{"id":21395468,"url":"https://github.com/a-jacobson/shiba","last_synced_at":"2026-05-18T07:35:00.270Z","repository":{"id":40953856,"uuid":"184832591","full_name":"A-Jacobson/shiba","owner":"A-Jacobson","description":"A simple, flexible, pytorch training loop.","archived":false,"fork":false,"pushed_at":"2023-02-11T07:49:06.000Z","size":4944,"stargazers_count":3,"open_issues_count":2,"forks_count":0,"subscribers_count":1,"default_branch":"master","last_synced_at":"2025-06-03T19:12:42.910Z","etag":null,"topics":["computer-vision","deep-learning","nlp","pytorch"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/A-Jacobson.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE.txt","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2019-05-03T23:41:52.000Z","updated_at":"2025-06-03T07:37:25.000Z","dependencies_parsed_at":"2025-03-16T14:40:43.251Z","dependency_job_id":null,"html_url":"https://github.com/A-Jacobson/shiba","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/A-Jacobson/shiba","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/A-Jacobson%2Fshiba","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/A-Jacobson%2Fshiba/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/A-Jacobson%2Fshiba/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/A-Jacobson%2Fshiba/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/A-Jacobson","download_url":"https://codeload.github.com/A-Jacobson/shiba/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/A-Jacobson%2Fshiba/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":33169364,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-18T05:43:36.989Z","status":"ssl_error","status_checked_at":"2026-05-18T05:43:19.133Z","response_time":71,"last_error":"SSL_read: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["computer-vision","deep-learning","nlp","pytorch"],"created_at":"2024-11-22T14:20:04.446Z","updated_at":"2026-05-18T07:35:00.249Z","avatar_url":"https://github.com/A-Jacobson.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# ARCHIVED\ncurrently using the lovely pytorch lightning https://github.com/PyTorchLightning/pytorch-lightning\n\n# shiba\nA simple, flexible, pytorch trainer. We try to be lighter (just a trainer) and lower level than `fastai` and higher level than `ignite`.\n\n### Features \n- `callbacks/fit api` (keras/sklearn)\n- `learning rate finder` (fastai)\n- `one_cycle` (fastai)\n- `mixed precision training` (apex)\n- `process_functions/process_function zoo` (ignite/me)\n- `output_transforms for metrics` (ignite)\n- `tensorboard prediction vis functions` (me)\n\n## Install\n```bash\npip install -U git+https://github.com/A-Jacobson/shiba.git\n```\n\n### Train resnet18 on CIFAR10 with tensorboard logging, Checkpointing, and a customer Metric.\n```python\nfrom torch import nn\nfrom torchvision.datasets import CIFAR10\nfrom torchvision.transforms import ToTensor\nfrom torchvision.models import resnet18\n\nfrom shiba import Trainer\nfrom shiba.callbacks import TensorBoard, Save, Metric\nfrom shiba.vis import vis_classify\nfrom shiba.metrics import categorical_accuracy\n\ntrain_dataset = CIFAR10('data', train=True, download=True, transform=ToTensor())\nval_dataset = CIFAR10('data', train=False, transform=ToTensor())\n\nmodel = resnet18()\nmodel.fc = nn.Linear(512, 10)\ncriterion = nn.CrossEntropyLoss()      \ntrainer = Trainer(model, criterion) \ntrainer.find_lr(train_dataset) # prints lr finder graph\n\ncallbacks = [TensorBoard(log_dir='runs/shiba-test-cifar', vis_function=vis_classify),\n             Metric(name='accuracy', score_func=categorical_accuracy),\n             Save('weights/cifar', monitor='val_loss')]\n\ntrainer.fit_one_cycle(train_dataset, val_dataset, epochs=10, max_lr=1e-3, callbacks=callbacks)\n```\n\n### Write your own training steps and validation steps.\nshiba comes with sensible default steps that can be easily overridden by passing your own\n `train_step` and/or `val_step` functions to the constructor. \n```python\ndef default_train_step(trainer, batch):\n    inputs, targets = batch\n    inputs = inputs.to(trainer.device, non_blocking=True)\n    targets = targets.to(trainer.device, non_blocking=True)\n    outputs = trainer.model(inputs)\n    loss = trainer.criterion(outputs, targets)\n    return dict(loss=loss,\n                inputs=inputs,\n                outputs=outputs,\n                targets=targets)\n                \ndef rnn_step(trainer, batch):\n    \"\"\"An Example RNN step, output is saved to trainer.out\"\"\"\n    hidden = repackage_hidden(trainer.out['hidden'])\n    inputs, targets = batch  # inputs.shape : (seq_len, batch_size)\n    outputs, hidden = trainer.model(inputs, hidden)\n    seq_len, batch_size, vocab_size = outputs.shape\n    loss = trainer.criterion(outputs.view(-1, vocab_size), targets.view(-1)) \n    return dict(loss=loss,\n                inputs=inputs,\n                outputs=outputs,\n                hidden=hidden,\n                targets=targets)\n\n\ntrainer = Trainer(model, criterion, train_step=rnn_step)\n```\n\n### Use Callbacks to easily add support for logging, Progress bars, metrics, and learning rate schedulers.\n```python\nclass ProgressBar(Callback):\n    def __init__(self):\n        self.train_pbar = None\n        self.epoch_pbar = None\n\n    def on_train_begin(self, trainer):\n        self.train_pbar = tqdm(total=trainer.epochs, unit='epochs')\n\n    def on_epoch_begin(self, trainer):\n        self.epoch_pbar = tqdm(total=trainer.num_batches, unit='b')\n\n    def on_epoch_end(self, trainer):\n        self.train_pbar.update()\n        self.epoch_pbar.close()\n\n    def on_batch_end(self, trainer):\n        self.epoch_pbar.update()\n        self.epoch_pbar.set_postfix(trainer.metrics)\n\n    def on_eval_end(self, trainer):\n        self.epoch_pbar.set_postfix(trainer.metrics)\n\n    def on_train_end(self, trainer):\n        self.train_pbar.close()\n\n ```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fa-jacobson%2Fshiba","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fa-jacobson%2Fshiba","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fa-jacobson%2Fshiba/lists"}