Anurag1734/cuda-error-resolution-analysis
07
1[2 {3 "post_stream": {4 "posts": [5 {6 "id": 218882,7 "name": "siahuat0727",8 "username": "siahuat0727",9 "avatar_template": "/user_avatar/discuss.pytorch.org/siahuat0727/{size}/27590_2.png",10 "created_at": "2020-08-07T09:13:02.981Z",11 "cooked": "<pre><code class=\"lang-python\">class Foo(nn.Module):\n def __init__(self, dim, n_iter):\n ...\n self.n_iter = n_iter\n self.attn = nn.MultiheadAttention(dim, 1)\n\n def forward(self, input):\n input = input.unsqueeze(0)\n\n for i in range(self.n_iter):\n assert input.size(0) == i+1\n output, _ = self.attn(input, input, input, need_weights=False)\n\n # For every iteration, I need only the last feature\n # since it uses only the previous input (like rnns)\n last = output[-1]\n\n # The input is concatenated with previous input,\n # that means some results of the previous `attn` can be reused.\n input = torch.cat([input, some_func(last)], dim=0)\n ...\n</code></pre>\n<p><strong>Updated</strong> to make the question clearer.</p>\n<p>I want to solve a problem iteratively, similar to RNNs but using self-attention.<br>\nI have a series of inputs, I want the first output depends on the first input, and the second output depends on both the first and second input and so on.<br>\nBut the things different is, the second input depends on the first output, so that’s why I do it iteratively.</p>\n<p>I think <code>MultiheadAttention</code> is good for solving this problem but it brings a lot of unnecessary calculation.</p>\n<ol>\n<li>For each iteration, I need only the last output.</li>\n<li>The later iteration can reuse some hidden state of the previous iteration (like qkv of the previous input).</li>\n</ol>\n<p>Do I need a custom modified MultiheadAttention layer to only calculate for the last output and save the hidden states so that can be reused later?<br>\nIf so, how to do it since the <code>forward</code> calls <code>F.multi_head_attention_forward</code> which is not written in python?<br>\nOr is there any other way better suited to my needs?</p>\n<p>I think anyone who tries to solve a time series using self-attention will meet a similar situation? (except for that the next input depends on the previous output)<br>\nBut I’m new in this domain and don’t know where to find the solution.<br>\nCan anyone help me?</p>\n<p>Any help would be appreciated!</p>",12 "post_number": 1,13 "post_type": 1,14 "posts_count": 3,15 "updated_at": "2020-08-08T02:07:56.564Z",16 "reply_count": 0,17 "reply_to_post_number": null,18 "quote_count": 0,19 "incoming_link_count": 71,20 "reads": 14,21 "readers_count": 13,22 "score": 357.8,23 "yours": false,24 "topic_id": 91940,25 "topic_slug": "do-i-need-a-custom-modified-multiheadattention-layer",26 "display_username": "siahuat0727",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": 8,35 "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": 35250,48 "hidden": false,49 "trust_level": 2,50 "deleted_at": null,51 "user_deleted": false,52 "edit_reason": null,53 "can_view_edit_history": true,54 "wiki": false,55 "post_url": "/t/do-i-need-a-custom-modified-multiheadattention-layer/91940/1",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": 218938,64 "name": "",65 "username": "SandPhoenix",66 "avatar_template": "/user_avatar/discuss.pytorch.org/sandphoenix/{size}/29137_2.png",67 "created_at": "2020-08-07T13:23:15.228Z",68 "cooked": "<p>If I understood well, I think that you can do that by simply write something like this in the forward function</p>\n<pre><code class=\"lang-auto\">def forward(self, x):\n previous = x\n x = self.layer_or_model(x)\n x += previous\n return x\n</code></pre>\n<p>and use it as a building block. I believe this kind of solution works because I used it before but I am not sure if this is what you asked for.</p>",69 "post_number": 2,70 "post_type": 1,71 "posts_count": 3,72 "updated_at": "2020-08-07T13:23:15.228Z",73 "reply_count": 1,74 "reply_to_post_number": null,75 "quote_count": 0,76 "incoming_link_count": 0,77 "reads": 11,78 "readers_count": 10,79 "score": 7.2,80 "yours": false,81 "topic_id": 91940,82 "topic_slug": "do-i-need-a-custom-modified-multiheadattention-layer",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 "moderator": false,102 "admin": false,103 "staff": false,104 "user_id": 29396,105 "hidden": false,106 "trust_level": 2,107 "deleted_at": null,108 "user_deleted": false,109 "edit_reason": null,110 "can_view_edit_history": true,111 "wiki": false,112 "post_url": "/t/do-i-need-a-custom-modified-multiheadattention-layer/91940/2",113 "can_accept_answer": false,114 "can_unaccept_answer": false,115 "accepted_answer": false,116 "topic_accepted_answer": null117 },118 {119 "id": 218970,120 "name": "siahuat0727",121 "username": "siahuat0727",122 "avatar_template": "/user_avatar/discuss.pytorch.org/siahuat0727/{size}/27590_2.png",123 "created_at": "2020-08-07T14:18:10.398Z",124 "cooked": "<p>Sorry, I didn’t make it clear.</p>\n<p>The way I want to solve the problem is similar to RNNs.<br>\nI have a series of inputs, I want the first output depends on the first input, the second output depends on the first and second input, and so on.<br>\nBut the things different is, the second input depends on the first output, so that’s why I do it iteratively.</p>\n<p>Thanks for your help!</p>",125 "post_number": 3,126 "post_type": 1,127 "posts_count": 3,128 "updated_at": "2020-08-07T14:18:10.398Z",129 "reply_count": 0,130 "reply_to_post_number": 2,131 "quote_count": 0,132 "incoming_link_count": 4,133 "reads": 11,134 "readers_count": 10,135 "score": 22.2,136 "yours": false,137 "topic_id": 91940,138 "topic_slug": "do-i-need-a-custom-modified-multiheadattention-layer",139 "display_username": "siahuat0727",140 "primary_group_name": null,141 "flair_name": null,142 "flair_url": null,143 "flair_bg_color": null,144 "flair_color": null,145 "flair_group_id": null,146 "badges_granted": [],147 "version": 1,148 "can_edit": false,149 "can_delete": false,150 "can_recover": false,151 "can_see_hidden_post": false,152 "can_wiki": false,153 "read": true,154 "user_title": null,155 "reply_to_user": {156 "id": 29396,157 "username": "SandPhoenix",158 "name": "",159 "avatar_template": "/user_avatar/discuss.pytorch.org/sandphoenix/{size}/29137_2.png"160 },161 "bookmarked": false,162 "actions_summary": [],163 "moderator": false,164 "admin": false,165 "staff": false,166 "user_id": 35250,167 "hidden": false,168 "trust_level": 2,169 "deleted_at": null,170 "user_deleted": false,171 "edit_reason": null,172 "can_view_edit_history": true,173 "wiki": false,174 "post_url": "/t/do-i-need-a-custom-modified-multiheadattention-layer/91940/3",175 "can_accept_answer": false,176 "can_unaccept_answer": false,177 "accepted_answer": false,178 "topic_accepted_answer": null179 }180 ],181 "stream": [182 218882,183 218938,184 218970185 ]186 },187 "timeline_lookup": [188 [189 1,190 1906191 ],192 [193 2,194 1905195 ]196 ],197 "suggested_topics": [198 {199 "fancy_title": "Size mismatch for model",200 "id": 212435,201 "title": "Size mismatch for model",202 "slug": "size-mismatch-for-model",203 "posts_count": 2,204 "reply_count": 0,205 "highest_post_number": 2,206 "image_url": null,207 "created_at": "2024-11-02T01:18:21.033Z",208 "last_posted_at": "2024-11-02T18:14:42.220Z",209 "bumped": true,210 "bumped_at": "2024-11-02T18:14:42.220Z",211 "archetype": "regular",212 "unseen": false,213 "pinned": false,214 "unpinned": null,215 "visible": true,216 "closed": false,217 "archived": false,218 "bookmarked": null,219 "liked": null,220 "tags_descriptions": {},221 "like_count": 0,222 "views": 817,223 "category_id": 1,224 "featured_link": null,225 "has_accepted_answer": false,226 "posters": [227 {228 "extras": null,229 "description": "Original Poster",230 "user": {231 "id": 80643,232 "username": "Andile_Zungu",233 "name": "Andile Zungu",234 "avatar_template": "/user_avatar/discuss.pytorch.org/andile_zungu/{size}/73200_2.png",235 "trust_level": 0236 }237 },238 {239 "extras": "latest",240 "description": "Most Recent Poster",241 "user": {242 "id": 3534,243 "username": "ptrblck",244 "name": "",245 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",246 "admin": true,247 "moderator": true,248 "trust_level": 2249 }250 }251 ]252 },253 {254 "fancy_title": "Pytorch GPU via conda not working",255 "id": 212586,256 "title": "Pytorch GPU via conda not working",257 "slug": "pytorch-gpu-via-conda-not-working",258 "posts_count": 5,259 "reply_count": 3,260 "highest_post_number": 5,261 "image_url": null,262 "created_at": "2024-11-06T04:00:56.989Z",263 "last_posted_at": "2024-11-06T15:45:21.971Z",264 "bumped": true,265 "bumped_at": "2024-11-06T15:45:21.971Z",266 "archetype": "regular",267 "unseen": false,268 "pinned": false,269 "unpinned": null,270 "visible": true,271 "closed": false,272 "archived": false,273 "bookmarked": null,274 "liked": null,275 "tags_descriptions": {},276 "like_count": 0,277 "views": 913,278 "category_id": 1,279 "featured_link": null,280 "has_accepted_answer": false,281 "posters": [282 {283 "extras": "latest",284 "description": "Original Poster, Most Recent Poster",285 "user": {286 "id": 80717,287 "username": "janko",288 "name": "",289 "avatar_template": "/user_avatar/discuss.pytorch.org/janko/{size}/73795_2.png",290 "trust_level": 1291 }292 },293 {294 "extras": null,295 "description": "Frequent Poster",296 "user": {297 "id": 3534,298 "username": "ptrblck",299 "name": "",300 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",301 "admin": true,302 "moderator": true,303 "trust_level": 2304 }305 }306 ]307 },308 {309 "fancy_title": "Would calling torch.sort(…, stable=False) ever fail reproducibility?",310 "id": 212956,311 "title": "Would calling torch.sort(..., stable=False) ever fail reproducibility?",312 "slug": "would-calling-torch-sort-stable-false-ever-fail-reproducibility",313 "posts_count": 1,314 "reply_count": 0,315 "highest_post_number": 1,316 "image_url": null,317 "created_at": "2024-11-14T01:08:23.042Z",318 "last_posted_at": "2024-11-14T01:08:23.096Z",319 "bumped": true,320 "bumped_at": "2024-11-14T01:13:57.489Z",321 "archetype": "regular",322 "unseen": false,323 "pinned": false,324 "unpinned": null,325 "visible": true,326 "closed": false,327 "archived": false,328 "bookmarked": null,329 "liked": null,330 "tags_descriptions": {},331 "like_count": 0,332 "views": 70,333 "category_id": 1,334 "featured_link": null,335 "has_accepted_answer": false,336 "posters": [337 {338 "extras": "latest single",339 "description": "Original Poster, Most Recent Poster",340 "user": {341 "id": 80644,342 "username": "Craig_Hicks1",343 "name": "Craig Hicks",344 "avatar_template": "/user_avatar/discuss.pytorch.org/craig_hicks1/{size}/73727_2.png",345 "trust_level": 0346 }347 }348 ]349 },350 {351 "fancy_title": "Dino training- input image",352 "id": 213737,353 "title": "Dino training- input image",354 "slug": "dino-training-input-image",355 "posts_count": 5,356 "reply_count": 3,357 "highest_post_number": 5,358 "image_url": null,359 "created_at": "2024-12-03T08:01:43.211Z",360 "last_posted_at": "2024-12-09T10:30:16.201Z",361 "bumped": true,362 "bumped_at": "2024-12-09T10:30:16.201Z",363 "archetype": "regular",364 "unseen": false,365 "pinned": false,366 "unpinned": null,367 "visible": true,368 "closed": false,369 "archived": false,370 "bookmarked": null,371 "liked": null,372 "tags_descriptions": {},373 "like_count": 1,374 "views": 354,375 "category_id": 1,376 "featured_link": null,377 "has_accepted_answer": false,378 "posters": [379 {380 "extras": "latest",381 "description": "Original Poster, Most Recent Poster",382 "user": {383 "id": 81269,384 "username": "Barbo",385 "name": "",386 "avatar_template": "/user_avatar/discuss.pytorch.org/barbo/{size}/74321_2.png",387 "trust_level": 1388 }389 },390 {391 "extras": null,392 "description": "Frequent Poster",393 "user": {394 "id": 72430,395 "username": "Eduardo_Lawson",396 "name": "Eduardo Lawson da Silva",397 "avatar_template": "/user_avatar/discuss.pytorch.org/eduardo_lawson/{size}/66899_2.png",398 "trust_level": 2399 }400 }401 ]402 },403 {404 "fancy_title": "nn.BatchNorm is slow",405 "id": 212195,406 "title": "nn.BatchNorm is slow",407 "slug": "nn-batchnorm-is-slow",408 "posts_count": 1,409 "reply_count": 0,410 "highest_post_number": 1,411 "image_url": null,412 "created_at": "2024-10-28T08:24:38.399Z",413 "last_posted_at": "2024-10-28T08:24:38.458Z",414 "bumped": true,415 "bumped_at": "2024-10-28T08:24:38.458Z",416 "archetype": "regular",417 "unseen": false,418 "pinned": false,419 "unpinned": null,420 "visible": true,421 "closed": false,422 "archived": false,423 "bookmarked": null,424 "liked": null,425 "tags_descriptions": {},426 "like_count": 0,427 "views": 131,428 "category_id": 1,429 "featured_link": null,430 "has_accepted_answer": false,431 "posters": [432 {433 "extras": "latest single",434 "description": "Original Poster, Most Recent Poster",435 "user": {436 "id": 80459,437 "username": "dailaila",438 "name": null,439 "avatar_template": "/letter_avatar_proxy/v4/letter/d/50afbb/{size}.png",440 "trust_level": 0441 }442 }443 ]444 }445 ],446 "tags_descriptions": {},447 "fancy_title": "Do I need a custom modified MultiheadAttention layer?",448 "id": 91940,449 "title": "Do I need a custom modified MultiheadAttention layer?",450 "posts_count": 3,451 "created_at": "2020-08-07T09:13:02.925Z",452 "views": 553,453 "reply_count": 1,454 "like_count": 0,455 "last_posted_at": "2020-08-07T14:18:10.398Z",456 "visible": true,457 "closed": false,458 "archived": false,459 "has_summary": false,460 "archetype": "regular",461 "slug": "do-i-need-a-custom-modified-multiheadattention-layer",462 "category_id": 1,463 "word_count": 461,464 "deleted_at": null,465 "user_id": 35250,466 "featured_link": null,467 "pinned_globally": false,468 "pinned_at": null,469 "pinned_until": null,470 "image_url": null,471 "slow_mode_seconds": 0,472 "draft": null,473 "draft_key": "topic_91940",474 "draft_sequence": null,475 "unpinned": null,476 "pinned": false,477 "current_post_number": 1,478 "highest_post_number": 3,479 "deleted_by": null,480 "actions_summary": [481 {482 "id": 4,483 "count": 0,484 "hidden": false,485 "can_act": false486 },487 {488 "id": 8,489 "count": 0,490 "hidden": false,491 "can_act": false492 },493 {494 "id": 10,495 "count": 0,496 "hidden": false,497 "can_act": false498 },499 {500 "id": 7,501 "count": 0,502 "hidden": false,503 "can_act": false504 }505 ],506 "chunk_size": 20,507 "bookmarked": false,508 "topic_timer": null,509 "message_bus_last_id": 0,510 "participant_count": 2,511 "show_read_indicator": false,512 "thumbnails": null,513 "slow_mode_enabled_until": null,514 "can_vote": false,515 "vote_count": 0,516 "user_voted": false,517 "discourse_zendesk_plugin_zendesk_id": null,518 "discourse_zendesk_plugin_zendesk_url": "https://your-url.zendesk.com/agent/tickets/",519 "details": {520 "can_edit": false,521 "notification_level": 1,522 "participants": [523 {524 "id": 35250,525 "username": "siahuat0727",526 "name": "siahuat0727",527 "avatar_template": "/user_avatar/discuss.pytorch.org/siahuat0727/{size}/27590_2.png",528 "post_count": 2,529 "primary_group_name": null,530 "flair_name": null,531 "flair_url": null,532 "flair_color": null,533 "flair_bg_color": null,534 "flair_group_id": null,535 "trust_level": 2536 },537 {538 "id": 29396,539 "username": "SandPhoenix",540 "name": "",541 "avatar_template": "/user_avatar/discuss.pytorch.org/sandphoenix/{size}/29137_2.png",542 "post_count": 1,543 "primary_group_name": null,544 "flair_name": null,545 "flair_url": null,546 "flair_color": null,547 "flair_bg_color": null,548 "flair_group_id": null,549 "trust_level": 2550 }551 ],552 "created_by": {553 "id": 35250,554 "username": "siahuat0727",555 "name": "siahuat0727",556 "avatar_template": "/user_avatar/discuss.pytorch.org/siahuat0727/{size}/27590_2.png"557 },558 "last_poster": {559 "id": 35250,560 "username": "siahuat0727",561 "name": "siahuat0727",562 "avatar_template": "/user_avatar/discuss.pytorch.org/siahuat0727/{size}/27590_2.png"563 }564 },565 "bookmarks": []566 },567 {568 "post_stream": {569 "posts": [570 {571 "id": 217383,572 "name": "sa_lu",573 "username": "saluei",574 "avatar_template": "/user_avatar/discuss.pytorch.org/saluei/{size}/44594_2.png",575 "created_at": "2020-08-01T02:53:22.238Z",576 "cooked": "<p>I have input tensor (N*2) , first column is category and second is the value, something like this:</p>\n<p>[ [14,50] , [18,1.5] , [14,250] , [18,2.5] , [10,1] , [14,252] , [18,5.3] ]</p>\n<p>each category may be have N row in input tensor,<br>\nhow to found index of row in input tensor that have max value in each category, something like this for above input tensor</p>\n<p>[ [14,5] , [18,6] , [10,4] ]</p>\n<p>[ category code, row index have max value ]</p>\n<p>but if it’s possible without using loop,<br>\nthank’s</p>",577 "post_number": 1,578 "post_type": 1,579 "posts_count": 3,580 "updated_at": "2020-08-01T02:54:59.297Z",581 "reply_count": 0,582 "reply_to_post_number": null,583 "quote_count": 0,584 "incoming_link_count": 107,585 "reads": 11,586 "readers_count": 10,587 "score": 537.2,588 "yours": false,589 "topic_id": 91279,590 "topic_slug": "find-max-row-index-with-some-condition",591 "display_username": "sa_lu",592 "primary_group_name": null,593 "flair_name": null,594 "flair_url": null,595 "flair_bg_color": null,596 "flair_color": null,597 "flair_group_id": null,598 "badges_granted": [],599 "version": 1,600 "can_edit": false,601 "can_delete": false,602 "can_recover": false,603 "can_see_hidden_post": false,604 "can_wiki": false,605 "read": true,606 "user_title": null,607 "bookmarked": false,608 "actions_summary": [],609 "moderator": false,610 "admin": false,611 "staff": false,612 "user_id": 28364,613 "hidden": false,614 "trust_level": 2,615 "deleted_at": null,616 "user_deleted": false,617 "edit_reason": null,618 "can_view_edit_history": true,619 "wiki": false,620 "post_url": "/t/find-max-row-index-with-some-condition/91279/1",621 "can_accept_answer": false,622 "can_unaccept_answer": false,623 "accepted_answer": false,624 "topic_accepted_answer": null,625 "can_vote": false626 },627 {628 "id": 217519,629 "name": "Kushajveer Singh",630 "username": "Kushaj",631 "avatar_template": "/user_avatar/discuss.pytorch.org/kushaj/{size}/58838_2.png",632 "created_at": "2020-08-01T23:26:06.544Z",633 "cooked": "<p>There is no groupby function in pytorch. So not possible without a for loop. You can convert to numpy and then use pandas groupby.</p>",634 "post_number": 2,635 "post_type": 1,636 "posts_count": 3,637 "updated_at": "2020-08-01T23:26:06.544Z",638 "reply_count": 1,639 "reply_to_post_number": null,640 "quote_count": 0,641 "incoming_link_count": 2,642 "reads": 8,643 "readers_count": 7,644 "score": 16.6,645 "yours": false,646 "topic_id": 91279,647 "topic_slug": "find-max-row-index-with-some-condition",648 "display_username": "Kushajveer Singh",649 "primary_group_name": null,650 "flair_name": null,651 "flair_url": null,652 "flair_bg_color": null,653 "flair_color": null,654 "flair_group_id": null,655 "badges_granted": [],656 "version": 1,657 "can_edit": false,658 "can_delete": false,659 "can_recover": false,660 "can_see_hidden_post": false,661 "can_wiki": false,662 "read": true,663 "user_title": "",664 "bookmarked": false,665 "actions_summary": [],666 "moderator": false,667 "admin": false,668 "staff": false,669 "user_id": 14623,670 "hidden": false,671 "trust_level": 2,672 "deleted_at": null,673 "user_deleted": false,674 "edit_reason": null,675 "can_view_edit_history": true,676 "wiki": false,677 "post_url": "/t/find-max-row-index-with-some-condition/91279/2",678 "can_accept_answer": false,679 "can_unaccept_answer": false,680 "accepted_answer": false,681 "topic_accepted_answer": null682 },683 {684 "id": 218966,685 "name": "sa_lu",686 "username": "saluei",687 "avatar_template": "/user_avatar/discuss.pytorch.org/saluei/{size}/44594_2.png",688 "created_at": "2020-08-07T13:52:58.268Z",689 "cooked": "<p>yes in fact I search something like group by in PyTorch and do not find,<br>\nThank’s for reply Kushaj</p>",690 "post_number": 3,691 "post_type": 1,692 "posts_count": 3,693 "updated_at": "2020-08-07T13:53:13.096Z",694 "reply_count": 0,695 "reply_to_post_number": 2,696 "quote_count": 0,697 "incoming_link_count": 0,698 "reads": 7,699 "readers_count": 6,700 "score": 16.4,701 "yours": false,702 "topic_id": 91279,703 "topic_slug": "find-max-row-index-with-some-condition",704 "display_username": "sa_lu",705 "primary_group_name": null,706 "flair_name": null,707 "flair_url": null,708 "flair_bg_color": null,709 "flair_color": null,710 "flair_group_id": null,711 "badges_granted": [],712 "version": 1,713 "can_edit": false,714 "can_delete": false,715 "can_recover": false,716 "can_see_hidden_post": false,717 "can_wiki": false,718 "read": true,719 "user_title": null,720 "reply_to_user": {721 "id": 14623,722 "username": "Kushaj",723 "name": "Kushajveer Singh",724 "avatar_template": "/user_avatar/discuss.pytorch.org/kushaj/{size}/58838_2.png"725 },726 "bookmarked": false,727 "actions_summary": [728 {729 "id": 2,730 "count": 1731 }732 ],733 "moderator": false,734 "admin": false,735 "staff": false,736 "user_id": 28364,737 "hidden": false,738 "trust_level": 2,739 "deleted_at": null,740 "user_deleted": false,741 "edit_reason": null,742 "can_view_edit_history": true,743 "wiki": false,744 "post_url": "/t/find-max-row-index-with-some-condition/91279/3",745 "can_accept_answer": false,746 "can_unaccept_answer": false,747 "accepted_answer": false,748 "topic_accepted_answer": null749 }750 ],751 "stream": [752 217383,753 217519,754 218966755 ]756 },757 "timeline_lookup": [758 [759 1,760 1912761 ],762 [763 2,764 1911765 ],766 [767 3,768 1905769 ]770 ],771 "suggested_topics": [772 {773 "fancy_title": "The meaning of some fields in the state_dict of a quantized model",774 "id": 213977,775 "title": "The meaning of some fields in the state_dict of a quantized model",776 "slug": "the-meaning-of-some-fields-in-the-state-dict-of-a-quantized-model",777 "posts_count": 1,778 "reply_count": 0,779 "highest_post_number": 1,780 "image_url": null,781 "created_at": "2024-12-09T05:56:44.856Z",782 "last_posted_at": "2024-12-09T05:56:44.912Z",783 "bumped": true,784 "bumped_at": "2024-12-09T05:56:44.912Z",785 "archetype": "regular",786 "unseen": false,787 "pinned": false,788 "unpinned": null,789 "visible": true,790 "closed": false,791 "archived": false,792 "bookmarked": null,793 "liked": null,794 "tags_descriptions": {},795 "like_count": 0,796 "views": 27,797 "category_id": 1,798 "featured_link": null,799 "has_accepted_answer": false,800 "posters": [801 {802 "extras": "latest single",803 "description": "Original Poster, Most Recent Poster",804 "user": {805 "id": 62571,806 "username": "jensenchen",807 "name": "jensenchen",808 "avatar_template": "/user_avatar/discuss.pytorch.org/jensenchen/{size}/74431_2.png",809 "trust_level": 1810 }811 }812 ]813 },814 {815 "fancy_title": "About the uniqueness of eigen decomposition with torch.eigh",816 "id": 212468,817 "title": "About the uniqueness of eigen decomposition with torch.eigh",818 "slug": "about-the-uniqueness-of-eigen-decomposition-with-torch-eigh",819 "posts_count": 2,820 "reply_count": 0,821 "highest_post_number": 2,822 "image_url": "https://discuss.pytorch.org/uploads/default/original/3X/2/1/21f5e2cb3119b8de74bce7c3b6be1a76554ed46d.png",823 "created_at": "2024-11-03T08:41:03.397Z",824 "last_posted_at": "2024-11-04T21:39:04.342Z",825 "bumped": true,826 "bumped_at": "2024-11-04T21:39:04.342Z",827 "archetype": "regular",828 "unseen": false,829 "pinned": false,830 "unpinned": null,831 "visible": true,832 "closed": false,833 "archived": false,834 "bookmarked": null,835 "liked": null,836 "tags_descriptions": {},837 "like_count": 1,838 "views": 91,839 "category_id": 1,840 "featured_link": null,841 "has_accepted_answer": true,842 "posters": [843 {844 "extras": null,845 "description": "Original Poster",846 "user": {847 "id": 80508,848 "username": "wasabi_linguist",849 "name": "",850 "avatar_template": "/user_avatar/discuss.pytorch.org/wasabi_linguist/{size}/73592_2.png",851 "trust_level": 1852 }853 },854 {855 "extras": "latest",856 "description": "Most Recent Poster, Accepted Answer",857 "user": {858 "id": 18088,859 "username": "KFrank",860 "name": "K. Frank",861 "avatar_template": "/letter_avatar_proxy/v4/letter/k/ecb155/{size}.png",862 "trust_level": 2863 }864 }865 ]866 },867 {868 "fancy_title": "torch.cudnn.benchmark=True picks slower algorithm",869 "id": 215906,870 "title": "torch.cudnn.benchmark=True picks slower algorithm",871 "slug": "torch-cudnn-benchmark-true-picks-slower-algorithm",872 "posts_count": 7,873 "reply_count": 4,874 "highest_post_number": 7,875 "image_url": null,876 "created_at": "2025-01-26T21:29:30.709Z",877 "last_posted_at": "2025-01-29T05:02:54.062Z",878 "bumped": true,879 "bumped_at": "2025-01-29T11:33:31.111Z",880 "archetype": "regular",881 "unseen": false,882 "pinned": false,883 "unpinned": null,884 "visible": true,885 "closed": false,886 "archived": false,887 "bookmarked": null,888 "liked": null,889 "tags_descriptions": {},890 "like_count": 1,891 "views": 161,892 "category_id": 1,893 "featured_link": null,894 "has_accepted_answer": false,895 "posters": [896 {897 "extras": "latest",898 "description": "Original Poster, Most Recent Poster",899 "user": {900 "id": 82331,901 "username": "nkem",902 "name": "Nico",903 "avatar_template": "/user_avatar/discuss.pytorch.org/nkem/{size}/75310_2.png",904 "trust_level": 1905 }906 },907 {908 "extras": null,909 "description": "Frequent Poster",910 "user": {911 "id": 3534,912 "username": "ptrblck",913 "name": "",914 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",915 "admin": true,916 "moderator": true,917 "trust_level": 2918 }919 },920 {921 "extras": null,922 "description": "Frequent Poster",923 "user": {924 "id": 35275,925 "username": "eqy",926 "name": "",927 "avatar_template": "/user_avatar/discuss.pytorch.org/eqy/{size}/46233_2.png",928 "trust_level": 2929 }930 }931 ]932 },933 {934 "fancy_title": "How can I combine my custom conv2D in CUDA C and autograd of pytorch",935 "id": 216106,936 "title": "How can I combine my custom conv2D in CUDA C and autograd of pytorch",937 "slug": "how-can-i-combine-my-custom-conv2d-in-cuda-c-and-autograd-of-pytorch",938 "posts_count": 1,939 "reply_count": 0,940 "highest_post_number": 1,941 "image_url": null,942 "created_at": "2025-01-31T22:01:03.923Z",943 "last_posted_at": "2025-01-31T22:01:03.968Z",944 "bumped": true,945 "bumped_at": "2025-01-31T22:01:03.968Z",946 "archetype": "regular",947 "unseen": false,948 "pinned": false,949 "unpinned": null,950 "visible": true,951 "closed": false,952 "archived": false,953 "bookmarked": null,954 "liked": null,955 "tags_descriptions": {},956 "like_count": 0,957 "views": 32,958 "category_id": 1,959 "featured_link": null,960 "has_accepted_answer": false,961 "posters": [962 {963 "extras": "latest single",964 "description": "Original Poster, Most Recent Poster",965 "user": {966 "id": 66124,967 "username": "Aakira",968 "name": "Aakira",969 "avatar_template": "/user_avatar/discuss.pytorch.org/aakira/{size}/58984_2.png",970 "trust_level": 1971 }972 }973 ]974 },975 {976 "fancy_title": "2 gpu model loading cuda error",977 "id": 216669,978 "title": "2 gpu model loading cuda error",979 "slug": "2-gpu-model-loading-cuda-error",980 "posts_count": 1,981 "reply_count": 0,982 "highest_post_number": 1,983 "image_url": null,984 "created_at": "2025-02-14T10:56:48.064Z",985 "last_posted_at": "2025-02-14T10:56:48.102Z",986 "bumped": true,987 "bumped_at": "2025-02-14T11:01:40.215Z",988 "archetype": "regular",989 "unseen": false,990 "pinned": false,991 "unpinned": null,992 "visible": true,993 "closed": false,994 "archived": false,995 "bookmarked": null,996 "liked": null,997 "tags_descriptions": {},998 "like_count": 0,999 "views": 87,1000 "category_id": 1,1001 "featured_link": null,1002 "has_accepted_answer": false,1003 "posters": [1004 {1005 "extras": "latest single",1006 "description": "Original Poster, Most Recent Poster",1007 "user": {1008 "id": 82364,1009 "username": "Sourabh_Yadav",1010 "name": "Sourabh Yadav",1011 "avatar_template": "/user_avatar/discuss.pytorch.org/sourabh_yadav/{size}/75350_2.png",1012 "trust_level": 11013 }1014 }1015 ]1016 }1017 ],1018 "tags_descriptions": {},1019 "fancy_title": "Find max row index with some condition",1020 "id": 91279,1021 "title": "Find max row index with some condition",1022 "posts_count": 3,1023 "created_at": "2020-08-01T02:53:22.189Z",1024 "views": 514,1025 "reply_count": 1,1026 "like_count": 1,1027 "last_posted_at": "2020-08-07T13:52:58.268Z",1028 "visible": true,1029 "closed": false,1030 "archived": false,1031 "has_summary": false,1032 "archetype": "regular",1033 "slug": "find-max-row-index-with-some-condition",1034 "category_id": 1,1035 "word_count": 135,1036 "deleted_at": null,1037 "user_id": 28364,1038 "featured_link": null,1039 "pinned_globally": false,1040 "pinned_at": null,1041 "pinned_until": null,1042 "image_url": null,1043 "slow_mode_seconds": 0,1044 "draft": null,1045 "draft_key": "topic_91279",1046 "draft_sequence": null,1047 "unpinned": null,1048 "pinned": false,1049 "current_post_number": 1,1050 "highest_post_number": 3,1051 "deleted_by": null,1052 "actions_summary": [1053 {1054 "id": 4,1055 "count": 0,1056 "hidden": false,1057 "can_act": false1058 },1059 {1060 "id": 8,1061 "count": 0,1062 "hidden": false,1063 "can_act": false1064 },1065 {1066 "id": 10,1067 "count": 0,1068 "hidden": false,1069 "can_act": false1070 },1071 {1072 "id": 7,1073 "count": 0,1074 "hidden": false,1075 "can_act": false1076 }1077 ],1078 "chunk_size": 20,1079 "bookmarked": false,1080 "topic_timer": null,1081 "message_bus_last_id": 0,1082 "participant_count": 2,1083 "show_read_indicator": false,1084 "thumbnails": null,1085 "slow_mode_enabled_until": null,1086 "can_vote": false,1087 "vote_count": 0,1088 "user_voted": false,1089 "discourse_zendesk_plugin_zendesk_id": null,1090 "discourse_zendesk_plugin_zendesk_url": "https://your-url.zendesk.com/agent/tickets/",1091 "details": {1092 "can_edit": false,1093 "notification_level": 1,1094 "participants": [1095 {1096 "id": 28364,1097 "username": "saluei",1098 "name": "sa_lu",1099 "avatar_template": "/user_avatar/discuss.pytorch.org/saluei/{size}/44594_2.png",1100 "post_count": 2,1101 "primary_group_name": null,1102 "flair_name": null,1103 "flair_url": null,1104 "flair_color": null,1105 "flair_bg_color": null,1106 "flair_group_id": null,1107 "trust_level": 21108 },1109 {1110 "id": 14623,1111 "username": "Kushaj",1112 "name": "Kushajveer Singh",1113 "avatar_template": "/user_avatar/discuss.pytorch.org/kushaj/{size}/58838_2.png",1114 "post_count": 1,1115 "primary_group_name": null,1116 "flair_name": null,1117 "flair_url": null,1118 "flair_color": null,1119 "flair_bg_color": null,1120 "flair_group_id": null,1121 "trust_level": 21122 }1123 ],1124 "created_by": {1125 "id": 28364,1126 "username": "saluei",1127 "name": "sa_lu",1128 "avatar_template": "/user_avatar/discuss.pytorch.org/saluei/{size}/44594_2.png"1129 },1130 "last_poster": {1131 "id": 28364,1132 "username": "saluei",1133 "name": "sa_lu",1134 "avatar_template": "/user_avatar/discuss.pytorch.org/saluei/{size}/44594_2.png"1135 }1136 },1137 "bookmarks": []1138 },1139 {1140 "post_stream": {1141 "posts": [1142 {1143 "id": 218936,1144 "name": "",1145 "username": "n4tman",1146 "avatar_template": "/letter_avatar_proxy/v4/letter/n/bc8723/{size}.png",1147 "created_at": "2020-08-07T13:20:40.828Z",1148 "cooked": "<p>Do the nn pool2d layers (nn.AvgPool2d, nn.MaxPool2d, nn.AdaptiveAvgPool2d, nn.AdpativeMaxPool2d) work as specified below, replacing the MAX by average when using average pooling?</p>\n<blockquote>\n<p>It is common to periodically insert a Pooling layer in-between successive Conv layers in a ConvNet architecture. Its function is to progressively reduce the spatial size of the representation to reduce the amount of parameters and computation in the network, and hence to also control overfitting. The Pooling Layer operates independently on every depth slice of the input and resizes it spatially, using the MAX operation. The most common form is a pooling layer with filters of size 2x2 applied with a stride of 2 downsamples every depth slice in the input by 2 along both width and height, discarding 75% of the activations. Every MAX operation would in this case be taking a MAX over 4 numbers (little 2x2 region in some depth slice). The depth dimension remains unchanged.</p>\n</blockquote>\n<p>Thanks!</p>",1149 "post_number": 1,1150 "post_type": 1,1151 "posts_count": 2,1152 "updated_at": "2020-08-07T13:21:02.769Z",1153 "reply_count": 0,1154 "reply_to_post_number": null,1155 "quote_count": 0,1156 "incoming_link_count": 28,1157 "reads": 7,1158 "readers_count": 6,1159 "score": 141.4,1160 "yours": false,1161 "topic_id": 91975,1162 "topic_slug": "confusion-about-pooling-layers",1163 "display_username": "",1164 "primary_group_name": null,1165 "flair_name": null,1166 "flair_url": null,1167 "flair_bg_color": null,1168 "flair_color": null,1169 "flair_group_id": null,1170 "badges_granted": [],1171 "version": 1,1172 "can_edit": false,1173 "can_delete": false,1174 "can_recover": false,1175 "can_see_hidden_post": false,1176 "can_wiki": false,1177 "read": true,1178 "user_title": null,1179 "bookmarked": false,1180 "actions_summary": [],1181 "moderator": false,1182 "admin": false,1183 "staff": false,1184 "user_id": 35011,1185 "hidden": false,1186 "trust_level": 1,1187 "deleted_at": null,1188 "user_deleted": false,1189 "edit_reason": null,1190 "can_view_edit_history": true,1191 "wiki": false,1192 "post_url": "/t/confusion-about-pooling-layers/91975/1",1193 "can_accept_answer": false,1194 "can_unaccept_answer": false,1195 "accepted_answer": false,1196 "topic_accepted_answer": null,1197 "can_vote": false1198 },1199 {1200 "id": 218940,