tokenizers/src/main/python/model_zoo_importer.py#L23)) slower than in default PyTorch format. ``` from sentence_transformers import CrossEncoder from transformers import AutoModelForSequenceClassification, AutoTokenizer import torch import time...
- **[tokenizer] Refactor djl_convert python code** - **[huggingface] Adds more option to convert onnx model** ## Description ## Brief description of what this PR is about - If this change...
## Description ## Brief description of what this PR is about - If this change is a backward incompatible change, why must this change be made? - Interesting edge cases...