pretrained_model_name_or_path (str or os. 19 juil. "every_save": push the model, its configuration, the tokenizer (if passed along to the Trainer) and a draft of a model card each time there is a model save. 4 Likes carted-ml March 30, 2022, 10:14am #6. 3 avr. An award-winning team of journalists, designers, and videographers who tell brand stories through Fast Company's distinctive lens The future of innovation and technology in government for the greater good Our annual g. But if i directly use this pytorch_model. PyTorchでのファインチューニング 「TF」で始まらない「Huggingface Transformers」のモデルクラスはPyTorchモジュールです。推論と最適化の両方でPyTorchのモデルと同じように利用できます。 テキスト分類のデータセット. Author: PL team License: CC BY-SA Generated: 2022-05-05T03:23:24. what does the number 3 mean in a dream. Thank you very much for the detailed answer!. If you aren’t familiar with fine-tuning a model with the Trainer, take a look at the basic tutorial here! At this point, only three steps remain: Define your training hyperparameters in Seq2SeqTrainingArguments. If load_best_model_at_end=True is passed to Trainer, then W&B will save the best performing model checkpoint to Artifacts instead of the final checkpoint. save_model("model_mlm_exp1") subprocess. Tokenizers huggingface from transformers import AutoTokenizer tokenizer = AutoTokenizer. . Save / Load 11:35 Model Hub 13:25 Finetune HuggingFace Tutorial . from_pretrained ( "/path/to/model-directory", local_files_only=True) I get HFValidationError: Repo id must be in the form 'repo_name' or 'namespace/repo_name': '/path/to/model-directory'. If not provided, a model_init must be passed. from_pretrained ("path/to/model") Share Follow edited May 4, 2022 at 18:06. Important attributes: model — Always points to the core model. 14 sept. sunfish sail height; antenna direction indicator. Modified 6 months ago. Trainer( model=model, args=args, train_dataset=train_dataset, eval_dataset=eval_dataset, ) . pretrained_model_name_or_path (str or os. The main novelty seems to be an extra layer of indirection with the prior network (whether it is an autoregressive transformer or a diffusion network), which predicts an image embedding based on the text embedding from CLIP. save_model () , i. 5 jan. This is known as fine-tuning, an incredibly powerful training technique. The pushes are asynchronous to. The authors highlight “the importance of exploring previously unexplored design choices of BERT”. solitaire grand harvest freebies 2020 emove cruiser. I validate the model as I train it, and save the model with the highest scores on the validation set using torch. save_model (output_dir=new_path). Parameters model ( PreTrainedModel, optional) - The model to train, evaluate or use for predictions. In addition to wrapping the model, DeepSpeed can construct and manage the training optimizer, data loader, and the learning rate scheduler based on the parameters passed to deepspeed. Dec 13, 2020 · The RoBERTa model (Liu et al. 2 mar. save_model (optional_output_dir), which will behind the scenes call the save_pretrained of your model ( optional_output_dir is optional and will default to the output_dir you set). The T5 model was proposed in Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer by Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, Peter J. 1 Like Tushar-Faroque July 14, 2021, 2:06pm #3 What if the pre-trained model is saved by using torch. 15 sept. sgugger October 20, 2020, 9:19pm #3 If you set the option load_best_model_at_end to True, the saves will be done at each evaluation (and the Trainer will reload the best model found during the fine-tuning). I validate the model as I train it, and save the model with the highest scores on the validation set using torch. Create a custom architecture Sharing custom models Train with a script Run training on Amazon SageMaker Converting from TensorFlow checkpoints Export to ONNX Export to TorchScript Troubleshoot Natural Language Processing Use tokenizers from 🤗 Tokenizers Inference for multilingual models Task guides Audio. The PyTorch framework is convenient and flexible, with examples that cover reinforcement learning, image classification, and machine translation as the more common use cases. Transformers Models from HuggingFace When specifying and running a language model for the first time in textEmbed() , the python package transformers will . In addition to wrapping the model, DeepSpeed can construct and manage the training optimizer, data loader, and the learning rate scheduler based on the parameters passed to deepspeed. Implementation of DALL-E 2, OpenAI's updated text-to-image synthesis neural network, in Pytorch. 8 is now with the Hub. . These models are based on a variety of transformer architecture - GPT, T5, BERT, etc. The role of the model is to split your “words” into tokens, using the rules it has learned. Dec 13, 2020 · The RoBERTa model (Liu et al. state_dict ()). save and torch. Finetune Transformers Models with PyTorch Lightning¶. save_model () , i. modelname [<ModelNAME>]: uppercase_modelname [<MODEL_NAME>]: lowercase_modelname [<model_name>]: camelcase_modelname [<ModelName>]: Fill in the authors with your team members: authors [The HuggingFace Team]: The checkpoint identifier is the checkpoint that will be used in the examples across the files. But a lot of them are obsolete or outdated. wendy watson nelson. We used the Huggingface's transformers library to load the pre-trained model DistilBERT and fine-tune it to our data. Our training scripts are now optimized for publishing your models on the Hub, taking care of . load ). 启智AI协作平台域名切换公告>>> 15万奖金，400个上榜名额，快来冲击第4期"我为开源打榜狂"，戳详情了解多重上榜加分渠道! >>> 第3期打榜活动领奖名单公示，快去确认你的奖金~>>> 可以查看启智AI协作平台资源说明啦>>> 关于启智集群V100不能访问外网的公告>>>. Bert Model with a language modeling head on top for CLM fine-tuning. " encoding = tokenizer (example) print ( type (encoding)) As mentioned previously, we get a BatchEncoding object in the tokenizer's output:. Details of these design choices can be found in the paper’s Experimental Setup section. 25 mar. huggingface-transformers is this different from Trainer. # Create and train a new model instance. initialize ensures that all of the necessary setup required for distributed data parallel or mixed precision training are done appropriately under the hood. Then I trained again and loaded the previously saved model instead of training from scratch, but it didn't work well, which made me feel like it wasn't saved or loaded successfully ?. Then I trained again and loaded the previously saved model instead of training from scratch, but it didn't work well, which made me feel like it wasn't saved or loaded successfully ?. Do you tried loading the by the trainer saved model in the folder: mitmovie_pt_distilbert_uncased/results. Finetune Transformers Models with PyTorch Lightning¶. 启智AI协作平台域名切换公告>>> 15万奖金，400个上榜名额，快来冲击第4期"我为开源打榜狂"，戳详情了解多重上榜加分渠道! >>> 第3期打榜活动领奖名单公示，快去确认你的奖金~>>> 可以查看启智AI协作平台资源说明啦>>> 关于启智集群V100不能访问外网的公告>>>. a string, the model id of a pretrained feature_extractor hosted inside a model repo on huggingface. 0 and pytorch version 1. Sep 07, 2020 · 以下の記事を参考に書いてます。 ・Huggingface Transformers : Training and fine-tuning 前回 1. huggingface / diffusers Public. 近日 HuggingFace 公司开源了最新的 Transformer2. state_dict ()). Fortunately, hugging face has a model hub, a collection of pre-trained and fine-tuned models for all the tasks mentioned above. If not provided, a model_init must be passed. Valid model ids can be located at the root-level, like bert-base-uncased, or namespaced under a user or organization name, like dbmdz/bert-base-german-cased. Model Once the input texts are normalized and pre-tokenized, the Tokenizer applies the model on the pre-tokens. 0 and pytorch version 1. In the case of a PyTorch checkpoint, from_pt should be set to True and a configuration object should be provided as config argument. sunfish sail height; antenna direction indicator. huggingface の Trainer クラスは huggingface で提供されるモデルの事前学習のときに使うものだと思ってて、下流タスクを学習させるとき（Fine Tuning）は普通に学習のコードを実装. 启智AI协作平台域名切换公告>>> 15万奖金，400个上榜名额，快来冲击第4期"我为开源打榜狂"，戳详情了解多重上榜加分渠道! >>> 第3期打榜活动领奖名单公示，快去确认你的奖金~>>> 可以查看启智AI协作平台资源说明啦>>> 关于启智集群V100不能访问外网的公告>>>. After using the Trainer to train the downloaded model, I save the model with trainer. save (model. ) This model is also a PyTorch torch. They now automatically use torch's `DataLoader` when possible leading to much better GPU utilization (90+% on most models)!. Source code for ray. Bert Model with a language modeling head on top for CLM fine-tuning. Viewed 77k times. checkpoint_fp = checkpoint_dir + "checkpoint_2. The T5 model was proposed in Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer by Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, Peter J. pretrained_model_name_or_path (str or os. Modified 5 months ago. Motivation: While working on a data science competition, I was fine-tuning a pre-trained model and realised how tedious it was to fine-tune a model using native PyTorch or Tensorflow. Check whether the cause is really due to your GPU memory, by a code below. If you make your model a subclass of PreTrainedModel, then you can use our methods save_pretrained and from_pretrained. 启智AI协作平台域名切换公告>>> 15万奖金，400个上榜名额，快来冲击第4期"我为开源打榜狂"，戳详情了解多重上榜加分渠道! >>> 第3期打榜活动领奖名单公示，快去确认你的奖金~>>> 可以查看启智AI协作平台资源说明啦>>> 关于启智集群V100不能访问外网的公告>>>. A pricing model is a method used by a company to determine the prices for its products or services. Now you can simply pass this model and optimizer to your training loop and you would notice that the model resumes training from where it left off. Parameters model ( PreTrainedModel, optional) - The model to train, evaluate. 3 Likes agemagician October 21, 2020, 10:03am #4. The authors highlight “the importance of exploring previously unexplored design choices of BERT”. The Transformer-XL model was proposed in Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context by Zihang Dai, Zhilin Yang, Yiming Yang, Jaime Carbonell, Quoc V. state_dict ()). Motivation: While working on a data science competition, I was fine-tuning a pre-trained model and realised how tedious it was to fine-tune a model using native PyTorch or Tensorflow. Alternatively, if you don’t want to delete the checkpoints, then you can avoid rm -r $save_path, and provide a new output_dir path to trainer. 3 avr. They now automatically use torch's `DataLoader` when possible leading to much better GPU utilization (90+% on most models)!. solitaire grand harvest freebies 2020 emove cruiser. to Trainer , then W&B will save the best performing model checkpoint to . state_dict(), output_model_file). With huggingface_hub, you can easily download and upload. Saving the best/last model in the trainer is confusing to me,. Hello! I'm using Huggingface Transformers to create an NLP model. But a lot of them are obsolete or outdated. Apr 07, 2022 · DALL-E 2 - Pytorch. fit(train_images, train_labels, epochs=5) # Save the entire model as a SavedModel. ( Trainer class will do all setup. This model inherits from PreTrainedModel. Questions & Help I first fine-tuned a bert-base-uncased model on SST-2 dataset with run_glue. Models The base classes PreTrainedModel, TFPreTrainedModel, and FlaxPreTrainedModel implement the common methods for loading/saving a model either from a local file or directory, or from a pretrained model configuration provided by the library (downloaded from HuggingFace's AWS S3 repository). Questions & Help I first fine-tuned a bert-base-uncased model on SST-2 dataset with run_glue. Jan 19, 2022 · In this demo, we will use the Hugging Faces transformers and datasets library together with Tensorflow & Keras to fine-tune a pre-trained seq2seq transformer for financial summarization. Hugging Face Transformers教程笔记(7)：Fine-tuning a pretrained model with the. Save your neuron model to disk and avoid recompilation. An award-winning team of journalists, designers, and videographers who tell brand stories through Fast Company's distinctive lens The future of innovation and technology in government for the greater good Our annual g. Any clue why that may be happening? Reproduction. ) This model is also a PyTorch torch. save and torch. You can search for more pretrained model to use from Huggingface Models page. metrics: max_train_samples = (data_args. e trained on steps x gradient_accumulation_step x per_device_train_size = 1000x8x10 = 80,000 samples). Fine-tuning pretrained NLP models with Huggingface's Trainer. Le, Ruslan Salakhutdinov. json # Save PyTorch model to. Loading a saved model If you. max_train_samples is not None else len (train_dataset)) metrics ["train_samples"] = min (max_train_samples, len (train_dataset)) trainer. 21 oct. Nov 03, 2022 · train_result = trainer. "every_save": push the model, its configuration, the tokenizer (if passed along to the Trainer) and a draft of a model card each time there is a model save. Asked 2 years, 3 months ago. The Trainer class is optimized for Transformers models and can have surprising. This is the part of the pipeline that needs training on your corpus (or that has been trained if you are using a pretrained tokenizer). 24 oct. fit(model, dm). model用于指定使用哪一种模型，例如model为bert，则相应的网络结构为bert的网络结构，configuration是模型具体的结构配置，例如可以配置多头的数量等，这里配置需要注意的地方就是，如果自定义配置不改变核心网络结构的则仍旧可以使用预训练模型权重，如果配置. OpenAI GPT-2 model was proposed in Language Models are Unsupervised Multitask Learners by Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei and Ilya Sutskever. does it save the same thing? – yulGM May 4, 2022 at 14:46 1 @yulGM, . max_train_samples is not None else len (train_dataset)) metrics ["train_samples"] = min (max_train_samples, len (train. When you use a pretrained model, you train it on a dataset specific to your task. As shown in the figure below. Perhaps you could use the Trainer callback mechanism and register handler for on_epoch_end. Trainer is a simple but feature-complete training and eval loop for PyTorch, optimized for 🤗 Transformers. Our training scripts are now optimized for publishing your models on the Hub, taking care of . Loading a saved model If you. 1 Like Tushar-Faroque July 14, 2021, 2:06pm 3 What if the pre-trained model is saved by using torch. 21 oct. model_init ( Callable [ [], PreTrainedModel], optional) - A function that instantiates the model to be used. You can set save_strategy to NO to avoid saving anything and save the final model once training is done with trainer. The pushes are asynchronous to not block training, and in case the save are very frequent, a new push is only attempted if the previous one is finished. In this Pytorch implementation, we will be training a multi-head attention model on the well-known MNIST dataset. I was able to get it to run through with batch 32. diffusers version: 0. Author: PL team License: CC BY-SA Generated: 2022-05-05T03:23:24. metrics: max_train_samples = (data_args. # Create and train a new model instance. does it save the same thing? – yulGM May 4, 2022 at 14:46 1 @yulGM, . ) trainer. Pytorch Of Clear Memory Out Cuda. Saving model checkpoint to test-trainer/checkpoint-500 . Learn how to get started with Hugging Face and the Transformers Library. save_model (optional_output_dir), which will behind the scenes call the save_pretrained of your model ( optional_output_dir is optional and will default to the output_dir you set). If you set save_strategy="epoch" and save_total_limit=1, you will have a save of the model for each trial and you should be able to access it at the end by looking at checkpoint- {trail_id}-xxx. load(checkpoint_fp, map. Describe the bug. Asked 2 years, 4 months ago. save_model (output_dir=new_path). save (model. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc. Train a transformer model to use it as a pretrained transformers model. Bert Model with a language modeling head on top for CLM fine-tuning. As long as the manufacturer is still in business (unlike Saab), this type of situation can present a great buying opportunity for those. 8 déc. pretrained_model_name_or_path (str or os. 19 juil. In this tutorial, we are going to use the transformers library by Huggingface in their newest version (3. If provided, will be used to automatically pad the inputs the maximum length when batching inputs, and it will be saved along the model to make it easier to rerun an interrupted training or reuse the fine-tuned model. The Transformer-XL model was proposed in Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context by Zihang Dai, Zhilin Yang, Yiming Yang, Jaime Carbonell, Quoc V. Check whether the cause is really due to your GPU memory, by a code below. 8 déc. Is there a way to save the model locally instead of pushing to the hub? So in addition to this: trainer. 近日 HuggingFace 公司开. Otherwise it’s regular PyTorch code to save and load (using torch. args ( TrainingArguments, optional) - The arguments to tweak for training. Run training. Viewed 16k times. We'll put having it being automatic on the roadmap so it becomes easier in a future version!. You can save models with trainer. save (model. 2 jan. Trainer is a simple but feature-complete training and eval loop for PyTorch, optimized for 🤗 Transformers. max_train_samples is not None else len (train_dataset)) metrics ["train_samples"] = min (max_train_samples, len (train. 24 jan. solitaire grand harvest freebies 2020 emove cruiser. Saving model checkpoint to test-trainer/checkpoint-500 . Implementation of DALL-E 2, OpenAI's updated text-to-image synthesis neural network, in Pytorch. humiliated in bondage, shop online buy now pay later
Ask Question. . Huggingface trainer save model
In this tutorial, you will fine-tune a pretrained model with a deep learning framework of your choice: Fine-tune a pretrained model with 🤗 Transformers Trainer. what does the number 3 mean in a dream. If you filter for translation, you will see there are 1423 models as of Nov 2021. Another cool thing you can do is you can push your model to the Hugging Face . 1 Answer. The pushes are asynchronous to not block training, and in case the save are very frequent, a new push is only attempted if the previous one is finished. py on a v3-8 TPU VM, and the script hangs at the model saving (save_progress) step. Thank you very much for the detailed answer!. IdoAmit198 December 12, 2022, 7:55am 17. KYIV, Ukraine — Ukraine's president has suggested he's open to peace talks with Russia, softening his refusal to negotiate with Moscow as long as President Vladimir Putin is in powerSep 20, 2022 · The Permissions API was created to be flexible and extensible for applications that require additional validation or permissions that aren't included in Xamarin. The T5 model was proposed in Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer by Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, Peter J. If not provided, a model_init must be passed. When I go and evaluate the model from this point (either manually or by making a Trainer and using trainer. I am running the textual_inversion. A company must consider factors such as the positioning of its products and services as well as production costs when setting the prices of. Perhaps you could use the Trainer callback mechanism and register handler for on_epoch_end. The section below illustrates the steps to save and restore the model. I am trying to reload a fine-tuned DistilBertForTokenClassification model. save and torch. state_dict ()). RoBERTa Model with a language modeling head on top for CLM fine-tuning. And I want to save the best model in a specified directory. Trainer is a simple but feature-complete training and eval loop for PyTorch, optimized for 🤗 Transformers. Saving model checkpoint to test-trainer/checkpoint-500 . The role of the model is to split your “words” into tokens, using the rules it has learned. 193004 This notebook will use HuggingFace’s datasets library to get data, which will be wrapped in a LightningDataModule. Valid model ids can be located at the root-level, like bert-base-uncased, or namespaced under a user or organization name, like dbmdz/bert-base-german-cased. Our training scripts are now optimized for publishing your models on the Hub, taking care of . a string, the model id of a pretrained feature_extractor hosted inside a model repo on huggingface. CLIP Overview The CLIP model was proposed in Learning Transferable Visual Models From Natural Language Supervision by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever. save and torch. If not provided, a model_init must be passed. Apr 07, 2022 · DALL-E 2 - Pytorch. json # Save PyTorch model to. An award-winning team of journalists, designers, and videographers who tell brand stories through Fast Company's distinctive lens The future of innovation and technology in government for the greater good Our annual g. model用于指定使用哪一种模型，例如model为bert，则相应的网络结构为bert的网络结构，configuration是模型具体的结构配置，例如可以配置多头的数量等，这里配置需要注意的地方就是，如果自定义配置不改变核心网络结构的则仍旧可以使用预训练模型权重，如果配置. Num examples = 14143 Batch size = 8 Saving model checkpoint to. TPU VM - tpu-vm-pt-1. 115 suzuki 4 stroke for sale. "every_save": push the model, its configuration, the tokenizer (if passed along to the Trainer) and a draft of a model card each time there is a model save. Ba 2014) and 1-. Jun 19, 2022 · 经过前面一系列的步骤后，我们终于可以开始进行模型训练了。Transformers 库提供了 Trainer 类，可以很简单方便地进行模型训练。首先，创建一个 Trainer，然后调用 train() 函数，就开始进行模型训练了。当模型训练完毕后，调用 save_model() 保存模型。. pretrained_model_name_or_path (str or os. load ). "every_save": push the model, its configuration, the tokenizer (if passed along to the Trainer) and a draft of a model card each time there is a model save. . a path to a directory containing model weights saved using save_pretrained(), e. You can see that integrations. A pricing model is a method used by a company to determine the prices for its products or services. However, since the logging method is fixed, I came across a TrainerCallback while looking for a way to do different logging depending on the situation. Pytorch Of Clear Memory Out Cuda. I am trying to reload a fine-tuned DistilBertForTokenClassification model. In this blog post, we will be explaining how to train a dataset with SSD-Mobilenet object detection model using PyTorch. Trainer is a simple but feature-complete training and eval loop for PyTorch, optimized for 🤗 Transformers. Yannic Kilcher summary | AssemblyAI explainer. In the various training scripts in examples, would it be better to checkpoint the model at the end of each epoch, as well as every save_steps iterations as specified by the user?. Otherwise it’s regular PyTorch code to save and load (using torch. Ask Question. py is integrated with. save_model # Saves the tokenizer too for. Deploy machine learning models and tens of thousands of pretrained Hugging Face transformers to a dedicated endpoint with Microsoft Azure. Model Once the input texts are normalized and pre-tokenized, the Tokenizer applies the model on the pre-tokens. However, since the logging method is fixed, I came across a TrainerCallback while looking for a way to do different logging depending on the situation. Storage space can be an issue when training models, especially when using a Google collab and saving the model to a google drive so it isn't lost when the collab disconnects. Unfortunately, there is currently no way to disable the saving of single files. Ba 2014) and 1-. "end": push the model, its configuration, the tokenizer (if passed along to the Trainer) and a draft of a model card when the save_model() method is called. Trainer(plugins=HFSaveCheckpoint(model=model)) trainer. I have also noticed this issue when trying to fine-tune a RoBERTa language model train_adapter(["sst-2"]) By calling train_adapter. using the k-fold technique with PyTorch-Ignite. The bare T5 Model transformer outputting encoder’s raw hidden-states without any specific head on top. After the training has completed, you can save model with Hugging Face libraries as follows . train (resume_from_checkpoint = checkpoint) metrics = train_result. I am trying to reload a fine-tuned DistilBertForTokenClassification model. If you set save_strategy="epoch" and save_total_limit=1, you will have a save of the model for each trial and you should be able to access it at the end by looking at checkpoint- {trail_id}-xxx. save and torch. Parameters model ( PreTrainedModel, optional) - The model to train, evaluate. save (model. wendy watson nelson. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc. In this tutorial, you will fine-tune a pretrained model with a deep learning framework of your choice: Fine-tune a pretrained model with 🤗 Transformers Trainer. ( Trainer class will do all setup. This tutorial will show you how to take a fine-tuned transformer model, like one of these, and upload the weights and/or the tokenizer to HuggingFace's . Create notebooks and keep track of their status here. This model inherits from PreTrainedModel. 115 suzuki 4 stroke for sale. save_model # Saves the tokenizer too for. "every_save": push the model, its configuration, the tokenizer (if passed along to the Trainer) and a draft of a model card each time there is a model save. Check whether the cause is really due to your GPU memory, by a code below. As long as the manufacturer is still in business (unlike Saab), this type of situation can present a great buying opportunity for those. The PyTorch framework is convenient and flexible, with examples that cover reinforcement learning, image classification, and machine translation as the more common use cases. save_model # Saves the tokenizer too for easy upload: metrics = train_result. If provided, each call to [`~Trainer. X or TF 2. As a result, we can watch how the loss is decreasing while training. 22 avr. If you filter for translation, you will see there are 1423 models as of Nov 2021. . I was able to get it to run through with batch 32. In this tutorial, we are going to use the transformers library by Huggingface in their newest. huggingface / diffusers Public. save_model () , i. Otherwise it’s regular PyTorch code to save and load (using torch. . yelp walmart