From c5edec0e736a6d039838eb8d2465bced44d032db Mon Sep 17 00:00:00 2001 From: wassname <1103714+wassname@users.noreply.github.com> Date: Sat, 11 Jan 2025 23:38:18 +0000 Subject: [PATCH] wip --- .vscode/extensions.json | 5 + nbs/fine_tuning_gemma2.ipynb | 1411 +++++++++++++++++++++++++++++++++- research_journal.md | 6 +- 3 files changed, 1381 insertions(+), 41 deletions(-) diff --git a/.vscode/extensions.json b/.vscode/extensions.json index e04bbab..220abf1 100644 --- a/.vscode/extensions.json +++ b/.vscode/extensions.json @@ -1,5 +1,10 @@ { "recommendations": [ + "github.copilot", + "jsynowiec.vscode-insertdatestring", + "visualstudioexptteam.vscodeintellicode", + "yzhang.markdown-all-in-one", + "charliermarsh.ruff", "ms-toolsai.jupyter", "ms-python.python" ] diff --git a/nbs/fine_tuning_gemma2.ipynb b/nbs/fine_tuning_gemma2.ipynb index 4369705..0b0c6d8 100644 --- a/nbs/fine_tuning_gemma2.ipynb +++ b/nbs/fine_tuning_gemma2.ipynb @@ -159,7 +159,50 @@ }, "trusted": true }, - "outputs": [], + "outputs": [ + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "5b32e6c2481b436cb3e7ec2b45c7532f", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + "README.md: 0%| | 0.00/3.21k [00:00/workspace/latent-gemma/nbs/wandb/run-20250111_103939-vy48g5pt" + "Run data is saved locally in /workspace/latent-gemma/nbs/wandb/run-20250111_233622-lznyx98u" ], "text/plain": [ "" @@ -1798,7 +1967,7 @@ { "data": { "text/html": [ - "Syncing run wobbly-jazz-8 to Weights & Biases (docs)
" + "Syncing run jolly-sponge-9 to Weights & Biases (docs)
" ], "text/plain": [ "" @@ -1822,7 +1991,7 @@ { "data": { "text/html": [ - " View run at https://wandb.ai/wassname/gemma2-finetuning/runs/vy48g5pt" + " View run at https://wandb.ai/wassname/gemma2-finetuning/runs/lznyx98u" ], "text/plain": [ "" @@ -1834,7 +2003,7 @@ { "data": { "application/vnd.jupyter.widget-view+json": { - "model_id": "ec8b8f1c0731455897b5dae1a12c0121", + "model_id": "ac3f630d668e49f7bce0ab25b63cdcb0", "version_major": 2, "version_minor": 0 }, @@ -1848,7 +2017,7 @@ { "data": { "application/vnd.jupyter.widget-view+json": { - "model_id": "e69a008afcb04cb29237b19305eb1f37", + "model_id": "67cd536f5e4e4c99baf67a2fa9d518ff", "version_major": 2, "version_minor": 0 }, @@ -1860,33 +2029,1195 @@ "output_type": "display_data" }, { - "name": "stderr", - "output_type": "stream", - "text": [ - "Detected kernel version 5.4.0, which is below the recommended minimum of 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher.\n" - ] - }, - { - "ename": "OutOfMemoryError", - "evalue": "CUDA out of memory. Tried to allocate 42.00 MiB. GPU 0 has a total capacity of 23.64 GiB of which 11.69 MiB is free. Process 2052815 has 23.62 GiB memory in use. Of the allocated memory 23.12 GiB is allocated by PyTorch, and 43.60 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)", - "output_type": "error", - "traceback": [ - "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m", - "\u001b[0;31mOutOfMemoryError\u001b[0m Traceback (most recent call last)", - "Cell \u001b[0;32mIn[16], line 100\u001b[0m\n\u001b[1;32m 98\u001b[0m \u001b[38;5;66;03m# Run training stages\u001b[39;00m\n\u001b[1;32m 99\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m stage \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mrange\u001b[39m(config[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mstages\u001b[39m\u001b[38;5;124m\"\u001b[39m] \u001b[38;5;241m+\u001b[39m \u001b[38;5;241m1\u001b[39m):\n\u001b[0;32m--> 100\u001b[0m \u001b[43mstage_trainer\u001b[49m\u001b[43m(\u001b[49m\u001b[43mstage\u001b[49m\u001b[43m)\u001b[49m\n", - "Cell \u001b[0;32mIn[16], line 86\u001b[0m, in \u001b[0;36mstage_trainer\u001b[0;34m(stage)\u001b[0m\n\u001b[1;32m 70\u001b[0m dataset_ \u001b[38;5;241m=\u001b[39m dataset_\u001b[38;5;241m.\u001b[39mmap(\n\u001b[1;32m 71\u001b[0m (\u001b[38;5;28;01mlambda\u001b[39;00m x: tokenizer_function(\n\u001b[1;32m 72\u001b[0m x, \n\u001b[0;32m (...)\u001b[0m\n\u001b[1;32m 77\u001b[0m remove_columns\u001b[38;5;241m=\u001b[39m[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124minput\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124minstruction\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124moutput\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mprompt\u001b[39m\u001b[38;5;124m\"\u001b[39m]\n\u001b[1;32m 78\u001b[0m )\n\u001b[1;32m 80\u001b[0m trainer \u001b[38;5;241m=\u001b[39m Trainer(\n\u001b[1;32m 81\u001b[0m model\u001b[38;5;241m=\u001b[39mmodel,\n\u001b[1;32m 82\u001b[0m args\u001b[38;5;241m=\u001b[39mtraining_args,\n\u001b[1;32m 83\u001b[0m train_dataset\u001b[38;5;241m=\u001b[39mdataset_[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtrain\u001b[39m\u001b[38;5;124m\"\u001b[39m]\n\u001b[1;32m 84\u001b[0m )\n\u001b[0;32m---> 86\u001b[0m \u001b[43mtrainer\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtrain\u001b[49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 88\u001b[0m \u001b[38;5;66;03m# Save checkpoints\u001b[39;00m\n\u001b[1;32m 89\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m folder \u001b[38;5;129;01min\u001b[39;00m os\u001b[38;5;241m.\u001b[39mlistdir(current_output_dir):\n", - "File \u001b[0;32m/workspace/latent-gemma/.venv/lib/python3.11/site-packages/transformers/trainer.py:2171\u001b[0m, in \u001b[0;36mTrainer.train\u001b[0;34m(self, resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)\u001b[0m\n\u001b[1;32m 2169\u001b[0m hf_hub_utils\u001b[38;5;241m.\u001b[39menable_progress_bars()\n\u001b[1;32m 2170\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 2171\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43minner_training_loop\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m 2172\u001b[0m \u001b[43m \u001b[49m\u001b[43margs\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 2173\u001b[0m \u001b[43m \u001b[49m\u001b[43mresume_from_checkpoint\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mresume_from_checkpoint\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 2174\u001b[0m \u001b[43m \u001b[49m\u001b[43mtrial\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtrial\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 2175\u001b[0m \u001b[43m \u001b[49m\u001b[43mignore_keys_for_eval\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mignore_keys_for_eval\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 2176\u001b[0m \u001b[43m \u001b[49m\u001b[43m)\u001b[49m\n", - "File \u001b[0;32m/workspace/latent-gemma/.venv/lib/python3.11/site-packages/transformers/trainer.py:2584\u001b[0m, in \u001b[0;36mTrainer._inner_training_loop\u001b[0;34m(self, batch_size, args, resume_from_checkpoint, trial, ignore_keys_for_eval)\u001b[0m\n\u001b[1;32m 2580\u001b[0m grad_norm \u001b[38;5;241m=\u001b[39m _grad_norm\n\u001b[1;32m 2582\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcontrol \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcallback_handler\u001b[38;5;241m.\u001b[39mon_pre_optimizer_step(args, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcontrol)\n\u001b[0;32m-> 2584\u001b[0m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43moptimizer\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mstep\u001b[49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 2586\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcontrol \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcallback_handler\u001b[38;5;241m.\u001b[39mon_optimizer_step(args, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcontrol)\n\u001b[1;32m 2588\u001b[0m optimizer_was_run \u001b[38;5;241m=\u001b[39m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39maccelerator\u001b[38;5;241m.\u001b[39moptimizer_step_was_skipped\n", - "File \u001b[0;32m/workspace/latent-gemma/.venv/lib/python3.11/site-packages/accelerate/optimizer.py:178\u001b[0m, in \u001b[0;36mAcceleratedOptimizer.step\u001b[0;34m(self, closure)\u001b[0m\n\u001b[1;32m 176\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_accelerate_step_called \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mFalse\u001b[39;00m\n\u001b[1;32m 177\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m--> 178\u001b[0m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43moptimizer\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mstep\u001b[49m\u001b[43m(\u001b[49m\u001b[43mclosure\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 179\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39maccelerator_state\u001b[38;5;241m.\u001b[39mdistributed_type \u001b[38;5;241m==\u001b[39m DistributedType\u001b[38;5;241m.\u001b[39mXLA:\n\u001b[1;32m 180\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mgradient_state\u001b[38;5;241m.\u001b[39mis_xla_gradients_synced \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mFalse\u001b[39;00m\n", - "File \u001b[0;32m/workspace/latent-gemma/.venv/lib/python3.11/site-packages/torch/optim/lr_scheduler.py:137\u001b[0m, in \u001b[0;36mLRScheduler.__init__..patch_track_step_called..wrap_step..wrapper\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m 135\u001b[0m opt \u001b[38;5;241m=\u001b[39m opt_ref()\n\u001b[1;32m 136\u001b[0m opt\u001b[38;5;241m.\u001b[39m_opt_called \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mTrue\u001b[39;00m \u001b[38;5;66;03m# type: ignore[union-attr]\u001b[39;00m\n\u001b[0;32m--> 137\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfunc\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[38;5;21;43m__get__\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43mopt\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mopt\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[38;5;18;43m__class__\u001b[39;49m\u001b[43m)\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n", - "File \u001b[0;32m/workspace/latent-gemma/.venv/lib/python3.11/site-packages/torch/optim/optimizer.py:487\u001b[0m, in \u001b[0;36mOptimizer.profile_hook_step..wrapper\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m 482\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[1;32m 483\u001b[0m \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mRuntimeError\u001b[39;00m(\n\u001b[1;32m 484\u001b[0m \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mfunc\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m must return None or a tuple of (new_args, new_kwargs), but got \u001b[39m\u001b[38;5;132;01m{\u001b[39;00mresult\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m.\u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[1;32m 485\u001b[0m )\n\u001b[0;32m--> 487\u001b[0m out \u001b[38;5;241m=\u001b[39m \u001b[43mfunc\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 488\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_optimizer_step_code()\n\u001b[1;32m 490\u001b[0m \u001b[38;5;66;03m# call optimizer step post hooks\u001b[39;00m\n", - "File \u001b[0;32m/workspace/latent-gemma/.venv/lib/python3.11/site-packages/torch/optim/optimizer.py:91\u001b[0m, in \u001b[0;36m_use_grad_for_differentiable.._use_grad\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m 89\u001b[0m torch\u001b[38;5;241m.\u001b[39mset_grad_enabled(\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mdefaults[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mdifferentiable\u001b[39m\u001b[38;5;124m\"\u001b[39m])\n\u001b[1;32m 90\u001b[0m torch\u001b[38;5;241m.\u001b[39m_dynamo\u001b[38;5;241m.\u001b[39mgraph_break()\n\u001b[0;32m---> 91\u001b[0m ret \u001b[38;5;241m=\u001b[39m \u001b[43mfunc\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 92\u001b[0m \u001b[38;5;28;01mfinally\u001b[39;00m:\n\u001b[1;32m 93\u001b[0m torch\u001b[38;5;241m.\u001b[39m_dynamo\u001b[38;5;241m.\u001b[39mgraph_break()\n", - "File \u001b[0;32m/workspace/latent-gemma/.venv/lib/python3.11/site-packages/torch/optim/adamw.py:220\u001b[0m, in \u001b[0;36mAdamW.step\u001b[0;34m(self, closure)\u001b[0m\n\u001b[1;32m 207\u001b[0m beta1, beta2 \u001b[38;5;241m=\u001b[39m cast(Tuple[\u001b[38;5;28mfloat\u001b[39m, \u001b[38;5;28mfloat\u001b[39m], group[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mbetas\u001b[39m\u001b[38;5;124m\"\u001b[39m])\n\u001b[1;32m 209\u001b[0m has_complex \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_init_group(\n\u001b[1;32m 210\u001b[0m group,\n\u001b[1;32m 211\u001b[0m params_with_grad,\n\u001b[0;32m (...)\u001b[0m\n\u001b[1;32m 217\u001b[0m state_steps,\n\u001b[1;32m 218\u001b[0m )\n\u001b[0;32m--> 220\u001b[0m \u001b[43madamw\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m 221\u001b[0m \u001b[43m \u001b[49m\u001b[43mparams_with_grad\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 222\u001b[0m \u001b[43m \u001b[49m\u001b[43mgrads\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 223\u001b[0m \u001b[43m \u001b[49m\u001b[43mexp_avgs\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 224\u001b[0m \u001b[43m \u001b[49m\u001b[43mexp_avg_sqs\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 225\u001b[0m \u001b[43m \u001b[49m\u001b[43mmax_exp_avg_sqs\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 226\u001b[0m \u001b[43m \u001b[49m\u001b[43mstate_steps\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 227\u001b[0m \u001b[43m \u001b[49m\u001b[43mamsgrad\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mamsgrad\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 228\u001b[0m \u001b[43m \u001b[49m\u001b[43mbeta1\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mbeta1\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 229\u001b[0m \u001b[43m \u001b[49m\u001b[43mbeta2\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mbeta2\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 230\u001b[0m \u001b[43m \u001b[49m\u001b[43mlr\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mgroup\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mlr\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m]\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 231\u001b[0m \u001b[43m \u001b[49m\u001b[43mweight_decay\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mgroup\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mweight_decay\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m]\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 232\u001b[0m \u001b[43m \u001b[49m\u001b[43meps\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mgroup\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43meps\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m]\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 233\u001b[0m \u001b[43m \u001b[49m\u001b[43mmaximize\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mgroup\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mmaximize\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m]\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 234\u001b[0m \u001b[43m \u001b[49m\u001b[43mforeach\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mgroup\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mforeach\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m]\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 235\u001b[0m \u001b[43m \u001b[49m\u001b[43mcapturable\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mgroup\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mcapturable\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m]\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 236\u001b[0m \u001b[43m \u001b[49m\u001b[43mdifferentiable\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mgroup\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mdifferentiable\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m]\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 237\u001b[0m \u001b[43m \u001b[49m\u001b[43mfused\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mgroup\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mfused\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m]\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 238\u001b[0m \u001b[43m \u001b[49m\u001b[43mgrad_scale\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mgetattr\u001b[39;49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mgrad_scale\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43;01mNone\u001b[39;49;00m\u001b[43m)\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 239\u001b[0m \u001b[43m \u001b[49m\u001b[43mfound_inf\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mgetattr\u001b[39;49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mfound_inf\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43;01mNone\u001b[39;49;00m\u001b[43m)\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 240\u001b[0m \u001b[43m \u001b[49m\u001b[43mhas_complex\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mhas_complex\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 241\u001b[0m \u001b[43m \u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 243\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m loss\n", - "File \u001b[0;32m/workspace/latent-gemma/.venv/lib/python3.11/site-packages/torch/optim/optimizer.py:154\u001b[0m, in \u001b[0;36m_disable_dynamo_if_unsupported..wrapper..maybe_fallback\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m 152\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m disabled_func(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)\n\u001b[1;32m 153\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m--> 154\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfunc\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n", - "File \u001b[0;32m/workspace/latent-gemma/.venv/lib/python3.11/site-packages/torch/optim/adamw.py:782\u001b[0m, in \u001b[0;36madamw\u001b[0;34m(params, grads, exp_avgs, exp_avg_sqs, max_exp_avg_sqs, state_steps, foreach, capturable, differentiable, fused, grad_scale, found_inf, has_complex, amsgrad, beta1, beta2, lr, weight_decay, eps, maximize)\u001b[0m\n\u001b[1;32m 779\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[1;32m 780\u001b[0m func \u001b[38;5;241m=\u001b[39m _single_tensor_adamw\n\u001b[0;32m--> 782\u001b[0m \u001b[43mfunc\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m 783\u001b[0m \u001b[43m \u001b[49m\u001b[43mparams\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 784\u001b[0m \u001b[43m \u001b[49m\u001b[43mgrads\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 785\u001b[0m \u001b[43m \u001b[49m\u001b[43mexp_avgs\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 786\u001b[0m \u001b[43m \u001b[49m\u001b[43mexp_avg_sqs\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 787\u001b[0m \u001b[43m \u001b[49m\u001b[43mmax_exp_avg_sqs\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 788\u001b[0m \u001b[43m \u001b[49m\u001b[43mstate_steps\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 789\u001b[0m \u001b[43m \u001b[49m\u001b[43mamsgrad\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mamsgrad\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 790\u001b[0m \u001b[43m \u001b[49m\u001b[43mbeta1\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mbeta1\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 791\u001b[0m \u001b[43m \u001b[49m\u001b[43mbeta2\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mbeta2\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 792\u001b[0m \u001b[43m \u001b[49m\u001b[43mlr\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mlr\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 793\u001b[0m \u001b[43m \u001b[49m\u001b[43mweight_decay\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mweight_decay\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 794\u001b[0m \u001b[43m \u001b[49m\u001b[43meps\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43meps\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 795\u001b[0m \u001b[43m \u001b[49m\u001b[43mmaximize\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mmaximize\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 796\u001b[0m \u001b[43m \u001b[49m\u001b[43mcapturable\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mcapturable\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 797\u001b[0m \u001b[43m \u001b[49m\u001b[43mdifferentiable\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mdifferentiable\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 798\u001b[0m \u001b[43m \u001b[49m\u001b[43mgrad_scale\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mgrad_scale\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 799\u001b[0m \u001b[43m \u001b[49m\u001b[43mfound_inf\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mfound_inf\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 800\u001b[0m \u001b[43m \u001b[49m\u001b[43mhas_complex\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mhas_complex\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 801\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n", - "File \u001b[0;32m/workspace/latent-gemma/.venv/lib/python3.11/site-packages/torch/optim/adamw.py:606\u001b[0m, in \u001b[0;36m_multi_tensor_adamw\u001b[0;34m(params, grads, exp_avgs, exp_avg_sqs, max_exp_avg_sqs, state_steps, grad_scale, found_inf, amsgrad, beta1, beta2, lr, weight_decay, eps, maximize, capturable, differentiable, has_complex)\u001b[0m\n\u001b[1;32m 604\u001b[0m exp_avg_sq_sqrt \u001b[38;5;241m=\u001b[39m torch\u001b[38;5;241m.\u001b[39m_foreach_sqrt(device_max_exp_avg_sqs)\n\u001b[1;32m 605\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m--> 606\u001b[0m exp_avg_sq_sqrt \u001b[38;5;241m=\u001b[39m \u001b[43mtorch\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_foreach_sqrt\u001b[49m\u001b[43m(\u001b[49m\u001b[43mdevice_exp_avg_sqs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 608\u001b[0m torch\u001b[38;5;241m.\u001b[39m_foreach_div_(exp_avg_sq_sqrt, bias_correction2_sqrt)\n\u001b[1;32m 609\u001b[0m torch\u001b[38;5;241m.\u001b[39m_foreach_add_(exp_avg_sq_sqrt, eps)\n", - "\u001b[0;31mOutOfMemoryError\u001b[0m: CUDA out of memory. Tried to allocate 42.00 MiB. GPU 0 has a total capacity of 23.64 GiB of which 11.69 MiB is free. Process 2052815 has 23.62 GiB memory in use. Of the allocated memory 23.12 GiB is allocated by PyTorch, and 43.60 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)" - ] + "data": { + "text/html": [ + "\n", + "
\n", + " \n", + " \n", + " [ 293/90000 01:10 < 6:01:03, 4.14 it/s, Epoch 0.01/3]\n", + "
\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
StepTraining Loss
178.845700
287.019100
378.956600
484.510600
577.455400
677.695400
779.805200
885.350900
982.701300
1080.544300
1182.508200
1280.840300
1383.641900
1478.476000
1577.079200
1680.595200
1786.005500
1884.214000
1975.761500
2075.340000
2187.324800
2299.237800
2377.487900
2477.282200
2582.492700
2680.137900
2776.222000
2878.914600
2979.661600
3081.387500
3179.175800
3277.114400
3381.905100
3485.060800
3573.431400
3681.307400
3791.982200
3882.673300
3981.406800
4083.990900
4177.555900
4276.819700
4376.154500
4484.111100
4587.089100
4680.613900
4794.783800
4886.942700
4977.689700
5077.111400
5177.302700
5285.260600
5380.604800
5481.129300
5587.161200
5678.547400
5791.405300
5880.070400
5980.946900
6079.321200
6178.648600
6281.552700
6375.980900
6469.450900
6589.388400
6675.781500
6782.913600
6887.410700
6979.577500
7074.273900
7184.209300
7275.919600
7373.108300
7478.527100
7583.970500
7676.333900
7778.671600
7874.168100
7981.640700
8076.637500
8176.290500
8270.558700
8377.890700
8476.121400
8581.717500
8682.563000
8774.597300
8879.479700
8977.527700
9075.307600
9186.773800
9282.851200
9379.547400
9483.117500
9579.647600
9686.809800
9784.604100
9877.641400
9975.033100
10094.384200
10182.481100
10278.418200
10387.593500
10475.218400
10578.940400
10673.944400
10778.180900
10878.876200
10975.849200
11088.152900
11185.070200
11273.411800
11382.399400
11472.965500
11578.870400
11673.651900
11779.633600
11877.217700
11975.806300
12075.851500
12176.224100
12274.574800
12376.063500
12483.738500
12581.505600
12680.363700
12784.645200
12868.278200
12970.693400
13077.646500
13174.208000
13275.622200
13371.742100
13483.608400
13572.555700
13669.425200
13771.866200
13892.751500
13975.826100
14082.484600
14173.899100
14272.839200
14371.951900
14470.943500
14578.037800
14674.408000
14792.414600
14864.599600
14974.672200
15072.482400
15174.897900
15268.468500
15370.043800
15474.629100
15568.695900
15678.637900
15768.528000
15874.401000
15977.724600
16079.024000
16177.561800
16272.664600
16371.449900
16470.451100
16573.722100
16680.396800
16777.762300
16878.052900
16975.001900
17076.887400
17169.595100
17277.151300
17354.731300
17475.822100
17572.619600
17673.582300
17777.569900
17872.680100
17971.146800
18067.335400
18174.677600
18269.761300
18372.366300
18473.325900
18568.711600
18683.478900
18796.225000
18881.374500
18967.654400
19069.617000
19171.264400
19289.342100
19371.249600
19468.632100
19566.474200
19671.751100
19769.111100
19860.833900
19967.816100
20067.681200
20164.842800
20264.041300
20360.572200
20469.861500
20561.489700
20670.064200
20761.345000
20899.887200
20973.527600
21065.796800
21169.729100
21268.035600
21375.500500
21464.245000
21564.363800
21664.755400
21770.981100
21863.046000
21957.376700
22060.001000
22190.157300
22260.843700
22360.424400
22464.971200
22561.912600
22664.053300
22762.532000
22862.300200
22966.131900
23063.364400
23151.262700
23261.773500
23358.392600
23487.031200
23568.935600
23655.670400
23775.800000
23858.985600
23964.245500
24066.865500
24164.841900
24263.263200
24353.591200
24459.279500
24564.325500
24663.230100
24749.011900
24862.295800
24960.083500
25062.083600
25158.933800
25257.998700
25360.781800
25457.126900
25546.494900
25657.466000
25765.292100
25852.747000
25951.339200
26057.799400
26155.079900
26259.348100
26348.604900
26452.619900
26559.665800
26648.814300
26743.831400
26853.765400
26955.431700
27040.151600
27150.624100
27244.376000
27353.412300
27449.834600
27562.822000
27653.230500
27742.881400
27861.869800
27939.555600
28046.256500
28152.069800
28250.417400
28344.161400
28457.288100
28552.256200
28650.828000
28743.257700
28845.735000
28942.440300
29052.033300
29146.914500

" + ], + "text/plain": [ + "" + ] + }, + "metadata": {}, + "output_type": "display_data" } ], "source": [ diff --git a/research_journal.md b/research_journal.md index d4eb723..d8a3b1f 100644 --- a/research_journal.md +++ b/research_journal.md @@ -11,4 +11,8 @@ Forked repo - [ ] make the latent space sparse, interpretable, compressed etc? -Wow 25GB or gpu ram was not enougth +Wow 25GB or gpu ram was not enougth? + +This page says, 17.22 GB of GPU RAM. to fine tune a 1b model https://lightning.ai/lightning-ai/studios/finetune-and-serve-llama-3-2-1b-and-3b presumably 2b is ~40. I know you can train in 8bit though, and use adam 8b + +But I eventually found it takes 25gb, but that's with a 128 seq len