Anurag1734/cuda-error-resolution-analysis
07
1[2 {3 "post_stream": {4 "posts": [5 {6 "id": 471672,7 "name": "Natalia ",8 "username": "NataliaZagalabs",9 "avatar_template": "/letter_avatar_proxy/v4/letter/n/85e7bf/{size}.png",10 "created_at": "2025-06-11T15:07:01.464Z",11 "cooked": "<p>Una empresa colombiana en crecimiento global busca un/a <strong>Desarrollador/a con experiencia en Machine Learning</strong>, altamente competente en <strong>Python</strong>, para sumarse a un equipo internacional.<br>\nSi tienes conocimientos sólidos en <strong>PyTorch, Torchaudio y Torchvideo</strong>, te apasiona la inteligencia artificial aplicada a audio y video, y hablas inglés con fluidez, ¡esta oportunidad es para ti!<br>\n<img src=\"https://discuss.pytorch.org/images/emoji/apple/round_pushpin.png?v=14\" title=\":round_pushpin:\" class=\"emoji\" alt=\":round_pushpin:\" loading=\"lazy\" width=\"20\" height=\"20\"> Modalidad: Trabajo <strong>100% remoto</strong></p>\n<ul>\n<li>Contratación desde <strong>Colombia o Argentina</strong></li>\n<li>Pagos en <strong>USD</strong></li>\n</ul>\n<p>CV <a href=\"mailto:natalia.ramirez@zagalabs.com\">natalia.ramirez@zagalabs.com</a></p>",12 "post_number": 1,13 "post_type": 1,14 "posts_count": 2,15 "updated_at": "2025-06-11T15:15:42.656Z",16 "reply_count": 0,17 "reply_to_post_number": null,18 "quote_count": 0,19 "incoming_link_count": 3,20 "reads": 18,21 "readers_count": 17,22 "score": 18.6,23 "yours": false,24 "topic_id": 220733,25 "topic_slug": "oferta-laboral-payton-con-ml-pytorch-torchaudio-torchvideo",26 "display_username": "Natalia ",27 "primary_group_name": null,28 "flair_name": null,29 "flair_url": null,30 "flair_bg_color": null,31 "flair_color": null,32 "flair_group_id": null,33 "badges_granted": [],34 "version": 2,35 "can_edit": false,36 "can_delete": false,37 "can_recover": false,38 "can_see_hidden_post": false,39 "can_wiki": false,40 "read": true,41 "user_title": null,42 "bookmarked": false,43 "actions_summary": [],44 "moderator": false,45 "admin": false,46 "staff": false,47 "user_id": 84655,48 "hidden": false,49 "trust_level": 0,50 "deleted_at": null,51 "user_deleted": false,52 "edit_reason": null,53 "can_view_edit_history": true,54 "wiki": false,55 "post_url": "/t/oferta-laboral-payton-con-ml-pytorch-torchaudio-torchvideo/220733/1",56 "can_accept_answer": false,57 "can_unaccept_answer": false,58 "accepted_answer": false,59 "topic_accepted_answer": null,60 "can_vote": false61 },62 {63 "id": 471673,64 "name": "",65 "username": "ptrblck",66 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",67 "created_at": "2025-06-11T15:15:36.963Z",68 "cooked": "<p>From Google Translate:</p>\n<pre><code class=\"lang-auto\">A Colombian company with global growth is seeking a Machine Learning Developer with extensive Python experience to join an international team.\nIf you have solid knowledge of PyTorch, Torchaudio, and Torchvideo, are passionate about artificial intelligence applied to audio and video, and are fluent in English, this opportunity is for you!\n:round_pushpin: Working Mode: 100% remote\n\nHiring from Colombia or Argentina\nPayments in USD```</code></pre>",69 "post_number": 2,70 "post_type": 1,71 "posts_count": 2,72 "updated_at": "2025-06-11T15:15:36.963Z",73 "reply_count": 0,74 "reply_to_post_number": null,75 "quote_count": 0,76 "incoming_link_count": 2,77 "reads": 18,78 "readers_count": 17,79 "score": 28.6,80 "yours": false,81 "topic_id": 220733,82 "topic_slug": "oferta-laboral-payton-con-ml-pytorch-torchaudio-torchvideo",83 "display_username": "",84 "primary_group_name": null,85 "flair_name": null,86 "flair_url": null,87 "flair_bg_color": null,88 "flair_color": null,89 "flair_group_id": null,90 "badges_granted": [],91 "version": 1,92 "can_edit": false,93 "can_delete": false,94 "can_recover": false,95 "can_see_hidden_post": false,96 "can_wiki": false,97 "read": true,98 "user_title": "",99 "bookmarked": false,100 "actions_summary": [101 {102 "id": 2,103 "count": 1104 }105 ],106 "moderator": true,107 "admin": true,108 "staff": true,109 "user_id": 3534,110 "hidden": false,111 "trust_level": 2,112 "deleted_at": null,113 "user_deleted": false,114 "edit_reason": null,115 "can_view_edit_history": true,116 "wiki": false,117 "post_url": "/t/oferta-laboral-payton-con-ml-pytorch-torchaudio-torchvideo/220733/2",118 "can_accept_answer": false,119 "can_unaccept_answer": false,120 "accepted_answer": false,121 "topic_accepted_answer": null122 }123 ],124 "stream": [125 471672,126 471673127 ]128 },129 "timeline_lookup": [130 [131 1,132 136133 ]134 ],135 "suggested_topics": [136 {137 "fancy_title": "We’re Hiring: ML Engineer (Audio Signal Processing) :musical_notes:",138 "id": 218871,139 "title": "We're Hiring: ML Engineer (Audio Signal Processing) 🎶",140 "slug": "were-hiring-ml-engineer-audio-signal-processing",141 "posts_count": 1,142 "reply_count": 0,143 "highest_post_number": 1,144 "image_url": null,145 "created_at": "2025-04-08T14:13:21.488Z",146 "last_posted_at": "2025-04-08T14:13:21.528Z",147 "bumped": true,148 "bumped_at": "2025-04-08T14:13:21.528Z",149 "archetype": "regular",150 "unseen": false,151 "pinned": false,152 "unpinned": null,153 "visible": true,154 "closed": false,155 "archived": false,156 "bookmarked": null,157 "liked": null,158 "tags_descriptions": {},159 "like_count": 0,160 "views": 233,161 "category_id": 24,162 "featured_link": null,163 "has_accepted_answer": false,164 "posters": [165 {166 "extras": "latest single",167 "description": "Original Poster, Most Recent Poster",168 "user": {169 "id": 83680,170 "username": "durai",171 "name": "",172 "avatar_template": "/letter_avatar_proxy/v4/letter/d/34f0e0/{size}.png",173 "trust_level": 0174 }175 }176 ]177 },178 {179 "fancy_title": "Looking for Pytorch Developer",180 "id": 216814,181 "title": "Looking for Pytorch Developer",182 "slug": "looking-for-pytorch-developer",183 "posts_count": 3,184 "reply_count": 1,185 "highest_post_number": 3,186 "image_url": null,187 "created_at": "2025-02-18T08:12:40.307Z",188 "last_posted_at": "2025-02-20T10:55:23.972Z",189 "bumped": true,190 "bumped_at": "2025-02-20T10:55:23.972Z",191 "archetype": "regular",192 "unseen": false,193 "pinned": false,194 "unpinned": null,195 "visible": true,196 "closed": false,197 "archived": false,198 "bookmarked": null,199 "liked": null,200 "tags_descriptions": {},201 "like_count": 0,202 "views": 254,203 "category_id": 24,204 "featured_link": null,205 "has_accepted_answer": false,206 "posters": [207 {208 "extras": null,209 "description": "Original Poster",210 "user": {211 "id": 82757,212 "username": "Aglowid_IT_Solutions",213 "name": "Aglowid IT Solutions",214 "avatar_template": "/user_avatar/discuss.pytorch.org/aglowid_it_solutions/{size}/75716_2.png",215 "trust_level": 0216 }217 },218 {219 "extras": "latest",220 "description": "Most Recent Poster",221 "user": {222 "id": 82779,223 "username": "Jay_G",224 "name": "Jay G",225 "avatar_template": "/user_avatar/discuss.pytorch.org/jay_g/{size}/75740_2.png",226 "trust_level": 0227 }228 }229 ]230 },231 {232 "fancy_title": "Hiring: ML Engineers @ Red Hat",233 "id": 217114,234 "title": "Hiring: ML Engineers @ Red Hat",235 "slug": "hiring-ml-engineers-red-hat",236 "posts_count": 1,237 "reply_count": 0,238 "highest_post_number": 1,239 "image_url": null,240 "created_at": "2025-02-24T20:13:02.436Z",241 "last_posted_at": "2025-02-24T20:13:02.482Z",242 "bumped": true,243 "bumped_at": "2025-02-25T14:45:06.281Z",244 "archetype": "regular",245 "unseen": false,246 "pinned": false,247 "unpinned": null,248 "visible": true,249 "closed": false,250 "archived": false,251 "bookmarked": null,252 "liked": null,253 "tags_descriptions": {},254 "like_count": 0,255 "views": 300,256 "category_id": 24,257 "featured_link": null,258 "has_accepted_answer": false,259 "posters": [260 {261 "extras": "latest single",262 "description": "Original Poster, Most Recent Poster",263 "user": {264 "id": 82907,265 "username": "mejones_redhat",266 "name": "",267 "avatar_template": "/letter_avatar_proxy/v4/letter/m/77aa72/{size}.png",268 "trust_level": 0269 }270 }271 ]272 },273 {274 "fancy_title": "Loading and retraining Peft model error ( grad and does not have a grad_fn)",275 "id": 216684,276 "title": "Loading and retraining Peft model error ( grad and does not have a grad_fn)",277 "slug": "loading-and-retraining-peft-model-error-grad-and-does-not-have-a-grad-fn",278 "posts_count": 1,279 "reply_count": 0,280 "highest_post_number": 1,281 "image_url": null,282 "created_at": "2025-02-14T17:01:40.154Z",283 "last_posted_at": "2025-02-14T17:01:40.194Z",284 "bumped": true,285 "bumped_at": "2025-02-15T02:11:43.680Z",286 "archetype": "regular",287 "unseen": false,288 "pinned": false,289 "unpinned": null,290 "visible": true,291 "closed": false,292 "archived": false,293 "bookmarked": null,294 "liked": null,295 "tags_descriptions": {},296 "like_count": 0,297 "views": 128,298 "category_id": 1,299 "featured_link": null,300 "has_accepted_answer": false,301 "posters": [302 {303 "extras": "latest single",304 "description": "Original Poster, Most Recent Poster",305 "user": {306 "id": 82364,307 "username": "Sourabh_Yadav",308 "name": "Sourabh Yadav",309 "avatar_template": "/user_avatar/discuss.pytorch.org/sourabh_yadav/{size}/75350_2.png",310 "trust_level": 1311 }312 }313 ]314 },315 {316 "fancy_title": "CNN predicts constant values for sparse amplitude regression — can’t learn true pixel values",317 "id": 220387,318 "title": "CNN predicts constant values for sparse amplitude regression — can't learn true pixel values",319 "slug": "cnn-predicts-constant-values-for-sparse-amplitude-regression-cant-learn-true-pixel-values",320 "posts_count": 10,321 "reply_count": 7,322 "highest_post_number": 10,323 "image_url": "https://discuss.pytorch.org/uploads/default/optimized/3X/a/3/a34222a0d2908564d4addbd58d278604e98dbfaa_2_1024x528.png",324 "created_at": "2025-05-27T22:23:28.017Z",325 "last_posted_at": "2025-05-30T12:25:47.403Z",326 "bumped": true,327 "bumped_at": "2025-05-30T12:25:47.403Z",328 "archetype": "regular",329 "unseen": false,330 "pinned": false,331 "unpinned": null,332 "visible": true,333 "closed": false,334 "archived": false,335 "bookmarked": null,336 "liked": null,337 "tags_descriptions": {},338 "like_count": 0,339 "views": 105,340 "category_id": 1,341 "featured_link": null,342 "has_accepted_answer": false,343 "posters": [344 {345 "extras": "latest",346 "description": "Original Poster, Most Recent Poster",347 "user": {348 "id": 84482,349 "username": "thunlok",350 "name": "",351 "avatar_template": "/letter_avatar_proxy/v4/letter/t/c4cdca/{size}.png",352 "trust_level": 0353 }354 },355 {356 "extras": null,357 "description": "Frequent Poster",358 "user": {359 "id": 18088,360 "username": "KFrank",361 "name": "K. Frank",362 "avatar_template": "/letter_avatar_proxy/v4/letter/k/ecb155/{size}.png",363 "trust_level": 2364 }365 }366 ]367 }368 ],369 "tags_descriptions": {},370 "fancy_title": "Oferta Laboral Payton con ML, Pytorch, Torchaudio, Torchvideo",371 "id": 220733,372 "title": "Oferta Laboral Payton con ML, Pytorch, Torchaudio, Torchvideo",373 "posts_count": 2,374 "created_at": "2025-06-11T15:07:01.426Z",375 "views": 108,376 "reply_count": 0,377 "like_count": 1,378 "last_posted_at": "2025-06-11T15:15:36.963Z",379 "visible": true,380 "closed": false,381 "archived": false,382 "has_summary": false,383 "archetype": "regular",384 "slug": "oferta-laboral-payton-con-ml-pytorch-torchaudio-torchvideo",385 "category_id": 24,386 "word_count": 139,387 "deleted_at": null,388 "user_id": 84655,389 "featured_link": null,390 "pinned_globally": false,391 "pinned_at": null,392 "pinned_until": null,393 "image_url": null,394 "slow_mode_seconds": 0,395 "draft": null,396 "draft_key": "topic_220733",397 "draft_sequence": null,398 "unpinned": null,399 "pinned": false,400 "current_post_number": 1,401 "highest_post_number": 2,402 "deleted_by": null,403 "actions_summary": [404 {405 "id": 4,406 "count": 0,407 "hidden": false,408 "can_act": false409 },410 {411 "id": 8,412 "count": 0,413 "hidden": false,414 "can_act": false415 },416 {417 "id": 10,418 "count": 0,419 "hidden": false,420 "can_act": false421 },422 {423 "id": 7,424 "count": 0,425 "hidden": false,426 "can_act": false427 }428 ],429 "chunk_size": 20,430 "bookmarked": false,431 "topic_timer": null,432 "message_bus_last_id": 0,433 "participant_count": 2,434 "show_read_indicator": false,435 "thumbnails": null,436 "slow_mode_enabled_until": null,437 "can_vote": false,438 "vote_count": 0,439 "user_voted": false,440 "discourse_zendesk_plugin_zendesk_id": null,441 "discourse_zendesk_plugin_zendesk_url": "https://your-url.zendesk.com/agent/tickets/",442 "details": {443 "can_edit": false,444 "notification_level": 1,445 "participants": [446 {447 "id": 3534,448 "username": "ptrblck",449 "name": "",450 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",451 "post_count": 1,452 "primary_group_name": null,453 "flair_name": null,454 "flair_url": null,455 "flair_color": null,456 "flair_bg_color": null,457 "flair_group_id": null,458 "admin": true,459 "moderator": true,460 "trust_level": 2461 },462 {463 "id": 84655,464 "username": "NataliaZagalabs",465 "name": "Natalia ",466 "avatar_template": "/letter_avatar_proxy/v4/letter/n/85e7bf/{size}.png",467 "post_count": 1,468 "primary_group_name": null,469 "flair_name": null,470 "flair_url": null,471 "flair_color": null,472 "flair_bg_color": null,473 "flair_group_id": null,474 "trust_level": 0475 }476 ],477 "created_by": {478 "id": 84655,479 "username": "NataliaZagalabs",480 "name": "Natalia ",481 "avatar_template": "/letter_avatar_proxy/v4/letter/n/85e7bf/{size}.png"482 },483 "last_poster": {484 "id": 3534,485 "username": "ptrblck",486 "name": "",487 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png"488 }489 },490 "bookmarks": []491 },492 {493 "post_stream": {494 "posts": [495 {496 "id": 471661,497 "name": "Jie Huang",498 "username": "shangxiaaabb",499 "avatar_template": "/user_avatar/discuss.pytorch.org/shangxiaaabb/{size}/76694_2.png",500 "created_at": "2025-06-11T14:11:37.246Z",501 "cooked": "<p>I ran the same code on both A100 and 4090 GPUs, but encountered an issue only on the A100.</p>\n<p><strong>Problem Description:</strong><br>\nOn the A100, GPU memory usage on the main device increases gradually with each epoch. For example, at <code>epoch=1</code>, CUDA memory usage is around <strong>75GB</strong>, but by <code>epoch=4</code>, it grows to <strong>80GB</strong>, eventually leading to an <strong>Out Of Memory (OOM)</strong> error.<br>\nHowever, the same code running on the <strong>4090</strong> does <strong>not</strong> exhibit this issue — GPU memory remains stable throughout training.<br>\nA100：CUDA Version: 12.9，Name: accelerate Version: 1.7.0，Name: torch Version: 2.7.0+cu128；<br>\n4090：CUDA Version: 12.4，Name: accelerate Version: 1.3.0；Name: torch Version: 2.5.1；<br>\nMy Code like that:</p>\n<pre data-code-wrap=\"python\"><code class=\"lang-python\">import sys\nimport os\nimport warnings\nsys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), '../')))\nsys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), './')))\nos.environ[\"TOKENIZERS_PARALLELISM\"] = \"false\"\nos.environ['CURL_CA_BUNDLE'] = ''\nwarnings.filterwarnings(\"ignore\")\n\nfrom tqdm import tqdm\nimport numpy as np\nimport torch\nfrom torch.utils.data import DataLoader\nfrom accelerate import Accelerator\nfrom accelerate.utils import DistributedDataParallelKwargs\nfrom sklearn.metrics import accuracy_score, f1_score, precision_score, recall_score, classification_report\n\nfrom config import MyConfig\nfrom model.model import MyModel\nfrom data_loader import MyDataset\nfrom evaluate import compute_acc_f1_recall\n\nconfig = MyConfig()\n\ndef train(epoch, model, optimizer, loss_function, data_loader, accelerator, lr_scheduler):\n model.train()\n progress_bar = tqdm(total=len(data_loader), \n disable=not accelerator.is_main_process, \n desc=f\"Epoch-TRAIN {epoch}\")\n\n for i, batch in enumerate(data_loader):\n with accelerator.accumulate(model):\n image_data, label_data, text_data, bbox_data, text_padding, bbox_padding = batch\n label_data = label_data.to(accelerator.device)\n image_data = image_data.to(accelerator.device)\n bbox_data = bbox_data.to(accelerator.device)\n\n if config.bbox_embedding:\n bbox_padding = bbox_padding.to(accelerator.device)\n out = model(image_data, bbox_data, text_data, bbox_padding)\n else:\n out = model(image_data, bbox_data, text_data)\n\n bbox_padding_mask = bbox_padding.to(dtype=torch.bool, device=accelerator.device)\n valid_mask = ~bbox_padding_mask\n valid_out = out[valid_mask]\n valid_labels = label_data[valid_mask]\n valid_labels = torch.argmax(valid_labels, dim=-1)\n\n loss = loss_function(valid_out, valid_labels)\n\n accelerator.backward(loss)\n if accelerator.sync_gradients:\n accelerator.clip_grad_norm_(model.parameters(), 1.0)\n optimizer.step()\n lr_scheduler.step()\n optimizer.zero_grad()\n\n if accelerator.is_main_process:\n acc, f1, recall, balanced_acc = compute_acc_f1_recall(preds= valid_out, labels= valid_labels)\n progress_bar.update(1)\n if accelerator.is_main_process:\n logs = {\"Train/loss\": loss.item(), \n \"Train/lr\": lr_scheduler.get_last_lr()[0],\n \"Train/ACC\": acc, \n \"Train/F1-Micro\": f1, \n \"Train/Recall\": recall, \n \"Train/ACC-Balance\": balanced_acc}\n progress_bar.set_postfix(\n loss=loss.item(), lr=lr_scheduler.get_last_lr()[0],\n acc=acc, f1=f1)\n accelerator.log(logs)\n\n\ndef test(epoch, model, loss_function, data_loader, accelerator):\n model.eval()\n progress_bar = tqdm(total=len(data_loader),\n disable=not accelerator.is_main_process, \n desc=f\"Epoch-TEST {epoch}\")\n\n mean_acc, mean_f1 = 0.0, 0.0\n with torch.no_grad():\n for i, batch in enumerate(data_loader):\n image_data, label_data, text_data, bbox_data, text_padding, bbox_padding = batch\n label_data = label_data.to(accelerator.device)\n image_data = image_data.to(accelerator.device)\n bbox_data = bbox_data.to(accelerator.device)\n\n if config.bbox_embedding:\n bbox_padding = bbox_padding.to(accelerator.device)\n out = model(image_data, bbox_data, text_data, bbox_padding)\n else:\n out = model(image_data, bbox_data, text_data)\n\n bbox_padding_mask = bbox_padding.to(dtype=torch.bool, device=accelerator.device)\n valid_mask = ~bbox_padding_mask \n valid_out = out[valid_mask] \n valid_labels = label_data[valid_mask] \n valid_labels = torch.argmax(valid_labels, dim=-1)\n\n loss = loss_function(valid_out, valid_labels)\n acc, f1, recall, balanced_acc = compute_acc_f1_recall(preds= valid_out, labels= valid_labels)\n\n progress_bar.update(1)\n logs = {\"Test/loss\": loss.item(), \n \"Test/ACC\": acc, \n \"Test/F1-Micro\": f1, \n \"Test/Recall\": recall, \n \"Test/ACC-Balance\": balanced_acc}\n progress_bar.set_postfix(\n loss=loss.item(),\n acc=acc, f1=f1)\n mean_acc += acc\n mean_f1 += f1\n accelerator.log(logs)\n return mean_acc/len(data_loader), mean_f1/ len(data_loader)\n\n\n@torch.no_grad\ndef evaluate(model, pth_path, data_loader, accelerator, data_type: str='test'):\n if config.data_type== 'latex':\n labels =....\n index_to_label = {idx: label for idx, label in enumerate(labels)}\n all_preds, all_labels = [], []\n if pth_path is not None:\n checkpoint = torch.load(pth_path, map_location=accelerator.device, weights_only= False)\n try:\n model.module.load_state_dict(checkpoint['model_state_dict'])\n except Exception:\n model.load_state_dict(checkpoint['model_state_dict'])\n model.eval()\n\n with tqdm(total= len(data_loader), desc=f'Eva-{data_type}') as pbar:\n for i, batch in enumerate(data_loader):\n image_data, label_data, text_data, bbox_data, text_padding, bbox_padding = batch\n label_data = label_data.to(accelerator.device)\n image_data = image_data.to(accelerator.device)\n bbox_data = bbox_data.to(accelerator.device)\n\n if config.bbox_embedding:\n bbox_padding = bbox_padding.to(accelerator.device)\n out = model(image_data, bbox_data, text_data, bbox_padding)\n else:\n out = model(image_data, bbox_data, text_data)\n\n bbox_padding_mask = bbox_padding.to(dtype=torch.bool, device=accelerator.device)\n valid_mask = ~bbox_padding_mask \n valid_out = out[valid_mask]\n valid_labels = label_data[valid_mask]\n valid_labels = torch.argmax(valid_labels, dim=-1)\n\n preds = torch.argmax(valid_out, dim=-1) \n \n all_preds.append(preds.detach().cpu().numpy())\n all_labels.append(valid_labels.detach().cpu().numpy())\n pbar.update(1)\n ......\n\ndef main(image_model_name= None, features_fushion=None):\n if image_model_name:\n config.image_model_name = image_model_name\n config.features_fushion = features_fushion\n config.output_dir = f\"{config.output_dir}/{config.features_fushion}\"\n\n kwargs_handlers=[DistributedDataParallelKwargs(find_unused_parameters=False)]\n log_writing = \"tensorboard\" # if config.small_dataset else [\"tensorboard\", \"wandb\"]\n accelerator = Accelerator(mixed_precision= config.mixed_precision, \n gradient_accumulation_steps= config.gradient_accumulation_steps,\n log_with= log_writing,\n project_dir=os.path.join(config.output_dir, f\"logs\"),\n kwargs_handlers= kwargs_handlers\n )\n if accelerator.is_main_process:\n os.makedirs(config.output_dir, exist_ok=True)\n accelerator.init_trackers(f\"Train-{config.pred_heads}\")\n \n # data\n train_dataset = MyDataset(...)\n test_dataset = MyDataset(...)\n train_dataloader = DataLoader(train_dataset, batch_size= config.batch_size, \n collate_fn= train_dataset.collate_fn, num_workers= 4)\n test_dataloader = DataLoader(test_dataset, batch_size= config.batch_size,\n collate_fn= test_dataset.collate_fn)\n\n # model\n model = MyModel(...)\n \n if config.lora:\n optimizer = torch.optim.AdamW([\n {'params': model.image_model.parameters(), 'lr': 2e-4, 'weight_decay': 1e-2},\n {'params': model.text_model.parameters(), 'lr': 4e-5, 'weight_decay': 1e-2},\n {'params': [p for n, p in model.named_parameters() \n if 'image_model' not in n and 'text_model' not in n]},\n ], lr= config.learning_rate)\n else:\n optimizer = torch.optim.AdamW(model.parameters(), lr= config.learning_rate)\n lr_scheduler = torch.optim.lr_scheduler.LinearLR(optimizer,\n start_factor=0.1,\n total_iters= 10 * len(train_dataloader))\n loss_function = torch.nn.CrossEntropyLoss()\n\n model, optimizer, train_dataloader, test_dataloader, lr_scheduler = accelerator.prepare(model, optimizer, \n train_dataloader, \n test_dataloader, \n lr_scheduler)\n \n best_acc, best_f1 = 0.0, 0.0\n for epoch in range(config.epochs):\n train(epoch, model, optimizer, loss_function, train_dataloader, \n accelerator, lr_scheduler)\n if accelerator.is_main_process:\n mean_acc, mean_f1 = test(epoch, model,loss_function, test_dataloader, accelerator)\n if mean_acc>= best_acc+ 0.01:\n best_acc = mean_acc\n # model_to_save = accelerator.unwrap_model(model)\n model_to_save = accelerator.get_state_dict(model)\n accelerator.save({\n 'model_state_dict': model_to_save,\n 'optimizer_state_dict': optimizer.state_dict(),\n 'lr_scheduler_state_dict': lr_scheduler.state_dict(),\n 'epoch': epoch,\n 'best_acc': best_acc,\n 'config': vars(config)\n }, f\"{config.output_dir}/model_acc_best.pth\")\n torch.cuda.empty_cache()\n if mean_f1>= best_f1+ 0.01:\n best_f1 = mean_f1\n # model_to_save = accelerator.unwrap_model(model)\n model_to_save = accelerator.get_state_dict(model)\n accelerator.save({\n 'model_state_dict': model_to_save,\n 'optimizer_state_dict': optimizer.state_dict(),\n 'lr_scheduler_state_dict': lr_scheduler.state_dict(),\n 'epoch': epoch,\n 'best_f1': best_f1,\n 'config': vars(config)\n }, f\"{config.output_dir}/model_f1_best.pth\")\n torch.cuda.empty_cache()\n\n if epoch% 5== 0 or epoch== config.epochs- 1:\n for _ in [(\"TEST-ACC\", f\"{config.output_dir}/model_acc_best.pth\"), (\"TEST-F1\", f\"{config.output_dir}/model_f1_best.pth\")]:\n checkpoint = torch.load(_[1], map_location=accelerator.device, weights_only= False)\n try:\n model.module.load_state_dict(checkpoint['model_state_dict'])\n except Exception:\n model.load_state_dict(checkpoint['model_state_dict'])\n evaluate(model, None, data_loader= test_dataloader, accelerator=accelerator,\n data_type= _[0])\n del checkpoint\n torch.cuda.empty_cache()\n torch.cuda.empty_cache()\n \n accelerator.end_training()\n\nif __name__ == '__main__':\n # CUDA_VISIBLE_DEVICES=1,2,3 accelerate launch --num_processes=3 train.py\n # CUDA_VISIBLE_DEVICES=2,3 accelerate launch --num_processes=2 train.py\n import argparse\n parser = argparse.ArgumentParser(description=\"Run main function with parameters\")\n parser.add_argument('--image_model_name', type=str, default=None, help='Name of the image model')\n parser.add_argument('--features_fusion', type=str, default=None, help='Type of features fusion')\n args = parser.parse_args()\n main(args.image_model_name, args.features_fusion)\n</code></pre>",502 "post_number": 1,503 "post_type": 1,504 "posts_count": 3,505 "updated_at": "2025-06-11T14:14:53.257Z",506 "reply_count": 0,507 "reply_to_post_number": null,508 "quote_count": 0,509 "incoming_link_count": 22,510 "reads": 6,511 "readers_count": 5,512 "score": 96.2,513 "yours": false,514 "topic_id": 220725,515 "topic_slug": "memory-issues-when-running-the-same-code-on-a100-and-4090ti-when-using-accelerate",516 "display_username": "Jie Huang",517 "primary_group_name": null,518 "flair_name": null,519 "flair_url": null,520 "flair_bg_color": null,521 "flair_color": null,522 "flair_group_id": null,523 "badges_granted": [],524 "version": 2,525 "can_edit": false,526 "can_delete": false,527 "can_recover": false,528 "can_see_hidden_post": false,529 "can_wiki": false,530 "read": true,531 "user_title": "",532 "bookmarked": false,533 "actions_summary": [],534 "moderator": false,535 "admin": false,536 "staff": false,537 "user_id": 83877,538 "hidden": false,539 "trust_level": 1,540 "deleted_at": null,541 "user_deleted": false,542 "edit_reason": null,543 "can_view_edit_history": true,544 "wiki": false,545 "post_url": "/t/memory-issues-when-running-the-same-code-on-a100-and-4090ti-when-using-accelerate/220725/1",546 "can_accept_answer": false,547 "can_unaccept_answer": false,548 "accepted_answer": false,549 "topic_accepted_answer": null,550 "can_vote": false551 },552 {553 "id": 471663,554 "name": "",555 "username": "ptrblck",556 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",557 "created_at": "2025-06-11T14:22:35.617Z",558 "cooked": "<p>It seems you are using different libs in your envs (PyTorch, accelerate). Do you see the same behavior on your 4090 when updating to the latest stack?</p>",559 "post_number": 2,560 "post_type": 1,561 "posts_count": 3,562 "updated_at": "2025-06-11T14:22:35.617Z",563 "reply_count": 1,564 "reply_to_post_number": null,565 "quote_count": 0,566 "incoming_link_count": 0,567 "reads": 4,568 "readers_count": 3,569 "score": 5.8,570 "yours": false,571 "topic_id": 220725,572 "topic_slug": "memory-issues-when-running-the-same-code-on-a100-and-4090ti-when-using-accelerate",573 "display_username": "",574 "primary_group_name": null,575 "flair_name": null,576 "flair_url": null,577 "flair_bg_color": null,578 "flair_color": null,579 "flair_group_id": null,580 "badges_granted": [],581 "version": 1,582 "can_edit": false,583 "can_delete": false,584 "can_recover": false,585 "can_see_hidden_post": false,586 "can_wiki": false,587 "read": true,588 "user_title": "",589 "bookmarked": false,590 "actions_summary": [],591 "moderator": true,592 "admin": true,593 "staff": true,594 "user_id": 3534,595 "hidden": false,596 "trust_level": 2,597 "deleted_at": null,598 "user_deleted": false,599 "edit_reason": null,600 "can_view_edit_history": true,601 "wiki": false,602 "post_url": "/t/memory-issues-when-running-the-same-code-on-a100-and-4090ti-when-using-accelerate/220725/2",603 "can_accept_answer": false,604 "can_unaccept_answer": false,605 "accepted_answer": false,606 "topic_accepted_answer": null607 },608 {609 "id": 471665,610 "name": "Jie Huang",611 "username": "shangxiaaabb",612 "avatar_template": "/user_avatar/discuss.pytorch.org/shangxiaaabb/{size}/76694_2.png",613 "created_at": "2025-06-11T14:26:13.424Z",614 "cooked": "<p>When I run the code on the <strong>4090</strong>, the GPU memory usage remains stable at around <strong>29GB</strong> throughout training, unlike on the <strong>A100</strong> where it keeps increasing over epochs.<br>\n<div class=\"lightbox-wrapper\"><a class=\"lightbox\" href=\"https://discuss.pytorch.org/uploads/default/original/3X/1/f/1f887098b4e2a32f81998e4c5b42736d5652fef7.png\" data-download-href=\"https://discuss.pytorch.org/uploads/default/1f887098b4e2a32f81998e4c5b42736d5652fef7\" title=\"image\"><img src=\"https://discuss.pytorch.org/uploads/default/original/3X/1/f/1f887098b4e2a32f81998e4c5b42736d5652fef7.png\" alt=\"image\" data-base62-sha1=\"4uX7BahLGrTb7FGCOlRfceDkOnd\" width=\"690\" height=\"383\" data-dominant-color=\"EEE9E9\"><div class=\"meta\"><svg class=\"fa d-icon d-icon-far-image svg-icon\" aria-hidden=\"true\"><use href=\"#far-image\"></use></svg><span class=\"filename\">image</span><span class=\"informations\">770×428 12.4 KB</span><svg class=\"fa d-icon d-icon-discourse-expand svg-icon\" aria-hidden=\"true\"><use href=\"#discourse-expand\"></use></svg></div></a></div></p>\n<p><strong>Wait I try it</strong></p>",615 "post_number": 3,616 "post_type": 1,617 "posts_count": 3,618 "updated_at": "2025-06-11T14:27:35.379Z",619 "reply_count": 0,620 "reply_to_post_number": 2,621 "quote_count": 0,622 "incoming_link_count": 3,623 "reads": 4,624 "readers_count": 3,625 "score": 10.8,626 "yours": false,627 "topic_id": 220725,628 "topic_slug": "memory-issues-when-running-the-same-code-on-a100-and-4090ti-when-using-accelerate",629 "display_username": "Jie Huang",630 "primary_group_name": null,631 "flair_name": null,632 "flair_url": null,633 "flair_bg_color": null,634 "flair_color": null,635 "flair_group_id": null,636 "badges_granted": [],637 "version": 3,638 "can_edit": false,639 "can_delete": false,640 "can_recover": false,641 "can_see_hidden_post": false,642 "can_wiki": false,643 "read": true,644 "user_title": "",645 "reply_to_user": {646 "id": 3534,647 "username": "ptrblck",648 "name": "",649 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png"650 },651 "bookmarked": false,652 "actions_summary": [],653 "moderator": false,654 "admin": false,655 "staff": false,656 "user_id": 83877,657 "hidden": false,658 "trust_level": 1,659 "deleted_at": null,660 "user_deleted": false,661 "edit_reason": null,662 "can_view_edit_history": true,663 "wiki": false,664 "post_url": "/t/memory-issues-when-running-the-same-code-on-a100-and-4090ti-when-using-accelerate/220725/3",665 "can_accept_answer": false,666 "can_unaccept_answer": false,667 "accepted_answer": false,668 "topic_accepted_answer": null669 }670 ],671 "stream": [672 471661,673 471663,674 471665675 ]676 },677 "timeline_lookup": [678 [679 1,680 136681 ]682 ],683 "suggested_topics": [684 {685 "fancy_title": "Torch::jit::load error file_name!=nullptr",686 "id": 217824,687 "title": "Torch::jit::load error file_name!=nullptr",688 "slug": "torch-load-error-file-name-nullptr",689 "posts_count": 1,690 "reply_count": 0,691 "highest_post_number": 1,692 "image_url": null,693 "created_at": "2025-03-13T22:18:25.486Z",694 "last_posted_at": "2025-03-13T22:18:25.523Z",695 "bumped": true,696 "bumped_at": "2025-03-13T22:18:25.523Z",697 "archetype": "regular",698 "unseen": false,699 "pinned": false,700 "unpinned": null,701 "visible": true,702 "closed": false,703 "archived": false,704 "bookmarked": null,705 "liked": null,706 "unicode_title": "Torch::jit::load error file_name!=nullptr",707 "tags_descriptions": {},708 "like_count": 0,709 "views": 37,710 "category_id": 1,711 "featured_link": null,712 "has_accepted_answer": false,713 "posters": [714 {715 "extras": "latest single",716 "description": "Original Poster, Most Recent Poster",717 "user": {718 "id": 83260,719 "username": "Sanjib",720 "name": "Sanjib",721 "avatar_template": "/letter_avatar_proxy/v4/letter/s/b4bc9f/{size}.png",722 "trust_level": 1723 }724 }725 ]726 },727 {728 "fancy_title": "Segmentation fault in 50K x 50K torch.matmul (fixed)",729 "id": 218513,730 "title": "Segmentation fault in 50K x 50K torch.matmul (fixed)",731 "slug": "segmentation-fault-in-50k-x-50k-torch-matmul-fixed",732 "posts_count": 2,733 "reply_count": 0,734 "highest_post_number": 2,735 "image_url": null,736 "created_at": "2025-04-02T05:10:32.606Z",737 "last_posted_at": "2025-04-02T14:44:43.440Z",738 "bumped": true,739 "bumped_at": "2025-04-02T14:44:43.440Z",740 "archetype": "regular",741 "unseen": false,742 "pinned": false,743 "unpinned": null,744 "visible": true,745 "closed": false,746 "archived": false,747 "bookmarked": null,748 "liked": null,749 "tags_descriptions": {},750 "like_count": 0,751 "views": 40,752 "category_id": 1,753 "featured_link": null,754 "has_accepted_answer": false,755 "posters": [756 {757 "extras": "latest single",758 "description": "Original Poster, Most Recent Poster",759 "user": {760 "id": 77475,761 "username": "Sanjeev_Rampal",762 "name": "Sanjeev Rampal",763 "avatar_template": "/user_avatar/discuss.pytorch.org/sanjeev_rampal/{size}/62377_2.png",764 "trust_level": 1765 }766 }767 ]768 },769 {770 "fancy_title": "How can I save the proper embeddings and weights after training?",771 "id": 212912,772 "title": "How can I save the proper embeddings and weights after training?",773 "slug": "how-can-i-save-the-proper-embeddings-and-weights-after-training",774 "posts_count": 2,775 "reply_count": 0,776 "highest_post_number": 2,777 "image_url": null,778 "created_at": "2024-11-13T08:45:07.754Z",779 "last_posted_at": "2024-11-13T12:37:00.417Z",780 "bumped": true,781 "bumped_at": "2024-11-13T15:11:25.864Z",782 "archetype": "regular",783 "unseen": false,784 "pinned": false,785 "unpinned": null,786 "visible": true,787 "closed": false,788 "archived": false,789 "bookmarked": null,790 "liked": null,791 "tags_descriptions": {},792 "like_count": 0,793 "views": 68,794 "category_id": 1,795 "featured_link": null,796 "has_accepted_answer": false,797 "posters": [798 {799 "extras": "latest single",800 "description": "Original Poster, Most Recent Poster",801 "user": {802 "id": 72736,803 "username": "songsong0425",804 "name": "Songyeon Lee",805 "avatar_template": "/user_avatar/discuss.pytorch.org/songsong0425/{size}/67200_2.png",806 "trust_level": 1807 }808 }809 ]810 },811 {812 "fancy_title": "Does PyTorch `.to(device)` propagate gradients back to original device?",813 "id": 217641,814 "title": "Does PyTorch `.to(device)` propagate gradients back to original device?",815 "slug": "does-pytorch-to-device-propagate-gradients-back-to-original-device",816 "posts_count": 5,817 "reply_count": 1,818 "highest_post_number": 5,819 "image_url": null,820 "created_at": "2025-03-10T05:14:24.681Z",821 "last_posted_at": "2025-03-12T07:57:27.711Z",822 "bumped": true,823 "bumped_at": "2025-03-12T07:57:27.711Z",824 "archetype": "regular",825 "unseen": false,826 "pinned": false,827 "unpinned": null,828 "visible": true,829 "closed": false,830 "archived": false,831 "bookmarked": null,832 "liked": null,833 "tags_descriptions": {},834 "like_count": 0,835 "views": 195,836 "category_id": 1,837 "featured_link": null,838 "has_accepted_answer": false,839 "posters": [840 {841 "extras": "latest",842 "description": "Original Poster, Most Recent Poster",843 "user": {844 "id": 83164,845 "username": "gammag4",846 "name": "Gabriel Maia",847 "avatar_template": "/user_avatar/discuss.pytorch.org/gammag4/{size}/76063_2.png",848 "trust_level": 1849 }850 },851 {852 "extras": null,853 "description": "Frequent Poster",854 "user": {855 "id": 64488,856 "username": "Naming-isDifficult",857 "name": "",858 "avatar_template": "/user_avatar/discuss.pytorch.org/naming-isdifficult/{size}/58644_2.png",859 "trust_level": 2860 }861 },862 {863 "extras": null,864 "description": "Frequent Poster",865 "user": {866 "id": 3534,867 "username": "ptrblck",868 "name": "",869 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",870 "admin": true,871 "moderator": true,872 "trust_level": 2873 }874 }875 ]876 },877 {878 "fancy_title": "Why is torch.rand different on different devices for big indices",879 "id": 219976,880 "title": "Why is torch.rand different on different devices for big indices",881 "slug": "why-is-torch-rand-different-on-different-devices-for-big-indices",882 "posts_count": 2,883 "reply_count": 0,884 "highest_post_number": 2,885 "image_url": null,886 "created_at": "2025-05-13T03:29:50.431Z",887 "last_posted_at": "2025-05-13T10:56:45.681Z",888 "bumped": true,889 "bumped_at": "2025-05-13T10:56:45.681Z",890 "archetype": "regular",891 "unseen": false,892 "pinned": false,893 "unpinned": null,894 "visible": true,895 "closed": false,896 "archived": false,897 "bookmarked": null,898 "liked": null,899 "tags_descriptions": {},900 "like_count": 1,901 "views": 39,902 "category_id": 1,903 "featured_link": null,904 "has_accepted_answer": false,905 "posters": [906 {907 "extras": null,908 "description": "Original Poster",909 "user": {910 "id": 53070,911 "username": "Jackmin801",912 "name": "Jackmin801",913 "avatar_template": "/user_avatar/discuss.pytorch.org/jackmin801/{size}/46475_2.png",914 "trust_level": 1915 }916 },917 {918 "extras": "latest",919 "description": "Most Recent Poster",920 "user": {921 "id": 3534,922 "username": "ptrblck",923 "name": "",924 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",925 "admin": true,926 "moderator": true,927 "trust_level": 2928 }929 }930 ]931 }932 ],933 "tags_descriptions": {},934 "fancy_title": "Memory issues when running the same code on A100 and 4090Ti when using accelerate",935 "id": 220725,936 "title": "Memory issues when running the same code on A100 and 4090Ti when using accelerate",937 "posts_count": 3,938 "created_at": "2025-06-11T14:11:37.200Z",939 "views": 41,940 "reply_count": 1,941 "like_count": 0,942 "last_posted_at": "2025-06-11T14:26:13.424Z",943 "visible": true,944 "closed": false,945 "archived": false,946 "has_summary": false,947 "archetype": "regular",948 "slug": "memory-issues-when-running-the-same-code-on-a100-and-4090ti-when-using-accelerate",949 "category_id": 1,950 "word_count": 1210,951 "deleted_at": null,952 "user_id": 83877,953 "featured_link": null,954 "pinned_globally": false,955 "pinned_at": null,956 "pinned_until": null,957 "image_url": null,958 "slow_mode_seconds": 0,959 "draft": null,960 "draft_key": "topic_220725",961 "draft_sequence": null,962 "unpinned": null,963 "pinned": false,964 "current_post_number": 1,965 "highest_post_number": 3,966 "deleted_by": null,967 "actions_summary": [968 {969 "id": 4,970 "count": 0,971 "hidden": false,972 "can_act": false973 },974 {975 "id": 8,976 "count": 0,977 "hidden": false,978 "can_act": false979 },980 {981 "id": 10,982 "count": 0,983 "hidden": false,984 "can_act": false985 },986 {987 "id": 7,988 "count": 0,989 "hidden": false,990 "can_act": false991 }992 ],993 "chunk_size": 20,994 "bookmarked": false,995 "topic_timer": null,996 "message_bus_last_id": 0,997 "participant_count": 2,998 "show_read_indicator": false,999 "thumbnails": null,1000 "slow_mode_enabled_until": null,1001 "can_vote": false,1002 "vote_count": 0,1003 "user_voted": false,1004 "discourse_zendesk_plugin_zendesk_id": null,1005 "discourse_zendesk_plugin_zendesk_url": "https://your-url.zendesk.com/agent/tickets/",1006 "details": {1007 "can_edit": false,1008 "notification_level": 1,1009 "participants": [1010 {1011 "id": 83877,1012 "username": "shangxiaaabb",1013 "name": "Jie Huang",1014 "avatar_template": "/user_avatar/discuss.pytorch.org/shangxiaaabb/{size}/76694_2.png",1015 "post_count": 2,1016 "primary_group_name": null,1017 "flair_name": null,1018 "flair_url": null,1019 "flair_color": null,1020 "flair_bg_color": null,1021 "flair_group_id": null,1022 "trust_level": 11023 },1024 {1025 "id": 3534,1026 "username": "ptrblck",1027 "name": "",1028 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",1029 "post_count": 1,1030 "primary_group_name": null,1031 "flair_name": null,1032 "flair_url": null,1033 "flair_color": null,1034 "flair_bg_color": null,1035 "flair_group_id": null,1036 "admin": true,1037 "moderator": true,1038 "trust_level": 21039 }1040 ],1041 "created_by": {1042 "id": 83877,1043 "username": "shangxiaaabb",1044 "name": "Jie Huang",1045 "avatar_template": "/user_avatar/discuss.pytorch.org/shangxiaaabb/{size}/76694_2.png"1046 },1047 "last_poster": {1048 "id": 83877,1049 "username": "shangxiaaabb",1050 "name": "Jie Huang",1051 "avatar_template": "/user_avatar/discuss.pytorch.org/shangxiaaabb/{size}/76694_2.png"1052 }1053 },1054 "bookmarks": []1055 },1056 {1057 "post_stream": {1058 "posts": [1059 {1060 "id": 468976,1061 "name": "nafe alhmdan",1062 "username": "nafe_alhmdan",1063 "avatar_template": "/user_avatar/discuss.pytorch.org/nafe_alhmdan/{size}/75984_2.png",1064 "created_at": "2025-04-10T16:04:34.876Z",1065 "cooked": "<p>C:\\Users\\USER\\Desktop\\RVC1006Nvidia>runtime\\python.exe infer-web.py --pycmd runtime\\python.exe --port 7897<br>\nC:\\Users\\USER\\Desktop\\RVC1006Nvidia\\runtime\\lib\\site-packages\\torch\\cuda_<em>init</em>_.py:173: UserWarning:<br>\nNVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.<br>\nThe current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90 compute_37.<br>\nIf you want to use the NVIDIA GeForce RTX 5090 GPU with PyTorch, please check the instructions at <a href=\"https://pytorch.org/get-started/locally/\" class=\"inline-onebox\" rel=\"noopener nofollow ugc\">Start Locally | PyTorch</a></p>\n<p>warnings.warn(incompatible_device_warn.format(device_name, capability, \" \".join(arch_list), device_name))<br>\n2025-04-10 18:55:22 | INFO | configs.config | Found GPU NVIDIA GeForce RTX 5090<br>\nis_half:True, device:cuda:0</p>",1066 "post_number": 1,1067 "post_type": 1,1068 "posts_count": 11,1069 "updated_at": "2025-04-10T16:04:34.876Z",1070 "reply_count": 0,1071 "reply_to_post_number": null,1072 "quote_count": 0,1073 "incoming_link_count": 10949,1074 "reads": 36,1075 "readers_count": 35,1076 "score": 53597.2,1077 "yours": false,1078 "topic_id": 218954,1079 "topic_slug": "nvidia-geforce-rtx-5090",1080 "display_username": "nafe alhmdan",1081 "primary_group_name": null,1082 "flair_name": null,1083 "flair_url": null,1084 "flair_bg_color": null,1085 "flair_color": null,1086 "flair_group_id": null,1087 "badges_granted": [],1088 "version": 1,1089 "can_edit": false,1090 "can_delete": false,1091 "can_recover": false,1092 "can_see_hidden_post": false,1093 "can_wiki": false,1094 "link_counts": [1095 {1096 "url": "https://pytorch.org/get-started/locally/",1097 "internal": false,1098 "reflection": false,1099 "title": "Start Locally | PyTorch",1100 "clicks": 2411101 }1102 ],1103 "read": true,1104 "user_title": null,1105 "bookmarked": false,1106 "actions_summary": [],1107 "moderator": false,1108 "admin": false,1109 "staff": false,1110 "user_id": 83067,1111 "hidden": false,1112 "trust_level": 1,1113 "deleted_at": null,1114 "user_deleted": false,1115 "edit_reason": null,1116 "can_view_edit_history": true,1117 "wiki": false,1118 "post_url": "/t/nvidia-geforce-rtx-5090/218954/1",1119 "can_accept_answer": false,1120 "can_unaccept_answer": false,1121 "accepted_answer": false,1122 "topic_accepted_answer": null,1123 "can_vote": false1124 },1125 {1126 "id": 468977,1127 "name": "",1128 "username": "ptrblck",1129 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",1130 "created_at": "2025-04-10T16:06:14.959Z",1131 "cooked": "<p>Install the latest nightly binaries with CUDA 12.8 and it will work as the Blackwell support requires CUDA >= 12.8.</p>",1132 "post_number": 2,1133 "post_type": 1,1134 "posts_count": 11,1135 "updated_at": "2025-04-10T16:06:14.959Z",1136 "reply_count": 0,1137 "reply_to_post_number": null,1138 "quote_count": 0,1139 "incoming_link_count": 36,1140 "reads": 38,1141 "readers_count": 37,1142 "score": 187.6,1143 "yours": false,1144 "topic_id": 218954,1145 "topic_slug": "nvidia-geforce-rtx-5090",1146 "display_username": "",1147 "primary_group_name": null,1148 "flair_name": null,1149 "flair_url": null,1150 "flair_bg_color": null,1151 "flair_color": null,1152 "flair_group_id": null,1153 "badges_granted": [],1154 "version": 1,1155 "can_edit": false,1156 "can_delete": false,1157 "can_recover": false,1158 "can_see_hidden_post": false,1159 "can_wiki": false,1160 "read": true,1161 "user_title": "",1162 "bookmarked": false,1163 "actions_summary": [],1164 "moderator": true,1165 "admin": true,1166 "staff": true,1167 "user_id": 3534,1168 "hidden": false,1169 "trust_level": 2,1170 "deleted_at": null,1171 "user_deleted": false,1172 "edit_reason": null,1173 "can_view_edit_history": true,1174 "wiki": false,1175 "post_url": "/t/nvidia-geforce-rtx-5090/218954/2",1176 "can_accept_answer": false,1177 "can_unaccept_answer": false,1178 "accepted_answer": false,1179 "topic_accepted_answer": null1180 },1181 {1182 "id": 468979,1183 "name": "nafe alhmdan",1184 "username": "nafe_alhmdan",1185 "avatar_template": "/user_avatar/discuss.pytorch.org/nafe_alhmdan/{size}/75984_2.png",1186 "created_at": "2025-04-10T16:42:59.136Z",1187 "cooked": "<p>Python 3.12.8 (tags/v3.12.8:2dc476b, Dec 3 2024, 19:30:04) [MSC v.1942 64 bit (AMD64)] on win32<br>\nType “help”, “copyright”, “credits” or “license” for more information.</p>\n<blockquote>\n<blockquote>\n<blockquote>\n<p>import torch<br>\nprint(torch.<strong>version</strong>)<br>\n2.8.0.dev20250408+cu128</p>\n</blockquote>\n</blockquote>\n</blockquote>",1188 "post_number": 3,1189 "post_type": 1,1190 "posts_count": 11,1191 "updated_at": "2025-04-10T16:42:59.136Z",1192 "reply_count": 0,1193 "reply_to_post_number": null,1194 "quote_count": 0,1195 "incoming_link_count": 46,1196 "reads": 36,1197 "readers_count": 35,1198 "score": 237.2,1199 "yours": false,1200 "topic_id": 218954,