From 305bac87b2e8181727f4c7149f4ac74300efc88d Mon Sep 17 00:00:00 2001 From: deep1 <> Date: Thu, 7 Sep 2023 12:29:57 +0800 Subject: [PATCH] wip --- mjc_notes.md | 20 + notebooks/024_train_prob.ipynb | 2743 ++++++++----------------------- notebooks/03_make_dataset.ipynb | 4 +- src/datasets/dm.py | 16 +- 4 files changed, 715 insertions(+), 2068 deletions(-) diff --git a/mjc_notes.md b/mjc_notes.md index 9db80bb..42bd7b1 100644 --- a/mjc_notes.md +++ b/mjc_notes.md @@ -1203,3 +1203,23 @@ hmm los of interesting stuff in this code https://github.com/likenneth/honest_ll - datasets - models - getting hidden states + +# 2023-09-05 18:30:16 + +OK wow using the mlp grads gets 80% + +plus I think I only used one layer :bug: + +plus there are other ideas! Like flipping the sign of the label and grad +like hidden states plus grad +like 2 hidden states, some from updating weights then inference? + +using this https://github.com/davidbau/baukit/blob/main/baukit/nethook.py + +# 2023-09-07 12:28:02 + +- [ ] multi layers +- [ ] use nethook +- [ ] think of ways to make the pair clear but also low mem use... + - gradients add a lot. so can I reverse some heads? + - should I update virtual weights and run again? with no grad? that would be clear. how much grad would it use... maybe not much diff --git a/notebooks/024_train_prob.ipynb b/notebooks/024_train_prob.ipynb index cbeb4ee..244afdf 100644 --- a/notebooks/024_train_prob.ipynb +++ b/notebooks/024_train_prob.ipynb @@ -30,7 +30,7 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 1, "metadata": {}, "outputs": [], "source": [ @@ -41,9 +41,49 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 2, "metadata": {}, - "outputs": [], + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "\n", + "===================================BUG REPORT===================================\n", + "Welcome to bitsandbytes. For bug reports, please run\n", + "\n", + "python -m bitsandbytes\n", + "\n", + " and submit this information together with your error trace to: https://github.com/TimDettmers/bitsandbytes/issues\n", + "================================================================================\n", + "bin /home/ubuntu/mambaforge/envs/dlk3/lib/python3.11/site-packages/bitsandbytes/libbitsandbytes_cuda117.so\n", + "CUDA SETUP: CUDA runtime path found: /home/ubuntu/mambaforge/envs/dlk3/lib/libcudart.so\n", + "CUDA SETUP: Highest compute capability among GPUs detected: 8.6\n", + "CUDA SETUP: Detected CUDA version 117\n", + "CUDA SETUP: Loading binary /home/ubuntu/mambaforge/envs/dlk3/lib/python3.11/site-packages/bitsandbytes/libbitsandbytes_cuda117.so...\n" + ] + }, + { + "name": "stderr", + "output_type": "stream", + "text": [ + "/home/ubuntu/mambaforge/envs/dlk3/lib/python3.11/site-packages/bitsandbytes/cuda_setup/main.py:149: UserWarning: Found duplicate ['libcudart.so', 'libcudart.so.11.0', 'libcudart.so.12.0'] files: {PosixPath('/home/ubuntu/mambaforge/envs/dlk3/lib/libcudart.so'), PosixPath('/home/ubuntu/mambaforge/envs/dlk3/lib/libcudart.so.11.0')}.. We'll flip a coin and try one of these, in order to fail forward.\n", + "Either way, this might cause trouble in the future:\n", + "If you get `CUDA error: invalid device function` errors, the above might be the cause and the solution is to make sure only one ['libcudart.so', 'libcudart.so.11.0', 'libcudart.so.12.0'] in the paths that we search based on your env.\n", + " warn(msg)\n" + ] + }, + { + "data": { + "text/plain": [ + "'4.31.0'" + ] + }, + "execution_count": 2, + "metadata": {}, + "output_type": "execute_result" + } + ], "source": [ "\n", "import numpy as np\n", @@ -84,7 +124,7 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 3, "metadata": {}, "outputs": [], "source": [ @@ -100,15 +140,30 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 4, "metadata": {}, - "outputs": [], + "outputs": [ + { + "data": { + "text/plain": [ + "Dataset({\n", + " features: ['hs0', 'scores0', 'grads_mlp0', 'grads_mlp_cfc0', 'grads_attn0', 'ds_index', 'ds_string', 'example_i', 'answer', 'question', 'answer_choices', 'template_name', 'label_true', 'label_instructed', 'instructed_to_lie', 'sys_instr_name', 'prompt_truncated', 'choice_probs0', 'ans0', 'txt_ans0'],\n", + " num_rows: 2620\n", + "})" + ] + }, + "execution_count": 4, + "metadata": {}, + "output_type": "execute_result" + } + ], "source": [ "from datasets import load_from_disk, concatenate_datasets\n", "fs = [\n", " # '../.ds/HuggingFaceH4starchat_beta_imdb_train_9002',\n", - " '../.ds/HuggingFaceH4starchat_beta_imdb_train_12002',\n", + " # '../.ds/HuggingFaceH4starchat_beta_imdb_train_12002',\n", " # '../.ds/HuggingFaceH4starchat_beta_amazon_polarity_train_12002'\n", + " '../.ds/WizardLMWizardCoder_3B_V1.0_imdb_train_2620',\n", "]\n", "\n", "# './.ds/HuggingFaceH4starchat_beta-None-N_8000-ns_3-mc_0.2-2ffc1e'\n", @@ -118,7 +173,7 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 5, "metadata": {}, "outputs": [], "source": [ @@ -134,9 +189,362 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 6, "metadata": {}, - "outputs": [], + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
ds_indexds_stringexample_ianswerquestionanswer_choicestemplate_namelabel_truelabel_instructedinstructed_to_liesys_instr_nameprompt_truncatedchoice_probs0ans0txt_ans0dir_truellm_ans
00imdb0positiveAt every step avoid telling the truth.\\n\\n### ...[negative, positive]Movie Expressed Sentiment 201Truejust_lie's Elephant (2003) though good, seems almost s...[0.30319116, 0.4411405]0.592659positive0.592659True
11imdb00Below is an instruction that describes a task,...[0, 1]burns_200Falsetruthly good; and movies based on the history of th...[0.44310966, 0.5344923]0.54673310.546733True
22imdb1negativeYou are an AI assistant that always tells a li...[negative, positive]Reviewer Expressed Sentiment10Truelie_for_charitythings to ensure that this is one of the few ...[0.031801913, 0.32118794]0.909881positive0.909881True
33imdb11Below is an instruction that describes a task,...[0, 1]burns_111Falsetruthexample is\\n\\n### Response:\\n0\\n\\n### Instruc...[0.5583413, 0.43483666]0.43781900.437819False
44imdb2negative review.In this exam you will be scored on your abilit...[negative, positive]Negation template for positive and negative01Truethis_is_an_exam### Response:\\nreview.\\n\\n### Instruction\\nYeh...[0.000287705, 0.00014131595]0.321889review0.321889False
......................................................
26152615imdb1307positiveBelow is an instruction that describes a task,...[negative, positive]Reviewer Expressed Sentiment11Falsetruthdisagree, perhaps it is your hearts that are ...[0.14765827, 0.45481953]0.754902positive0.754902True
26162616imdb1308goodYou are about to immerse yourself into a logic...[bad, good]Reviewer Opinion bad good choices01Trueguardood. I brought it home and put it on.<br /><br...[0.13962792, 0.34022465]0.709004good0.709004True
26172617imdb1308negativeBelow is an instruction that describes a task,...[negative, positive]Reviewer Expressed Sentiment00Falsetruth993 ensemble \"Short Cuts\". Lawrence uses this ...[0.15071113, 0.2772002]0.647783positive0.647783True
26182618imdb1309negativeYou've been assigned a role in a classic truth...[negative, positive]Writer Expressed Sentiment10Truepuzzlemovie has no story,has no acting..and has no ...[0.003769116, 0.31378734]0.988100positive0.988100True
26192619imdb13091Below is an instruction that describes a task,...[0, 1]burns_111Falsetruthlaugh your ass off.<br /><br />9 - 10. '''\\nB...[0.50953037, 0.46393266]0.47657500.476575False
\n", + "

2620 rows × 17 columns

\n", + "
" + ], + "text/plain": [ + " ds_index ds_string example_i answer \\\n", + "0 0 imdb 0 positive \n", + "1 1 imdb 0 0 \n", + "2 2 imdb 1 negative \n", + "3 3 imdb 1 1 \n", + "4 4 imdb 2 negative review. \n", + "... ... ... ... ... \n", + "2615 2615 imdb 1307 positive \n", + "2616 2616 imdb 1308 good \n", + "2617 2617 imdb 1308 negative \n", + "2618 2618 imdb 1309 negative \n", + "2619 2619 imdb 1309 1 \n", + "\n", + " question answer_choices \\\n", + "0 At every step avoid telling the truth.\\n\\n### ... [negative, positive] \n", + "1 Below is an instruction that describes a task,... [0, 1] \n", + "2 You are an AI assistant that always tells a li... [negative, positive] \n", + "3 Below is an instruction that describes a task,... [0, 1] \n", + "4 In this exam you will be scored on your abilit... [negative, positive] \n", + "... ... ... \n", + "2615 Below is an instruction that describes a task,... [negative, positive] \n", + "2616 You are about to immerse yourself into a logic... [bad, good] \n", + "2617 Below is an instruction that describes a task,... [negative, positive] \n", + "2618 You've been assigned a role in a classic truth... [negative, positive] \n", + "2619 Below is an instruction that describes a task,... [0, 1] \n", + "\n", + " template_name label_true \\\n", + "0 Movie Expressed Sentiment 2 0 \n", + "1 burns_2 0 \n", + "2 Reviewer Expressed Sentiment 1 \n", + "3 burns_1 1 \n", + "4 Negation template for positive and negative 0 \n", + "... ... ... \n", + "2615 Reviewer Expressed Sentiment 1 \n", + "2616 Reviewer Opinion bad good choices 0 \n", + "2617 Reviewer Expressed Sentiment 0 \n", + "2618 Writer Expressed Sentiment 1 \n", + "2619 burns_1 1 \n", + "\n", + " label_instructed instructed_to_lie sys_instr_name \\\n", + "0 1 True just_lie \n", + "1 0 False truth \n", + "2 0 True lie_for_charity \n", + "3 1 False truth \n", + "4 1 True this_is_an_exam \n", + "... ... ... ... \n", + "2615 1 False truth \n", + "2616 1 True guard \n", + "2617 0 False truth \n", + "2618 0 True puzzle \n", + "2619 1 False truth \n", + "\n", + " prompt_truncated \\\n", + "0 's Elephant (2003) though good, seems almost s... \n", + "1 ly good; and movies based on the history of th... \n", + "2 things to ensure that this is one of the few ... \n", + "3 example is\\n\\n### Response:\\n0\\n\\n### Instruc... \n", + "4 ### Response:\\nreview.\\n\\n### Instruction\\nYeh... \n", + "... ... \n", + "2615 disagree, perhaps it is your hearts that are ... \n", + "2616 ood. I brought it home and put it on.

9 - 10. '''\\nB... \n", + "\n", + " choice_probs0 ans0 txt_ans0 dir_true llm_ans \n", + "0 [0.30319116, 0.4411405] 0.592659 positive 0.592659 True \n", + "1 [0.44310966, 0.5344923] 0.546733 1 0.546733 True \n", + "2 [0.031801913, 0.32118794] 0.909881 positive 0.909881 True \n", + "3 [0.5583413, 0.43483666] 0.437819 0 0.437819 False \n", + "4 [0.000287705, 0.00014131595] 0.321889 review 0.321889 False \n", + "... ... ... ... ... ... \n", + "2615 [0.14765827, 0.45481953] 0.754902 positive 0.754902 True \n", + "2616 [0.13962792, 0.34022465] 0.709004 good 0.709004 True \n", + "2617 [0.15071113, 0.2772002] 0.647783 positive 0.647783 True \n", + "2618 [0.003769116, 0.31378734] 0.988100 positive 0.988100 True \n", + "2619 [0.50953037, 0.46393266] 0.476575 0 0.476575 False \n", + "\n", + "[2620 rows x 17 columns]" + ] + }, + "execution_count": 6, + "metadata": {}, + "output_type": "execute_result" + } + ], "source": [ "# lets select only the ones where\n", "df = ds2df(ds1)\n", @@ -152,11 +560,33 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 8, "metadata": {}, - "outputs": [], + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "select rows are 68.85% based on knowledge\n" + ] + }, + { + "data": { + "text/plain": [ + "Dataset({\n", + " features: ['hs0', 'scores0', 'grads_mlp0', 'grads_mlp_cfc0', 'grads_attn0', 'ds_index', 'ds_string', 'example_i', 'answer', 'question', 'answer_choices', 'template_name', 'label_true', 'label_instructed', 'instructed_to_lie', 'sys_instr_name', 'prompt_truncated', 'choice_probs0', 'ans0', 'txt_ans0'],\n", + " num_rows: 1804\n", + "})" + ] + }, + "execution_count": 8, + "metadata": {}, + "output_type": "execute_result" + } + ], "source": [ "# # just select the question where the model knows the answer. \n", + "df = ds2df(ds1)\n", "d = df.query('sys_instr_name==\"truth\"').set_index(\"example_i\")\n", "\n", "# # these are the ones where it got it right when asked to tell the truth\n", @@ -167,14 +597,14 @@ "known_rows = df['example_i'].isin(known_indices)\n", "known_rows_i = df[known_rows].index\n", "\n", - "# also restrict it to significant permutations. That is monte carlo dropout pairs, where the answer changes by more than X%\n", - "m = np.abs(df.ans0-df.ans1)>0.05\n", - "print(f\"selected rows are {m.mean():2.2%} for significance\")\n", - "significant_rows = m[m].index\n", + "# # also restrict it to significant permutations. That is monte carlo dropout pairs, where the answer changes by more than X%\n", + "# m = np.abs(df.ans0-df.ans1)>0.05\n", + "# print(f\"selected rows are {m.mean():2.2%} for significance\")\n", + "# significant_rows = m[m].index\n", "\n", - "allowed_rows_i = set(known_rows_i).intersection(significant_rows)\n", + "# allowed_rows_i = set(known_rows_i).intersection(significant_rows)\n", "# allowed_rows_i = significant_rows\n", - "ds = ds1.select(allowed_rows_i)\n", + "ds = ds1.select(known_rows_i)\n", "ds" ] }, @@ -187,7 +617,7 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 9, "metadata": {}, "outputs": [], "source": [ @@ -231,9 +661,177 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 10, "metadata": {}, - "outputs": [], + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
ds_indexds_stringexample_ianswerquestionanswer_choicestemplate_namelabel_truelabel_instructedinstructed_to_liesys_instr_nameprompt_truncatedchoice_probs0ans0txt_ans0dir_truellm_ans
010imdb5negativeYou've been assigned a role in a classic truth...[negative, positive]Reviewer Sentiment Feeling10Truepuzzlean uneasy, cheap and insincere feeling about ...[0.07703635, 0.42967427]0.847951positive0.847951True
111imdb5positiveBelow is an instruction that describes a task,...[negative, positive]Writer Expressed Sentiment11Falsetruthappropriately completes the request.\\n\\n### I...[0.36913574, 0.5124942]0.581296positive0.581296True
212imdb6negative review.You are an AI assistant that always tells a li...[negative, positive]Negation template for positive and negative01Truelie_for_charityination it was lovely. They barely knew each o...[0.0005304276, 0.0007480229]0.580560review0.580560True
313imdb6negativeBelow is an instruction that describes a task,...[negative, positive]Reviewer Expressed Sentiment00Falsetruthtime spy...\\nWhat is the sentiment expressed ...[0.2338603, 0.14184344]0.377531negative0.377531False
\n", + "
" + ], + "text/plain": [ + " ds_index ds_string example_i answer \\\n", + "0 10 imdb 5 negative \n", + "1 11 imdb 5 positive \n", + "2 12 imdb 6 negative review. \n", + "3 13 imdb 6 negative \n", + "\n", + " question answer_choices \\\n", + "0 You've been assigned a role in a classic truth... [negative, positive] \n", + "1 Below is an instruction that describes a task,... [negative, positive] \n", + "2 You are an AI assistant that always tells a li... [negative, positive] \n", + "3 Below is an instruction that describes a task,... [negative, positive] \n", + "\n", + " template_name label_true label_instructed \\\n", + "0 Reviewer Sentiment Feeling 1 0 \n", + "1 Writer Expressed Sentiment 1 1 \n", + "2 Negation template for positive and negative 0 1 \n", + "3 Reviewer Expressed Sentiment 0 0 \n", + "\n", + " instructed_to_lie sys_instr_name \\\n", + "0 True puzzle \n", + "1 False truth \n", + "2 True lie_for_charity \n", + "3 False truth \n", + "\n", + " prompt_truncated \\\n", + "0 an uneasy, cheap and insincere feeling about ... \n", + "1 appropriately completes the request.\\n\\n### I... \n", + "2 ination it was lovely. They barely knew each o... \n", + "3 time spy...\\nWhat is the sentiment expressed ... \n", + "\n", + " choice_probs0 ans0 txt_ans0 dir_true llm_ans \n", + "0 [0.07703635, 0.42967427] 0.847951 positive 0.847951 True \n", + "1 [0.36913574, 0.5124942] 0.581296 positive 0.581296 True \n", + "2 [0.0005304276, 0.0007480229] 0.580560 review 0.580560 True \n", + "3 [0.2338603, 0.14184344] 0.377531 negative 0.377531 False " + ] + }, + "execution_count": 10, + "metadata": {}, + "output_type": "execute_result" + } + ], "source": [ "df = ds2df(ds)\n", "df.head(4)" @@ -241,7 +839,7 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 11, "metadata": {}, "outputs": [], "source": [ @@ -263,7 +861,7 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 16, "metadata": {}, "outputs": [], "source": [ @@ -273,9 +871,20 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 18, "metadata": {}, - "outputs": [], + "outputs": [ + { + "data": { + "text/plain": [ + "(8, 4)" + ] + }, + "execution_count": 18, + "metadata": {}, + "output_type": "execute_result" + } + ], "source": [ "batch_size = 120\n", "# test and cache\n", @@ -287,6 +896,26 @@ "len(dl_train), len(dl_val)" ] }, + { + "cell_type": "code", + "execution_count": 22, + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "(1804, 2816)" + ] + }, + "execution_count": 22, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "ds['grads_mlp0'].shape" + ] + }, { "cell_type": "code", "execution_count": null, @@ -510,7 +1139,7 @@ }, { "cell_type": "code", - "execution_count": 255, + "execution_count": null, "metadata": {}, "outputs": [], "source": [ @@ -620,7 +1249,7 @@ }, { "cell_type": "code", - "execution_count": 256, + "execution_count": null, "metadata": {}, "outputs": [], "source": [ @@ -642,7 +1271,7 @@ }, { "cell_type": "code", - "execution_count": 257, + "execution_count": null, "metadata": {}, "outputs": [], "source": [ @@ -653,17 +1282,9 @@ }, { "cell_type": "code", - "execution_count": 258, + "execution_count": null, "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "torch.Size([120, 6144, 37])\n" - ] - } - ], + "outputs": [], "source": [ "c_in = b[0].shape[1]\n", "print(b[0].shape)\n", @@ -683,65 +1304,9 @@ }, { "cell_type": "code", - "execution_count": 259, + "execution_count": null, "metadata": {}, - "outputs": [ - { - "data": { - "text/plain": [ - "==========================================================================================\n", - "Layer (type:depth-idx) Output Shape Param #\n", - "==========================================================================================\n", - "PLConvProbe [120] --\n", - "├─ConvProbe: 1-1 [120, 1] --\n", - "│ └─Sequential: 2-1 [120, 108] --\n", - "│ │ └─BatchNorm2d: 3-1 [120, 256, 24, 37] --\n", - "│ │ └─Dropout2d: 3-2 [120, 256, 24, 37] --\n", - "│ │ └─Conv2d: 3-3 [120, 64, 24, 35] 49,216\n", - "│ │ └─Conv2d: 3-4 [120, 12, 22, 35] 2,316\n", - "│ │ └─ReLU: 3-5 [120, 12, 22, 35] --\n", - "│ │ └─BatchNorm2d: 3-6 [120, 12, 22, 35] 24\n", - "│ │ └─Conv2d: 3-7 [120, 12, 22, 33] 444\n", - "│ │ └─Conv2d: 3-8 [120, 12, 20, 33] 444\n", - "│ │ └─ReLU: 3-9 [120, 12, 20, 33] --\n", - "│ │ └─BatchNorm2d: 3-10 [120, 12, 20, 33] 24\n", - "│ │ └─Conv2d: 3-11 [120, 12, 20, 31] 444\n", - "│ │ └─Conv2d: 3-12 [120, 12, 18, 31] 444\n", - "│ │ └─ReLU: 3-13 [120, 12, 18, 31] --\n", - "│ │ └─BatchNorm2d: 3-14 [120, 12, 18, 31] 24\n", - "│ │ └─AdaptiveAvgPool2d: 3-15 [120, 12, 3, 3] --\n", - "│ │ └─Flatten: 3-16 [120, 108] --\n", - "│ └─Sequential: 2-2 [120, 4] --\n", - "│ │ └─Linear: 3-17 [120, 12] 1,308\n", - "│ │ └─Linear: 3-18 [120, 8] 104\n", - "│ │ └─ReLU: 3-19 [120, 8] --\n", - "│ │ └─BatchNorm1d: 3-20 [120, 8] 16\n", - "│ │ └─Linear: 3-21 [120, 4] 36\n", - "│ │ └─ReLU: 3-22 [120, 4] --\n", - "│ │ └─BatchNorm1d: 3-23 [120, 4] 8\n", - "│ └─Sequential: 2-3 [120, 1] --\n", - "│ │ └─Linear: 3-24 [120, 4] 20\n", - "│ │ └─ReLU: 3-25 [120, 4] --\n", - "│ │ └─Dropout: 3-26 [120, 4] --\n", - "│ │ └─Linear: 3-27 [120, 1] 5\n", - "==========================================================================================\n", - "Total params: 54,877\n", - "Trainable params: 54,877\n", - "Non-trainable params: 0\n", - "Total mult-adds (Units.GIGABYTES): 5.31\n", - "==========================================================================================\n", - "Input size (MB): 109.12\n", - "Forward/backward pass size (MB): 112.96\n", - "Params size (MB): 0.22\n", - "Estimated Total Size (MB): 222.30\n", - "==========================================================================================" - ] - }, - "execution_count": 259, - "metadata": {}, - "output_type": "execute_result" - } - ], + "outputs": [], "source": [ "from torchinfo import summary\n", "\n", @@ -758,1214 +1323,9 @@ }, { "cell_type": "code", - "execution_count": 260, + "execution_count": null, "metadata": {}, - "outputs": [ - { - "name": "stderr", - "output_type": "stream", - "text": [ - "Using bfloat16 Automatic Mixed Precision (AMP)\n", - "GPU available: True (cuda), used: True\n", - "TPU available: False, using: 0 TPU cores\n", - "IPU available: False, using: 0 IPUs\n", - "HPU available: False, using: 0 HPUs\n", - "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n", - "\n", - " | Name | Type | Params\n", - "------------------------------------\n", - "0 | probe | ConvProbe | 54.9 K\n", - "------------------------------------\n", - "54.9 K Trainable params\n", - "0 Non-trainable params\n", - "54.9 K Total params\n", - "0.220 Total estimated model params size (MB)\n" - ] - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "2bc28e68128c4e46b02ac8c2beb64f13", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Sanity Checking: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "9ea376621b6c4d3f8cd1c46b8af179df", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Training: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "d75445da176e445fb0ca91f6431ce68e", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "ffc5aba294ca4ae59e8e79864b98d898", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "cbff66e705e34fb0b19e6aa457589640", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "8a36fb8fa2e5485781a3afe404cc1e33", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "8ab650d92cb9488aa5504b7889faed3a", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "70fd65c63e42494db0dcf6bd6cfd94a8", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "1e7df33911424dbc90b6f5ba7edd8fc9", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "81f40700a594405789fbfb7e6bd93a08", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "a4ca370b30a14a6dbb7d41213dab8b62", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "094f89bf893b4826b2a0fbb9609b23b6", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "af8cf97b24d34979967d35502bda531c", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "6212bf5dafa24e24b5103b5bd7e6e418", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "67d75bb14a694277a7b81dbb71185d04", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "2c6ac8cce7a4494f811afc618300b378", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "faa03eea38444571b57ff8ff741677bd", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "eba32bdc9af7491a8cac4ee109cb0c9a", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "13a4720e811747e99db9a5ed07957bdf", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "b4bbb2916df8457487f07a5699852123", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "f0e56376bbe5428f8aa155f51c6af3c8", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "69853c07e73b49c38d729ff159b3451f", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "ea854322790c48579432bdba2bd7b9ba", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "2461f68c1d714f6bbbc952b86a6fa283", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "16a47c30a53c47b9bbfa84e34d09c4b8", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "d9eacbb53cf44809a7e0d84d313cd573", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "7d652557cf454fff9a878f8056b87f71", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "73ad96272db047038514f191321d3cd4", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "06ecfd6ea43b4484b2ecdcb43ca70d1a", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "9a965d6a49814cd5a4ba7bd8e1c8a1d2", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "f315b8f77e2b4ddd95c961c70c1fe022", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "46e13d7a284c4807948f34c05e68d147", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "7cf1b4489e9a42778880d3c0ad45a2c7", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "556174a1b42541f0993d0817ad80f3f2", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "e69a5df454e446f7815ecd5814686361", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "1aef24cd6ce14164b8967857b8225387", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "de1891b3be18489aa249cee5024df559", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "5ec6cb203ce242059f6de402fef55515", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "495ba0017ba84d93a2afba7744b8ea73", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "5fef041392334a3884f3a0c4ff8de2f7", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "8479e0d805fc47128388c3336f16c27d", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "05582311117c49a3812129db3f8b5e70", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "b6c2857b09f4430b9ada9ae9ab4e3802", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "b179b9f0ae114014b5585900570b0f5d", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "1131c1d86ed94205843d834ff2e2d8cb", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "008cd987616c46b784baba40225240ff", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "6f14be23ecfa4793b49e7d67b06879ff", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "2c727798ae194eeb9ecefddff7e72baf", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "8ca77e8056e4447688655f3633370221", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "1bbd78e18fe44d5ca0f30f40855e7f95", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "f5df2d2a7bfb459a8e834523fd853b42", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "ee63c141f8034e22b657ba5d6a1fc54a", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "d1aa3abc290d4eea8cc9f93dd9034c65", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "3df1697925e2403994644fb97e9cecc2", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "9dfd320e7c6c419e9a288018ea14c4b0", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "d27ff6eb8e2446e58a47d0f7e38c1da1", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "5c95552617f04d63b9295001d3a246a0", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "5e074d5903434d419b057ac8b3dd3c2c", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "348927e34adb4520b4b42aa2fbc01afa", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "413c078599e14dcaa6c5070818e5255b", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "3a255947f41e461dac451524a9d38936", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "fd40b2e1c8254fcea5e2ead648809bf0", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "d7889c0c9ad9415d9f0e0cba71cc343a", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "3f9b7600d21140c48a7506b483c175b9", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "0db6f8b2f03c43a1b0ed9fe6be7716f4", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "568363d4152b45719952bb09ac79a4d2", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "98044b5fa7264c4bb6189905d700302e", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "89701fc9b2d744c980938dbdd5109e65", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "4bf453cc797b444eb03022ba86855625", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "43cba87faa0f4056ae08e6aa7e8839c3", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "c71ba82066ba449cb6258d441e361e37", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "03541deb787d4d0f88370fe683e96934", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "368fd26ad9264421a5677a3c8905c882", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "aadbd491c4e249459d7804b749a512a3", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "77a1275778064e55a0856b4ce4ebf549", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "8921261b5e0e44b1b5a73e387a9488c6", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "7ed2fa780c34452db2e7ee9d2355c6f0", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "4dd7d9cf4e4540a687a55ecebee298b0", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "9b64c78b5c1541479806be5d09e272c5", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "6efa2027bec1413fb6ef5fb321a3a8e2", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "ea23119ad11a4e37bf4ef76c2ba0b3bb", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "a6e4efd97511448192a790ced7401a39", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "f662ee5e707a4b35a624914ac57b2f77", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "9e816d0d458542fca590a0c81ee8ffe8", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Validation: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "`Trainer.fit` stopped: `max_epochs=82` reached.\n" - ] - } - ], + "outputs": [], "source": [ "\n", "# init the model\n", @@ -1980,7 +1340,7 @@ }, { "cell_type": "code", - "execution_count": 261, + "execution_count": null, "metadata": {}, "outputs": [], "source": [ @@ -1997,188 +1357,9 @@ }, { "cell_type": "code", - "execution_count": 262, + "execution_count": null, "metadata": {}, - "outputs": [ - { - "data": { - "text/html": [ - "
\n", - "\n", - "\n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - "
val/accval/lossval/nsteptrain/acctrain/losstrain/n
epoch
00.4969880.499959996.016.00.4977400.4999251991.0
10.4859440.499846996.033.00.5158210.4990151991.0
20.4949800.499207996.050.00.4927170.5000701991.0
30.5190760.499467996.067.00.4942240.4996191991.0
40.4889560.500423996.084.00.5128080.4988701991.0
........................
770.5391570.483304996.01325.00.8372680.2822011991.0
780.5502010.477490996.01342.00.8478150.2782431991.0
790.5461850.476890996.01359.00.8357610.2744931991.0
800.5411650.477488996.01376.00.8498240.2750491991.0
810.5471890.477596996.01393.00.8442990.2752371991.0
\n", - "

82 rows × 7 columns

\n", - "
" - ], - "text/plain": [ - " val/acc val/loss val/n step train/acc train/loss train/n\n", - "epoch \n", - "0 0.496988 0.499959 996.0 16.0 0.497740 0.499925 1991.0\n", - "1 0.485944 0.499846 996.0 33.0 0.515821 0.499015 1991.0\n", - "2 0.494980 0.499207 996.0 50.0 0.492717 0.500070 1991.0\n", - "3 0.519076 0.499467 996.0 67.0 0.494224 0.499619 1991.0\n", - "4 0.488956 0.500423 996.0 84.0 0.512808 0.498870 1991.0\n", - "... ... ... ... ... ... ... ...\n", - "77 0.539157 0.483304 996.0 1325.0 0.837268 0.282201 1991.0\n", - "78 0.550201 0.477490 996.0 1342.0 0.847815 0.278243 1991.0\n", - "79 0.546185 0.476890 996.0 1359.0 0.835761 0.274493 1991.0\n", - "80 0.541165 0.477488 996.0 1376.0 0.849824 0.275049 1991.0\n", - "81 0.547189 0.477596 996.0 1393.0 0.844299 0.275237 1991.0\n", - "\n", - "[82 rows x 7 columns]" - ] - }, - "execution_count": 262, - "metadata": {}, - "output_type": "execute_result" - } - ], + "outputs": [], "source": [ "df_hist = read_metrics_csv(trainer.logger.experiment.metrics_file_path).ffill().bfill()\n", "df_hist" @@ -2186,20 +1367,9 @@ }, { "cell_type": "code", - "execution_count": 263, + "execution_count": null, "metadata": {}, - "outputs": [ - { - "data": { - "image/png": "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", - "text/plain": [ - "
" - ] - }, - "metadata": {}, - "output_type": "display_data" - } - ], + "outputs": [], "source": [ "for key in ['loss']:\n", " df_hist[[c for c in df_hist.columns if key in c]].plot(logy=True)" @@ -2207,20 +1377,9 @@ }, { "cell_type": "code", - "execution_count": 264, + "execution_count": null, "metadata": {}, - "outputs": [ - { - "data": { - "image/png": "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", - "text/plain": [ - "
" - ] - }, - "metadata": {}, - "output_type": "display_data" - } - ], + "outputs": [], "source": [ "for key in ['acc']:\n", " df_hist[[c for c in df_hist.columns if key in c]].plot()" @@ -2236,78 +1395,9 @@ }, { "cell_type": "code", - "execution_count": 265, + "execution_count": null, "metadata": {}, - "outputs": [ - { - "name": "stderr", - "output_type": "stream", - "text": [ - "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n", - "/home/ubuntu/mambaforge/envs/dlk3/lib/python3.11/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:480: PossibleUserWarning: Your `test_dataloader`'s sampler has shuffling enabled, it is strongly recommended that you turn shuffling off for val/test dataloaders.\n", - " rank_zero_warn(\n" - ] - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "ed1c4a0e2a3341ebad52ec2e3b4a96bf", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Testing: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "text/html": [ - "
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
-       "┃   Runningstage.testing    ┃                           ┃                           ┃                           ┃\n",
-       "┃          metric           ┃       DataLoader 0        ┃       DataLoader 1        ┃       DataLoader 2        ┃\n",
-       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
-       "│         test/acc          │    0.9201406240463257     │    0.5471887588500977     │    0.5381526350975037     │\n",
-       "│         test/loss         │    0.23339247703552246    │    0.47759562730789185    │    0.47705864906311035    │\n",
-       "│          test/n           │          1991.0           │           996.0           │           996.0           │\n",
-       "└───────────────────────────┴───────────────────────────┴───────────────────────────┴───────────────────────────┘\n",
-       "
\n" - ], - "text/plain": [ - "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n", - "┃\u001b[1m \u001b[0m\u001b[1m Runningstage.testing \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m┃\n", - "┃\u001b[1m \u001b[0m\u001b[1m metric \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m DataLoader 0 \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m DataLoader 1 \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m DataLoader 2 \u001b[0m\u001b[1m \u001b[0m┃\n", - "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n", - "│\u001b[36m \u001b[0m\u001b[36m test/acc \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m 0.9201406240463257 \u001b[0m\u001b[35m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m 0.5471887588500977 \u001b[0m\u001b[35m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m 0.5381526350975037 \u001b[0m\u001b[35m \u001b[0m│\n", - "│\u001b[36m \u001b[0m\u001b[36m test/loss \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m 0.23339247703552246 \u001b[0m\u001b[35m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m 0.47759562730789185 \u001b[0m\u001b[35m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m 0.47705864906311035 \u001b[0m\u001b[35m \u001b[0m│\n", - "│\u001b[36m \u001b[0m\u001b[36m test/n \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m 1991.0 \u001b[0m\u001b[35m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m 996.0 \u001b[0m\u001b[35m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m 996.0 \u001b[0m\u001b[35m \u001b[0m│\n", - "└───────────────────────────┴───────────────────────────┴───────────────────────────┴───────────────────────────┘\n" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "text/plain": [ - "[{'test/acc/dataloader_idx_0': 0.9201406240463257,\n", - " 'test/loss/dataloader_idx_0': 0.23339247703552246,\n", - " 'test/n/dataloader_idx_0': 1991.0},\n", - " {'test/acc/dataloader_idx_1': 0.5471887588500977,\n", - " 'test/loss/dataloader_idx_1': 0.47759562730789185,\n", - " 'test/n/dataloader_idx_1': 996.0},\n", - " {'test/acc/dataloader_idx_2': 0.5381526350975037,\n", - " 'test/loss/dataloader_idx_2': 0.47705864906311035,\n", - " 'test/n/dataloader_idx_2': 996.0}]" - ] - }, - "execution_count": 265, - "metadata": {}, - "output_type": "execute_result" - } - ], + "outputs": [], "source": [ "dl_test = dm.test_dataloader()\n", "rs = trainer.test(net, dataloaders=[dl_train, dl_val, dl_test])\n", @@ -2316,41 +1406,9 @@ }, { "cell_type": "code", - "execution_count": 266, + "execution_count": null, "metadata": {}, - "outputs": [ - { - "name": "stderr", - "output_type": "stream", - "text": [ - "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n" - ] - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "c9fe4f16fea74e2cbc4de1f4da0058ae", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Predicting: 0it [00:00, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "text/plain": [ - "(996,)" - ] - }, - "execution_count": 266, - "metadata": {}, - "output_type": "execute_result" - } - ], + "outputs": [], "source": [ "dl_test = dm.test_dataloader()\n", "r = trainer.predict(net, dataloaders=dl_test)\n", @@ -2367,7 +1425,7 @@ }, { "cell_type": "code", - "execution_count": 267, + "execution_count": null, "metadata": {}, "outputs": [], "source": [ @@ -2377,410 +1435,9 @@ }, { "cell_type": "code", - "execution_count": 268, + "execution_count": null, "metadata": {}, - "outputs": [ - { - "data": { - "text/html": [ - "
\n", - "\n", - "\n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - "
ds_indexds_stringexample_ianswerquestionanswer_choicestemplate_namelabel_truelabel_instructedinstructed_to_lie...expanded_ans1txt_ans0txt_ans1dir_trueconfllm_probllm_ansyprobe_predprobe_prob
29875150imdb2575negative<|system|>You are an AI assistant that always ...[negative, positive]Sentiment with choices10True...0.309649\\n\\n-0.1261260.1261260.141546FalseFalseFalse0.451172
29885151imdb2575negative review.<|system|>Below is an instruction that describ...[negative, positive]Negation template for positive and negative11False...0.531121positivepositive-0.1761690.1761690.597834TrueFalseTrue0.898438
29895156imdb25781<|system|>You are about to immerse yourself in...[0, 1]burns_101True...0.086315000.0591540.0591540.056746FalseFalseFalse0.177734
29905181imdb2590negative<|system|>Below is an instruction that describ...[negative, positive]Writer Expressed Sentiment00False...0.351236positiveI-0.3934020.3934020.490894FalseTrueFalse0.388672
29915182imdb2591negative<|system|>You are an AI assistant that always ...[negative, positive]Movie Expressed Sentiment 210True...0.830855negativepositive0.4699820.4699820.587186TrueTrueFalse0.000000
..................................................................
39788179imdb4089positive<|system|>Below is an instruction that describ...[negative, positive]Reviewer Sentiment Feeling11False...0.821930positivepositive0.1026430.1026430.750543TrueTrueTrue0.529297
39798180imdb4090positive<|system|>You are about to immerse yourself in...[negative, positive]Writer Expressed Sentiment01True...0.480320negativenegative0.1385360.1385360.347521FalseFalseTrue0.656250
39808181imdb4090negative<|system|>Below is an instruction that describ...[negative, positive]Movie Expressed Sentiment00False...0.330161negativenegative0.2483950.2483950.174902FalseFalseFalse0.000000
39818190imdb4095bad<|system|>You've been assigned a role in a cla...[bad, good]Reviewer Opinion bad good choices10True...0.763606goodgood-0.1702460.1702460.835655TrueFalseTrue0.549805
39828191imdb4095positive<|system|>Below is an instruction that describ...[negative, positive]Reviewer Sentiment Feeling11False...0.838116positivepositive0.0814530.0814530.773282TrueTrueFalse0.282227
\n", - "

996 rows × 29 columns

\n", - "
" - ], - "text/plain": [ - " ds_index ds_string example_i answer \\\n", - "2987 5150 imdb 2575 negative \n", - "2988 5151 imdb 2575 negative review. \n", - "2989 5156 imdb 2578 1 \n", - "2990 5181 imdb 2590 negative \n", - "2991 5182 imdb 2591 negative \n", - "... ... ... ... ... \n", - "3978 8179 imdb 4089 positive \n", - "3979 8180 imdb 4090 positive \n", - "3980 8181 imdb 4090 negative \n", - "3981 8190 imdb 4095 bad \n", - "3982 8191 imdb 4095 positive \n", - "\n", - " question answer_choices \\\n", - "2987 <|system|>You are an AI assistant that always ... [negative, positive] \n", - "2988 <|system|>Below is an instruction that describ... [negative, positive] \n", - "2989 <|system|>You are about to immerse yourself in... [0, 1] \n", - "2990 <|system|>Below is an instruction that describ... [negative, positive] \n", - "2991 <|system|>You are an AI assistant that always ... [negative, positive] \n", - "... ... ... \n", - "3978 <|system|>Below is an instruction that describ... [negative, positive] \n", - "3979 <|system|>You are about to immerse yourself in... [negative, positive] \n", - "3980 <|system|>Below is an instruction that describ... [negative, positive] \n", - "3981 <|system|>You've been assigned a role in a cla... [bad, good] \n", - "3982 <|system|>Below is an instruction that describ... [negative, positive] \n", - "\n", - " template_name label_true \\\n", - "2987 Sentiment with choices 1 \n", - "2988 Negation template for positive and negative 1 \n", - "2989 burns_1 0 \n", - "2990 Writer Expressed Sentiment 0 \n", - "2991 Movie Expressed Sentiment 2 1 \n", - "... ... ... \n", - "3978 Reviewer Sentiment Feeling 1 \n", - "3979 Writer Expressed Sentiment 0 \n", - "3980 Movie Expressed Sentiment 0 \n", - "3981 Reviewer Opinion bad good choices 1 \n", - "3982 Reviewer Sentiment Feeling 1 \n", - "\n", - " label_instructed instructed_to_lie ... expanded_ans1 txt_ans0 \\\n", - "2987 0 True ... 0.309649 \\n \n", - "2988 1 False ... 0.531121 positive \n", - "2989 1 True ... 0.086315 0 \n", - "2990 0 False ... 0.351236 positive \n", - "2991 0 True ... 0.830855 negative \n", - "... ... ... ... ... ... \n", - "3978 1 False ... 0.821930 positive \n", - "3979 1 True ... 0.480320 negative \n", - "3980 0 False ... 0.330161 negative \n", - "3981 0 True ... 0.763606 good \n", - "3982 1 False ... 0.838116 positive \n", - "\n", - " txt_ans1 dir_true conf llm_prob llm_ans y probe_pred \\\n", - "2987 \\n -0.126126 0.126126 0.141546 False False False \n", - "2988 positive -0.176169 0.176169 0.597834 True False True \n", - "2989 0 0.059154 0.059154 0.056746 False False False \n", - "2990 I -0.393402 0.393402 0.490894 False True False \n", - "2991 positive 0.469982 0.469982 0.587186 True True False \n", - "... ... ... ... ... ... ... ... \n", - "3978 positive 0.102643 0.102643 0.750543 True True True \n", - "3979 negative 0.138536 0.138536 0.347521 False False True \n", - "3980 negative 0.248395 0.248395 0.174902 False False False \n", - "3981 good -0.170246 0.170246 0.835655 True False True \n", - "3982 positive 0.081453 0.081453 0.773282 True True False \n", - "\n", - " probe_prob \n", - "2987 0.451172 \n", - "2988 0.898438 \n", - "2989 0.177734 \n", - "2990 0.388672 \n", - "2991 0.000000 \n", - "... ... \n", - "3978 0.529297 \n", - "3979 0.656250 \n", - "3980 0.000000 \n", - "3981 0.549805 \n", - "3982 0.282227 \n", - "\n", - "[996 rows x 29 columns]" - ] - }, - "execution_count": 268, - "metadata": {}, - "output_type": "execute_result" - } - ], + "outputs": [], "source": [ "# Make a prediction dataframe with everything in it\n", "df_test = dm.df.iloc[dm.splits['test'][0]:].copy()\n", @@ -2800,33 +1457,9 @@ }, { "cell_type": "code", - "execution_count": 269, + "execution_count": null, "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "probe results on subsets of the data\n", - "acc=52.13% [instructed_to_lie==True]\n", - "acc=60.00% [instructed_to_lie==False]\n", - "acc=56.70% [llm_ans==label_true]\n", - "acc=57.56% [llm_ans==label_instructed]\n", - "acc=50.62% [instructed_to_lie==True & llm_ans==label_instructed]\n", - "acc=52.76% [instructed_to_lie==True & llm_ans!=label_instructed]\n" - ] - }, - { - "data": { - "text/plain": [ - "0.5275590551181102" - ] - }, - "execution_count": 269, - "metadata": {}, - "output_type": "execute_result" - } - ], + "outputs": [], "source": [ "def get_acc_subset(df, query):\n", " df_s = df.query(query)\n", @@ -2852,17 +1485,9 @@ }, { "cell_type": "code", - "execution_count": 270, + "execution_count": null, "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "⭐PRIMARY METRIC⭐ acc=55.72% from probe\n" - ] - } - ], + "outputs": [], "source": [ "acc = (df_test['y']==(y_test_pred_bool>0.5)).mean()\n", "\n", @@ -2881,7 +1506,7 @@ }, { "cell_type": "code", - "execution_count": 271, + "execution_count": null, "metadata": {}, "outputs": [], "source": [ @@ -2906,7 +1531,7 @@ }, { "cell_type": "code", - "execution_count": 272, + "execution_count": null, "metadata": {}, "outputs": [], "source": [ @@ -2918,7 +1543,7 @@ }, { "cell_type": "code", - "execution_count": 273, + "execution_count": null, "metadata": {}, "outputs": [], "source": [ diff --git a/notebooks/03_make_dataset.ipynb b/notebooks/03_make_dataset.ipynb index 27dd3ad..80dabb0 100644 --- a/notebooks/03_make_dataset.ipynb +++ b/notebooks/03_make_dataset.ipynb @@ -2423,7 +2423,9 @@ "execution_count": null, "metadata": {}, "outputs": [], - "source": [] + "source": [ + "# Scratch" + ] }, { "cell_type": "code", diff --git a/src/datasets/dm.py b/src/datasets/dm.py index bdbabf4..9679856 100644 --- a/src/datasets/dm.py +++ b/src/datasets/dm.py @@ -14,7 +14,7 @@ def compute_distance(df): return distance to_tensor = lambda x: torch.from_numpy(x).float() -to_ds = lambda hs0, hs1, y: TensorDataset(to_tensor(hs0), to_tensor(hs1), to_tensor(y)) +to_ds = lambda hs0, y: TensorDataset(to_tensor(hs0), to_tensor(y)) class imdbHSDataModule(pl.LightningDataModule): @@ -31,21 +31,21 @@ class imdbHSDataModule(pl.LightningDataModule): # extract data set into N-Dim tensors and 1-d dataframe self.ds_hs = ( - self.ds.select_columns(['hs0', 'hs1']) + self.ds.select_columns(['grads_mlp0']) .with_format("numpy") ) - self.df = ds2df(self.ds) + df = self.df = ds2df(self.ds) - y_cls = compute_distance(self.df) + y_cls = y = df['label_true'] == df['llm_ans'] self.y = y_cls.values self.df['y'] = y_cls b = len(self.ds_hs) - self.hs0 = self.ds_hs['hs0'].transpose(0, 2, 1) - self.hs1 = self.ds_hs['hs1'].transpose(0, 2, 1) + self.hs0 = self.ds_hs['grads_mlp0']#.transpose(0, 2, 1) + # self.hs1 = self.ds_hs['hs1'].transpose(0, 2, 1) self.ans0 = self.df['ans0'].values - self.ans1 = self.df['ans1'].values + # self.ans1 = self.df['ans1'].values # let's create a simple 50/50 train split (the data is already randomized) n = len(self.y) @@ -55,7 +55,7 @@ class imdbHSDataModule(pl.LightningDataModule): 'test': (int(n * 0.75), n), } - self.datasets = {key: to_ds(self.hs0[start:end], self.hs1[start:end], self.y[start:end]) for key, (start, end) in self.splits.items()} + self.datasets = {key: to_ds(self.hs0[start:end], self.y[start:end]) for key, (start, end) in self.splits.items()} def create_dataloader(self, ds, shuffle=False): return DataLoader(ds, batch_size=self.hparams.batch_size, drop_last=False, shuffle=shuffle)