Anurag1734/cuda-error-resolution-analysis
07
1[2 {3 "post_stream": {4 "posts": [5 {6 "id": 336629,7 "name": "Federico Ottomano",8 "username": "Federico_Ottomano",9 "avatar_template": "/user_avatar/discuss.pytorch.org/federico_ottomano/{size}/43102_2.png",10 "created_at": "2022-03-17T17:29:45.243Z",11 "cooked": "<p>I have some perplexities about the implementation of Variational autoencoder loss. This is the one I’ve been using so far:</p>\n<pre><code>def vae_loss(recon_loss, mu, logvar): \n \n KLD = -0.5 * torch.sum(1 + logvar - mu.pow(2) - logvar.exp(),dim=1)\n \n return recon_loss + KLD\n</code></pre>\n<p>After having noticed problems in my loss convergence, even in simple tasks of <code>1d</code> vectors reconstruction, I started googling around and I have find a variation of this:</p>\n<pre><code>def vae_loss(recon_loss, mu, logvar):\n \n KLD = torch.mean(-0.5 * torch.sum(1 + logvar - mu.pow(2) - logvar.exp(),dim=1),dim=0)\n\n return recon_loss + KLD\n</code></pre>\n<p>With this second vae loss the performance increases noticeably, and it’s clearly visible in reconstructed vectors as well. What I’m doing with the second implementation is taking the average over batch of samples I guess.</p>\n<p>My perplexity arises from the fact that I’ve found these two separate implementations on different blogposts and I don’t know which one is correct.</p>",12 "post_number": 1,13 "post_type": 1,14 "posts_count": 6,15 "updated_at": "2022-03-17T17:29:45.243Z",16 "reply_count": 0,17 "reply_to_post_number": null,18 "quote_count": 0,19 "incoming_link_count": 18661,20 "reads": 149,21 "readers_count": 148,22 "score": 92924.6,23 "yours": false,24 "topic_id": 146750,25 "topic_slug": "correct-implementation-of-vae-loss",26 "display_username": "Federico Ottomano",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": 1,35 "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 {45 "id": 2,46 "count": 247 }48 ],49 "moderator": false,50 "admin": false,51 "staff": false,52 "user_id": 49912,53 "hidden": false,54 "trust_level": 1,55 "deleted_at": null,56 "user_deleted": false,57 "edit_reason": null,58 "can_view_edit_history": true,59 "wiki": false,60 "post_url": "/t/correct-implementation-of-vae-loss/146750/1",61 "can_accept_answer": false,62 "can_unaccept_answer": false,63 "accepted_answer": false,64 "topic_accepted_answer": null,65 "can_vote": false66 },67 {68 "id": 336646,69 "name": "Thomas V",70 "username": "tom",71 "avatar_template": "/user_avatar/discuss.pytorch.org/tom/{size}/3162_2.png",72 "created_at": "2022-03-17T18:55:59.214Z",73 "cooked": "<p>If you don’t reduce over the batch dimension in the loss function, you are bound to do so elsewhere because taking gradients only works for scalar losses.</p>\n<p>Typically, we might think of the optimization as minimizing the expected loss function on the training distribution, so taking the sample mean would seem to be a good way to reduce the batch dimension.</p>\n<p>Best regards</p>\n<p>Thomas</p>",74 "post_number": 2,75 "post_type": 1,76 "posts_count": 6,77 "updated_at": "2022-03-17T18:55:59.214Z",78 "reply_count": 1,79 "reply_to_post_number": null,80 "quote_count": 0,81 "incoming_link_count": 53,82 "reads": 147,83 "readers_count": 146,84 "score": 299.2,85 "yours": false,86 "topic_id": 146750,87 "topic_slug": "correct-implementation-of-vae-loss",88 "display_username": "Thomas V",89 "primary_group_name": null,90 "flair_name": null,91 "flair_url": null,92 "flair_bg_color": null,93 "flair_color": null,94 "flair_group_id": null,95 "badges_granted": [],96 "version": 1,97 "can_edit": false,98 "can_delete": false,99 "can_recover": false,100 "can_see_hidden_post": false,101 "can_wiki": false,102 "read": true,103 "user_title": null,104 "bookmarked": false,105 "actions_summary": [],106 "moderator": false,107 "admin": false,108 "staff": false,109 "user_id": 616,110 "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/correct-implementation-of-vae-loss/146750/2",118 "can_accept_answer": false,119 "can_unaccept_answer": false,120 "accepted_answer": false,121 "topic_accepted_answer": null122 },123 {124 "id": 336770,125 "name": "Federico Ottomano",126 "username": "Federico_Ottomano",127 "avatar_template": "/user_avatar/discuss.pytorch.org/federico_ottomano/{size}/43102_2.png",128 "created_at": "2022-03-18T09:55:49.884Z",129 "cooked": "<p>Thanks for your answer but could you elaborate a bit more on that please? The second way I’ve proposed is wrong then?</p>",130 "post_number": 3,131 "post_type": 1,132 "posts_count": 6,133 "updated_at": "2022-03-18T09:55:49.884Z",134 "reply_count": 1,135 "reply_to_post_number": 2,136 "quote_count": 0,137 "incoming_link_count": 73,138 "reads": 137,139 "readers_count": 136,140 "score": 397.4,141 "yours": false,142 "topic_id": 146750,143 "topic_slug": "correct-implementation-of-vae-loss",144 "display_username": "Federico Ottomano",145 "primary_group_name": null,146 "flair_name": null,147 "flair_url": null,148 "flair_bg_color": null,149 "flair_color": null,150 "flair_group_id": null,151 "badges_granted": [],152 "version": 1,153 "can_edit": false,154 "can_delete": false,155 "can_recover": false,156 "can_see_hidden_post": false,157 "can_wiki": false,158 "read": true,159 "user_title": null,160 "reply_to_user": {161 "id": 616,162 "username": "tom",163 "name": "Thomas V",164 "avatar_template": "/user_avatar/discuss.pytorch.org/tom/{size}/3162_2.png"165 },166 "bookmarked": false,167 "actions_summary": [],168 "moderator": false,169 "admin": false,170 "staff": false,171 "user_id": 49912,172 "hidden": false,173 "trust_level": 1,174 "deleted_at": null,175 "user_deleted": false,176 "edit_reason": null,177 "can_view_edit_history": true,178 "wiki": false,179 "post_url": "/t/correct-implementation-of-vae-loss/146750/3",180 "can_accept_answer": false,181 "can_unaccept_answer": false,182 "accepted_answer": false,183 "topic_accepted_answer": null184 },185 {186 "id": 336773,187 "name": "Thomas V",188 "username": "tom",189 "avatar_template": "/user_avatar/discuss.pytorch.org/tom/{size}/3162_2.png",190 "created_at": "2022-03-18T11:04:28.894Z",191 "cooked": "<p>No. But upon further thought you might have some broadcasting interaction between recon_loss and KLD.</p>\n<p><em>If</em> <code>recon_loss</code> is a scalar (maybe you can print shapes of these things to further the discussion):</p>\n<ul>\n<li>And KLD is not (in the first variant) and you add the two, you are essentially adding <code>recon_loss</code> to every entry in KLD.</li>\n<li>In the second part <code>recon_loss</code> is added once to the scalar KLD.</li>\n</ul>\n<p>So in the first, you have a “<code>batch_size</code>” times higher weighting of <code>recon_loss</code> than in the second.</p>\n<p>Now, if <code>recon_loss</code> is not a scalar, it would work the other way round, i.e. the second broadcasts KLD, so you again have that the weight is shifted between the two.<br>\nSo at any rate, it looks to me that KLD is weighted much higher in the second than in the first regime relative to <code>recon_loss</code>.<br>\nDepending on your optimizer (yes for SGD, no for ADAM and co), the total loss scale might also have an influence.</p>\n<p>I would recommend that to print KLD, recon_loss and the sum of the two to better see what’s going on.</p>\n<p>Best regards</p>\n<p>Thomas</p>",192 "post_number": 4,193 "post_type": 1,194 "posts_count": 6,195 "updated_at": "2022-03-18T11:04:28.894Z",196 "reply_count": 0,197 "reply_to_post_number": 3,198 "quote_count": 0,199 "incoming_link_count": 77,200 "reads": 136,201 "readers_count": 135,202 "score": 412.0,203 "yours": false,204 "topic_id": 146750,205 "topic_slug": "correct-implementation-of-vae-loss",206 "display_username": "Thomas V",207 "primary_group_name": null,208 "flair_name": null,209 "flair_url": null,210 "flair_bg_color": null,211 "flair_color": null,212 "flair_group_id": null,213 "badges_granted": [],214 "version": 1,215 "can_edit": false,216 "can_delete": false,217 "can_recover": false,218 "can_see_hidden_post": false,219 "can_wiki": false,220 "read": true,221 "user_title": null,222 "reply_to_user": {223 "id": 49912,224 "username": "Federico_Ottomano",225 "name": "Federico Ottomano",226 "avatar_template": "/user_avatar/discuss.pytorch.org/federico_ottomano/{size}/43102_2.png"227 },228 "bookmarked": false,229 "actions_summary": [],230 "moderator": false,231 "admin": false,232 "staff": false,233 "user_id": 616,234 "hidden": false,235 "trust_level": 2,236 "deleted_at": null,237 "user_deleted": false,238 "edit_reason": null,239 "can_view_edit_history": true,240 "wiki": false,241 "post_url": "/t/correct-implementation-of-vae-loss/146750/4",242 "can_accept_answer": false,243 "can_unaccept_answer": false,244 "accepted_answer": false,245 "topic_accepted_answer": null246 },247 {248 "id": 336828,249 "name": "Federico Ottomano",250 "username": "Federico_Ottomano",251 "avatar_template": "/user_avatar/discuss.pytorch.org/federico_ottomano/{size}/43102_2.png",252 "created_at": "2022-03-18T14:40:54.918Z",253 "cooked": "<p>Alright, as requested I’m including additional details of my problem. I’m trying to reconstruct simple <code>1d</code> vectors with 8 dimensions (plus a 1 dimension <em>label</em> vector) with a conditional variational autoencoder model:</p>\n<pre><code class=\"lang-auto\">class CVAE(BaseModel):\n \n def __init__(self, in_size, target_size):\n \n super().__init__()\n \n #the input is concatenated to the target property\n self.encoder = nn.Sequential(\n nn.Linear(in_size + target_size, 512),\n nn.ReLU(),\n nn.LayerNorm(512),\n nn.Linear(512,256),\n nn.ReLU(),\n nn.LayerNorm(256),\n nn.Linear(256,128),\n nn.ReLU(),\n nn.LayerNorm(128),\n nn.Linear(128,latent_size*2),\n )\n \n self.decoder = nn.Sequential(\n nn.Linear(latent_size + target_size, 128),\n nn.ReLU(),\n nn.LayerNorm(128),\n nn.Linear(128,256),\n nn.ReLU(),\n nn.LayerNorm(256),\n nn.Linear(256,512),\n nn.ReLU(),\n nn.LayerNorm(512),\n nn.Linear(512,in_size),\n )\n\n def reparameterise(self, mu, logvar):\n if self.training:\n std = logvar.mul(0.5).exp_()\n eps = std.data.new(std.size()).normal_()\n return eps.mul(std).add_(mu)\n else:\n return mu\n\n def encode(self, x,cond):\n x = torch.cat([x,cond],dim=1)\n mu_logvar = self.encoder(x).view(-1, 2, latent_size)\n mu = mu_logvar[:, 0, :]\n logvar = mu_logvar[:, 1, :]\n return mu, logvar\n \n\n def decode(self, z):\n return self.decoder(z)\n\n def forward(self,x,cond):\n \n mu, logvar = self.encode(x,cond)\n z = self.reparameterise(mu, logvar)\n z = torch.cat([z,cond],dim=1)\n x_hat = self.decode(z)\n \n part1 = nn.ReLU()(x_hat[:,:4])\n part2 = nn.Softmax(dim=1)(x_hat[:,4:])\n \n x_hat = torch.cat([part1,part2],dim=1)\n \n return x_hat, mu, logvar\n</code></pre>\n<p>Additional method/class that I use to calculate <code>recon_loss</code> are the following:</p>\n<pre><code>def fit(self, dataloader, optimizer, criterion):\n self.train()\n running_loss = 0.0\n for i, data in tqdm(enumerate(dataloader), total=int(len(dataloader.dataset)/dataloader.batch_size)):\n optimizer.zero_grad()\n reconstruction, mu, logvar = self.forward(data[0],data[1])\n loss = criterion(reconstruction, data[0])\n loss = vae_loss(loss, mu, logvar)\n running_loss += loss.item()\n loss.backward()\n optimizer.step()\n \n train_loss = running_loss/len(dataloader.dataset)\n \n return train_loss\n</code></pre>\n<h1>\n<a name=\"h-1\" class=\"anchor\" href=\"#h-1\"></a>------------------------------------</h1>\n<pre><code class=\"lang-auto\">class CustomLoss(_Loss):\n \n def __init__(self):\n \n super().__init__()\n \n def forward(self, input, target):\n \n \"\"\" loss function called at runtime \"\"\"\n \n # Class 1\n class_1_loss = nn.MSELoss()(\n input[:,:4], \n target[:,:4])\n \n # Class 2\n loss_cos = nn.CosineEmbeddingLoss()\n class_2_loss = loss_cos(\n input[:,4:], \n target[:,4:],torch.ones(input.shape[0]))\n \n return class_1_loss + class_2_loss\n</code></pre>\n<p>Given that we take a batch of data in <code>DataLoader</code> and calculating recon_loss:</p>\n<pre><code class=\"lang-auto\">for data in data_loader:\n break\n\ncriterion= CustomLoss()\n\nrecon, mu, logvar = model(data[0],data[1])\n\nrecon_loss = criterion(recon, data[0])\n\n</code></pre>\n<p>where</p>\n<pre><code class=\"lang-auto\">recon_loss\nOut[18]: tensor(1281.3059, grad_fn=<AddBackward0>)\n\n</code></pre>\n<p>next, in `vae_loss() we compute</p>\n<pre><code class=\"lang-auto\">KLD = -0.5 * torch.sum(1 + logvar - mu.pow(2) - logvar.exp(),dim=1)\n\nKLD\nOut[23]: \ntensor([10.0125, 9.2514, 8.6930, 11.5028, 6.2790, 11.4213, 11.1161, 8.7398,\n 8.7730, 11.4224, 10.7966, 8.1383, 7.9794, 8.4988, 8.1407, 7.8110,\n 7.8110, 7.9266, 7.5567, 7.7040, 9.9480, 7.6494, 10.5388, 9.4197,\n 10.1101, 8.4504, 8.7865, 8.8465, 10.7121, 8.4979, 8.6045, 8.3936,\n 8.5527, 7.2297, 8.4476, 10.8846, 10.4522, 11.5168, 7.7652, 9.0135,\n 8.9415, 10.0520, 9.4360, 10.9811, 9.3636, 10.0256, 7.2043, 7.6451,\n 7.7242, 7.6227, 10.8506, 8.6256, 7.5179, 10.1579, 8.7330, 10.3774,\n 9.8613, 8.9309, 10.0385, 9.0261, 9.4356, 9.6858, 10.1660, 8.4929,\n 8.7968, 7.5675, 8.2790, 8.7619, 9.4661, 10.3707, 10.6991, 10.1204,\n 11.2257, 11.0965, 8.7320, 10.7721, 9.4106, 9.6219, 8.5730, 12.0483,\n 6.5600, 10.1521, 10.1500, 10.1347, 10.1345, 10.1373, 8.4438, 6.3176,\n 8.5711, 8.7008, 9.7572, 11.4712, 10.6697, 11.0056, 10.6899, 10.3529,\n 8.3172, 10.5426, 8.4198, 10.1392, 11.1788, 8.7461, 8.5806, 7.9725,\n 8.7498, 10.9897, 11.0135, 8.6260, 9.3328, 9.3445, 9.6393, 6.3542,\n 9.4578, 9.5768, 11.3704, 9.4054, 11.4350, 9.7938, 9.7804, 10.1121,\n 11.1820, 6.0557, 10.4342, 7.9478, 6.3706, 11.0652, 9.2101, 8.6107],\n grad_fn=<MulBackward0>)\n</code></pre>\n<p>where I’ve just realized that if we do not compute a <code>torch.mean</code> over these values we won’t be able to perform the final operation into <code>vae_loss()</code> :<code> recon_loss + KLD</code>…<br>\nSo taking the mean could be reasonable here but I still don’t understand if it makes sense from a more conceptual standopoint.</p>",254 "post_number": 5,255 "post_type": 1,256 "posts_count": 6,257 "updated_at": "2022-03-18T14:40:54.918Z",258 "reply_count": 1,259 "reply_to_post_number": null,260 "quote_count": 0,261 "incoming_link_count": 410,262 "reads": 114,263 "readers_count": 113,264 "score": 2072.6,265 "yours": false,266 "topic_id": 146750,267 "topic_slug": "correct-implementation-of-vae-loss",268 "display_username": "Federico Ottomano",269 "primary_group_name": null,270 "flair_name": null,271 "flair_url": null,272 "flair_bg_color": null,273 "flair_color": null,274 "flair_group_id": null,275 "badges_granted": [],276 "version": 1,277 "can_edit": false,278 "can_delete": false,279 "can_recover": false,280 "can_see_hidden_post": false,281 "can_wiki": false,282 "read": true,283 "user_title": null,284 "bookmarked": false,285 "actions_summary": [],286 "moderator": false,287 "admin": false,288 "staff": false,289 "user_id": 49912,290 "hidden": false,291 "trust_level": 1,292 "deleted_at": null,293 "user_deleted": false,294 "edit_reason": null,295 "can_view_edit_history": true,296 "wiki": false,297 "post_url": "/t/correct-implementation-of-vae-loss/146750/5",298 "can_accept_answer": false,299 "can_unaccept_answer": false,300 "accepted_answer": false,301 "topic_accepted_answer": null302 },303 {304 "id": 336867,305 "name": "Thomas V",306 "username": "tom",307 "avatar_template": "/user_avatar/discuss.pytorch.org/tom/{size}/3162_2.png",308 "created_at": "2022-03-18T19:47:44.509Z",309 "cooked": "<p>Taking the mean can be seen as an estimator for the expectation. I’d say it does make sense here. (That is in contrast to e.g. taking the mean over sequences, which could be much more touchy, because summation would give you the likelihood of the sequence under the data.)<br>\nAnd you really want recon_loss and KLD to either be both reduced or both un-reduced to not accidentally introduce an imbalance in the loss weighting, so taking the mean of KLD first is a good option in my opinion.</p>\n<p>Best regards</p>\n<p>Thomas</p>",310 "post_number": 6,311 "post_type": 1,312 "posts_count": 6,313 "updated_at": "2022-03-18T19:47:44.509Z",314 "reply_count": 0,315 "reply_to_post_number": 5,316 "quote_count": 0,317 "incoming_link_count": 88,318 "reads": 86,319 "readers_count": 85,320 "score": 457.0,321 "yours": false,322 "topic_id": 146750,323 "topic_slug": "correct-implementation-of-vae-loss",324 "display_username": "Thomas V",325 "primary_group_name": null,326 "flair_name": null,327 "flair_url": null,328 "flair_bg_color": null,329 "flair_color": null,330 "flair_group_id": null,331 "badges_granted": [],332 "version": 1,333 "can_edit": false,334 "can_delete": false,335 "can_recover": false,336 "can_see_hidden_post": false,337 "can_wiki": false,338 "read": true,339 "user_title": null,340 "reply_to_user": {341 "id": 49912,342 "username": "Federico_Ottomano",343 "name": "Federico Ottomano",344 "avatar_template": "/user_avatar/discuss.pytorch.org/federico_ottomano/{size}/43102_2.png"345 },346 "bookmarked": false,347 "actions_summary": [],348 "moderator": false,349 "admin": false,350 "staff": false,351 "user_id": 616,352 "hidden": false,353 "trust_level": 2,354 "deleted_at": null,355 "user_deleted": false,356 "edit_reason": null,357 "can_view_edit_history": true,358 "wiki": false,359 "post_url": "/t/correct-implementation-of-vae-loss/146750/6",360 "can_accept_answer": false,361 "can_unaccept_answer": false,362 "accepted_answer": false,363 "topic_accepted_answer": null364 }365 ],366 "stream": [367 336629,368 336646,369 336770,370 336773,371 336828,372 336867373 ]374 },375 "timeline_lookup": [376 [377 1,378 1318379 ],380 [381 3,382 1317383 ]384 ],385 "suggested_topics": [386 {387 "fancy_title": "nn.BatchNorm is slow",388 "id": 212195,389 "title": "nn.BatchNorm is slow",390 "slug": "nn-batchnorm-is-slow",391 "posts_count": 1,392 "reply_count": 0,393 "highest_post_number": 1,394 "image_url": null,395 "created_at": "2024-10-28T08:24:38.399Z",396 "last_posted_at": "2024-10-28T08:24:38.458Z",397 "bumped": true,398 "bumped_at": "2024-10-28T08:24:38.458Z",399 "archetype": "regular",400 "unseen": false,401 "pinned": false,402 "unpinned": null,403 "visible": true,404 "closed": false,405 "archived": false,406 "bookmarked": null,407 "liked": null,408 "tags_descriptions": {},409 "like_count": 0,410 "views": 131,411 "category_id": 1,412 "featured_link": null,413 "has_accepted_answer": false,414 "posters": [415 {416 "extras": "latest single",417 "description": "Original Poster, Most Recent Poster",418 "user": {419 "id": 80459,420 "username": "dailaila",421 "name": null,422 "avatar_template": "/letter_avatar_proxy/v4/letter/d/50afbb/{size}.png",423 "trust_level": 0424 }425 }426 ]427 },428 {429 "fancy_title": "Memory leaking while using shared tensor",430 "id": 214252,431 "title": "Memory leaking while using shared tensor",432 "slug": "memory-leaking-while-using-shared-tensor",433 "posts_count": 1,434 "reply_count": 0,435 "highest_post_number": 1,436 "image_url": null,437 "created_at": "2024-12-16T04:35:35.930Z",438 "last_posted_at": "2024-12-16T04:35:35.968Z",439 "bumped": true,440 "bumped_at": "2024-12-16T04:35:35.968Z",441 "archetype": "regular",442 "unseen": false,443 "pinned": false,444 "unpinned": null,445 "visible": true,446 "closed": false,447 "archived": false,448 "bookmarked": null,449 "liked": null,450 "tags_descriptions": {},451 "like_count": 0,452 "views": 191,453 "category_id": 1,454 "featured_link": null,455 "has_accepted_answer": false,456 "posters": [457 {458 "extras": "latest single",459 "description": "Original Poster, Most Recent Poster",460 "user": {461 "id": 43235,462 "username": "AKT_TARAFDER",463 "name": "AKT TARAFDER",464 "avatar_template": "/user_avatar/discuss.pytorch.org/akt_tarafder/{size}/35960_2.png",465 "trust_level": 1466 }467 }468 ]469 },470 {471 "fancy_title": "Torch_sparse cannot find cuda lib file from pytorch",472 "id": 215733,473 "title": "Torch_sparse cannot find cuda lib file from pytorch",474 "slug": "torch-sparse-cannot-find-cuda-lib-file-from-pytorch",475 "posts_count": 6,476 "reply_count": 3,477 "highest_post_number": 6,478 "image_url": null,479 "created_at": "2025-01-22T21:56:45.806Z",480 "last_posted_at": "2025-01-23T18:24:01.921Z",481 "bumped": true,482 "bumped_at": "2025-01-23T18:24:01.921Z",483 "archetype": "regular",484 "unseen": false,485 "pinned": false,486 "unpinned": null,487 "visible": true,488 "closed": false,489 "archived": false,490 "bookmarked": null,491 "liked": null,492 "tags_descriptions": {},493 "like_count": 1,494 "views": 336,495 "category_id": 1,496 "featured_link": null,497 "has_accepted_answer": true,498 "posters": [499 {500 "extras": null,501 "description": "Original Poster",502 "user": {503 "id": 78812,504 "username": "c-earth",505 "name": "Abhijatmedhi Chotrattanapituk",506 "avatar_template": "/user_avatar/discuss.pytorch.org/c-earth/{size}/72657_2.png",507 "trust_level": 1508 }509 },510 {511 "extras": "latest",512 "description": "Most Recent Poster, Accepted Answer",513 "user": {514 "id": 3534,515 "username": "ptrblck",516 "name": "",517 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",518 "admin": true,519 "moderator": true,520 "trust_level": 2521 }522 }523 ]524 },525 {526 "fancy_title": "Why the memory usage is higher than expected when loading nvidia/NV-Embed-v2 model with FP16 precision?",527 "id": 213900,528 "title": "Why the memory usage is higher than expected when loading nvidia/NV-Embed-v2 model with FP16 precision?",529 "slug": "why-the-memory-usage-is-higher-than-expected-when-loading-nvidia-nv-embed-v2-model-with-fp16-precision",530 "posts_count": 4,531 "reply_count": 1,532 "highest_post_number": 4,533 "image_url": null,534 "created_at": "2024-12-06T08:07:56.582Z",535 "last_posted_at": "2024-12-11T19:10:19.309Z",536 "bumped": true,537 "bumped_at": "2024-12-11T19:10:19.309Z",538 "archetype": "regular",539 "unseen": false,540 "pinned": false,541 "unpinned": null,542 "visible": true,543 "closed": false,544 "archived": false,545 "bookmarked": null,546 "liked": null,547 "tags_descriptions": {},548 "like_count": 0,549 "views": 169,550 "category_id": 1,551 "featured_link": null,552 "has_accepted_answer": false,553 "posters": [554 {555 "extras": null,556 "description": "Original Poster",557 "user": {558 "id": 78017,559 "username": "Arunima_Ghosh",560 "name": "Arunima Ghosh",561 "avatar_template": "/user_avatar/discuss.pytorch.org/arunima_ghosh/{size}/75750_2.png",562 "trust_level": 1563 }564 },565 {566 "extras": "latest",567 "description": "Most Recent Poster",568 "user": {569 "id": 3534,570 "username": "ptrblck",571 "name": "",572 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",573 "admin": true,574 "moderator": true,575 "trust_level": 2576 }577 }578 ]579 },580 {581 "fancy_title": "Pytorch and Cuda version compatibility",582 "id": 216242,583 "title": "Pytorch and Cuda version compatibility",584 "slug": "pytorch-and-cuda-version-compatibility",585 "posts_count": 4,586 "reply_count": 2,587 "highest_post_number": 4,588 "image_url": null,589 "created_at": "2025-02-04T21:43:58.736Z",590 "last_posted_at": "2025-02-06T01:58:24.857Z",591 "bumped": true,592 "bumped_at": "2025-02-06T01:58:24.857Z",593 "archetype": "regular",594 "unseen": false,595 "pinned": false,596 "unpinned": null,597 "visible": true,598 "closed": false,599 "archived": false,600 "bookmarked": null,601 "liked": null,602 "tags_descriptions": {},603 "like_count": 0,604 "views": 814,605 "category_id": 1,606 "featured_link": null,607 "has_accepted_answer": false,608 "posters": [609 {610 "extras": null,611 "description": "Original Poster",612 "user": {613 "id": 82490,614 "username": "pooja-kabra",615 "name": "Pooja Kabra",616 "avatar_template": "/user_avatar/discuss.pytorch.org/pooja-kabra/{size}/75471_2.png",617 "trust_level": 1618 }619 },620 {621 "extras": "latest",622 "description": "Most Recent Poster",623 "user": {624 "id": 3534,625 "username": "ptrblck",626 "name": "",627 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",628 "admin": true,629 "moderator": true,630 "trust_level": 2631 }632 }633 ]634 }635 ],636 "tags_descriptions": {},637 "fancy_title": "Correct implementation of VAE loss",638 "id": 146750,639 "title": "Correct implementation of VAE loss",640 "posts_count": 6,641 "created_at": "2022-03-17T17:29:45.182Z",642 "views": 17292,643 "reply_count": 3,644 "like_count": 2,645 "last_posted_at": "2022-03-18T19:47:44.509Z",646 "visible": true,647 "closed": false,648 "archived": false,649 "has_summary": false,650 "archetype": "regular",651 "slug": "correct-implementation-of-vae-loss",652 "category_id": 1,653 "word_count": 1265,654 "deleted_at": null,655 "user_id": 49912,656 "featured_link": null,657 "pinned_globally": false,658 "pinned_at": null,659 "pinned_until": null,660 "image_url": null,661 "slow_mode_seconds": 0,662 "draft": null,663 "draft_key": "topic_146750",664 "draft_sequence": null,665 "unpinned": null,666 "pinned": false,667 "current_post_number": 1,668 "highest_post_number": 6,669 "deleted_by": null,670 "actions_summary": [671 {672 "id": 4,673 "count": 0,674 "hidden": false,675 "can_act": false676 },677 {678 "id": 8,679 "count": 0,680 "hidden": false,681 "can_act": false682 },683 {684 "id": 10,685 "count": 0,686 "hidden": false,687 "can_act": false688 },689 {690 "id": 7,691 "count": 0,692 "hidden": false,693 "can_act": false694 }695 ],696 "chunk_size": 20,697 "bookmarked": false,698 "topic_timer": null,699 "message_bus_last_id": 0,700 "participant_count": 2,701 "show_read_indicator": false,702 "thumbnails": null,703 "slow_mode_enabled_until": null,704 "can_vote": false,705 "vote_count": 0,706 "user_voted": false,707 "discourse_zendesk_plugin_zendesk_id": null,708 "discourse_zendesk_plugin_zendesk_url": "https://your-url.zendesk.com/agent/tickets/",709 "details": {710 "can_edit": false,711 "notification_level": 1,712 "participants": [713 {714 "id": 616,715 "username": "tom",716 "name": "Thomas V",717 "avatar_template": "/user_avatar/discuss.pytorch.org/tom/{size}/3162_2.png",718 "post_count": 3,719 "primary_group_name": null,720 "flair_name": null,721 "flair_url": null,722 "flair_color": null,723 "flair_bg_color": null,724 "flair_group_id": null,725 "trust_level": 2726 },727 {728 "id": 49912,729 "username": "Federico_Ottomano",730 "name": "Federico Ottomano",731 "avatar_template": "/user_avatar/discuss.pytorch.org/federico_ottomano/{size}/43102_2.png",732 "post_count": 3,733 "primary_group_name": null,734 "flair_name": null,735 "flair_url": null,736 "flair_color": null,737 "flair_bg_color": null,738 "flair_group_id": null,739 "trust_level": 1740 }741 ],742 "created_by": {743 "id": 49912,744 "username": "Federico_Ottomano",745 "name": "Federico Ottomano",746 "avatar_template": "/user_avatar/discuss.pytorch.org/federico_ottomano/{size}/43102_2.png"747 },748 "last_poster": {749 "id": 616,750 "username": "tom",751 "name": "Thomas V",752 "avatar_template": "/user_avatar/discuss.pytorch.org/tom/{size}/3162_2.png"753 }754 },755 "bookmarks": []756 },757 {758 "post_stream": {759 "posts": [760 {761 "id": 336840,762 "name": "",763 "username": "Mukesh1729",764 "avatar_template": "/letter_avatar_proxy/v4/letter/m/90db22/{size}.png",765 "created_at": "2022-03-18T16:41:45.357Z",766 "cooked": "<p>I am using an LSTM network multiclass problem setting but i get the above error. Any hints on where to look would be helpful. Thanks</p>\n<pre><code class=\"lang-auto\">class SequenceModel(nn.Module):\n def __init__(self, n_features, n_classes, n_hidden=256, n_layers=3):\n super().__init__()\n\n self.lstm = torch.nn.LSTM(input_size=n_features,\n hidden_size=n_hidden,\n num_layers=n_layers, \n batch_first=True,\n dropout=0.25)\n self.classifier = torch.nn.Linear(n_hidden, n_classes)\n\n def forward(self, x):\n self.lstm.flatten_parameters()\n _, (hidden, _) = self.lstm(x)\n out = hidden[-1] # get the last state of the last layer\n return self.classifier(out)\n\n\n#Print smaple output\nsample = next(iter(train_set))\nout = SequenceModel(39, 5)(sample[\"sequence\"])\nc = torch.nn.CrossEntropyLoss()\nc(out, sample[\"label\"])\n</code></pre>\n<pre><code class=\"lang-auto\">#data module for lightning\nclass PostDataModule(pl.LightningDataModule):\n def __init__(self, train_seq, test_seq, batch_size):\n super().__init__()\n self.train_seq = train_seq\n self.test_seq = test_seq\n self.batch_size= batch_size\n\n def setup(self, stage=None):\n self.train_dataset = Post_data(self.train_seq)\n self.test_dataset = Post_data(self.test_seq) \n\n def train_dataloader(self):\n return DataLoader(\n self.train_dataset,\n batch_size=self.batch_size,\n shuffle=True,\n num_workers=cpu_count()\n )\n\n def val_dataloader(self):\n return DataLoader(\n self.test_dataset,\n batch_size=self.batch_size,\n shuffle=False,\n num_workers=cpu_count()\n )\n\n def test_dataloader(self):\n return DataLoader(\n self.test_dataset,\n batch_size=self.batch_size,\n shuffle=False,\n num_workers=cpu_count()\n )\n</code></pre>\n<pre><code class=\"lang-auto\">from multiprocessing import cpu_count\n# from pytorch_lightning.metrics.functional import accuracy\n\nclass LamePrediction(pl.LightningModule):\n def __init__(self, n_features:int, n_classes:int):\n super().__init__()\n self.model = SequenceModel(n_features, n_classes)\n self.criterion = torch.nn.CrossEntropyLoss()\n\n def forward(self, x, labels):\n output = self.model(x)\n loss = 0\n if labels is not None:\n loss = self.criterion(output, labels)\n return loss, output\n\n def training_step(self, batch, batch_idx):\n sequence = batch[\"sequence\"]\n labels = batch[\"label\"]\n loss, outputs = self(sequences, labels)\n predictions = torch.argmax(outputs, dim=1)\n accuracy = torchmetrics.Accuracy()\n step_accuracy = accuracy(predictions, labels)\n\n self.log(\"train_loss\", loss, prog_bar=True, logger=True)\n self.log(\"train_accuracy\", step_accuracy, prog_bar=True, logger=True)\n\n return {\"loss\":loss, \"accuracy\":step_accuracy}\n\n def validation_step(self, batch, batch_idx):\n sequence = batch[\"sequence\"]\n labels = batch[\"label\"]\n loss, outputs = self(sequences, labels)\n predictions = torch.argmax(outputs, dim=1)\n accuracy = torchmetrics.Accuracy()\n step_accuracy = accuracy(predictions, labels)\n\n self.log(\"validation_loss\", loss, prog_bar=True, logger=True)\n self.log(\"validation_accuracy\", step_accuracy, prog_bar=True, logger=True)\n\n return {\"loss\":loss, \"accuracy\":step_accuracy}\n\n def test_step(self, batch, batch_idx):\n sequence = batch[\"sequence\"]\n labels = batch[\"label\"]\n loss, outputs = self(sequences, labels)\n predictions = torch.argmax(outputs, dim=1)\n accuracy = torchmetrics.Accuracy()\n step_accuracy = accuracy(predictions, labels)\n\n self.log(\"test_loss\", loss, prog_bar=True, logger=True)\n self.log(\"test_accuracy\", step_accuracy, prog_bar=True, logger=True)\n\n return {\"loss\":loss, \"accuracy\":step_accuracy}\n\n def configure_optimizers(self):\n return torch.optim.Adam(self.parameters(), lr=1e-3)\n\nFEATURE_COLUMNS = 39\nmodel = LamePrediction(n_features=39, \n n_classes=5)\n</code></pre>\n<p> in ()<br>\n----> 1 trainer.fit(model, data_module)</p>\n<p>22 frames<br>\n/usr/local/lib/python3.7/dist-packages/torch/nn/modules/rnn.py in forward(self, input, hx)<br>\n668 else:<br>\n669 batch_sizes = None<br>\n → 670 max_batch_size = input.size(0) if self.batch_first else input.size(1)<br>\n671 sorted_indices = None<br>\n672 unsorted_indices = None</p>\n<p>AttributeError: ‘list’ object has no attribute ‘size’</p>",767 "post_number": 1,768 "post_type": 1,769 "posts_count": 4,770 "updated_at": "2022-03-18T16:41:45.357Z",771 "reply_count": 0,772 "reply_to_post_number": null,773 "quote_count": 0,774 "incoming_link_count": 4425,775 "reads": 35,776 "readers_count": 34,777 "score": 22132.0,778 "yours": false,779 "topic_id": 146834,780 "topic_slug": "attributeerror-list-object-has-no-attribute-size",781 "display_username": "",782 "primary_group_name": null,783 "flair_name": null,784 "flair_url": null,785 "flair_bg_color": null,786 "flair_color": null,787 "flair_group_id": null,788 "badges_granted": [],789 "version": 1,790 "can_edit": false,791 "can_delete": false,792 "can_recover": false,793 "can_see_hidden_post": false,794 "can_wiki": false,795 "read": true,796 "user_title": "",797 "bookmarked": false,798 "actions_summary": [],799 "moderator": false,800 "admin": false,801 "staff": false,802 "user_id": 27130,803 "hidden": false,804 "trust_level": 1,805 "deleted_at": null,806 "user_deleted": false,807 "edit_reason": null,808 "can_view_edit_history": true,809 "wiki": false,810 "post_url": "/t/attributeerror-list-object-has-no-attribute-size/146834/1",811 "can_accept_answer": false,812 "can_unaccept_answer": false,813 "accepted_answer": false,814 "topic_accepted_answer": null,815 "can_vote": false816 },817 {818 "id": 336859,819 "name": "",820 "username": "ptrblck",821 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",822 "created_at": "2022-03-18T18:34:52.559Z",823 "cooked": "<p>Based on the error message you are passing a <code>list</code> to the model while tensors are expected.<br>\nMaybe unwrap the <code>list</code> and it would work.</p>",824 "post_number": 2,825 "post_type": 1,826 "posts_count": 4,827 "updated_at": "2022-03-18T18:34:52.559Z",828 "reply_count": 1,829 "reply_to_post_number": null,830 "quote_count": 0,831 "incoming_link_count": 12,832 "reads": 32,833 "readers_count": 31,834 "score": 86.4,835 "yours": false,836 "topic_id": 146834,837 "topic_slug": "attributeerror-list-object-has-no-attribute-size",838 "display_username": "",839 "primary_group_name": null,840 "flair_name": null,841 "flair_url": null,842 "flair_bg_color": null,843 "flair_color": null,844 "flair_group_id": null,845 "badges_granted": [],846 "version": 1,847 "can_edit": false,848 "can_delete": false,849 "can_recover": false,850 "can_see_hidden_post": false,851 "can_wiki": false,852 "read": true,853 "user_title": "",854 "bookmarked": false,855 "actions_summary": [856 {857 "id": 2,858 "count": 1859 }860 ],861 "moderator": true,862 "admin": true,863 "staff": true,864 "user_id": 3534,865 "hidden": false,866 "trust_level": 2,867 "deleted_at": null,868 "user_deleted": false,869 "edit_reason": null,870 "can_view_edit_history": true,871 "wiki": false,872 "post_url": "/t/attributeerror-list-object-has-no-attribute-size/146834/2",873 "can_accept_answer": false,874 "can_unaccept_answer": false,875 "accepted_answer": false,876 "topic_accepted_answer": null877 },878 {879 "id": 336864,880 "name": "",881 "username": "Mukesh1729",882 "avatar_template": "/letter_avatar_proxy/v4/letter/m/90db22/{size}.png",883 "created_at": "2022-03-18T19:27:07.120Z",884 "cooked": "<p>I figured that my data loader returns a sequence as i wanted and when I pass it through the model it works but when I put it in this class below i get a class list back somehow</p>\n<pre><code class=\"lang-auto\">#data module for lightning\nclass PostDataModule(pl.LightningDataModule):\n def __init__(self, train_seq, test_seq, batch_size):\n super().__init__()\n self.train_seq = train_seq\n self.test_seq = test_seq\n self.batch_size= batch_size\n\n def setup(self, stage=None):\n self.train_dataset = Post_data(self.train_seq)\n self.test_dataset = Post_data(self.test_seq) \n\n def train_dataloader(self):\n return DataLoader(\n self.train_dataset,\n batch_size=self.batch_size,\n shuffle=True,\n num_workers=cpu_count()\n )\n\n def val_dataloader(self):\n return DataLoader(\n self.test_dataset,\n batch_size=self.batch_size,\n shuffle=False,\n num_workers=cpu_count()\n )\n\n def test_dataloader(self):\n return DataLoader(\n self.test_dataset,\n batch_size=self.batch_size,\n shuffle=False,\n num_workers=cpu_count()\n )\n</code></pre>",885 "post_number": 3,886 "post_type": 1,887 "posts_count": 4,888 "updated_at": "2022-03-18T19:27:07.120Z",889 "reply_count": 1,890 "reply_to_post_number": 2,891 "quote_count": 0,892 "incoming_link_count": 49,893 "reads": 31,894 "readers_count": 30,895 "score": 256.2,896 "yours": false,897 "topic_id": 146834,898 "topic_slug": "attributeerror-list-object-has-no-attribute-size",899 "display_username": "",900 "primary_group_name": null,901 "flair_name": null,902 "flair_url": null,903 "flair_bg_color": null,904 "flair_color": null,905 "flair_group_id": null,906 "badges_granted": [],907 "version": 1,908 "can_edit": false,909 "can_delete": false,910 "can_recover": false,911 "can_see_hidden_post": false,912 "can_wiki": false,913 "read": true,914 "user_title": "",915 "reply_to_user": {916 "id": 3534,917 "username": "ptrblck",918 "name": "",919 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png"920 },921 "bookmarked": false,922 "actions_summary": [],923 "moderator": false,924 "admin": false,925 "staff": false,926 "user_id": 27130,927 "hidden": false,928 "trust_level": 1,929 "deleted_at": null,930 "user_deleted": false,931 "edit_reason": null,932 "can_view_edit_history": true,933 "wiki": false,934 "post_url": "/t/attributeerror-list-object-has-no-attribute-size/146834/3",935 "can_accept_answer": false,936 "can_unaccept_answer": false,937 "accepted_answer": false,938 "topic_accepted_answer": null939 },940 {941 "id": 336866,942 "name": "",943 "username": "ptrblck",944 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",945 "created_at": "2022-03-18T19:47:15.984Z",946 "cooked": "<p>I’m not familiar enough with Lightning and don’t know where the type is changed.</p>",947 "post_number": 4,948 "post_type": 1,949 "posts_count": 4,950 "updated_at": "2022-03-18T19:47:15.984Z",951 "reply_count": 0,952 "reply_to_post_number": 3,953 "quote_count": 0,954 "incoming_link_count": 13,955 "reads": 24,956 "readers_count": 23,957 "score": 69.8,958 "yours": false,959 "topic_id": 146834,960 "topic_slug": "attributeerror-list-object-has-no-attribute-size",961 "display_username": "",962 "primary_group_name": null,963 "flair_name": null,964 "flair_url": null,965 "flair_bg_color": null,966 "flair_color": null,967 "flair_group_id": null,968 "badges_granted": [],969 "version": 1,970 "can_edit": false,971 "can_delete": false,972 "can_recover": false,973 "can_see_hidden_post": false,974 "can_wiki": false,975 "read": true,976 "user_title": "",977 "reply_to_user": {978 "id": 27130,979 "username": "Mukesh1729",980 "name": "",981 "avatar_template": "/letter_avatar_proxy/v4/letter/m/90db22/{size}.png"982 },983 "bookmarked": false,984 "actions_summary": [],985 "moderator": true,986 "admin": true,987 "staff": true,988 "user_id": 3534,989 "hidden": false,990 "trust_level": 2,991 "deleted_at": null,992 "user_deleted": false,993 "edit_reason": null,994 "can_view_edit_history": true,995 "wiki": false,996 "post_url": "/t/attributeerror-list-object-has-no-attribute-size/146834/4",997 "can_accept_answer": false,998 "can_unaccept_answer": false,999 "accepted_answer": false,1000 "topic_accepted_answer": null1001 }1002 ],1003 "stream": [1004 336840,1005 336859,1006 336864,1007 3368661008 ]1009 },1010 "timeline_lookup": [1011 [1012 1,1013 13171014 ]1015 ],1016 "suggested_topics": [1017 {1018 "fancy_title": "What’s the theoreticl basis of torch.testing tolerance table?",1019 "id": 213255,1020 "title": "What's the theoreticl basis of torch.testing tolerance table?",1021 "slug": "whats-the-theoreticl-basis-of-torch-testing-tolerance-table",1022 "posts_count": 1,1023 "reply_count": 0,1024 "highest_post_number": 1,1025 "image_url": "https://discuss.pytorch.org/uploads/default/optimized/3X/7/0/704c9ddf26dc72ffe8f69312a123846fdab78ded_2_1024x736.png",1026 "created_at": "2024-11-21T07:42:12.294Z",1027 "last_posted_at": "2024-11-21T07:42:12.347Z",1028 "bumped": true,1029 "bumped_at": "2024-11-21T07:42:12.347Z",1030 "archetype": "regular",1031 "unseen": false,1032 "pinned": false,1033 "unpinned": null,1034 "visible": true,1035 "closed": false,1036 "archived": false,1037 "bookmarked": null,1038 "liked": null,1039 "tags_descriptions": {},1040 "like_count": 0,1041 "views": 25,1042 "category_id": 1,1043 "featured_link": null,1044 "has_accepted_answer": false,1045 "posters": [1046 {1047 "extras": "latest single",1048 "description": "Original Poster, Most Recent Poster",1049 "user": {1050 "id": 81043,1051 "username": "Jiexin-Zheng",1052 "name": "Jiexin Zheng",1053 "avatar_template": "/user_avatar/discuss.pytorch.org/jiexin-zheng/{size}/74108_2.png",1054 "trust_level": 01055 }1056 }1057 ]1058 },1059 {1060 "fancy_title": "GAN: RuntimeError: mat1 and mat2 shapes cannot be multiplied (32x189 and 10x512)",1061 "id": 213088,1062 "title": "GAN: RuntimeError: mat1 and mat2 shapes cannot be multiplied (32x189 and 10x512)",1063 "slug": "gan-runtimeerror-mat1-and-mat2-shapes-cannot-be-multiplied-32x189-and-10x512",1064 "posts_count": 1,1065 "reply_count": 0,1066 "highest_post_number": 1,1067 "image_url": null,1068 "created_at": "2024-11-17T21:59:16.402Z",1069 "last_posted_at": "2024-11-17T21:59:16.460Z",1070 "bumped": true,1071 "bumped_at": "2024-11-17T22:01:14.585Z",1072 "archetype": "regular",1073 "unseen": false,1074 "pinned": false,1075 "unpinned": null,1076 "visible": true,1077 "closed": false,1078 "archived": false,1079 "bookmarked": null,1080 "liked": null,1081 "tags_descriptions": {},1082 "like_count": 0,1083 "views": 22,1084 "category_id": 1,1085 "featured_link": null,1086 "has_accepted_answer": false,1087 "posters": [1088 {1089 "extras": "latest single",1090 "description": "Original Poster, Most Recent Poster",1091 "user": {1092 "id": 80971,1093 "username": "Soham_Lotlikar",1094 "name": "Soham Lotlikar",1095 "avatar_template": "/user_avatar/discuss.pytorch.org/soham_lotlikar/{size}/73112_2.png",1096 "trust_level": 01097 }1098 }1099 ]1100 },1101 {1102 "fancy_title": "Accessing secrets variables preset in the repository in CI",1103 "id": 215423,1104 "title": "Accessing secrets variables preset in the repository in CI",1105 "slug": "accessing-secrets-variables-preset-in-the-repository-in-ci",1106 "posts_count": 1,1107 "reply_count": 0,1108 "highest_post_number": 1,1109 "image_url": null,1110 "created_at": "2025-01-15T14:01:39.216Z",1111 "last_posted_at": "2025-01-15T14:01:39.292Z",1112 "bumped": true,1113 "bumped_at": "2025-01-15T14:01:39.292Z",1114 "archetype": "regular",1115 "unseen": false,1116 "pinned": false,1117 "unpinned": null,1118 "visible": true,1119 "closed": false,1120 "archived": false,1121 "bookmarked": null,1122 "liked": null,1123 "tags_descriptions": {},1124 "like_count": 0,1125 "views": 16,1126 "category_id": 1,1127 "featured_link": null,1128 "has_accepted_answer": false,1129 "posters": [1130 {1131 "extras": "latest single",1132 "description": "Original Poster, Most Recent Poster",1133 "user": {1134 "id": 82099,1135 "username": "swgu98",1136 "name": "Gu Shiwei",1137 "avatar_template": "/user_avatar/discuss.pytorch.org/swgu98/{size}/75112_2.png",1138 "trust_level": 01139 }1140 }1141 ]1142 },1143 {1144 "fancy_title": "Pytorch CUDA / GPU error",1145 "id": 216558,1146 "title": "Pytorch CUDA / GPU error",1147 "slug": "pytorch-cuda-gpu-error",1148 "posts_count": 13,1149 "reply_count": 6,1150 "highest_post_number": 13,1151 "image_url": null,1152 "created_at": "2025-02-12T04:28:39.969Z",1153 "last_posted_at": "2025-08-04T04:22:42.474Z",1154 "bumped": true,1155 "bumped_at": "2025-08-04T04:22:42.474Z",1156 "archetype": "regular",1157 "unseen": false,1158 "pinned": false,1159 "unpinned": null,1160 "visible": true,1161 "closed": false,1162 "archived": false,1163 "bookmarked": null,1164 "liked": null,1165 "tags_descriptions": {},1166 "like_count": 0,1167 "views": 459,1168 "category_id": 1,1169 "featured_link": null,1170 "has_accepted_answer": false,1171 "posters": [1172 {1173 "extras": null,1174 "description": "Original Poster",1175 "user": {1176 "id": 82635,1177 "username": "automasterz",1178 "name": "",1179 "avatar_template": "/letter_avatar_proxy/v4/letter/a/41988e/{size}.png",1180 "trust_level": 11181 }1182 },1183 {1184 "extras": null,1185 "description": "Frequent Poster",1186 "user": {1187 "id": 3534,1188 "username": "ptrblck",1189 "name": "",1190 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",1191 "admin": true,1192 "moderator": true,1193 "trust_level": 21194 }1195 },1196 {1197 "extras": "latest",1198 "description": "Most Recent Poster",1199 "user": {1200 "id": 85328,