LoKr experiments during hyper parameter search

lokr--llama-3.2-3B-rank32-factor4-all-linear-lr0.003--2026-08-05T18-01-26+00-00.json

@@ -0,0 +1,367 @@
+{
+  "run_info": {
+    "created_at": "2026-08-05T18:01:26+00:00",
+    "total_time": 4800.918897571013,
+    "experiment_name": "lokr/llama-3.2-3B-rank32-factor4-all-linear-lr0.003",
+    "peft_branch": "auto-research",
+    "train_config": {
+      "model_id": "meta-llama/Llama-3.2-3B",
+      "dtype": "bfloat16",
+      "max_seq_length": 768,
+      "batch_size": 4,
+      "batch_size_eval": 50,
+      "max_steps": 5000,
+      "eval_steps": 250,
+      "compile": false,
+      "use_gc": false,
+      "query_template": "Question: {query} Think step by step.\nAnswer:",
+      "seed": 0,
+      "grad_norm_clip": 1.0,
+      "optimizer_type": "AdamW",
+      "optimizer_kwargs": {
+        "lr": 0.003,
+        "weight_decay": 0.1
+      },
+      "lr_scheduler": "cosine",
+      "use_amp": false,
+      "autocast_adapter_dtype": true,
+      "generation_kwargs": {
+        "max_length": 800,
+        "max_new_tokens": 300
+      },
+      "attn_implementation": null,
+      "init_kv_cache_prefix": null
+    },
+    "peft_config": {
+      "task_type": null,
+      "peft_type": "LOKR",
+      "auto_mapping": null,
+      "peft_version": "0.20.1.dev0@UNKNOWN",
+      "base_model_name_or_path": "meta-llama/Llama-3.2-3B",
+      "revision": null,
+      "inference_mode": false,
+      "rank_pattern": {},
+      "alpha_pattern": {},
+      "r": 32,
+      "alpha": 64,
+      "rank_dropout": 0.0,
+      "module_dropout": 0.0,
+      "use_effective_conv2d": false,
+      "decompose_both": false,
+      "decompose_factor": 4,
+      "rank_dropout_scale": false,
+      "target_modules": [
+        "k_proj",
+        "o_proj",
+        "v_proj",
+        "gate_proj",
+        "down_proj",
+        "up_proj",
+        "q_proj"
+      ],
+      "exclude_modules": null,
+      "init_weights": true,
+      "layers_to_transform": null,
+      "layers_pattern": null,
+      "modules_to_save": null
+    },
+    "error_msg": ""
+  },
+  "train_info": {
+    "accelerator_memory_reserved_avg": 29442615699,
+    "accelerator_memory_max": 40982544384,
+    "accelerator_memory_reserved_99th": 37618712576,
+    "train_time": 4286.277680464031,
+    "file_size": 48714440,
+    "num_trainable_params": 12160064,
+    "num_total_params": 3224909888,
+    "status": "success",
+    "metrics": [
+      {
+        "step": 250,
+        "valid accuracy": 0.42,
+        "train loss": 0.8480014306306839,
+        "train samples": 1000,
+        "train time": 173.18856941291597,
+        "eval time": 30.864087088993983,
+        "tokens / sec": 1222.476753042632,
+        "mem allocated avg": 6973740474.368,
+        "mem reserved avg": 29612415385.6,
+        "elapsed time": 236.343513550004
+      },
+      {
+        "step": 500,
+        "valid accuracy": 0.38,
+        "train loss": 0.6955219099521637,
+        "train samples": 2000,
+        "train time": 172.0803752382344,
+        "eval time": 19.406204919010634,
+        "tokens / sec": 1208.7084289073873,
+        "mem allocated avg": 6965646647.296,
+        "mem reserved avg": 29059522232.32,
+        "elapsed time": 435.6121215989988
+      },
+      {
+        "step": 750,
+        "valid accuracy": 0.32,
+        "train loss": 0.6928771364688874,
+        "train samples": 3000,
+        "train time": 173.00618223295896,
+        "eval time": 30.71565401001135,
+        "tokens / sec": 1239.267852933148,
+        "mem allocated avg": 6976378224.64,
+        "mem reserved avg": 29390310211.584,
+        "elapsed time": 647.0163669160102
+      },
+      {
+        "step": 1000,
+        "valid accuracy": 0.42,
+        "train loss": 0.672966784954071,
+        "train samples": 4000,
+        "train time": 170.4940585368604,
+        "eval time": 30.606684819009388,
+        "tokens / sec": 1221.954605268302,
+        "mem allocated avg": 6967910549.504,
+        "mem reserved avg": 29517187907.584,
+        "elapsed time": 855.9565882030001
+      },
+      {
+        "step": 1250,
+        "valid accuracy": 0.34,
+        "train loss": 0.6651446379423142,
+        "train samples": 5000,
+        "train time": 169.89519203378586,
+        "eval time": 25.17170481197536,
+        "tokens / sec": 1227.4508625207563,
+        "mem allocated avg": 6967487967.232,
+        "mem reserved avg": 29469733552.128,
+        "elapsed time": 1058.731640137994
+      },
+      {
+        "step": 1500,
+        "valid accuracy": 0.44,
+        "train loss": 0.6554510877132416,
+        "train samples": 6000,
+        "train time": 171.36726364697097,
+        "eval time": 18.756061262014555,
+        "tokens / sec": 1221.5343557754245,
+        "mem allocated avg": 6969545603.072,
+        "mem reserved avg": 29379455352.832,
+        "elapsed time": 1256.6927439470019
+      },
+      {
+        "step": 1750,
+        "valid accuracy": 0.3,
+        "train loss": 0.6429227473735809,
+        "train samples": 7000,
+        "train time": 173.86972841914394,
+        "eval time": 21.52584348298842,
+        "tokens / sec": 1204.0911428544507,
+        "mem allocated avg": 6970501986.304,
+        "mem reserved avg": 29966490140.672,
+        "elapsed time": 1459.8789294830058
+      },
+      {
+        "step": 2000,
+        "valid accuracy": 0.4,
+        "train loss": 0.6416963441371918,
+        "train samples": 8000,
+        "train time": 168.05695327802096,
+        "eval time": 30.620164482999826,
+        "tokens / sec": 1235.866746057232,
+        "mem allocated avg": 6966375342.08,
+        "mem reserved avg": 29494203121.664,
+        "elapsed time": 1666.4743399640138
+      },
+      {
+        "step": 2250,
+        "valid accuracy": 0.44,
+        "train loss": 0.6263820115327835,
+        "train samples": 9000,
+        "train time": 173.70071922219358,
+        "eval time": 20.667255882988684,
+        "tokens / sec": 1237.4617731147328,
+        "mem allocated avg": 6978335006.72,
+        "mem reserved avg": 29878275538.944,
+        "elapsed time": 1868.6638691370026
+      },
+      {
+        "step": 2500,
+        "valid accuracy": 0.4,
+        "train loss": 0.6203293619155884,
+        "train samples": 10000,
+        "train time": 168.98416426277254,
+        "eval time": 20.466615880985046,
+        "tokens / sec": 1218.8538547299538,
+        "mem allocated avg": 6963728519.168,
+        "mem reserved avg": 29039137914.88,
+        "elapsed time": 2065.9307697149925
+      },
+      {
+        "step": 2750,
+        "valid accuracy": 0.42,
+        "train loss": 0.6047633152008056,
+        "train samples": 11000,
+        "train time": 176.7334707181144,
+        "eval time": 30.730138742015697,
+        "tokens / sec": 1198.8730778560052,
+        "mem allocated avg": 6973704886.272,
+        "mem reserved avg": 29623337353.216,
+        "elapsed time": 2281.1155376670067
+      },
+      {
+        "step": 3000,
+        "valid accuracy": 0.46,
+        "train loss": 0.5908040742874145,
+        "train samples": 12000,
+        "train time": 170.17627633034135,
+        "eval time": 30.608639862999553,
+        "tokens / sec": 1226.557570191613,
+        "mem allocated avg": 6969421531.136,
+        "mem reserved avg": 29619881246.72,
+        "elapsed time": 2489.831897511991
+      },
+      {
+        "step": 3250,
+        "valid accuracy": 0.52,
+        "train loss": 0.5951890103816986,
+        "train samples": 13000,
+        "train time": 172.44382934289752,
+        "eval time": 26.391422343993327,
+        "tokens / sec": 1223.0127387198759,
+        "mem allocated avg": 6971181449.216,
+        "mem reserved avg": 29373121953.792,
+        "elapsed time": 2696.2320239050023
+      },
+      {
+        "step": 3500,
+        "valid accuracy": 0.58,
+        "train loss": 0.5717675786018371,
+        "train samples": 14000,
+        "train time": 171.12018194136908,
+        "eval time": 23.351411000010557,
+        "tokens / sec": 1225.7467098291577,
+        "mem allocated avg": 6969213763.584,
+        "mem reserved avg": 29377182040.064,
+        "elapsed time": 2898.503444966016
+      },
+      {
+        "step": 3750,
+        "valid accuracy": 0.44,
+        "train loss": 0.5641409082412719,
+        "train samples": 15000,
+        "train time": 172.8447106151143,
+        "eval time": 27.62292090099072,
+        "tokens / sec": 1253.7438908532647,
+        "mem allocated avg": 6980701048.832,
+        "mem reserved avg": 29832146583.552,
+        "elapsed time": 3106.7359900620067
+      },
+      {
+        "step": 4000,
+        "valid accuracy": 0.56,
+        "train loss": 0.5671093467473983,
+        "train samples": 16000,
+        "train time": 169.0208053018432,
+        "eval time": 23.257049122999888,
+        "tokens / sec": 1209.1588348252371,
+        "mem allocated avg": 6962095366.144,
+        "mem reserved avg": 29339097759.744,
+        "elapsed time": 3306.899890332017
+      },
+      {
+        "step": 4250,
+        "valid accuracy": 0.56,
+        "train loss": 0.5516804548501968,
+        "train samples": 17000,
+        "train time": 172.25935157504864,
+        "eval time": 30.59878140699584,
+        "tokens / sec": 1227.1554378161213,
+        "mem allocated avg": 6972005851.136,
+        "mem reserved avg": 29362913017.856,
+        "elapsed time": 3517.3524070070125
+      },
+      {
+        "step": 4500,
+        "valid accuracy": 0.56,
+        "train loss": 0.5537665351629257,
+        "train samples": 18000,
+        "train time": 169.23266626289114,
+        "eval time": 18.641614267020486,
+        "tokens / sec": 1228.001688971615,
+        "mem allocated avg": 6967714617.344,
+        "mem reserved avg": 29272676761.6,
+        "elapsed time": 3713.0376706779934
+      },
+      {
+        "step": 4750,
+        "valid accuracy": 0.58,
+        "train loss": 0.5407684857845306,
+        "train samples": 19000,
+        "train time": 173.53581094602123,
+        "eval time": 20.048838881979464,
+        "tokens / sec": 1209.7733537275606,
+        "mem allocated avg": 6969948028.928,
+        "mem reserved avg": 29378473885.696,
+        "elapsed time": 3914.253921037016
+      },
+      {
+        "step": 5000,
+        "valid accuracy": 0.5,
+        "train loss": 0.5477373919487,
+        "train samples": 20000,
+        "train time": 166.469098860136,
+        "eval time": 30.707921049994184,
+        "tokens / sec": 1251.1631373399377,
+        "mem allocated avg": 6967096588.288,
+        "mem reserved avg": 28866752020.48,
+        "elapsed time": 4119.086701190012
+      },
+      {
+        "step": 5000,
+        "test accuracy": 0.513267626990144,
+        "train loss": 0.5477373919487,
+        "train samples": 20000,
+        "train total tokens": 4198051,
+        "forgetting": 0.14642632007598877
+      }
+    ]
+  },
+  "meta_info": {
+    "model_info": {
+      "sha": "13afe5124825b4f3751f836b40dafda64c1ed062",
+      "created_at": "2024-09-18T15:23:48+00:00"
+    },
+    "dataset_info": {
+      "metamath": {
+        "sha": "aa4f34d3d2d3231299b5b03d9b3e5a20da45aa18",
+        "created_at": "2023-09-21T17:22:46+00:00"
+      },
+      "gsm8k": {
+        "sha": "740312add88f781978c0658806c59bc2815b9866",
+        "created_at": "2022-04-12T10:22:10+00:00"
+      }
+    },
+    "package_info": {
+      "transformers-version": "5.12.1",
+      "transformers-commit-hash": null,
+      "peft-version": "0.20.1.dev0",
+      "peft-commit-hash": "c702de402330f48c0d8b1f260d59413aa3110b65",
+      "datasets-version": "4.8.4",
+      "datasets-commit-hash": null,
+      "bitsandbytes-version": "0.49.2",
+      "bitsandbytes-commit-hash": null,
+      "torch-version": "2.11.0+cu130",
+      "torch-commit-hash": null
+    },
+    "system_info": {
+      "system": "Linux",
+      "release": "7.0.0-1009-aws",
+      "version": "#9~24.04.1-Ubuntu SMP PREEMPT Mon Jul 13 22:43:41 UTC 2026",
+      "machine": "x86_64",
+      "processor": "x86_64",
+      "accelerator": "NVIDIA L40S"
+    },
+    "pytorch_info": "PyTorch built with:\n  - GCC 13.3\n  - C++ Version: 201703\n  - Intel(R) oneAPI Math Kernel Library Version 2024.2-Product Build 20240605 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v3.10.2 (Git Hash f1d471933dc852f956fd05389f9313c7148783d5)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 13.0\n  - NVCC architecture flags: -gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_120,code=sm_120\n  - CuDNN 90.7.1  (built against CUDA 12.8)\n    - Built with CuDNN 91.9\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, COMMIT_SHA=70d99e998b4955e0049d13a98d77ae1b14db1f45, CUDA_VERSION=13.0, CUDNN_VERSION=9.19.0, CXX_COMPILER=/opt/rh/gcc-toolset-13/root/usr/bin/c++, CXX_FLAGS= -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DLIBKINETO_NOXPUPTI=ON -DUSE_FBGEMM -DUSE_MSLK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -DC10_NODEPRECATED -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=range-loop-construct -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-unknown-pragmas -Wno-unused-parameter -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=old-style-cast -faligned-new -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-dangling-reference -Wno-error=dangling-reference -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, TORCH_VERSION=2.11.0, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, USE_XCCL=OFF, USE_XPU=OFF, \n"
+  }
+}

lokr--llama-3.2-3B-rank32-factor4-all-linear-lr0.003-10k--2026-08-05T20-43-48+00-00.json

@@ -0,0 +1,607 @@
+{
+  "run_info": {
+    "created_at": "2026-08-05T20:43:48+00:00",
+    "total_time": 8920.977496453008,
+    "experiment_name": "lokr/llama-3.2-3B-rank32-factor4-all-linear-lr0.003-10k",
+    "peft_branch": "auto-research",
+    "train_config": {
+      "model_id": "meta-llama/Llama-3.2-3B",
+      "dtype": "bfloat16",
+      "max_seq_length": 768,
+      "batch_size": 4,
+      "batch_size_eval": 50,
+      "max_steps": 10000,
+      "eval_steps": 250,
+      "compile": false,
+      "use_gc": false,
+      "query_template": "Question: {query} Think step by step.\nAnswer:",
+      "seed": 0,
+      "grad_norm_clip": 1.0,
+      "optimizer_type": "AdamW",
+      "optimizer_kwargs": {
+        "lr": 0.003,
+        "weight_decay": 0.1
+      },
+      "lr_scheduler": "cosine",
+      "use_amp": false,
+      "autocast_adapter_dtype": true,
+      "generation_kwargs": {
+        "max_length": 800,
+        "max_new_tokens": 300
+      },
+      "attn_implementation": null,
+      "init_kv_cache_prefix": null
+    },
+    "peft_config": {
+      "task_type": null,
+      "peft_type": "LOKR",
+      "auto_mapping": null,
+      "peft_version": "0.20.1.dev0@UNKNOWN",
+      "base_model_name_or_path": "meta-llama/Llama-3.2-3B",
+      "revision": null,
+      "inference_mode": false,
+      "rank_pattern": {},
+      "alpha_pattern": {},
+      "r": 32,
+      "alpha": 64,
+      "rank_dropout": 0.0,
+      "module_dropout": 0.0,
+      "use_effective_conv2d": false,
+      "decompose_both": false,
+      "decompose_factor": 4,
+      "rank_dropout_scale": false,
+      "target_modules": [
+        "v_proj",
+        "o_proj",
+        "down_proj",
+        "q_proj",
+        "gate_proj",
+        "up_proj",
+        "k_proj"
+      ],
+      "exclude_modules": null,
+      "init_weights": true,
+      "layers_to_transform": null,
+      "layers_pattern": null,
+      "modules_to_save": null
+    },
+    "error_msg": ""
+  },
+  "train_info": {
+    "accelerator_memory_reserved_avg": 29425366623,
+    "accelerator_memory_max": 40980447232,
+    "accelerator_memory_reserved_99th": 37662752768,
+    "train_time": 7889.369267331087,
+    "file_size": 48714440,
+    "num_trainable_params": 12160064,
+    "num_total_params": 3224909888,
+    "status": "success",
+    "metrics": [
+      {
+        "step": 250,
+        "valid accuracy": 0.42,
+        "train loss": 0.8949494372606277,
+        "train samples": 1000,
+        "train time": 172.23825368791586,
+        "eval time": 30.842159669991815,
+        "tokens / sec": 1229.2217057868027,
+        "mem allocated avg": 6974068426.752,
+        "mem reserved avg": 29637765758.976,
+        "elapsed time": 234.25320881602238
+      },
+      {
+        "step": 500,
+        "valid accuracy": 0.44,
+        "train loss": 0.6856197105646133,
+        "train samples": 2000,
+        "train time": 171.0570292349148,
+        "eval time": 20.454432144993916,
+        "tokens / sec": 1215.9395081879845,
+        "mem allocated avg": 6965838184.448,
+        "mem reserved avg": 29036730384.384,
+        "elapsed time": 433.53717456100276
+      },
+      {
+        "step": 750,
+        "valid accuracy": 0.32,
+        "train loss": 0.6662836416959762,
+        "train samples": 3000,
+        "train time": 174.30962963579805,
+        "eval time": 19.629886487993645,
+        "tokens / sec": 1230.0008923658936,
+        "mem allocated avg": 6976710492.16,
+        "mem reserved avg": 29298144575.488,
+        "elapsed time": 635.1746712150052
+      },
+      {
+        "step": 1000,
+        "valid accuracy": 0.24,
+        "train loss": 0.6707030124664307,
+        "train samples": 4000,
+        "train time": 170.08980607590638,
+        "eval time": 19.962032829003874,
+        "tokens / sec": 1224.8588249140892,
+        "mem allocated avg": 6968188157.952,
+        "mem reserved avg": 29486888255.488,
+        "elapsed time": 833.0557732230227
+      },
+      {
+        "step": 1250,
+        "valid accuracy": 0.26,
+        "train loss": 0.6850654114484787,
+        "train samples": 5000,
+        "train time": 168.75559069094015,
+        "eval time": 30.746524874004535,
+        "tokens / sec": 1235.739800655953,
+        "mem allocated avg": 6967430975.488,
+        "mem reserved avg": 29512901328.896,
+        "elapsed time": 1040.2997735380195
+      },
+      {
+        "step": 1500,
+        "valid accuracy": 0.34,
+        "train loss": 0.6718376358747482,
+        "train samples": 6000,
+        "train time": 169.33231035905192,
+        "eval time": 18.44809911600896,
+        "tokens / sec": 1236.214161113936,
+        "mem allocated avg": 6969443014.656,
+        "mem reserved avg": 29386233348.096,
+        "elapsed time": 1235.9416526080167
+      },
+      {
+        "step": 1750,
+        "valid accuracy": 0.4,
+        "train loss": 0.6636214228868484,
+        "train samples": 7000,
+        "train time": 170.86081746208947,
+        "eval time": 30.7044950290001,
+        "tokens / sec": 1225.2955540637724,
+        "mem allocated avg": 6970556262.4,
+        "mem reserved avg": 29963210194.944,
+        "elapsed time": 1445.2593260980211
+      },
+      {
+        "step": 2000,
+        "valid accuracy": 0.44,
+        "train loss": 0.6641189997196197,
+        "train samples": 8000,
+        "train time": 169.15395127126249,
+        "eval time": 30.652748754015192,
+        "tokens / sec": 1227.8518972751033,
+        "mem allocated avg": 6966487302.144,
+        "mem reserved avg": 29441472331.776,
+        "elapsed time": 1652.9759574920172
+      },
+      {
+        "step": 2250,
+        "valid accuracy": 0.4,
+        "train loss": 0.6514279855489731,
+        "train samples": 9000,
+        "train time": 174.06954159290763,
+        "eval time": 30.6216060250008,
+        "tokens / sec": 1234.839811910885,
+        "mem allocated avg": 6978265659.392,
+        "mem reserved avg": 29786906820.608,
+        "elapsed time": 1865.4745528470085
+      },
+      {
+        "step": 2500,
+        "valid accuracy": 0.38,
+        "train loss": 0.648625570654869,
+        "train samples": 10000,
+        "train time": 170.52176274178782,
+        "eval time": 29.434755503985798,
+        "tokens / sec": 1207.8634227578625,
+        "mem allocated avg": 6963658600.448,
+        "mem reserved avg": 29001515008.0,
+        "elapsed time": 2073.2369456640154
+      },
+      {
+        "step": 2750,
+        "valid accuracy": 0.34,
+        "train loss": 0.6369520275592804,
+        "train samples": 11000,
+        "train time": 173.95405193630722,
+        "eval time": 30.711821962991962,
+        "tokens / sec": 1218.0285405342534,
+        "mem allocated avg": 6973873780.736,
+        "mem reserved avg": 29700227334.144,
+        "elapsed time": 2285.583145290002
+      },
+      {
+        "step": 3000,
+        "valid accuracy": 0.4,
+        "train loss": 0.6276206455230713,
+        "train samples": 12000,
+        "train time": 171.57854383610538,
+        "eval time": 25.434444301004987,
+        "tokens / sec": 1216.533229232807,
+        "mem allocated avg": 6969537423.36,
+        "mem reserved avg": 29600050577.408,
+        "elapsed time": 2490.512576183013
+      },
+      {
+        "step": 3250,
+        "valid accuracy": 0.46,
+        "train loss": 0.6353357185125351,
+        "train samples": 13000,
+        "train time": 172.03060490611824,
+        "eval time": 30.666481404012302,
+        "tokens / sec": 1225.9504645414365,
+        "mem allocated avg": 6971543304.192,
+        "mem reserved avg": 29378851373.056,
+        "elapsed time": 2700.8043426890217
+      },
+      {
+        "step": 3500,
+        "valid accuracy": 0.36,
+        "train loss": 0.6162731720209121,
+        "train samples": 14000,
+        "train time": 170.90886393698747,
+        "eval time": 21.72865568700945,
+        "tokens / sec": 1227.2622681368528,
+        "mem allocated avg": 6969227558.912,
+        "mem reserved avg": 29302649257.984,
+        "elapsed time": 2901.162421387009
+      },
+      {
+        "step": 3750,
+        "valid accuracy": 0.48,
+        "train loss": 0.6119594979286194,
+        "train samples": 15000,
+        "train time": 176.6123301939515,
+        "eval time": 22.168275571981212,
+        "tokens / sec": 1226.998136325034,
+        "mem allocated avg": 6980796526.592,
+        "mem reserved avg": 29780892188.672,
+        "elapsed time": 3107.718649386021
+      },
+      {
+        "step": 4000,
+        "valid accuracy": 0.4,
+        "train loss": 0.6219998714923859,
+        "train samples": 16000,
+        "train time": 169.44954544273787,
+        "eval time": 21.188790128013352,
+        "tokens / sec": 1206.0994289835012,
+        "mem allocated avg": 6962208227.328,
+        "mem reserved avg": 29357410091.008,
+        "elapsed time": 3306.224600540998
+      },
+      {
+        "step": 4250,
+        "valid accuracy": 0.44,
+        "train loss": 0.6031627696752548,
+        "train samples": 17000,
+        "train time": 172.79886594458367,
+        "eval time": 19.71278954198351,
+        "tokens / sec": 1223.3240006782924,
+        "mem allocated avg": 6971919788.032,
+        "mem reserved avg": 29329702518.784,
+        "elapsed time": 3506.296442656021
+      },
+      {
+        "step": 4500,
+        "valid accuracy": 0.42,
+        "train loss": 0.6039325877428054,
+        "train samples": 18000,
+        "train time": 170.91172845792607,
+        "eval time": 30.605959645996336,
+        "tokens / sec": 1215.9376180620586,
+        "mem allocated avg": 6967363680.256,
+        "mem reserved avg": 29285041569.792,
+        "elapsed time": 3715.659448362014
+      },
+      {
+        "step": 4750,
+        "valid accuracy": 0.42,
+        "train loss": 0.5903248875141144,
+        "train samples": 19000,
+        "train time": 171.77074804794393,
+        "eval time": 30.74948202399537,
+        "tokens / sec": 1222.2046092586306,
+        "mem allocated avg": 6969873039.36,
+        "mem reserved avg": 29383817428.992,
+        "elapsed time": 3925.832625826006
+      },
+      {
+        "step": 5000,
+        "valid accuracy": 0.46,
+        "train loss": 0.5919159969091415,
+        "train samples": 20000,
+        "train time": 172.8896059941908,
+        "eval time": 27.89795029599918,
+        "tokens / sec": 1204.6993733504046,
+        "mem allocated avg": 6966943275.008,
+        "mem reserved avg": 28902328107.008,
+        "elapsed time": 4134.299820120999
+      },
+      {
+        "step": 5250,
+        "valid accuracy": 0.46,
+        "train loss": 0.5757226728200913,
+        "train samples": 21000,
+        "train time": 174.73050247496576,
+        "eval time": 19.57498934297473,
+        "tokens / sec": 1213.0566615314801,
+        "mem allocated avg": 6974017998.848,
+        "mem reserved avg": 29520870506.496,
+        "elapsed time": 4336.321290748019
+      },
+      {
+        "step": 5500,
+        "valid accuracy": 0.46,
+        "train loss": 0.5817540099620819,
+        "train samples": 22000,
+        "train time": 168.02078990414157,
+        "eval time": 30.746748675999697,
+        "tokens / sec": 1218.5753924666747,
+        "mem allocated avg": 6961666510.848,
+        "mem reserved avg": 28999191363.584,
+        "elapsed time": 4542.736618685012
+      },
+      {
+        "step": 5750,
+        "valid accuracy": 0.46,
+        "train loss": 0.5647573419809342,
+        "train samples": 23000,
+        "train time": 173.413503956981,
+        "eval time": 22.96224321599584,
+        "tokens / sec": 1240.5896604993861,
+        "mem allocated avg": 6978015358.976,
+        "mem reserved avg": 29507129966.592,
+        "elapsed time": 4746.8634043490165
+      },
+      {
+        "step": 6000,
+        "valid accuracy": 0.46,
+        "train loss": 0.5738131380081177,
+        "train samples": 24000,
+        "train time": 170.20395927492063,
+        "eval time": 30.678496007982176,
+        "tokens / sec": 1228.7845764044876,
+        "mem allocated avg": 6968983029.76,
+        "mem reserved avg": 29429233352.704,
+        "elapsed time": 4955.5860632859985
+      },
+      {
+        "step": 6250,
+        "valid accuracy": 0.52,
+        "train loss": 0.5661250177621842,
+        "train samples": 25000,
+        "train time": 172.2749537579366,
+        "eval time": 30.716671487985877,
+        "tokens / sec": 1205.9094805615593,
+        "mem allocated avg": 6968072611.84,
+        "mem reserved avg": 29586763022.336,
+        "elapsed time": 5166.3077040040225
+      },
+      {
+        "step": 6500,
+        "valid accuracy": 0.6,
+        "train loss": 0.5527391213178635,
+        "train samples": 26000,
+        "train time": 171.18087319325423,
+        "eval time": 23.657987303013215,
+        "tokens / sec": 1216.57283384045,
+        "mem allocated avg": 6967868983.296,
+        "mem reserved avg": 29137318182.912,
+        "elapsed time": 5368.990365178004
+      },
+      {
+        "step": 6750,
+        "valid accuracy": 0.5,
+        "train loss": 0.5544149683713913,
+        "train samples": 27000,
+        "train time": 169.88396649708739,
+        "eval time": 20.297047789004864,
+        "tokens / sec": 1220.444779310911,
+        "mem allocated avg": 6966295676.928,
+        "mem reserved avg": 29471553880.064,
+        "elapsed time": 5566.813652031007
+      },
+      {
+        "step": 7000,
+        "valid accuracy": 0.58,
+        "train loss": 0.5394300025701523,
+        "train samples": 28000,
+        "train time": 170.22449614675133,
+        "eval time": 30.696163899992825,
+        "tokens / sec": 1213.5562429387894,
+        "mem allocated avg": 6963857641.472,
+        "mem reserved avg": 29040555589.632,
+        "elapsed time": 5775.5202143070055
+      },
+      {
+        "step": 7250,
+        "valid accuracy": 0.52,
+        "train loss": 0.5215741968154908,
+        "train samples": 29000,
+        "train time": 176.06252912833588,
+        "eval time": 22.185015341994585,
+        "tokens / sec": 1239.0541081060178,
+        "mem allocated avg": 6982538385.408,
+        "mem reserved avg": 30031778676.736,
+        "elapsed time": 5981.63492844702
+      },
+      {
+        "step": 7500,
+        "valid accuracy": 0.6,
+        "train loss": 0.5314810944795608,
+        "train samples": 30000,
+        "train time": 170.96637836899026,
+        "eval time": 30.624822825018782,
+        "tokens / sec": 1224.626748237742,
+        "mem allocated avg": 6967260755.968,
+        "mem reserved avg": 28773529419.776,
+        "elapsed time": 6190.8903621690115
+      },
+      {
+        "step": 7750,
+        "valid accuracy": 0.58,
+        "train loss": 0.5229091964960099,
+        "train samples": 31000,
+        "train time": 168.64462656035903,
+        "eval time": 22.662872864981182,
+        "tokens / sec": 1217.672950442346,
+        "mem allocated avg": 6962802417.664,
+        "mem reserved avg": 29103579201.536,
+        "elapsed time": 6389.794873889012
+      },
+      {
+        "step": 8000,
+        "valid accuracy": 0.58,
+        "train loss": 0.5148865422010421,
+        "train samples": 32000,
+        "train time": 167.9298267000995,
+        "eval time": 18.765550082025584,
+        "tokens / sec": 1237.6895997825552,
+        "mem allocated avg": 6965727940.608,
+        "mem reserved avg": 29159053066.24,
+        "elapsed time": 6584.261041696998
+      },
+      {
+        "step": 8250,
+        "valid accuracy": 0.62,
+        "train loss": 0.5125223807692528,
+        "train samples": 33000,
+        "train time": 169.50851931003854,
+        "eval time": 20.5418613619986,
+        "tokens / sec": 1229.572418238078,
+        "mem allocated avg": 6968558663.68,
+        "mem reserved avg": 29544434106.368,
+        "elapsed time": 6782.0812423590105
+      },
+      {
+        "step": 8500,
+        "valid accuracy": 0.62,
+        "train loss": 0.5045390973091125,
+        "train samples": 34000,
+        "train time": 171.07040929992218,
+        "eval time": 22.437327421997907,
+        "tokens / sec": 1212.3487682571083,
+        "mem allocated avg": 6966485200.896,
+        "mem reserved avg": 29454936047.616,
+        "elapsed time": 6983.421357783023
+      },
+      {
+        "step": 8750,
+        "valid accuracy": 0.54,
+        "train loss": 0.5028502044677734,
+        "train samples": 35000,
+        "train time": 172.2428093130875,
+        "eval time": 27.035157548991265,
+        "tokens / sec": 1236.14449188981,
+        "mem allocated avg": 6974847162.368,
+        "mem reserved avg": 29512960049.152,
+        "elapsed time": 7190.340802373015
+      },
+      {
+        "step": 9000,
+        "valid accuracy": 0.62,
+        "train loss": 0.4954327570199966,
+        "train samples": 36000,
+        "train time": 172.43629023668473,
+        "eval time": 30.530970638006693,
+        "tokens / sec": 1212.784151833425,
+        "mem allocated avg": 6970138710.016,
+        "mem reserved avg": 29700848091.136,
+        "elapsed time": 7401.228956966021
+      },
+      {
+        "step": 9250,
+        "valid accuracy": 0.58,
+        "train loss": 0.490826522231102,
+        "train samples": 37000,
+        "train time": 175.1379066670488,
+        "eval time": 20.243021284986753,
+        "tokens / sec": 1216.584142489856,
+        "mem allocated avg": 6974603661.312,
+        "mem reserved avg": 29744686956.544,
+        "elapsed time": 7604.393750663003
+      },
+      {
+        "step": 9500,
+        "valid accuracy": 0.62,
+        "train loss": 0.49826363229751586,
+        "train samples": 38000,
+        "train time": 167.72828063127236,
+        "eval time": 19.732628641999327,
+        "tokens / sec": 1217.6181573635126,
+        "mem allocated avg": 6961817538.56,
+        "mem reserved avg": 29554399772.672,
+        "elapsed time": 7799.745701444015
+      },
+      {
+        "step": 9750,
+        "valid accuracy": 0.62,
+        "train loss": 0.48890409409999847,
+        "train samples": 39000,
+        "train time": 171.1529068540258,
+        "eval time": 30.650341822998598,
+        "tokens / sec": 1239.4647797652067,
+        "mem allocated avg": 6973635262.464,
+        "mem reserved avg": 29785061326.848,
+        "elapsed time": 8009.401820736006
+      },
+      {
+        "step": 10000,
+        "valid accuracy": 0.58,
+        "train loss": 0.4887709869146347,
+        "train samples": 40000,
+        "train time": 169.20751953776926,
+        "eval time": 30.604743445990607,
+        "tokens / sec": 1219.7448468236137,
+        "mem allocated avg": 6965170036.736,
+        "mem reserved avg": 29384043921.408,
+        "elapsed time": 8216.999875714013
+      },
+      {
+        "step": 10000,
+        "test accuracy": 0.5564821834723275,
+        "train loss": 0.4887709869146347,
+        "train samples": 40000,
+        "train total tokens": 8383359,
+        "forgetting": 0.27742505073547363
+      }
+    ]
+  },
+  "meta_info": {
+    "model_info": {
+      "sha": "13afe5124825b4f3751f836b40dafda64c1ed062",
+      "created_at": "2024-09-18T15:23:48+00:00"
+    },
+    "dataset_info": {
+      "metamath": {
+        "sha": "aa4f34d3d2d3231299b5b03d9b3e5a20da45aa18",
+        "created_at": "2023-09-21T17:22:46+00:00"
+      },
+      "gsm8k": {
+        "sha": "740312add88f781978c0658806c59bc2815b9866",
+        "created_at": "2022-04-12T10:22:10+00:00"
+      }
+    },
+    "package_info": {
+      "transformers-version": "5.12.1",
+      "transformers-commit-hash": null,
+      "peft-version": "0.20.1.dev0",
+      "peft-commit-hash": "c702de402330f48c0d8b1f260d59413aa3110b65",
+      "datasets-version": "4.8.4",
+      "datasets-commit-hash": null,
+      "bitsandbytes-version": "0.49.2",
+      "bitsandbytes-commit-hash": null,
+      "torch-version": "2.11.0+cu130",
+      "torch-commit-hash": null
+    },
+    "system_info": {
+      "system": "Linux",
+      "release": "7.0.0-1009-aws",
+      "version": "#9~24.04.1-Ubuntu SMP PREEMPT Mon Jul 13 22:43:41 UTC 2026",
+      "machine": "x86_64",
+      "processor": "x86_64",
+      "accelerator": "NVIDIA L40S"
+    },
+    "pytorch_info": "PyTorch built with:\n  - GCC 13.3\n  - C++ Version: 201703\n  - Intel(R) oneAPI Math Kernel Library Version 2024.2-Product Build 20240605 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v3.10.2 (Git Hash f1d471933dc852f956fd05389f9313c7148783d5)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 13.0\n  - NVCC architecture flags: -gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_120,code=sm_120\n  - CuDNN 90.7.1  (built against CUDA 12.8)\n    - Built with CuDNN 91.9\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, COMMIT_SHA=70d99e998b4955e0049d13a98d77ae1b14db1f45, CUDA_VERSION=13.0, CUDNN_VERSION=9.19.0, CXX_COMPILER=/opt/rh/gcc-toolset-13/root/usr/bin/c++, CXX_FLAGS= -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DLIBKINETO_NOXPUPTI=ON -DUSE_FBGEMM -DUSE_MSLK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -DC10_NODEPRECATED -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=range-loop-construct -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-unknown-pragmas -Wno-unused-parameter -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=old-style-cast -faligned-new -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-dangling-reference -Wno-error=dangling-reference -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, TORCH_VERSION=2.11.0, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, USE_XCCL=OFF, USE_XPU=OFF, \n"
+  }
+}

lokr--llama-3.2-3B-rank32-factor4-all-linear-lr0.003-10k-wd0--2026-08-05T23-12-59+00-00.json

@@ -0,0 +1,607 @@
+{
+  "run_info": {
+    "created_at": "2026-08-05T23:12:59+00:00",
+    "total_time": 9007.000587568997,
+    "experiment_name": "lokr/llama-3.2-3B-rank32-factor4-all-linear-lr0.003-10k-wd0",
+    "peft_branch": "auto-research",
+    "train_config": {
+      "model_id": "meta-llama/Llama-3.2-3B",
+      "dtype": "bfloat16",
+      "max_seq_length": 768,
+      "batch_size": 4,
+      "batch_size_eval": 50,
+      "max_steps": 10000,
+      "eval_steps": 250,
+      "compile": false,
+      "use_gc": false,
+      "query_template": "Question: {query} Think step by step.\nAnswer:",
+      "seed": 0,
+      "grad_norm_clip": 1.0,
+      "optimizer_type": "AdamW",
+      "optimizer_kwargs": {
+        "lr": 0.003,
+        "weight_decay": 0.0
+      },
+      "lr_scheduler": "cosine",
+      "use_amp": false,
+      "autocast_adapter_dtype": true,
+      "generation_kwargs": {
+        "max_length": 800,
+        "max_new_tokens": 300
+      },
+      "attn_implementation": null,
+      "init_kv_cache_prefix": null
+    },
+    "peft_config": {
+      "task_type": null,
+      "peft_type": "LOKR",
+      "auto_mapping": null,
+      "peft_version": "0.20.1.dev0@UNKNOWN",
+      "base_model_name_or_path": "meta-llama/Llama-3.2-3B",
+      "revision": null,
+      "inference_mode": false,
+      "rank_pattern": {},
+      "alpha_pattern": {},
+      "r": 32,
+      "alpha": 64,
+      "rank_dropout": 0.0,
+      "module_dropout": 0.0,
+      "use_effective_conv2d": false,
+      "decompose_both": false,
+      "decompose_factor": 4,
+      "rank_dropout_scale": false,
+      "target_modules": [
+        "k_proj",
+        "q_proj",
+        "gate_proj",
+        "down_proj",
+        "v_proj",
+        "up_proj",
+        "o_proj"
+      ],
+      "exclude_modules": null,
+      "init_weights": true,
+      "layers_to_transform": null,
+      "layers_pattern": null,
+      "modules_to_save": null
+    },
+    "error_msg": ""
+  },
+  "train_info": {
+    "accelerator_memory_reserved_avg": 29439301569,
+    "accelerator_memory_max": 40982544384,
+    "accelerator_memory_reserved_99th": 37675335680,
+    "train_time": 7933.3307930609735,
+    "file_size": 48714440,
+    "num_trainable_params": 12160064,
+    "num_total_params": 3224909888,
+    "status": "success",
+    "metrics": [
+      {
+        "step": 250,
+        "valid accuracy": 0.44,
+        "train loss": 0.8949982010126114,
+        "train samples": 1000,
+        "train time": 171.3976422761043,
+        "eval time": 24.5194029819977,
+        "tokens / sec": 1235.2503639399079,
+        "mem allocated avg": 6973625303.04,
+        "mem reserved avg": 29615586279.424,
+        "elapsed time": 226.39103153799078
+      },
+      {
+        "step": 500,
+        "valid accuracy": 0.36,
+        "train loss": 0.685373764514923,
+        "train samples": 2000,
+        "train time": 169.98101295909146,
+        "eval time": 30.617474483005935,
+        "tokens / sec": 1223.6366661143336,
+        "mem allocated avg": 6965853036.544,
+        "mem reserved avg": 29103000387.584,
+        "elapsed time": 434.7747755720047
+      },
+      {
+        "step": 750,
+        "valid accuracy": 0.34,
+        "train loss": 0.6668877815008163,
+        "train samples": 3000,
+        "train time": 174.92015415464994,
+        "eval time": 30.83243089399184,
+        "tokens / sec": 1225.7078152952254,
+        "mem allocated avg": 6976431538.176,
+        "mem reserved avg": 29361335959.552,
+        "elapsed time": 648.2199807830038
+      },
+      {
+        "step": 1000,
+        "valid accuracy": 0.3,
+        "train loss": 0.6703291549682617,
+        "train samples": 4000,
+        "train time": 168.94037354181637,
+        "eval time": 19.526054535992444,
+        "tokens / sec": 1233.1924905353212,
+        "mem allocated avg": 6967942184.96,
+        "mem reserved avg": 29474406006.784,
+        "elapsed time": 844.5394411569869
+      },
+      {
+        "step": 1250,
+        "valid accuracy": 0.3,
+        "train loss": 0.6888422293663025,
+        "train samples": 5000,
+        "train time": 169.8963452410535,
+        "eval time": 30.673396903002867,
+        "tokens / sec": 1227.442530939207,
+        "mem allocated avg": 6967469674.496,
+        "mem reserved avg": 29498984628.224,
+        "elapsed time": 1052.865548398986
+      },
+      {
+        "step": 1500,
+        "valid accuracy": 0.36,
+        "train loss": 0.6771859076023102,
+        "train samples": 6000,
+        "train time": 172.2039898619114,
+        "eval time": 19.118000183982076,
+        "tokens / sec": 1215.5990123565684,
+        "mem allocated avg": 6969083271.168,
+        "mem reserved avg": 29424594452.48,
+        "elapsed time": 1252.1105965079914
+      },
+      {
+        "step": 1750,
+        "valid accuracy": 0.4,
+        "train loss": 0.6689975430965424,
+        "train samples": 7000,
+        "train time": 169.03733924406697,
+        "eval time": 27.43203862401424,
+        "tokens / sec": 1238.5133422960462,
+        "mem allocated avg": 6970877360.128,
+        "mem reserved avg": 29957036179.456,
+        "elapsed time": 1456.3577687070065
+      },
+      {
+        "step": 2000,
+        "valid accuracy": 0.38,
+        "train loss": 0.6676360635757447,
+        "train samples": 8000,
+        "train time": 169.73538258188637,
+        "eval time": 30.58883438099292,
+        "tokens / sec": 1223.6458706528092,
+        "mem allocated avg": 6966508093.44,
+        "mem reserved avg": 29495159422.976,
+        "elapsed time": 1664.644099119003
+      },
+      {
+        "step": 2250,
+        "valid accuracy": 0.36,
+        "train loss": 0.6558140796422959,
+        "train samples": 9000,
+        "train time": 175.10481430220534,
+        "eval time": 30.702585085993633,
+        "tokens / sec": 1227.5390648542143,
+        "mem allocated avg": 6978239281.152,
+        "mem reserved avg": 29797459689.472,
+        "elapsed time": 1878.3001420579967
+      },
+      {
+        "step": 2500,
+        "valid accuracy": 0.38,
+        "train loss": 0.6527049542665482,
+        "train samples": 10000,
+        "train time": 171.3574726433726,
+        "eval time": 27.830049729003804,
+        "tokens / sec": 1201.9726763165818,
+        "mem allocated avg": 6963613773.824,
+        "mem reserved avg": 29037191757.824,
+        "elapsed time": 2085.3302400379907
+      },
+      {
+        "step": 2750,
+        "valid accuracy": 0.34,
+        "train loss": 0.6389935193061829,
+        "train samples": 11000,
+        "train time": 172.8635393739387,
+        "eval time": 30.60976703799679,
+        "tokens / sec": 1225.7124941868665,
+        "mem allocated avg": 6973982658.56,
+        "mem reserved avg": 29671278247.936,
+        "elapsed time": 2296.559379327984
+      },
+      {
+        "step": 3000,
+        "valid accuracy": 0.36,
+        "train loss": 0.6307049007415771,
+        "train samples": 12000,
+        "train time": 170.22398920581327,
+        "eval time": 30.546133575000567,
+        "tokens / sec": 1226.21377265239,
+        "mem allocated avg": 6969360951.296,
+        "mem reserved avg": 29585789943.808,
+        "elapsed time": 2505.2557326619863
+      },
+      {
+        "step": 3250,
+        "valid accuracy": 0.4,
+        "train loss": 0.6370149928331376,
+        "train samples": 13000,
+        "train time": 170.96555595166865,
+        "eval time": 23.28644464397803,
+        "tokens / sec": 1233.5876593740377,
+        "mem allocated avg": 6971444633.6,
+        "mem reserved avg": 29368172675.072,
+        "elapsed time": 2707.1341430539906
+      },
+      {
+        "step": 3500,
+        "valid accuracy": 0.4,
+        "train loss": 0.6177912712097168,
+        "train samples": 14000,
+        "train time": 172.09483984508552,
+        "eval time": 30.673877114983043,
+        "tokens / sec": 1218.8047020399363,
+        "mem allocated avg": 6969167626.24,
+        "mem reserved avg": 29359574351.872,
+        "elapsed time": 2917.7247013830056
+      },
+      {
+        "step": 3750,
+        "valid accuracy": 0.48,
+        "train loss": 0.612206832408905,
+        "train samples": 15000,
+        "train time": 176.50543513981393,
+        "eval time": 30.85041491300217,
+        "tokens / sec": 1227.7412297720161,
+        "mem allocated avg": 6980901793.792,
+        "mem reserved avg": 29857186578.432,
+        "elapsed time": 3132.8426076890028
+      },
+      {
+        "step": 4000,
+        "valid accuracy": 0.48,
+        "train loss": 0.6191495014429093,
+        "train samples": 16000,
+        "train time": 171.08480115508428,
+        "eval time": 30.740958007983863,
+        "tokens / sec": 1194.571339009482,
+        "mem allocated avg": 6961991802.88,
+        "mem reserved avg": 29372434087.936,
+        "elapsed time": 3342.560094543005
+      },
+      {
+        "step": 4250,
+        "valid accuracy": 0.5,
+        "train loss": 0.6020842252969741,
+        "train samples": 17000,
+        "train time": 170.72924799792236,
+        "eval time": 30.68499062000774,
+        "tokens / sec": 1238.1534065128221,
+        "mem allocated avg": 6972223258.624,
+        "mem reserved avg": 29325340442.624,
+        "elapsed time": 3551.5764037730114
+      },
+      {
+        "step": 4500,
+        "valid accuracy": 0.48,
+        "train loss": 0.603379097700119,
+        "train samples": 18000,
+        "train time": 171.22398967514164,
+        "eval time": 30.78706062101992,
+        "tokens / sec": 1213.72011243452,
+        "mem allocated avg": 6967436220.416,
+        "mem reserved avg": 29257954754.56,
+        "elapsed time": 3761.4645812119998
+      },
+      {
+        "step": 4750,
+        "valid accuracy": 0.48,
+        "train loss": 0.587617068529129,
+        "train samples": 19000,
+        "train time": 172.58199024686473,
+        "eval time": 30.659869527007686,
+        "tokens / sec": 1216.4594909335501,
+        "mem allocated avg": 6969999155.2,
+        "mem reserved avg": 29373851762.688,
+        "elapsed time": 3972.4039701599977
+      },
+      {
+        "step": 5000,
+        "valid accuracy": 0.42,
+        "train loss": 0.5884629808664322,
+        "train samples": 20000,
+        "train time": 169.2553744801262,
+        "eval time": 21.032124360994203,
+        "tokens / sec": 1230.5665367480312,
+        "mem allocated avg": 6966949509.12,
+        "mem reserved avg": 28873857171.456,
+        "elapsed time": 4170.4182277559885
+      },
+      {
+        "step": 5250,
+        "valid accuracy": 0.56,
+        "train loss": 0.5720609217882157,
+        "train samples": 21000,
+        "train time": 173.28161718780757,
+        "eval time": 27.016861401003553,
+        "tokens / sec": 1223.1995721177616,
+        "mem allocated avg": 6973880750.08,
+        "mem reserved avg": 29549282721.792,
+        "elapsed time": 4378.443800904992
+      },
+      {
+        "step": 5500,
+        "valid accuracy": 0.5,
+        "train loss": 0.5772967861890793,
+        "train samples": 22000,
+        "train time": 168.6424825040449,
+        "eval time": 30.556040593015496,
+        "tokens / sec": 1214.083171451708,
+        "mem allocated avg": 6961765470.208,
+        "mem reserved avg": 29048482824.192,
+        "elapsed time": 4585.332566566009
+      },
+      {
+        "step": 5750,
+        "valid accuracy": 0.5,
+        "train loss": 0.5600146712064743,
+        "train samples": 23000,
+        "train time": 174.04982690158067,
+        "eval time": 18.295675226981984,
+        "tokens / sec": 1236.05408767026,
+        "mem allocated avg": 6978040623.104,
+        "mem reserved avg": 29432387469.312,
+        "elapsed time": 4785.4279907169985
+      },
+      {
+        "step": 6000,
+        "valid accuracy": 0.58,
+        "train loss": 0.5662243200540542,
+        "train samples": 24000,
+        "train time": 170.55430918792263,
+        "eval time": 30.779600164009025,
+        "tokens / sec": 1226.260426932737,
+        "mem allocated avg": 6968833345.536,
+        "mem reserved avg": 29413118836.736,
+        "elapsed time": 4994.6479376729985
+      },
+      {
+        "step": 6250,
+        "valid accuracy": 0.52,
+        "train loss": 0.5612275519371033,
+        "train samples": 25000,
+        "train time": 172.04471145616844,
+        "eval time": 30.784128725004848,
+        "tokens / sec": 1207.5233132227236,
+        "mem allocated avg": 6967910082.56,
+        "mem reserved avg": 29668577116.16,
+        "elapsed time": 5205.249400614994
+      },
+      {
+        "step": 6500,
+        "valid accuracy": 0.52,
+        "train loss": 0.5485070507526397,
+        "train samples": 26000,
+        "train time": 171.21381389093585,
+        "eval time": 25.085028658999363,
+        "tokens / sec": 1216.3387711966918,
+        "mem allocated avg": 6967761831.936,
+        "mem reserved avg": 29132687671.296,
+        "elapsed time": 5409.425448774011
+      },
+      {
+        "step": 6750,
+        "valid accuracy": 0.66,
+        "train loss": 0.5476124639511109,
+        "train samples": 27000,
+        "train time": 172.53885023278417,
+        "eval time": 22.688375061989063,
+        "tokens / sec": 1201.66559427208,
+        "mem allocated avg": 6966259134.464,
+        "mem reserved avg": 29525173862.4,
+        "elapsed time": 5612.382705447002
+      },
+      {
+        "step": 7000,
+        "valid accuracy": 0.58,
+        "train loss": 0.5320072404742241,
+        "train samples": 28000,
+        "train time": 169.6953858086781,
+        "eval time": 28.003433011006564,
+        "tokens / sec": 1217.340112199066,
+        "mem allocated avg": 6963905171.456,
+        "mem reserved avg": 29061552275.456,
+        "elapsed time": 5817.900293373008
+      },
+      {
+        "step": 7250,
+        "valid accuracy": 0.58,
+        "train loss": 0.5134797089099884,
+        "train samples": 29000,
+        "train time": 177.20017832212034,
+        "eval time": 24.322306215006392,
+        "tokens / sec": 1231.099212572111,
+        "mem allocated avg": 6982032873.472,
+        "mem reserved avg": 30056919334.912,
+        "elapsed time": 6027.303013839002
+      },
+      {
+        "step": 7500,
+        "valid accuracy": 0.5,
+        "train loss": 0.5264345513582229,
+        "train samples": 30000,
+        "train time": 171.69255989571684,
+        "eval time": 30.784176111017587,
+        "tokens / sec": 1219.4471334527705,
+        "mem allocated avg": 6967294406.656,
+        "mem reserved avg": 28893562011.648,
+        "elapsed time": 6237.538969763991
+      },
+      {
+        "step": 7750,
+        "valid accuracy": 0.48,
+        "train loss": 0.5177160662412643,
+        "train samples": 31000,
+        "train time": 170.80944362809532,
+        "eval time": 28.494459347013617,
+        "tokens / sec": 1202.2403190254445,
+        "mem allocated avg": 6962667085.824,
+        "mem reserved avg": 29122763948.032,
+        "elapsed time": 6444.514497463999
+      },
+      {
+        "step": 8000,
+        "valid accuracy": 0.6,
+        "train loss": 0.511238426566124,
+        "train samples": 32000,
+        "train time": 172.3884374545887,
+        "eval time": 19.91309441899648,
+        "tokens / sec": 1205.678310384079,
+        "mem allocated avg": 6965754892.288,
+        "mem reserved avg": 29171350765.568,
+        "elapsed time": 6644.613732709986
+      },
+      {
+        "step": 8250,
+        "valid accuracy": 0.58,
+        "train loss": 0.5103235207796096,
+        "train samples": 33000,
+        "train time": 171.33711948918062,
+        "eval time": 30.648292024008697,
+        "tokens / sec": 1216.4497723633158,
+        "mem allocated avg": 6968335892.48,
+        "mem reserved avg": 29565531455.488,
+        "elapsed time": 6854.38911728299
+      },
+      {
+        "step": 8500,
+        "valid accuracy": 0.6,
+        "train loss": 0.5028567316532135,
+        "train samples": 34000,
+        "train time": 171.2165831530874,
+        "eval time": 19.94937202098663,
+        "tokens / sec": 1211.313741815319,
+        "mem allocated avg": 6966661298.176,
+        "mem reserved avg": 29505955561.472,
+        "elapsed time": 7053.429109546007
+      },
+      {
+        "step": 8750,
+        "valid accuracy": 0.62,
+        "train loss": 0.5015089976787567,
+        "train samples": 35000,
+        "train time": 173.58059258732828,
+        "eval time": 20.065320283989422,
+        "tokens / sec": 1226.6175430463609,
+        "mem allocated avg": 6974720499.712,
+        "mem reserved avg": 29524913815.552,
+        "elapsed time": 7254.774949451996
+      },
+      {
+        "step": 9000,
+        "valid accuracy": 0.56,
+        "train loss": 0.49307151556015016,
+        "train samples": 36000,
+        "train time": 172.55944810275105,
+        "eval time": 18.78639112599194,
+        "tokens / sec": 1211.9185724068502,
+        "mem allocated avg": 6969916526.592,
+        "mem reserved avg": 29658443677.696,
+        "elapsed time": 7454.073133013997
+      },
+      {
+        "step": 9250,
+        "valid accuracy": 0.6,
+        "train loss": 0.4912492474317551,
+        "train samples": 37000,
+        "train time": 174.3095906781091,
+        "eval time": 30.712087610998424,
+        "tokens / sec": 1222.3653280987178,
+        "mem allocated avg": 6974721978.368,
+        "mem reserved avg": 29724134866.944,
+        "elapsed time": 7666.895846384985
+      },
+      {
+        "step": 9500,
+        "valid accuracy": 0.64,
+        "train loss": 0.49961484169960024,
+        "train samples": 38000,
+        "train time": 168.3479381708894,
+        "eval time": 21.436372572003165,
+        "tokens / sec": 1213.1363307383538,
+        "mem allocated avg": 6961836355.584,
+        "mem reserved avg": 29566395482.112,
+        "elapsed time": 7864.577572510985
+      },
+      {
+        "step": 9750,
+        "valid accuracy": 0.58,
+        "train loss": 0.4892568769454956,
+        "train samples": 39000,
+        "train time": 176.4860098841018,
+        "eval time": 18.936853881023126,
+        "tokens / sec": 1202.0102904434796,
+        "mem allocated avg": 6973540751.36,
+        "mem reserved avg": 29748671545.344,
+        "elapsed time": 8067.85907531
+      },
+      {
+        "step": 10000,
+        "valid accuracy": 0.6,
+        "train loss": 0.4894174293279648,
+        "train samples": 40000,
+        "train time": 173.01108982303413,
+        "eval time": 30.8168013850227,
+        "tokens / sec": 1192.9293099714462,
+        "mem allocated avg": 6965466226.688,
+        "mem reserved avg": 29391962767.36,
+        "elapsed time": 8279.503795614
+      },
+      {
+        "step": 10000,
+        "test accuracy": 0.5458680818802123,
+        "train loss": 0.4894174293279648,
+        "train samples": 40000,
+        "train total tokens": 8383359,
+        "forgetting": 0.43333709239959717
+      }
+    ]
+  },
+  "meta_info": {
+    "model_info": {
+      "sha": "13afe5124825b4f3751f836b40dafda64c1ed062",
+      "created_at": "2024-09-18T15:23:48+00:00"
+    },
+    "dataset_info": {
+      "metamath": {
+        "sha": "aa4f34d3d2d3231299b5b03d9b3e5a20da45aa18",
+        "created_at": "2023-09-21T17:22:46+00:00"
+      },
+      "gsm8k": {
+        "sha": "740312add88f781978c0658806c59bc2815b9866",
+        "created_at": "2022-04-12T10:22:10+00:00"
+      }
+    },
+    "package_info": {
+      "transformers-version": "5.12.1",
+      "transformers-commit-hash": null,
+      "peft-version": "0.20.1.dev0",
+      "peft-commit-hash": "c702de402330f48c0d8b1f260d59413aa3110b65",
+      "datasets-version": "4.8.4",
+      "datasets-commit-hash": null,
+      "bitsandbytes-version": "0.49.2",
+      "bitsandbytes-commit-hash": null,
+      "torch-version": "2.11.0+cu130",
+      "torch-commit-hash": null
+    },
+    "system_info": {
+      "system": "Linux",
+      "release": "7.0.0-1009-aws",
+      "version": "#9~24.04.1-Ubuntu SMP PREEMPT Mon Jul 13 22:43:41 UTC 2026",
+      "machine": "x86_64",
+      "processor": "x86_64",
+      "accelerator": "NVIDIA L40S"
+    },
+    "pytorch_info": "PyTorch built with:\n  - GCC 13.3\n  - C++ Version: 201703\n  - Intel(R) oneAPI Math Kernel Library Version 2024.2-Product Build 20240605 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v3.10.2 (Git Hash f1d471933dc852f956fd05389f9313c7148783d5)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 13.0\n  - NVCC architecture flags: -gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_120,code=sm_120\n  - CuDNN 90.7.1  (built against CUDA 12.8)\n    - Built with CuDNN 91.9\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, COMMIT_SHA=70d99e998b4955e0049d13a98d77ae1b14db1f45, CUDA_VERSION=13.0, CUDNN_VERSION=9.19.0, CXX_COMPILER=/opt/rh/gcc-toolset-13/root/usr/bin/c++, CXX_FLAGS= -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DLIBKINETO_NOXPUPTI=ON -DUSE_FBGEMM -DUSE_MSLK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -DC10_NODEPRECATED -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=range-loop-construct -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-unknown-pragmas -Wno-unused-parameter -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=old-style-cast -faligned-new -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-dangling-reference -Wno-error=dangling-reference -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, TORCH_VERSION=2.11.0, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, USE_XCCL=OFF, USE_XPU=OFF, \n"
+  }
+}

lokr--llama-3.2-3B-rank32-factor4-all-linear-lr0.003-20k--2026-08-06T01-43-49+00-00.json

@@ -0,0 +1,1087 @@
+{
+  "run_info": {
+    "created_at": "2026-08-06T01:43:49+00:00",
+    "total_time": 17144.428262664005,
+    "experiment_name": "lokr/llama-3.2-3B-rank32-factor4-all-linear-lr0.003-20k",
+    "peft_branch": "auto-research",
+    "train_config": {
+      "model_id": "meta-llama/Llama-3.2-3B",
+      "dtype": "bfloat16",
+      "max_seq_length": 768,
+      "batch_size": 4,
+      "batch_size_eval": 50,
+      "max_steps": 20000,
+      "eval_steps": 250,
+      "compile": false,
+      "use_gc": false,
+      "query_template": "Question: {query} Think step by step.\nAnswer:",
+      "seed": 0,
+      "grad_norm_clip": 1.0,
+      "optimizer_type": "AdamW",
+      "optimizer_kwargs": {
+        "lr": 0.003,
+        "weight_decay": 0.1
+      },
+      "lr_scheduler": "cosine",
+      "use_amp": false,
+      "autocast_adapter_dtype": true,
+      "generation_kwargs": {
+        "max_length": 800,
+        "max_new_tokens": 300
+      },
+      "attn_implementation": null,
+      "init_kv_cache_prefix": null
+    },
+    "peft_config": {
+      "task_type": null,
+      "peft_type": "LOKR",
+      "auto_mapping": null,
+      "peft_version": "0.20.1.dev0@UNKNOWN",
+      "base_model_name_or_path": "meta-llama/Llama-3.2-3B",
+      "revision": null,
+      "inference_mode": false,
+      "rank_pattern": {},
+      "alpha_pattern": {},
+      "r": 32,
+      "alpha": 64,
+      "rank_dropout": 0.0,
+      "module_dropout": 0.0,
+      "use_effective_conv2d": false,
+      "decompose_both": false,
+      "decompose_factor": 4,
+      "rank_dropout_scale": false,
+      "target_modules": [
+        "gate_proj",
+        "v_proj",
+        "up_proj",
+        "q_proj",
+        "o_proj",
+        "down_proj",
+        "k_proj"
+      ],
+      "exclude_modules": null,
+      "init_weights": true,
+      "layers_to_transform": null,
+      "layers_pattern": null,
+      "modules_to_save": null
+    },
+    "error_msg": ""
+  },
+  "train_info": {
+    "accelerator_memory_reserved_avg": 29407510737,
+    "accelerator_memory_max": 40982544384,
+    "accelerator_memory_reserved_99th": 37908119552,
+    "train_time": 15059.725224568945,
+    "file_size": 48714440,
+    "num_trainable_params": 12160064,
+    "num_total_params": 3224909888,
+    "status": "success",
+    "metrics": [
+      {
+        "step": 250,
+        "valid accuracy": 0.42,
+        "train loss": 0.9610863697528839,
+        "train samples": 1000,
+        "train time": 171.79683673282852,
+        "eval time": 24.92885081001441,
+        "tokens / sec": 1232.3800835125783,
+        "mem allocated avg": 6973737287.68,
+        "mem reserved avg": 29583348858.88,
+        "elapsed time": 228.51951690900023
+      },
+      {
+        "step": 500,
+        "valid accuracy": 0.46,
+        "train loss": 0.6889648658037185,
+        "train samples": 2000,
+        "train time": 169.9671751902206,
+        "eval time": 30.64026821500738,
+        "tokens / sec": 1223.7362877109663,
+        "mem allocated avg": 6965520924.672,
+        "mem reserved avg": 29093630312.448,
+        "elapsed time": 436.939744237985
+      },
+      {
+        "step": 750,
+        "valid accuracy": 0.42,
+        "train loss": 0.659138729095459,
+        "train samples": 3000,
+        "train time": 171.3070431847591,
+        "eval time": 18.82749630298349,
+        "tokens / sec": 1251.5597491736692,
+        "mem allocated avg": 6976247709.696,
+        "mem reserved avg": 29367191207.936,
+        "elapsed time": 634.7782167390105
+      },
+      {
+        "step": 1000,
+        "valid accuracy": 0.44,
+        "train loss": 0.6462710534334183,
+        "train samples": 4000,
+        "train time": 170.5948522099934,
+        "eval time": 22.919337497005472,
+        "tokens / sec": 1221.2326298307596,
+        "mem allocated avg": 6967841570.816,
+        "mem reserved avg": 29504093290.496,
+        "elapsed time": 836.158662383008
+      },
+      {
+        "step": 1250,
+        "valid accuracy": 0.4,
+        "train loss": 0.6491906325817108,
+        "train samples": 5000,
+        "train time": 172.19864580288413,
+        "eval time": 22.46991980599705,
+        "tokens / sec": 1211.0315910307074,
+        "mem allocated avg": 6967527682.048,
+        "mem reserved avg": 29526608314.368,
+        "elapsed time": 1038.596893237991
+      },
+      {
+        "step": 1500,
+        "valid accuracy": 0.44,
+        "train loss": 0.6491728991270065,
+        "train samples": 6000,
+        "train time": 172.17588627900113,
+        "eval time": 20.18095650698524,
+        "tokens / sec": 1215.7974297329367,
+        "mem allocated avg": 6969189701.632,
+        "mem reserved avg": 29389219692.544,
+        "elapsed time": 1238.8270268719934
+      },
+      {
+        "step": 1750,
+        "valid accuracy": 0.38,
+        "train loss": 0.654322525024414,
+        "train samples": 7000,
+        "train time": 173.59733828672324,
+        "eval time": 30.70192297300673,
+        "tokens / sec": 1205.980472201811,
+        "mem allocated avg": 6970588690.432,
+        "mem reserved avg": 29948270084.096,
+        "elapsed time": 1450.9282776280015
+      },
+      {
+        "step": 2000,
+        "valid accuracy": 0.34,
+        "train loss": 0.6699976642131805,
+        "train samples": 8000,
+        "train time": 172.16915632793098,
+        "eval time": 23.291683515009936,
+        "tokens / sec": 1206.3484797730027,
+        "mem allocated avg": 6966354518.016,
+        "mem reserved avg": 29498741358.592,
+        "elapsed time": 1654.3528160580026
+      },
+      {
+        "step": 2250,
+        "valid accuracy": 0.38,
+        "train loss": 0.670303931593895,
+        "train samples": 9000,
+        "train time": 171.96220255299704,
+        "eval time": 30.68109203100903,
+        "tokens / sec": 1249.9723590930116,
+        "mem allocated avg": 6978359050.24,
+        "mem reserved avg": 29852145025.024,
+        "elapsed time": 1864.8035511719936
+      },
+      {
+        "step": 2500,
+        "valid accuracy": 0.36,
+        "train loss": 0.6671720509529113,
+        "train samples": 10000,
+        "train time": 169.09803840817767,
+        "eval time": 30.70332172201597,
+        "tokens / sec": 1218.0330531264124,
+        "mem allocated avg": 6963769954.304,
+        "mem reserved avg": 29051737604.096,
+        "elapsed time": 2072.4014120810025
+      },
+      {
+        "step": 2750,
+        "valid accuracy": 0.34,
+        "train loss": 0.6551967885494232,
+        "train samples": 11000,
+        "train time": 176.2429425897717,
+        "eval time": 30.72074737199,
+        "tokens / sec": 1202.2098410668307,
+        "mem allocated avg": 6973763858.432,
+        "mem reserved avg": 29628244688.896,
+        "elapsed time": 2287.0825515759934
+      },
+      {
+        "step": 3000,
+        "valid accuracy": 0.32,
+        "train loss": 0.6460895957946777,
+        "train samples": 12000,
+        "train time": 173.26054636220215,
+        "eval time": 30.63482701999601,
+        "tokens / sec": 1204.7232008817903,
+        "mem allocated avg": 6969272338.432,
+        "mem reserved avg": 29611391975.424,
+        "elapsed time": 2498.9181774450117
+      },
+      {
+        "step": 3250,
+        "valid accuracy": 0.44,
+        "train loss": 0.6532680690288544,
+        "train samples": 13000,
+        "train time": 172.25852237379877,
+        "eval time": 30.719604269019328,
+        "tokens / sec": 1224.3283937055232,
+        "mem allocated avg": 6971559395.328,
+        "mem reserved avg": 29344307085.312,
+        "elapsed time": 2709.468509438011
+      },
+      {
+        "step": 3500,
+        "valid accuracy": 0.48,
+        "train loss": 0.6353620156049729,
+        "train samples": 14000,
+        "train time": 172.01096387507278,
+        "eval time": 30.71934367599897,
+        "tokens / sec": 1219.3990154740145,
+        "mem allocated avg": 6969082503.168,
+        "mem reserved avg": 29335717150.72,
+        "elapsed time": 2920.007108977996
+      },
+      {
+        "step": 3750,
+        "valid accuracy": 0.42,
+        "train loss": 0.630974931716919,
+        "train samples": 15000,
+        "train time": 178.07678941314225,
+        "eval time": 17.86254966899287,
+        "tokens / sec": 1216.9076088700367,
+        "mem allocated avg": 6980761587.712,
+        "mem reserved avg": 29835023876.096,
+        "elapsed time": 3123.7434894150065
+      },
+      {
+        "step": 4000,
+        "valid accuracy": 0.36,
+        "train loss": 0.6445099025964737,
+        "train samples": 16000,
+        "train time": 169.99613859973033,
+        "eval time": 30.74968663099571,
+        "tokens / sec": 1202.2214250478523,
+        "mem allocated avg": 6961925545.984,
+        "mem reserved avg": 29382282313.728,
+        "elapsed time": 3332.329720507987
+      },
+      {
+        "step": 4250,
+        "valid accuracy": 0.42,
+        "train loss": 0.6274779469966888,
+        "train samples": 17000,
+        "train time": 171.6437004666368,
+        "eval time": 30.72802247700747,
+        "tokens / sec": 1231.5569952483556,
+        "mem allocated avg": 6971945697.28,
+        "mem reserved avg": 29341111025.664,
+        "elapsed time": 3542.2687234650075
+      },
+      {
+        "step": 4500,
+        "valid accuracy": 0.46,
+        "train loss": 0.6312829607725143,
+        "train samples": 18000,
+        "train time": 170.1949475101137,
+        "eval time": 30.68255225199391,
+        "tokens / sec": 1221.0585745364185,
+        "mem allocated avg": 6967646959.616,
+        "mem reserved avg": 29272357994.496,
+        "elapsed time": 3751.0097267999954
+      },
+      {
+        "step": 4750,
+        "valid accuracy": 0.4,
+        "train loss": 0.6233677245378494,
+        "train samples": 19000,
+        "train time": 172.96381472694338,
+        "eval time": 30.79086076997919,
+        "tokens / sec": 1213.7741083672852,
+        "mem allocated avg": 6970038394.88,
+        "mem reserved avg": 29374774509.568,
+        "elapsed time": 3962.4423660789907
+      },
+      {
+        "step": 5000,
+        "valid accuracy": 0.38,
+        "train loss": 0.6259740364551544,
+        "train samples": 20000,
+        "train time": 171.76812939089723,
+        "eval time": 23.789134004997322,
+        "tokens / sec": 1212.564872998132,
+        "mem allocated avg": 6967149254.656,
+        "mem reserved avg": 28875283234.816,
+        "elapsed time": 4165.6995755399985
+      },
+      {
+        "step": 5250,
+        "valid accuracy": 0.54,
+        "train loss": 0.6133968497514725,
+        "train samples": 21000,
+        "train time": 170.67316381281125,
+        "eval time": 30.68046033201972,
+        "tokens / sec": 1241.8941283145639,
+        "mem allocated avg": 6974006718.464,
+        "mem reserved avg": 29507507453.952,
+        "elapsed time": 4374.723204473994
+      },
+      {
+        "step": 5500,
+        "valid accuracy": 0.46,
+        "train loss": 0.6253580703735352,
+        "train samples": 22000,
+        "train time": 167.74530908471206,
+        "eval time": 18.557162634009728,
+        "tokens / sec": 1220.576605791119,
+        "mem allocated avg": 6961618006.016,
+        "mem reserved avg": 29010499207.168,
+        "elapsed time": 4568.675552848988
+      },
+      {
+        "step": 5750,
+        "valid accuracy": 0.4,
+        "train loss": 0.6091819525957107,
+        "train samples": 23000,
+        "train time": 173.64858763973461,
+        "eval time": 30.684087104018545,
+        "tokens / sec": 1238.9101628994326,
+        "mem allocated avg": 6977952616.448,
+        "mem reserved avg": 29454810218.496,
+        "elapsed time": 4780.747373077
+      },
+      {
+        "step": 6000,
+        "valid accuracy": 0.48,
+        "train loss": 0.6196607483625411,
+        "train samples": 24000,
+        "train time": 174.06167243191157,
+        "eval time": 30.775191562977852,
+        "tokens / sec": 1201.5511345946175,
+        "mem allocated avg": 6968932239.36,
+        "mem reserved avg": 29413848645.632,
+        "elapsed time": 4993.417520110001
+      },
+      {
+        "step": 6250,
+        "valid accuracy": 0.44,
+        "train loss": 0.61952920627594,
+        "train samples": 25000,
+        "train time": 171.89116472413298,
+        "eval time": 30.72593003101065,
+        "tokens / sec": 1208.601968189659,
+        "mem allocated avg": 6967978575.872,
+        "mem reserved avg": 29655549607.936,
+        "elapsed time": 5203.786160051997
+      },
+      {
+        "step": 6500,
+        "valid accuracy": 0.46,
+        "train loss": 0.6097409170866013,
+        "train samples": 26000,
+        "train time": 170.64674598391866,
+        "eval time": 30.631961806007894,
+        "tokens / sec": 1220.380727445136,
+        "mem allocated avg": 6967718619.136,
+        "mem reserved avg": 29146503708.672,
+        "elapsed time": 5412.883332634985
+      },
+      {
+        "step": 6750,
+        "valid accuracy": 0.52,
+        "train loss": 0.6144000781774521,
+        "train samples": 27000,
+        "train time": 169.8698111831327,
+        "eval time": 30.71524023701204,
+        "tokens / sec": 1220.5464794240456,
+        "mem allocated avg": 6966486200.32,
+        "mem reserved avg": 29525350023.168,
+        "elapsed time": 5621.148808142985
+      },
+      {
+        "step": 7000,
+        "valid accuracy": 0.46,
+        "train loss": 0.6016047027111053,
+        "train samples": 28000,
+        "train time": 168.5002364088141,
+        "eval time": 20.469277928990778,
+        "tokens / sec": 1225.9745410612027,
+        "mem allocated avg": 6963706146.816,
+        "mem reserved avg": 29035295932.416,
+        "elapsed time": 5817.903433194006
+      },
+      {
+        "step": 7250,
+        "valid accuracy": 0.44,
+        "train loss": 0.5895852934122086,
+        "train samples": 29000,
+        "train time": 174.2408317917434,
+        "eval time": 20.638584872009233,
+        "tokens / sec": 1252.0084859370909,
+        "mem allocated avg": 6982119364.608,
+        "mem reserved avg": 30067371540.48,
+        "elapsed time": 6020.640226660995
+      },
+      {
+        "step": 7500,
+        "valid accuracy": 0.36,
+        "train loss": 0.6044657553434372,
+        "train samples": 30000,
+        "train time": 169.2932935403078,
+        "eval time": 28.75819454298471,
+        "tokens / sec": 1236.7294393156228,
+        "mem allocated avg": 6967078899.712,
+        "mem reserved avg": 28892169502.72,
+        "elapsed time": 6226.409051725001
+      },
+      {
+        "step": 7750,
+        "valid accuracy": 0.42,
+        "train loss": 0.5994244650602341,
+        "train samples": 31000,
+        "train time": 170.44431554916082,
+        "eval time": 30.204051584005356,
+        "tokens / sec": 1204.815774221407,
+        "mem allocated avg": 6962793979.904,
+        "mem reserved avg": 29118066327.552,
+        "elapsed time": 6434.709696039994
+      },
+      {
+        "step": 8000,
+        "valid accuracy": 0.44,
+        "train loss": 0.5919614787101746,
+        "train samples": 32000,
+        "train time": 170.99229166895384,
+        "eval time": 27.106889552989742,
+        "tokens / sec": 1215.5226295369741,
+        "mem allocated avg": 6965711171.584,
+        "mem reserved avg": 29176660754.432,
+        "elapsed time": 6640.569886065001
+      },
+      {
+        "step": 8250,
+        "valid accuracy": 0.52,
+        "train loss": 0.5939536187648773,
+        "train samples": 33000,
+        "train time": 172.01891086486285,
+        "eval time": 24.93397859900142,
+        "tokens / sec": 1211.628413132647,
+        "mem allocated avg": 6968187305.984,
+        "mem reserved avg": 29569859977.216,
+        "elapsed time": 6845.286798370013
+      },
+      {
+        "step": 8500,
+        "valid accuracy": 0.54,
+        "train loss": 0.5879585744142533,
+        "train samples": 34000,
+        "train time": 168.8742023879604,
+        "eval time": 30.59148289699806,
+        "tokens / sec": 1228.115348983499,
+        "mem allocated avg": 6966380304.384,
+        "mem reserved avg": 29469481893.888,
+        "elapsed time": 7052.57047268399
+      },
+      {
+        "step": 8750,
+        "valid accuracy": 0.44,
+        "train loss": 0.5851555465459823,
+        "train samples": 35000,
+        "train time": 174.13013303573825,
+        "eval time": 30.731486786011374,
+        "tokens / sec": 1222.7464384713999,
+        "mem allocated avg": 6975052922.88,
+        "mem reserved avg": 29534225170.432,
+        "elapsed time": 7265.109816590004
+      },
+      {
+        "step": 9000,
+        "valid accuracy": 0.42,
+        "train loss": 0.5804140285253525,
+        "train samples": 36000,
+        "train time": 172.553419456759,
+        "eval time": 30.76102186201024,
+        "tokens / sec": 1211.9609142396996,
+        "mem allocated avg": 6969977663.488,
+        "mem reserved avg": 29691041808.384,
+        "elapsed time": 7476.346765204013
+      },
+      {
+        "step": 9250,
+        "valid accuracy": 0.54,
+        "train loss": 0.5720908064842224,
+        "train samples": 37000,
+        "train time": 172.8666125019081,
+        "eval time": 21.360682764003286,
+        "tokens / sec": 1232.568839732705,
+        "mem allocated avg": 6974648111.104,
+        "mem reserved avg": 29688013520.896,
+        "elapsed time": 7678.343583943002
+      },
+      {
+        "step": 9500,
+        "valid accuracy": 0.48,
+        "train loss": 0.5810169658660889,
+        "train samples": 38000,
+        "train time": 168.65665205937694,
+        "eval time": 28.963550884014694,
+        "tokens / sec": 1210.9157718137292,
+        "mem allocated avg": 6962060840.96,
+        "mem reserved avg": 29558887677.952,
+        "elapsed time": 7883.88087289501
+      },
+      {
+        "step": 9750,
+        "valid accuracy": 0.44,
+        "train loss": 0.5680638591051101,
+        "train samples": 39000,
+        "train time": 173.65742951087304,
+        "eval time": 30.74409557101899,
+        "tokens / sec": 1221.5889674142481,
+        "mem allocated avg": 6973438425.088,
+        "mem reserved avg": 29760432373.76,
+        "elapsed time": 8096.062616196985
+      },
+      {
+        "step": 10000,
+        "valid accuracy": 0.48,
+        "train loss": 0.5634032604694367,
+        "train samples": 40000,
+        "train time": 169.49355823599035,
+        "eval time": 30.60387931301375,
+        "tokens / sec": 1217.68639556577,
+        "mem allocated avg": 6965239169.024,
+        "mem reserved avg": 29368080400.384,
+        "elapsed time": 8303.941635889001
+      },
+      {
+        "step": 10250,
+        "valid accuracy": 0.52,
+        "train loss": 0.5606767899990082,
+        "train samples": 41000,
+        "train time": 168.76627820308204,
+        "eval time": 25.141207000997383,
+        "tokens / sec": 1219.3431187264396,
+        "mem allocated avg": 6963630600.192,
+        "mem reserved avg": 29278255185.92,
+        "elapsed time": 8505.426422146993
+      },
+      {
+        "step": 10500,
+        "valid accuracy": 0.46,
+        "train loss": 0.5653843686580659,
+        "train samples": 42000,
+        "train time": 171.55485373476404,
+        "eval time": 23.03960806300165,
+        "tokens / sec": 1214.8825606661678,
+        "mem allocated avg": 6967405594.624,
+        "mem reserved avg": 29249675198.464,
+        "elapsed time": 8707.701591034012
+      },
+      {
+        "step": 10750,
+        "valid accuracy": 0.52,
+        "train loss": 0.5517334004640579,
+        "train samples": 43000,
+        "train time": 171.51006954509648,
+        "eval time": 30.613034264999442,
+        "tokens / sec": 1232.796421579231,
+        "mem allocated avg": 6973592811.52,
+        "mem reserved avg": 29845467693.056,
+        "elapsed time": 8917.593452545
+      },
+      {
+        "step": 11000,
+        "valid accuracy": 0.46,
+        "train loss": 0.558027804851532,
+        "train samples": 44000,
+        "train time": 168.21222303493414,
+        "eval time": 30.721659283008194,
+        "tokens / sec": 1224.2213810897254,
+        "mem allocated avg": 6964628082.688,
+        "mem reserved avg": 29235624280.064,
+        "elapsed time": 9124.31383405099
+      },
+      {
+        "step": 11250,
+        "valid accuracy": 0.46,
+        "train loss": 0.557011668086052,
+        "train samples": 45000,
+        "train time": 169.0620566032012,
+        "eval time": 26.104685962985968,
+        "tokens / sec": 1240.473493660487,
+        "mem allocated avg": 6968411570.176,
+        "mem reserved avg": 29232973479.936,
+        "elapsed time": 9327.120027102006
+      },
+      {
+        "step": 11500,
+        "valid accuracy": 0.4,
+        "train loss": 0.5513826822042465,
+        "train samples": 46000,
+        "train time": 171.40321559412405,
+        "eval time": 19.5690330209909,
+        "tokens / sec": 1213.7403564973251,
+        "mem allocated avg": 6965893273.6,
+        "mem reserved avg": 28864562593.792,
+        "elapsed time": 9525.791539052996
+      },
+      {
+        "step": 11750,
+        "valid accuracy": 0.48,
+        "train loss": 0.5313005843162537,
+        "train samples": 47000,
+        "train time": 175.4224885792646,
+        "eval time": 30.735795768006938,
+        "tokens / sec": 1219.0284252145595,
+        "mem allocated avg": 6974539284.48,
+        "mem reserved avg": 29290510942.208,
+        "elapsed time": 9739.656438242004
+      },
+      {
+        "step": 12000,
+        "valid accuracy": 0.52,
+        "train loss": 0.5336237369775773,
+        "train samples": 48000,
+        "train time": 167.62555019688443,
+        "eval time": 30.5678302939923,
+        "tokens / sec": 1216.6045078478162,
+        "mem allocated avg": 6961333362.688,
+        "mem reserved avg": 29227898372.096,
+        "elapsed time": 9945.537856065988
+      },
+      {
+        "step": 12250,
+        "valid accuracy": 0.56,
+        "train loss": 0.5234718472957611,
+        "train samples": 49000,
+        "train time": 169.3317394120968,
+        "eval time": 30.705751812987728,
+        "tokens / sec": 1252.5236009289376,
+        "mem allocated avg": 6973541867.52,
+        "mem reserved avg": 29555297353.728,
+        "elapsed time": 10153.286983998987
+      },
+      {
+        "step": 12500,
+        "valid accuracy": 0.52,
+        "train loss": 0.524735977768898,
+        "train samples": 50000,
+        "train time": 168.84534780608374,
+        "eval time": 30.6525500860007,
+        "tokens / sec": 1204.1492563567942,
+        "mem allocated avg": 6960001439.744,
+        "mem reserved avg": 28973018906.624,
+        "elapsed time": 10360.513763035997
+      },
+      {
+        "step": 12750,
+        "valid accuracy": 0.52,
+        "train loss": 0.5225069969892502,
+        "train samples": 51000,
+        "train time": 173.843651830015,
+        "eval time": 24.634994040010497,
+        "tokens / sec": 1244.2732174742152,
+        "mem allocated avg": 6980235247.616,
+        "mem reserved avg": 29903474917.376,
+        "elapsed time": 10566.833168787009
+      },
+      {
+        "step": 13000,
+        "valid accuracy": 0.48,
+        "train loss": 0.5218035694360733,
+        "train samples": 52000,
+        "train time": 168.3541679691116,
+        "eval time": 25.71232538498589,
+        "tokens / sec": 1211.9034679167182,
+        "mem allocated avg": 6960411924.48,
+        "mem reserved avg": 28957684531.2,
+        "elapsed time": 10768.591327572009
+      },
+      {
+        "step": 13250,
+        "valid accuracy": 0.66,
+        "train loss": 0.5213898307085038,
+        "train samples": 53000,
+        "train time": 174.92087974085007,
+        "eval time": 19.067908728000475,
+        "tokens / sec": 1222.3126268102599,
+        "mem allocated avg": 6975828649.984,
+        "mem reserved avg": 29767403307.008,
+        "elapsed time": 10970.29764865301
+      },
+      {
+        "step": 13500,
+        "valid accuracy": 0.58,
+        "train loss": 0.5148020983934403,
+        "train samples": 54000,
+        "train time": 169.88899307019892,
+        "eval time": 19.94494408200262,
+        "tokens / sec": 1213.1745340019554,
+        "mem allocated avg": 6965132003.328,
+        "mem reserved avg": 29332042940.416,
+        "elapsed time": 11167.967929117993
+      },
+      {
+        "step": 13750,
+        "valid accuracy": 0.56,
+        "train loss": 0.505045391201973,
+        "train samples": 55000,
+        "train time": 176.1165522079682,
+        "eval time": 22.387796136987163,
+        "tokens / sec": 1201.05689867366,
+        "mem allocated avg": 6974318190.592,
+        "mem reserved avg": 29852153413.632,
+        "elapsed time": 11374.266127874987
+      },
+      {
+        "step": 14000,
+        "valid accuracy": 0.54,
+        "train loss": 0.49443184566497805,
+        "train samples": 56000,
+        "train time": 171.27784125340986,
+        "eval time": 20.43928565402166,
+        "tokens / sec": 1203.5182046404743,
+        "mem allocated avg": 6962779381.76,
+        "mem reserved avg": 28876046598.144,
+        "elapsed time": 11573.688829808001
+      },
+      {
+        "step": 14250,
+        "valid accuracy": 0.62,
+        "train loss": 0.5036366237401962,
+        "train samples": 57000,
+        "train time": 172.08854516083375,
+        "eval time": 21.94274735599174,
+        "tokens / sec": 1230.1690377017674,
+        "mem allocated avg": 6971923554.304,
+        "mem reserved avg": 29041285398.528,
+        "elapsed time": 11775.33258013401
+      },
+      {
+        "step": 14500,
+        "valid accuracy": 0.6,
+        "train loss": 0.4969416060447693,
+        "train samples": 58000,
+        "train time": 167.78399256194825,
+        "eval time": 23.040681084996322,
+        "tokens / sec": 1215.6702012242995,
+        "mem allocated avg": 6961757241.344,
+        "mem reserved avg": 29157811552.256,
+        "elapsed time": 11973.882213180012
+      },
+      {
+        "step": 14750,
+        "valid accuracy": 0.6,
+        "train loss": 0.4927418532371521,
+        "train samples": 59000,
+        "train time": 168.31322940738755,
+        "eval time": 22.590195139986463,
+        "tokens / sec": 1251.4346064276451,
+        "mem allocated avg": 6970914009.088,
+        "mem reserved avg": 29198865399.808,
+        "elapsed time": 12172.35395660199
+      },
+      {
+        "step": 15000,
+        "valid accuracy": 0.62,
+        "train loss": 0.4801807880997658,
+        "train samples": 60000,
+        "train time": 170.72411192808067,
+        "eval time": 20.901687847013818,
+        "tokens / sec": 1214.679037764501,
+        "mem allocated avg": 6965701888.0,
+        "mem reserved avg": 29037988675.584,
+        "elapsed time": 12371.769835898012
+      },
+      {
+        "step": 15250,
+        "valid accuracy": 0.56,
+        "train loss": 0.4763737685680389,
+        "train samples": 61000,
+        "train time": 174.21619906811975,
+        "eval time": 23.08776327999658,
+        "tokens / sec": 1220.982900199923,
+        "mem allocated avg": 6973737148.416,
+        "mem reserved avg": 29813205106.688,
+        "elapsed time": 12576.942732289986
+      },
+      {
+        "step": 15500,
+        "valid accuracy": 0.6,
+        "train loss": 0.47083090579509734,
+        "train samples": 62000,
+        "train time": 168.79248122885474,
+        "eval time": 21.987459181022132,
+        "tokens / sec": 1244.17271712041,
+        "mem allocated avg": 6969051408.384,
+        "mem reserved avg": 29202061459.456,
+        "elapsed time": 12775.533608391008
+      },
+      {
+        "step": 15750,
+        "valid accuracy": 0.6,
+        "train loss": 0.4586274927854538,
+        "train samples": 63000,
+        "train time": 170.14473092296976,
+        "eval time": 26.397917327994946,
+        "tokens / sec": 1224.9336131035227,
+        "mem allocated avg": 6967867320.32,
+        "mem reserved avg": 29691922612.224,
+        "elapsed time": 12979.855119065003
+      },
+      {
+        "step": 16000,
+        "valid accuracy": 0.62,
+        "train loss": 0.46891966664791107,
+        "train samples": 64000,
+        "train time": 172.07654142184765,
+        "eval time": 27.683391362981638,
+        "tokens / sec": 1219.1609516703372,
+        "mem allocated avg": 6969539037.184,
+        "mem reserved avg": 29515157864.448,
+        "elapsed time": 13187.581383618002
+      },
+      {
+        "step": 16250,
+        "valid accuracy": 0.72,
+        "train loss": 0.45970044922828673,
+        "train samples": 65000,
+        "train time": 171.93470184094622,
+        "eval time": 22.664465795998694,
+        "tokens / sec": 1210.9306485005168,
+        "mem allocated avg": 6967731288.064,
+        "mem reserved avg": 29620225179.648,
+        "elapsed time": 13389.917077972
+      },
+      {
+        "step": 16500,
+        "valid accuracy": 0.64,
+        "train loss": 0.4485881251692772,
+        "train samples": 66000,
+        "train time": 170.40854759063222,
+        "eval time": 21.17048738899757,
+        "tokens / sec": 1207.844341790017,
+        "mem allocated avg": 6963923716.096,
+        "mem reserved avg": 29116782870.528,
+        "elapsed time": 13589.23178718699
+      },
+      {
+        "step": 16750,
+        "valid accuracy": 0.68,
+        "train loss": 0.4502259805202484,
+        "train samples": 67000,
+        "train time": 169.258822891803,
+        "eval time": 25.066891107999254,
+        "tokens / sec": 1227.8414587158554,
+        "mem allocated avg": 6967624613.888,
+        "mem reserved avg": 29815470030.848,
+        "elapsed time": 13791.351112853998
+      },
+      {
+        "step": 17000,
+        "valid accuracy": 0.6,
+        "train loss": 0.43991368901729583,
+        "train samples": 68000,
+        "train time": 170.40273850920494,
+        "eval time": 22.699942364997696,
+        "tokens / sec": 1226.2068193740492,
+        "mem allocated avg": 6967878397.952,
+        "mem reserved avg": 29168062431.232,
+        "elapsed time": 13992.276433043007
+      },
+      {
+        "step": 17250,
+        "valid accuracy": 0.62,
+        "train loss": 0.4485655171871185,
+        "train samples": 69000,
+        "train time": 175.55842464990565,
+        "eval time": 29.480302722979104,
+        "tokens / sec": 1236.52852566262,
+        "mem allocated avg": 6980605931.52,
+        "mem reserved avg": 29673761275.904,
+        "elapsed time": 14205.057181976998
+      },
+      {
+        "step": 17500,
+        "valid accuracy": 0.68,
+        "train loss": 0.4372402468919754,
+        "train samples": 70000,
+        "train time": 168.44072627069545,
+        "eval time": 21.076059184997575,
+        "tokens / sec": 1224.8047403242072,
+        "mem allocated avg": 6964315848.704,
+        "mem reserved avg": 29363256950.784,
+        "elapsed time": 14402.422432071995
+      },
+      {
+        "step": 17750,
+        "valid accuracy": 0.66,
+        "train loss": 0.43826009291410445,
+        "train samples": 71000,
+        "train time": 172.937114319182,
+        "eval time": 29.079054092988372,
+        "tokens / sec": 1215.8350208846862,
+        "mem allocated avg": 6970201442.304,
+        "mem reserved avg": 29584221274.112,
+        "elapsed time": 14612.144416022988
+      },
+      {
+        "step": 18000,
+        "valid accuracy": 0.68,
+        "train loss": 0.42263948208093643,
+        "train samples": 72000,
+        "train time": 167.03616468189284,
+        "eval time": 20.53648081200663,
+        "tokens / sec": 1231.5057663814946,
+        "mem allocated avg": 6963220590.592,
+        "mem reserved avg": 29116355051.52,
+        "elapsed time": 14807.464960957994
+      },
+      {
+        "step": 18250,
+        "valid accuracy": 0.64,
+        "train loss": 0.4186551516652107,
+        "train samples": 73000,
+        "train time": 173.9695659817953,
+        "eval time": 20.841062296007294,
+        "tokens / sec": 1223.7945114048212,
+        "mem allocated avg": 6973775902.72,
+        "mem reserved avg": 29595227127.808,
+        "elapsed time": 15009.944732523
+      },
+      {
+        "step": 18500,
+        "valid accuracy": 0.68,
+        "train loss": 0.4335354580283165,
+        "train samples": 74000,
+        "train time": 169.157907269313,
+        "eval time": 21.018800074001774,
+        "tokens / sec": 1222.8751427544196,
+        "mem allocated avg": 6963947921.408,
+        "mem reserved avg": 28941125419.008,
+        "elapsed time": 15207.840157419007
+      },
+      {
+        "step": 18750,
+        "valid accuracy": 0.72,
+        "train loss": 0.434735699236393,
+        "train samples": 75000,
+        "train time": 173.53714782529278,
+        "eval time": 19.690483615995618,
+        "tokens / sec": 1235.187985317105,
+        "mem allocated avg": 6977327067.136,
+        "mem reserved avg": 29618413240.32,
+        "elapsed time": 15408.781935703999
+      },
+      {
+        "step": 19000,
+        "valid accuracy": 0.64,
+        "train loss": 0.42556005668640134,
+        "train samples": 76000,
+        "train time": 169.8566626807442,
+        "eval time": 27.14048506799736,
+        "tokens / sec": 1222.4012689467395,
+        "mem allocated avg": 6965633824.768,
+        "mem reserved avg": 29325524992.0,
+        "elapsed time": 15613.64058382099
+      },
+      {
+        "step": 19250,
+        "valid accuracy": 0.72,
+        "train loss": 0.4332517256140709,
+        "train samples": 77000,
+        "train time": 172.7995636741107,
+        "eval time": 30.646320986998035,
+        "tokens / sec": 1233.5100590993839,
+        "mem allocated avg": 6974624098.304,
+        "mem reserved avg": 29485294419.968,
+        "elapsed time": 15824.803288046009
+      },
+      {
+        "step": 19500,
+        "valid accuracy": 0.64,
+        "train loss": 0.43008509677648543,
+        "train samples": 78000,
+        "train time": 169.32743887693505,
+        "eval time": 25.64423408100265,
+        "tokens / sec": 1196.5042484771045,
+        "mem allocated avg": 6960381396.992,
+        "mem reserved avg": 29205274296.32,
+        "elapsed time": 16027.579781459994
+      },
+      {
+        "step": 19750,
+        "valid accuracy": 0.7,
+        "train loss": 0.41904746931791303,
+        "train samples": 79000,
+        "train time": 174.864879222936,
+        "eval time": 24.811317635001615,
+        "tokens / sec": 1228.2626503071326,
+        "mem allocated avg": 6979446730.752,
+        "mem reserved avg": 30267087519.744,
+        "elapsed time": 16235.117151182989
+      },
+      {
+        "step": 20000,
+        "valid accuracy": 0.58,
+        "train loss": 0.4246500636935234,
+        "train samples": 80000,
+        "train time": 166.44995271685184,
+        "eval time": 20.16140359599376,
+        "tokens / sec": 1220.943573025268,
+        "mem allocated avg": 6959579637.76,
+        "mem reserved avg": 29145253806.08,
+        "elapsed time": 16429.555045605986
+      },
+      {
+        "step": 20000,
+        "test accuracy": 0.5913570887035633,
+        "train loss": 0.4246500636935234,
+        "train samples": 80000,
+        "train total tokens": 16744048,
+        "forgetting": 0.40216314792633057
+      }
+    ]
+  },
+  "meta_info": {
+    "model_info": {
+      "sha": "13afe5124825b4f3751f836b40dafda64c1ed062",
+      "created_at": "2024-09-18T15:23:48+00:00"
+    },
+    "dataset_info": {
+      "metamath": {
+        "sha": "aa4f34d3d2d3231299b5b03d9b3e5a20da45aa18",
+        "created_at": "2023-09-21T17:22:46+00:00"
+      },
+      "gsm8k": {
+        "sha": "740312add88f781978c0658806c59bc2815b9866",
+        "created_at": "2022-04-12T10:22:10+00:00"
+      }
+    },
+    "package_info": {
+      "transformers-version": "5.12.1",
+      "transformers-commit-hash": null,
+      "peft-version": "0.20.1.dev0",
+      "peft-commit-hash": "c702de402330f48c0d8b1f260d59413aa3110b65",
+      "datasets-version": "4.8.4",
+      "datasets-commit-hash": null,
+      "bitsandbytes-version": "0.49.2",
+      "bitsandbytes-commit-hash": null,
+      "torch-version": "2.11.0+cu130",
+      "torch-commit-hash": null
+    },
+    "system_info": {
+      "system": "Linux",
+      "release": "7.0.0-1009-aws",
+      "version": "#9~24.04.1-Ubuntu SMP PREEMPT Mon Jul 13 22:43:41 UTC 2026",
+      "machine": "x86_64",
+      "processor": "x86_64",
+      "accelerator": "NVIDIA L40S"
+    },
+    "pytorch_info": "PyTorch built with:\n  - GCC 13.3\n  - C++ Version: 201703\n  - Intel(R) oneAPI Math Kernel Library Version 2024.2-Product Build 20240605 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v3.10.2 (Git Hash f1d471933dc852f956fd05389f9313c7148783d5)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 13.0\n  - NVCC architecture flags: -gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_120,code=sm_120\n  - CuDNN 90.7.1  (built against CUDA 12.8)\n    - Built with CuDNN 91.9\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, COMMIT_SHA=70d99e998b4955e0049d13a98d77ae1b14db1f45, CUDA_VERSION=13.0, CUDNN_VERSION=9.19.0, CXX_COMPILER=/opt/rh/gcc-toolset-13/root/usr/bin/c++, CXX_FLAGS= -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DLIBKINETO_NOXPUPTI=ON -DUSE_FBGEMM -DUSE_MSLK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -DC10_NODEPRECATED -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=range-loop-construct -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-unknown-pragmas -Wno-unused-parameter -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=old-style-cast -faligned-new -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-dangling-reference -Wno-error=dangling-reference -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, TORCH_VERSION=2.11.0, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, USE_XCCL=OFF, USE_XPU=OFF, \n"
+  }
+}

lokr--llama-3.2-3B-rank32-factor4-all-linear-lr0.005--2026-08-05T19-21-57+00-00.json

@@ -0,0 +1,367 @@
+{
+  "run_info": {
+    "created_at": "2026-08-05T19:21:57+00:00",
+    "total_time": 4876.834152824013,
+    "experiment_name": "lokr/llama-3.2-3B-rank32-factor4-all-linear-lr0.005",
+    "peft_branch": "auto-research",
+    "train_config": {
+      "model_id": "meta-llama/Llama-3.2-3B",
+      "dtype": "bfloat16",
+      "max_seq_length": 768,
+      "batch_size": 4,
+      "batch_size_eval": 50,
+      "max_steps": 5000,
+      "eval_steps": 250,
+      "compile": false,
+      "use_gc": false,
+      "query_template": "Question: {query} Think step by step.\nAnswer:",
+      "seed": 0,
+      "grad_norm_clip": 1.0,
+      "optimizer_type": "AdamW",
+      "optimizer_kwargs": {
+        "lr": 0.005,
+        "weight_decay": 0.1
+      },
+      "lr_scheduler": "cosine",
+      "use_amp": false,
+      "autocast_adapter_dtype": true,
+      "generation_kwargs": {
+        "max_length": 800,
+        "max_new_tokens": 300
+      },
+      "attn_implementation": null,
+      "init_kv_cache_prefix": null
+    },
+    "peft_config": {
+      "task_type": null,
+      "peft_type": "LOKR",
+      "auto_mapping": null,
+      "peft_version": "0.20.1.dev0@UNKNOWN",
+      "base_model_name_or_path": "meta-llama/Llama-3.2-3B",
+      "revision": null,
+      "inference_mode": false,
+      "rank_pattern": {},
+      "alpha_pattern": {},
+      "r": 32,
+      "alpha": 64,
+      "rank_dropout": 0.0,
+      "module_dropout": 0.0,
+      "use_effective_conv2d": false,
+      "decompose_both": false,
+      "decompose_factor": 4,
+      "rank_dropout_scale": false,
+      "target_modules": [
+        "up_proj",
+        "gate_proj",
+        "v_proj",
+        "q_proj",
+        "k_proj",
+        "o_proj",
+        "down_proj"
+      ],
+      "exclude_modules": null,
+      "init_weights": true,
+      "layers_to_transform": null,
+      "layers_pattern": null,
+      "modules_to_save": null
+    },
+    "error_msg": ""
+  },
+  "train_info": {
+    "accelerator_memory_reserved_avg": 29428636922,
+    "accelerator_memory_max": 40980447232,
+    "accelerator_memory_reserved_99th": 37629198336,
+    "train_time": 4337.368219144992,
+    "file_size": 48714440,
+    "num_trainable_params": 12160064,
+    "num_total_params": 3224909888,
+    "status": "success",
+    "metrics": [
+      {
+        "step": 250,
+        "valid accuracy": 0.4,
+        "train loss": 0.8243188563585281,
+        "train samples": 1000,
+        "train time": 173.05660409928532,
+        "eval time": 30.925348686025245,
+        "tokens / sec": 1223.4089597559273,
+        "mem allocated avg": 6974068426.752,
+        "mem reserved avg": 29637765758.976,
+        "elapsed time": 234.67427716800012
+      },
+      {
+        "step": 500,
+        "valid accuracy": 0.32,
+        "train loss": 0.7367534626722336,
+        "train samples": 2000,
+        "train time": 168.92729432377382,
+        "eval time": 19.947831519006286,
+        "tokens / sec": 1231.2693507145577,
+        "mem allocated avg": 6965838184.448,
+        "mem reserved avg": 29036730384.384,
+        "elapsed time": 431.30717777099926
+      },
+      {
+        "step": 750,
+        "valid accuracy": 0.22,
+        "train loss": 0.7591401909589768,
+        "train samples": 3000,
+        "train time": 173.5782248750329,
+        "eval time": 30.72412709100172,
+        "tokens / sec": 1235.1837343327904,
+        "mem allocated avg": 6976710492.16,
+        "mem reserved avg": 29298144575.488,
+        "elapsed time": 643.3134106380166
+      },
+      {
+        "step": 1000,
+        "valid accuracy": 0.16,
+        "train loss": 0.7286917352676392,
+        "train samples": 4000,
+        "train time": 173.25569501990685,
+        "eval time": 15.424150008999277,
+        "tokens / sec": 1202.4770670658904,
+        "mem allocated avg": 6968188157.952,
+        "mem reserved avg": 29486888255.488,
+        "elapsed time": 839.8342328320141
+      },
+      {
+        "step": 1250,
+        "valid accuracy": 0.24,
+        "train loss": 0.7228857347965241,
+        "train samples": 5000,
+        "train time": 171.9056585111539,
+        "eval time": 30.715734964003786,
+        "tokens / sec": 1213.0956118961567,
+        "mem allocated avg": 6967430975.488,
+        "mem reserved avg": 29512901328.896,
+        "elapsed time": 1050.2152808600222
+      },
+      {
+        "step": 1500,
+        "valid accuracy": 0.28,
+        "train loss": 0.7035419543981553,
+        "train samples": 6000,
+        "train time": 171.0979807379772,
+        "eval time": 30.68885849401704,
+        "tokens / sec": 1223.4568701343917,
+        "mem allocated avg": 6969443014.656,
+        "mem reserved avg": 29386233348.096,
+        "elapsed time": 1259.8666626070044
+      },
+      {
+        "step": 1750,
+        "valid accuracy": 0.38,
+        "train loss": 0.6936728912591934,
+        "train samples": 7000,
+        "train time": 172.45823883989942,
+        "eval time": 30.67647124498035,
+        "tokens / sec": 1213.9460625847714,
+        "mem allocated avg": 6970556262.4,
+        "mem reserved avg": 29963210194.944,
+        "elapsed time": 1470.7583586190012
+      },
+      {
+        "step": 2000,
+        "valid accuracy": 0.34,
+        "train loss": 0.6850932078361511,
+        "train samples": 8000,
+        "train time": 172.2125422480749,
+        "eval time": 21.635232597007416,
+        "tokens / sec": 1206.044561497795,
+        "mem allocated avg": 6966487302.144,
+        "mem reserved avg": 29441472331.776,
+        "elapsed time": 1672.5279613650055
+      },
+      {
+        "step": 2250,
+        "valid accuracy": 0.38,
+        "train loss": 0.6678359975814819,
+        "train samples": 9000,
+        "train time": 174.90155081104604,
+        "eval time": 30.710825373011176,
+        "tokens / sec": 1228.9656609861506,
+        "mem allocated avg": 6978265659.392,
+        "mem reserved avg": 29786906820.608,
+        "elapsed time": 1885.9478800620127
+      },
+      {
+        "step": 2500,
+        "valid accuracy": 0.44,
+        "train loss": 0.6567883492708206,
+        "train samples": 10000,
+        "train time": 168.23802922523464,
+        "eval time": 30.644450194988167,
+        "tokens / sec": 1224.2594670688538,
+        "mem allocated avg": 6963658600.448,
+        "mem reserved avg": 29001515008.0,
+        "elapsed time": 2092.6179425250157
+      },
+      {
+        "step": 2750,
+        "valid accuracy": 0.46,
+        "train loss": 0.634885214805603,
+        "train samples": 11000,
+        "train time": 173.90507797230384,
+        "eval time": 30.730780140991556,
+        "tokens / sec": 1218.371553438734,
+        "mem allocated avg": 6973873780.736,
+        "mem reserved avg": 29700227334.144,
+        "elapsed time": 2304.930961822014
+      },
+      {
+        "step": 3000,
+        "valid accuracy": 0.5,
+        "train loss": 0.6183034368753433,
+        "train samples": 12000,
+        "train time": 173.86735759672592,
+        "eval time": 30.715003296005307,
+        "tokens / sec": 1200.5186188205496,
+        "mem allocated avg": 6969537423.36,
+        "mem reserved avg": 29600050577.408,
+        "elapsed time": 2517.428843543021
+      },
+      {
+        "step": 3250,
+        "valid accuracy": 0.48,
+        "train loss": 0.6171810357570648,
+        "train samples": 13000,
+        "train time": 172.11153198982356,
+        "eval time": 20.41151063601137,
+        "tokens / sec": 1225.3740209137754,
+        "mem allocated avg": 6971543304.192,
+        "mem reserved avg": 29378851373.056,
+        "elapsed time": 2717.537005299004
+      },
+      {
+        "step": 3500,
+        "valid accuracy": 0.54,
+        "train loss": 0.5918659865856171,
+        "train samples": 14000,
+        "train time": 171.7280791127414,
+        "eval time": 20.10971114897984,
+        "tokens / sec": 1221.407710863037,
+        "mem allocated avg": 6969227558.912,
+        "mem reserved avg": 29302649257.984,
+        "elapsed time": 2917.0929282610014
+      },
+      {
+        "step": 3750,
+        "valid accuracy": 0.52,
+        "train loss": 0.5810757111310959,
+        "train samples": 15000,
+        "train time": 177.90301188721787,
+        "eval time": 30.743899679015158,
+        "tokens / sec": 1218.09629697208,
+        "mem allocated avg": 6980796526.592,
+        "mem reserved avg": 29780892188.672,
+        "elapsed time": 3133.516127213021
+      },
+      {
+        "step": 4000,
+        "valid accuracy": 0.46,
+        "train loss": 0.5831188062429428,
+        "train samples": 16000,
+        "train time": 169.06811870416277,
+        "eval time": 30.76702547399327,
+        "tokens / sec": 1208.8204539474061,
+        "mem allocated avg": 6962208227.328,
+        "mem reserved avg": 29357410091.008,
+        "elapsed time": 3341.2088356550084
+      },
+      {
+        "step": 4250,
+        "valid accuracy": 0.54,
+        "train loss": 0.564371225476265,
+        "train samples": 17000,
+        "train time": 173.65914930703002,
+        "eval time": 30.680932629999006,
+        "tokens / sec": 1217.2638230898128,
+        "mem allocated avg": 6971919788.032,
+        "mem reserved avg": 29329702518.784,
+        "elapsed time": 3553.1107871610147
+      },
+      {
+        "step": 4500,
+        "valid accuracy": 0.54,
+        "train loss": 0.5646625064611435,
+        "train samples": 18000,
+        "train time": 168.5221932909335,
+        "eval time": 21.17475585200009,
+        "tokens / sec": 1233.1788231667917,
+        "mem allocated avg": 6967363680.256,
+        "mem reserved avg": 29285041569.792,
+        "elapsed time": 3750.635109158
+      },
+      {
+        "step": 4750,
+        "valid accuracy": 0.5,
+        "train loss": 0.5495809932947159,
+        "train samples": 19000,
+        "train time": 171.79509386298014,
+        "eval time": 20.51026588899549,
+        "tokens / sec": 1222.0314054337464,
+        "mem allocated avg": 6969873039.36,
+        "mem reserved avg": 29383817428.992,
+        "elapsed time": 3950.5909262890054
+      },
+      {
+        "step": 5000,
+        "valid accuracy": 0.54,
+        "train loss": 0.5584390600919723,
+        "train samples": 20000,
+        "train time": 170.36222512708628,
+        "eval time": 27.515913671988528,
+        "tokens / sec": 1222.5714934436196,
+        "mem allocated avg": 6966943275.008,
+        "mem reserved avg": 28902328107.008,
+        "elapsed time": 4156.128801064013
+      },
+      {
+        "step": 5000,
+        "test accuracy": 0.5064442759666414,
+        "train loss": 0.5584390600919723,
+        "train samples": 20000,
+        "train total tokens": 4198051,
+        "forgetting": 0.15337693691253662
+      }
+    ]
+  },
+  "meta_info": {
+    "model_info": {
+      "sha": "13afe5124825b4f3751f836b40dafda64c1ed062",
+      "created_at": "2024-09-18T15:23:48+00:00"
+    },
+    "dataset_info": {
+      "metamath": {
+        "sha": "aa4f34d3d2d3231299b5b03d9b3e5a20da45aa18",
+        "created_at": "2023-09-21T17:22:46+00:00"
+      },
+      "gsm8k": {
+        "sha": "740312add88f781978c0658806c59bc2815b9866",
+        "created_at": "2022-04-12T10:22:10+00:00"
+      }
+    },
+    "package_info": {
+      "transformers-version": "5.12.1",
+      "transformers-commit-hash": null,
+      "peft-version": "0.20.1.dev0",
+      "peft-commit-hash": "c702de402330f48c0d8b1f260d59413aa3110b65",
+      "datasets-version": "4.8.4",
+      "datasets-commit-hash": null,
+      "bitsandbytes-version": "0.49.2",
+      "bitsandbytes-commit-hash": null,
+      "torch-version": "2.11.0+cu130",
+      "torch-commit-hash": null
+    },
+    "system_info": {
+      "system": "Linux",
+      "release": "7.0.0-1009-aws",
+      "version": "#9~24.04.1-Ubuntu SMP PREEMPT Mon Jul 13 22:43:41 UTC 2026",
+      "machine": "x86_64",
+      "processor": "x86_64",
+      "accelerator": "NVIDIA L40S"
+    },
+    "pytorch_info": "PyTorch built with:\n  - GCC 13.3\n  - C++ Version: 201703\n  - Intel(R) oneAPI Math Kernel Library Version 2024.2-Product Build 20240605 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v3.10.2 (Git Hash f1d471933dc852f956fd05389f9313c7148783d5)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 13.0\n  - NVCC architecture flags: -gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_120,code=sm_120\n  - CuDNN 90.7.1  (built against CUDA 12.8)\n    - Built with CuDNN 91.9\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, COMMIT_SHA=70d99e998b4955e0049d13a98d77ae1b14db1f45, CUDA_VERSION=13.0, CUDNN_VERSION=9.19.0, CXX_COMPILER=/opt/rh/gcc-toolset-13/root/usr/bin/c++, CXX_FLAGS= -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DLIBKINETO_NOXPUPTI=ON -DUSE_FBGEMM -DUSE_MSLK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -DC10_NODEPRECATED -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=range-loop-construct -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-unknown-pragmas -Wno-unused-parameter -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=old-style-cast -faligned-new -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-dangling-reference -Wno-error=dangling-reference -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, TORCH_VERSION=2.11.0, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, USE_XCCL=OFF, USE_XPU=OFF, \n"
+  }
+}

lokr--llama-3.2-3B-rank32-factor4-lr0.001--2026-08-05T17-07-38+00-00.json

@@ -0,0 +1,362 @@
+{
+  "run_info": {
+    "created_at": "2026-08-05T17:07:38+00:00",
+    "total_time": 1546.6131941369968,
+    "experiment_name": "lokr/llama-3.2-3B-rank32-factor4-lr0.001",
+    "peft_branch": "auto-research",
+    "train_config": {
+      "model_id": "meta-llama/Llama-3.2-3B",
+      "dtype": "bfloat16",
+      "max_seq_length": 768,
+      "batch_size": 4,
+      "batch_size_eval": 50,
+      "max_steps": 5000,
+      "eval_steps": 250,
+      "compile": false,
+      "use_gc": false,
+      "query_template": "Question: {query} Think step by step.\nAnswer:",
+      "seed": 0,
+      "grad_norm_clip": 1.0,
+      "optimizer_type": "AdamW",
+      "optimizer_kwargs": {
+        "lr": 0.001,
+        "weight_decay": 0.1
+      },
+      "lr_scheduler": "cosine",
+      "use_amp": false,
+      "autocast_adapter_dtype": true,
+      "generation_kwargs": {
+        "max_length": 800,
+        "max_new_tokens": 300
+      },
+      "attn_implementation": null,
+      "init_kv_cache_prefix": null
+    },
+    "peft_config": {
+      "task_type": null,
+      "peft_type": "LOKR",
+      "auto_mapping": null,
+      "peft_version": "0.20.1.dev0@UNKNOWN",
+      "base_model_name_or_path": "meta-llama/Llama-3.2-3B",
+      "revision": null,
+      "inference_mode": false,
+      "rank_pattern": {},
+      "alpha_pattern": {},
+      "r": 32,
+      "alpha": 64,
+      "rank_dropout": 0.0,
+      "module_dropout": 0.0,
+      "use_effective_conv2d": false,
+      "decompose_both": false,
+      "decompose_factor": 4,
+      "rank_dropout_scale": false,
+      "target_modules": [
+        "v_proj",
+        "q_proj"
+      ],
+      "exclude_modules": null,
+      "init_weights": true,
+      "layers_to_transform": null,
+      "layers_pattern": null,
+      "modules_to_save": null
+    },
+    "error_msg": ""
+  },
+  "train_info": {
+    "accelerator_memory_reserved_avg": 15813729006,
+    "accelerator_memory_max": 23603445760,
+    "accelerator_memory_reserved_99th": 21462253568,
+    "train_time": 1314.7368741209793,
+    "file_size": 9199776,
+    "num_trainable_params": 2294656,
+    "num_total_params": 3215044480,
+    "status": "success",
+    "metrics": [
+      {
+        "step": 250,
+        "valid accuracy": 0.3,
+        "train loss": 1.0161663095951081,
+        "train samples": 1000,
+        "train time": 44.565753069211496,
+        "eval time": 13.766088383999886,
+        "tokens / sec": 4750.710700909647,
+        "mem allocated avg": 6816982290.432,
+        "mem reserved avg": 15925411577.856,
+        "elapsed time": 85.46226108600968
+      },
+      {
+        "step": 500,
+        "valid accuracy": 0.38,
+        "train loss": 0.7154890269041061,
+        "train samples": 2000,
+        "train time": 44.36160620805458,
+        "eval time": 13.70357837001211,
+        "tokens / sec": 4688.62644477997,
+        "mem allocated avg": 6809787807.744,
+        "mem reserved avg": 15590773227.52,
+        "elapsed time": 147.68463174402132
+      },
+      {
+        "step": 750,
+        "valid accuracy": 0.44,
+        "train loss": 0.6800865457057953,
+        "train samples": 3000,
+        "train time": 45.21799132809974,
+        "eval time": 10.416135988984024,
+        "tokens / sec": 4741.497658405829,
+        "mem allocated avg": 6820353296.384,
+        "mem reserved avg": 15783870595.072,
+        "elapsed time": 207.43689091701526
+      },
+      {
+        "step": 1000,
+        "valid accuracy": 0.38,
+        "train loss": 0.6610932334661483,
+        "train samples": 4000,
+        "train time": 43.50991478617652,
+        "eval time": 13.727248279988999,
+        "tokens / sec": 4788.241967924749,
+        "mem allocated avg": 6811382177.792,
+        "mem reserved avg": 15812651909.12,
+        "elapsed time": 268.92848025701824
+      },
+      {
+        "step": 1250,
+        "valid accuracy": 0.42,
+        "train loss": 0.6575651440620423,
+        "train samples": 5000,
+        "train time": 43.86161071388051,
+        "eval time": 12.016878391004866,
+        "tokens / sec": 4754.453760495525,
+        "mem allocated avg": 6811189080.064,
+        "mem reserved avg": 15849394012.16,
+        "elapsed time": 328.95342484102
+      },
+      {
+        "step": 1500,
+        "valid accuracy": 0.4,
+        "train loss": 0.6488472428321839,
+        "train samples": 6000,
+        "train time": 44.450407360622194,
+        "eval time": 12.069276472990168,
+        "tokens / sec": 4709.31567177138,
+        "mem allocated avg": 6813455890.432,
+        "mem reserved avg": 15758318895.104,
+        "elapsed time": 389.6669018280227
+      },
+      {
+        "step": 1750,
+        "valid accuracy": 0.4,
+        "train loss": 0.6409421992301941,
+        "train samples": 7000,
+        "train time": 43.89553755291854,
+        "eval time": 8.112704949016916,
+        "tokens / sec": 4769.391415872531,
+        "mem allocated avg": 6814575978.496,
+        "mem reserved avg": 16176843325.44,
+        "elapsed time": 445.87124504201347
+      },
+      {
+        "step": 2000,
+        "valid accuracy": 0.42,
+        "train loss": 0.6436473038196564,
+        "train samples": 8000,
+        "train time": 43.46780985026271,
+        "eval time": 13.70299476198852,
+        "tokens / sec": 4778.156541943756,
+        "mem allocated avg": 6810022129.664,
+        "mem reserved avg": 15811108405.248,
+        "elapsed time": 507.305808789999
+      },
+      {
+        "step": 2250,
+        "valid accuracy": 0.46,
+        "train loss": 0.6351276015043259,
+        "train samples": 9000,
+        "train time": 45.941694433102384,
+        "eval time": 13.683427158015547,
+        "tokens / sec": 4678.712935000574,
+        "mem allocated avg": 6821545334.784,
+        "mem reserved avg": 16136871608.32,
+        "elapsed time": 571.1446734110068
+      },
+      {
+        "step": 2500,
+        "valid accuracy": 0.44,
+        "train loss": 0.6321750155687332,
+        "train samples": 10000,
+        "train time": 42.98558233407675,
+        "eval time": 10.8900341510016,
+        "tokens / sec": 4791.536808766692,
+        "mem allocated avg": 6807899869.184,
+        "mem reserved avg": 15569608769.536,
+        "elapsed time": 629.1753076720051
+      },
+      {
+        "step": 2750,
+        "valid accuracy": 0.46,
+        "train loss": 0.623354480266571,
+        "train samples": 11000,
+        "train time": 43.88556467826129,
+        "eval time": 8.260783003992401,
+        "tokens / sec": 4828.034037008876,
+        "mem allocated avg": 6817548113.92,
+        "mem reserved avg": 15957439283.2,
+        "elapsed time": 685.4198551190202
+      },
+      {
+        "step": 3000,
+        "valid accuracy": 0.4,
+        "train loss": 0.6137815948724746,
+        "train samples": 12000,
+        "train time": 43.619943306664936,
+        "eval time": 9.229391172993928,
+        "tokens / sec": 4785.219424347734,
+        "mem allocated avg": 6813387800.576,
+        "mem reserved avg": 15859359678.464,
+        "elapsed time": 742.4976472580165
+      },
+      {
+        "step": 3250,
+        "valid accuracy": 0.5,
+        "train loss": 0.6225851331949234,
+        "train samples": 13000,
+        "train time": 44.45619710412575,
+        "eval time": 13.625753461994464,
+        "tokens / sec": 4744.018016341469,
+        "mem allocated avg": 6815676938.24,
+        "mem reserved avg": 15804858892.288,
+        "elapsed time": 804.7041690270125
+      },
+      {
+        "step": 3500,
+        "valid accuracy": 0.44,
+        "train loss": 0.6047447552680969,
+        "train samples": 14000,
+        "train time": 45.048323025373975,
+        "eval time": 13.649868922017049,
+        "tokens / sec": 4656.11116937374,
+        "mem allocated avg": 6813028171.776,
+        "mem reserved avg": 15719949402.112,
+        "elapsed time": 867.6115258440259
+      },
+      {
+        "step": 3750,
+        "valid accuracy": 0.5,
+        "train loss": 0.6024837223291397,
+        "train samples": 15000,
+        "train time": 45.74933339908603,
+        "eval time": 10.695509596989723,
+        "tokens / sec": 4736.746612450735,
+        "mem allocated avg": 6824250314.752,
+        "mem reserved avg": 16002880372.736,
+        "elapsed time": 928.2686512710061
+      },
+      {
+        "step": 4000,
+        "valid accuracy": 0.44,
+        "train loss": 0.6132795907258988,
+        "train samples": 16000,
+        "train time": 43.87121415033471,
+        "eval time": 8.80364583901246,
+        "tokens / sec": 4658.476040796805,
+        "mem allocated avg": 6805346758.656,
+        "mem reserved avg": 15795564314.624,
+        "elapsed time": 985.1485650210234
+      },
+      {
+        "step": 4250,
+        "valid accuracy": 0.4,
+        "train loss": 0.5986514059305191,
+        "train samples": 17000,
+        "train time": 43.865029192122165,
+        "eval time": 9.22572749099345,
+        "tokens / sec": 4819.078064992235,
+        "mem allocated avg": 6816118788.096,
+        "mem reserved avg": 15759350693.888,
+        "elapsed time": 1042.2822316290112
+      },
+      {
+        "step": 4500,
+        "valid accuracy": 0.54,
+        "train loss": 0.6039946029186248,
+        "train samples": 18000,
+        "train time": 43.74795002306928,
+        "eval time": 13.635402487008832,
+        "tokens / sec": 4750.348299529758,
+        "mem allocated avg": 6811553531.904,
+        "mem reserved avg": 15635811663.872,
+        "elapsed time": 1103.884899763012
+      },
+      {
+        "step": 4750,
+        "valid accuracy": 0.42,
+        "train loss": 0.5956217089891434,
+        "train samples": 19000,
+        "train time": 44.844085472112056,
+        "eval time": 10.099945699999807,
+        "tokens / sec": 4681.531528400959,
+        "mem allocated avg": 6813660581.888,
+        "mem reserved avg": 15827466190.848,
+        "elapsed time": 1162.9690284210083
+      },
+      {
+        "step": 5000,
+        "valid accuracy": 0.5,
+        "train loss": 0.603061123251915,
+        "train samples": 20000,
+        "train time": 43.117642779659946,
+        "eval time": 8.968568419018993,
+        "tokens / sec": 4830.505254295876,
+        "mem allocated avg": 6810714724.352,
+        "mem reserved avg": 15497047310.336,
+        "elapsed time": 1219.1641551900248
+      },
+      {
+        "step": 5000,
+        "test accuracy": 0.45564821834723274,
+        "train loss": 0.603061123251915,
+        "train samples": 20000,
+        "train total tokens": 4198051,
+        "forgetting": 0.08525705337524414
+      }
+    ]
+  },
+  "meta_info": {
+    "model_info": {
+      "sha": "13afe5124825b4f3751f836b40dafda64c1ed062",
+      "created_at": "2024-09-18T15:23:48+00:00"
+    },
+    "dataset_info": {
+      "metamath": {
+        "sha": "aa4f34d3d2d3231299b5b03d9b3e5a20da45aa18",
+        "created_at": "2023-09-21T17:22:46+00:00"
+      },
+      "gsm8k": {
+        "sha": "740312add88f781978c0658806c59bc2815b9866",
+        "created_at": "2022-04-12T10:22:10+00:00"
+      }
+    },
+    "package_info": {
+      "transformers-version": "5.12.1",
+      "transformers-commit-hash": null,
+      "peft-version": "0.20.1.dev0",
+      "peft-commit-hash": "c702de402330f48c0d8b1f260d59413aa3110b65",
+      "datasets-version": "4.8.4",
+      "datasets-commit-hash": null,
+      "bitsandbytes-version": "0.49.2",
+      "bitsandbytes-commit-hash": null,
+      "torch-version": "2.11.0+cu130",
+      "torch-commit-hash": null
+    },
+    "system_info": {
+      "system": "Linux",
+      "release": "7.0.0-1009-aws",
+      "version": "#9~24.04.1-Ubuntu SMP PREEMPT Mon Jul 13 22:43:41 UTC 2026",
+      "machine": "x86_64",
+      "processor": "x86_64",
+      "accelerator": "NVIDIA L40S"
+    },
+    "pytorch_info": "PyTorch built with:\n  - GCC 13.3\n  - C++ Version: 201703\n  - Intel(R) oneAPI Math Kernel Library Version 2024.2-Product Build 20240605 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v3.10.2 (Git Hash f1d471933dc852f956fd05389f9313c7148783d5)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 13.0\n  - NVCC architecture flags: -gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_120,code=sm_120\n  - CuDNN 90.7.1  (built against CUDA 12.8)\n    - Built with CuDNN 91.9\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, COMMIT_SHA=70d99e998b4955e0049d13a98d77ae1b14db1f45, CUDA_VERSION=13.0, CUDNN_VERSION=9.19.0, CXX_COMPILER=/opt/rh/gcc-toolset-13/root/usr/bin/c++, CXX_FLAGS= -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DLIBKINETO_NOXPUPTI=ON -DUSE_FBGEMM -DUSE_MSLK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -DC10_NODEPRECATED -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=range-loop-construct -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-unknown-pragmas -Wno-unused-parameter -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=old-style-cast -faligned-new -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-dangling-reference -Wno-error=dangling-reference -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, TORCH_VERSION=2.11.0, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, USE_XCCL=OFF, USE_XPU=OFF, \n"
+  }
+}

lokr--llama-3.2-3B-rank32-factor8-all-linear-lr0.003--2026-08-06T10-13-59+00-00.json

@@ -0,0 +1,367 @@
+{
+  "run_info": {
+    "created_at": "2026-08-06T10:13:59+00:00",
+    "total_time": 4814.026151216996,
+    "experiment_name": "lokr/llama-3.2-3B-rank32-factor8-all-linear-lr0.003",
+    "peft_branch": "auto-research",
+    "train_config": {
+      "model_id": "meta-llama/Llama-3.2-3B",
+      "dtype": "bfloat16",
+      "max_seq_length": 768,
+      "batch_size": 4,
+      "batch_size_eval": 50,
+      "max_steps": 5000,
+      "eval_steps": 250,
+      "compile": false,
+      "use_gc": false,
+      "query_template": "Question: {query} Think step by step.\nAnswer:",
+      "seed": 0,
+      "grad_norm_clip": 1.0,
+      "optimizer_type": "AdamW",
+      "optimizer_kwargs": {
+        "lr": 0.003,
+        "weight_decay": 0.1
+      },
+      "lr_scheduler": "cosine",
+      "use_amp": false,
+      "autocast_adapter_dtype": true,
+      "generation_kwargs": {
+        "max_length": 800,
+        "max_new_tokens": 300
+      },
+      "attn_implementation": null,
+      "init_kv_cache_prefix": null
+    },
+    "peft_config": {
+      "task_type": null,
+      "peft_type": "LOKR",
+      "auto_mapping": null,
+      "peft_version": "0.20.1.dev0@UNKNOWN",
+      "base_model_name_or_path": "meta-llama/Llama-3.2-3B",
+      "revision": null,
+      "inference_mode": false,
+      "rank_pattern": {},
+      "alpha_pattern": {},
+      "r": 32,
+      "alpha": 64,
+      "rank_dropout": 0.0,
+      "module_dropout": 0.0,
+      "use_effective_conv2d": false,
+      "decompose_both": false,
+      "decompose_factor": 8,
+      "rank_dropout_scale": false,
+      "target_modules": [
+        "down_proj",
+        "k_proj",
+        "up_proj",
+        "o_proj",
+        "v_proj",
+        "q_proj",
+        "gate_proj"
+      ],
+      "exclude_modules": null,
+      "init_weights": true,
+      "layers_to_transform": null,
+      "layers_pattern": null,
+      "modules_to_save": null
+    },
+    "error_msg": ""
+  },
+  "train_info": {
+    "accelerator_memory_reserved_avg": 28733233902,
+    "accelerator_memory_max": 40372273152,
+    "accelerator_memory_reserved_99th": 36979081216,
+    "train_time": 4292.985215063964,
+    "file_size": 24437952,
+    "num_trainable_params": 6091008,
+    "num_total_params": 3218840832,
+    "status": "success",
+    "metrics": [
+      {
+        "step": 250,
+        "valid accuracy": 0.38,
+        "train loss": 0.8615508460998536,
+        "train samples": 1000,
+        "train time": 171.43761316602468,
+        "eval time": 30.569398073013872,
+        "tokens / sec": 1234.9623638014941,
+        "mem allocated avg": 6873432496.128,
+        "mem reserved avg": 28930102788.096,
+        "elapsed time": 233.96793232500204
+      },
+      {
+        "step": 500,
+        "valid accuracy": 0.36,
+        "train loss": 0.7002903352975846,
+        "train samples": 2000,
+        "train time": 171.33122572375578,
+        "eval time": 30.383617023995612,
+        "tokens / sec": 1213.993532827219,
+        "mem allocated avg": 6865498011.648,
+        "mem reserved avg": 28395521966.08,
+        "elapsed time": 442.96696859801887
+      },
+      {
+        "step": 750,
+        "valid accuracy": 0.26,
+        "train loss": 0.7056614835262298,
+        "train samples": 3000,
+        "train time": 172.8551740029361,
+        "eval time": 30.535659058979945,
+        "tokens / sec": 1240.3504913098996,
+        "mem allocated avg": 6876121849.856,
+        "mem reserved avg": 28633364168.704,
+        "elapsed time": 653.4886630160036
+      },
+      {
+        "step": 1000,
+        "valid accuracy": 0.32,
+        "train loss": 0.6885825423002243,
+        "train samples": 4000,
+        "train time": 169.4359879681724,
+        "eval time": 30.406957479019184,
+        "tokens / sec": 1229.5852994296272,
+        "mem allocated avg": 6867443728.384,
+        "mem reserved avg": 28759604330.496,
+        "elapsed time": 860.6749559840246
+      },
+      {
+        "step": 1250,
+        "valid accuracy": 0.34,
+        "train loss": 0.6797382447719574,
+        "train samples": 5000,
+        "train time": 168.55452173313824,
+        "eval time": 30.731183161988156,
+        "tokens / sec": 1237.2139166350285,
+        "mem allocated avg": 6867681554.432,
+        "mem reserved avg": 28730613301.248,
+        "elapsed time": 1067.156196824013
+      },
+      {
+        "step": 1500,
+        "valid accuracy": 0.42,
+        "train loss": 0.6683728358745575,
+        "train samples": 6000,
+        "train time": 174.49907591292867,
+        "eval time": 30.453326868999284,
+        "tokens / sec": 1199.610937220388,
+        "mem allocated avg": 6869211674.624,
+        "mem reserved avg": 28693585985.536,
+        "elapsed time": 1279.4822671050206
+      },
+      {
+        "step": 1750,
+        "valid accuracy": 0.44,
+        "train loss": 0.6583834865093231,
+        "train samples": 7000,
+        "train time": 170.7857195379038,
+        "eval time": 25.239067126007285,
+        "tokens / sec": 1225.834341222752,
+        "mem allocated avg": 6870617331.712,
+        "mem reserved avg": 29237360721.92,
+        "elapsed time": 1482.7154341220157
+      },
+      {
+        "step": 2000,
+        "valid accuracy": 0.4,
+        "train loss": 0.6530058711767197,
+        "train samples": 8000,
+        "train time": 170.49540531635284,
+        "eval time": 20.073756367026363,
+        "tokens / sec": 1218.1911859420595,
+        "mem allocated avg": 6866557816.832,
+        "mem reserved avg": 28706730934.272,
+        "elapsed time": 1680.7453745120147
+      },
+      {
+        "step": 2250,
+        "valid accuracy": 0.38,
+        "train loss": 0.6375795097351075,
+        "train samples": 9000,
+        "train time": 172.48475469491677,
+        "eval time": 30.480793218011968,
+        "tokens / sec": 1246.185498423848,
+        "mem allocated avg": 6878300321.792,
+        "mem reserved avg": 29174731374.592,
+        "elapsed time": 1890.9911642649968
+      },
+      {
+        "step": 2500,
+        "valid accuracy": 0.44,
+        "train loss": 0.6300644111633301,
+        "train samples": 10000,
+        "train time": 170.72695773094893,
+        "eval time": 30.564193555997917,
+        "tokens / sec": 1206.4117040296962,
+        "mem allocated avg": 6863159779.328,
+        "mem reserved avg": 28354786885.632,
+        "elapsed time": 2099.5891719350184
+      },
+      {
+        "step": 2750,
+        "valid accuracy": 0.46,
+        "train loss": 0.6156140941381455,
+        "train samples": 11000,
+        "train time": 175.56682728099986,
+        "eval time": 30.624924142000964,
+        "tokens / sec": 1206.839602226668,
+        "mem allocated avg": 6873532012.544,
+        "mem reserved avg": 29056410058.752,
+        "elapsed time": 2313.028025173
+      },
+      {
+        "step": 3000,
+        "valid accuracy": 0.4,
+        "train loss": 0.6006648186445236,
+        "train samples": 12000,
+        "train time": 168.66434494018904,
+        "eval time": 18.980127498012735,
+        "tokens / sec": 1237.5526082529132,
+        "mem allocated avg": 6868978239.488,
+        "mem reserved avg": 28823433248.768,
+        "elapsed time": 2508.053776479006
+      },
+      {
+        "step": 3250,
+        "valid accuracy": 0.56,
+        "train loss": 0.6037812237739563,
+        "train samples": 13000,
+        "train time": 171.0474435731303,
+        "eval time": 24.763687291997485,
+        "tokens / sec": 1232.9970889616397,
+        "mem allocated avg": 6870784204.8,
+        "mem reserved avg": 28628398112.768,
+        "elapsed time": 2710.9462912730232
+      },
+      {
+        "step": 3500,
+        "valid accuracy": 0.46,
+        "train loss": 0.5813168592453003,
+        "train samples": 14000,
+        "train time": 171.50457438497688,
+        "eval time": 22.86701555599575,
+        "tokens / sec": 1222.9994491527293,
+        "mem allocated avg": 6869285494.784,
+        "mem reserved avg": 28652557303.808,
+        "elapsed time": 2912.6192737650126
+      },
+      {
+        "step": 3750,
+        "valid accuracy": 0.62,
+        "train loss": 0.5727146596908569,
+        "train samples": 15000,
+        "train time": 174.36638085806044,
+        "eval time": 24.90966416199808,
+        "tokens / sec": 1242.8026488454955,
+        "mem allocated avg": 6880991285.248,
+        "mem reserved avg": 29152602226.688,
+        "elapsed time": 3119.1520819069992
+      },
+      {
+        "step": 4000,
+        "valid accuracy": 0.48,
+        "train loss": 0.5771222689151764,
+        "train samples": 16000,
+        "train time": 166.84103391700773,
+        "eval time": 21.578396992990747,
+        "tokens / sec": 1224.9564462760516,
+        "mem allocated avg": 6861483413.504,
+        "mem reserved avg": 28693292384.256,
+        "elapsed time": 3314.887631963007
+      },
+      {
+        "step": 4250,
+        "valid accuracy": 0.48,
+        "train loss": 0.5596228052377701,
+        "train samples": 17000,
+        "train time": 168.76604135675007,
+        "eval time": 26.615190682001412,
+        "tokens / sec": 1252.5564876712988,
+        "mem allocated avg": 6871832551.424,
+        "mem reserved avg": 28660518092.8,
+        "elapsed time": 3517.33625098801
+      },
+      {
+        "step": 4500,
+        "valid accuracy": 0.56,
+        "train loss": 0.5638360705375671,
+        "train samples": 18000,
+        "train time": 170.94274596704054,
+        "eval time": 19.347630154981744,
+        "tokens / sec": 1215.7169865521487,
+        "mem allocated avg": 6867000530.944,
+        "mem reserved avg": 28579878404.096,
+        "elapsed time": 3714.970551975013
+      },
+      {
+        "step": 4750,
+        "valid accuracy": 0.54,
+        "train loss": 0.5517311846017837,
+        "train samples": 19000,
+        "train time": 172.1988175440929,
+        "eval time": 17.573275297996588,
+        "tokens / sec": 1219.1663275866772,
+        "mem allocated avg": 6869788155.904,
+        "mem reserved avg": 28620730925.056,
+        "elapsed time": 3911.9516312760243
+      },
+      {
+        "step": 5000,
+        "valid accuracy": 0.52,
+        "train loss": 0.5587065149545669,
+        "train samples": 20000,
+        "train time": 168.7623130355787,
+        "eval time": 20.3918305770203,
+        "tokens / sec": 1234.161799833178,
+        "mem allocated avg": 6866212829.184,
+        "mem reserved avg": 28180454834.176,
+        "elapsed time": 4108.278737300017
+      },
+      {
+        "step": 5000,
+        "test accuracy": 0.5079605761940864,
+        "train loss": 0.5587065149545669,
+        "train samples": 20000,
+        "train total tokens": 4198051,
+        "forgetting": 0.18798315525054932
+      }
+    ]
+  },
+  "meta_info": {
+    "model_info": {
+      "sha": "13afe5124825b4f3751f836b40dafda64c1ed062",
+      "created_at": "2024-09-18T15:23:48+00:00"
+    },
+    "dataset_info": {
+      "metamath": {
+        "sha": "aa4f34d3d2d3231299b5b03d9b3e5a20da45aa18",
+        "created_at": "2023-09-21T17:22:46+00:00"
+      },
+      "gsm8k": {
+        "sha": "740312add88f781978c0658806c59bc2815b9866",
+        "created_at": "2022-04-12T10:22:10+00:00"
+      }
+    },
+    "package_info": {
+      "transformers-version": "5.12.1",
+      "transformers-commit-hash": null,
+      "peft-version": "0.20.1.dev0",
+      "peft-commit-hash": "c702de402330f48c0d8b1f260d59413aa3110b65",
+      "datasets-version": "4.8.4",
+      "datasets-commit-hash": null,
+      "bitsandbytes-version": "0.49.2",
+      "bitsandbytes-commit-hash": null,
+      "torch-version": "2.11.0+cu130",
+      "torch-commit-hash": null
+    },
+    "system_info": {
+      "system": "Linux",
+      "release": "7.0.0-1009-aws",
+      "version": "#9~24.04.1-Ubuntu SMP PREEMPT Mon Jul 13 22:43:41 UTC 2026",
+      "machine": "x86_64",
+      "processor": "x86_64",
+      "accelerator": "NVIDIA L40S"
+    },
+    "pytorch_info": "PyTorch built with:\n  - GCC 13.3\n  - C++ Version: 201703\n  - Intel(R) oneAPI Math Kernel Library Version 2024.2-Product Build 20240605 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v3.10.2 (Git Hash f1d471933dc852f956fd05389f9313c7148783d5)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 13.0\n  - NVCC architecture flags: -gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_120,code=sm_120\n  - CuDNN 90.7.1  (built against CUDA 12.8)\n    - Built with CuDNN 91.9\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, COMMIT_SHA=70d99e998b4955e0049d13a98d77ae1b14db1f45, CUDA_VERSION=13.0, CUDNN_VERSION=9.19.0, CXX_COMPILER=/opt/rh/gcc-toolset-13/root/usr/bin/c++, CXX_FLAGS= -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DLIBKINETO_NOXPUPTI=ON -DUSE_FBGEMM -DUSE_MSLK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -DC10_NODEPRECATED -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=range-loop-construct -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-unknown-pragmas -Wno-unused-parameter -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=old-style-cast -faligned-new -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-dangling-reference -Wno-error=dangling-reference -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, TORCH_VERSION=2.11.0, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, USE_XCCL=OFF, USE_XPU=OFF, \n"
+  }
+}

lokr--llama-3.2-3B-rank32-factor8-lr0.003--2026-08-06T12-43-21+00-00.json

@@ -0,0 +1,362 @@
+{
+  "run_info": {
+    "created_at": "2026-08-06T12:43:21+00:00",
+    "total_time": 1532.9845309570082,
+    "experiment_name": "lokr/llama-3.2-3B-rank32-factor8-lr0.003",
+    "peft_branch": "auto-research",
+    "train_config": {
+      "model_id": "meta-llama/Llama-3.2-3B",
+      "dtype": "bfloat16",
+      "max_seq_length": 768,
+      "batch_size": 4,
+      "batch_size_eval": 50,
+      "max_steps": 5000,
+      "eval_steps": 250,
+      "compile": false,
+      "use_gc": false,
+      "query_template": "Question: {query} Think step by step.\nAnswer:",
+      "seed": 0,
+      "grad_norm_clip": 1.0,
+      "optimizer_type": "AdamW",
+      "optimizer_kwargs": {
+        "lr": 0.003,
+        "weight_decay": 0.1
+      },
+      "lr_scheduler": "cosine",
+      "use_amp": false,
+      "autocast_adapter_dtype": true,
+      "generation_kwargs": {
+        "max_length": 800,
+        "max_new_tokens": 300
+      },
+      "attn_implementation": null,
+      "init_kv_cache_prefix": null
+    },
+    "peft_config": {
+      "task_type": null,
+      "peft_type": "LOKR",
+      "auto_mapping": null,
+      "peft_version": "0.20.1.dev0@UNKNOWN",
+      "base_model_name_or_path": "meta-llama/Llama-3.2-3B",
+      "revision": null,
+      "inference_mode": false,
+      "rank_pattern": {},
+      "alpha_pattern": {},
+      "r": 32,
+      "alpha": 64,
+      "rank_dropout": 0.0,
+      "module_dropout": 0.0,
+      "use_effective_conv2d": false,
+      "decompose_both": false,
+      "decompose_factor": 8,
+      "rank_dropout_scale": false,
+      "target_modules": [
+        "q_proj",
+        "v_proj"
+      ],
+      "exclude_modules": null,
+      "init_weights": true,
+      "layers_to_transform": null,
+      "layers_pattern": null,
+      "modules_to_save": null
+    },
+    "error_msg": ""
+  },
+  "train_info": {
+    "accelerator_memory_reserved_avg": 15824497462,
+    "accelerator_memory_max": 23595057152,
+    "accelerator_memory_reserved_99th": 21386756096,
+    "train_time": 1307.3613977189816,
+    "file_size": 4622968,
+    "num_trainable_params": 1150464,
+    "num_total_params": 3213900288,
+    "status": "success",
+    "metrics": [
+      {
+        "step": 250,
+        "valid accuracy": 0.32,
+        "train loss": 0.9270877740383148,
+        "train samples": 1000,
+        "train time": 43.66479822300607,
+        "eval time": 13.922552911011735,
+        "tokens / sec": 4848.73418900742,
+        "mem allocated avg": 6799740565.504,
+        "mem reserved avg": 15942247514.112,
+        "elapsed time": 85.75519717100542
+      },
+      {
+        "step": 500,
+        "valid accuracy": 0.34,
+        "train loss": 0.7118661236763001,
+        "train samples": 2000,
+        "train time": 43.944341480208095,
+        "eval time": 13.804070371988928,
+        "tokens / sec": 4733.146361828587,
+        "mem allocated avg": 6792644532.224,
+        "mem reserved avg": 15596393594.88,
+        "elapsed time": 147.75061033500242
+      },
+      {
+        "step": 750,
+        "valid accuracy": 0.28,
+        "train loss": 0.693268860578537,
+        "train samples": 3000,
+        "train time": 45.00842113170074,
+        "eval time": 13.829937142989365,
+        "tokens / sec": 4763.575229014002,
+        "mem allocated avg": 6802870536.192,
+        "mem reserved avg": 15802895958.016,
+        "elapsed time": 210.7075444869988
+      },
+      {
+        "step": 1000,
+        "valid accuracy": 0.32,
+        "train loss": 0.6756969265937806,
+        "train samples": 4000,
+        "train time": 43.577776318008546,
+        "eval time": 10.035979694977868,
+        "tokens / sec": 4780.785473762346,
+        "mem allocated avg": 6793522110.464,
+        "mem reserved avg": 15816049295.36,
+        "elapsed time": 268.5969132730097
+      },
+      {
+        "step": 1250,
+        "valid accuracy": 0.32,
+        "train loss": 0.6747316555976868,
+        "train samples": 5000,
+        "train time": 43.6757008919958,
+        "eval time": 10.33713387500029,
+        "tokens / sec": 4774.691550244076,
+        "mem allocated avg": 6794507005.952,
+        "mem reserved avg": 15854636892.16,
+        "elapsed time": 326.7247922690003
+      },
+      {
+        "step": 1500,
+        "valid accuracy": 0.42,
+        "train loss": 0.6669513373374939,
+        "train samples": 6000,
+        "train time": 44.722955610981444,
+        "eval time": 8.636160207999637,
+        "tokens / sec": 4680.616411420718,
+        "mem allocated avg": 6795964114.944,
+        "mem reserved avg": 15784088698.88,
+        "elapsed time": 384.3359192000062
+      },
+      {
+        "step": 1750,
+        "valid accuracy": 0.46,
+        "train loss": 0.6600646764039994,
+        "train samples": 7000,
+        "train time": 44.147544838750036,
+        "eval time": 7.810703630006174,
+        "tokens / sec": 4742.166314450195,
+        "mem allocated avg": 6797846458.368,
+        "mem reserved avg": 16197387026.432,
+        "elapsed time": 440.50814956301474
+      },
+      {
+        "step": 2000,
+        "valid accuracy": 0.46,
+        "train loss": 0.6624095524549485,
+        "train samples": 8000,
+        "train time": 43.98306571893045,
+        "eval time": 7.598463203001302,
+        "tokens / sec": 4722.181062303872,
+        "mem allocated avg": 6793247461.376,
+        "mem reserved avg": 15799850893.312,
+        "elapsed time": 496.3889749700029
+      },
+      {
+        "step": 2250,
+        "valid accuracy": 0.36,
+        "train loss": 0.6519601926803589,
+        "train samples": 9000,
+        "train time": 45.462686326907715,
+        "eval time": 13.740995499014389,
+        "tokens / sec": 4728.009217369544,
+        "mem allocated avg": 6804674760.704,
+        "mem reserved avg": 16098552446.976,
+        "elapsed time": 559.7353278819937
+      },
+      {
+        "step": 2500,
+        "valid accuracy": 0.38,
+        "train loss": 0.6490274410247803,
+        "train samples": 10000,
+        "train time": 42.95588268118445,
+        "eval time": 13.741985282016685,
+        "tokens / sec": 4794.84967236438,
+        "mem allocated avg": 6790849527.808,
+        "mem reserved avg": 15565380911.104,
+        "elapsed time": 620.6423299649905
+      },
+      {
+        "step": 2750,
+        "valid accuracy": 0.38,
+        "train loss": 0.638805770277977,
+        "train samples": 11000,
+        "train time": 44.075625506957294,
+        "eval time": 9.21670454001287,
+        "tokens / sec": 4807.214816873213,
+        "mem allocated avg": 6800556472.32,
+        "mem reserved avg": 15965735616.512,
+        "elapsed time": 678.0697450590087
+      },
+      {
+        "step": 3000,
+        "valid accuracy": 0.44,
+        "train loss": 0.6281948779821396,
+        "train samples": 12000,
+        "train time": 43.90088801126694,
+        "eval time": 10.656771742011188,
+        "tokens / sec": 4754.5963067177645,
+        "mem allocated avg": 6795873429.504,
+        "mem reserved avg": 15874777939.968,
+        "elapsed time": 736.8918209109979
+      },
+      {
+        "step": 3250,
+        "valid accuracy": 0.5,
+        "train loss": 0.6351900390386581,
+        "train samples": 13000,
+        "train time": 44.91159329522634,
+        "eval time": 13.774559659010265,
+        "tokens / sec": 4695.91445161258,
+        "mem allocated avg": 6798974142.464,
+        "mem reserved avg": 15770885029.888,
+        "elapsed time": 799.6646873739955
+      },
+      {
+        "step": 3500,
+        "valid accuracy": 0.52,
+        "train loss": 0.6157620249986648,
+        "train samples": 14000,
+        "train time": 43.8277162159211,
+        "eval time": 10.279233909997856,
+        "tokens / sec": 4785.784387364565,
+        "mem allocated avg": 6796146567.168,
+        "mem reserved avg": 15755340939.264,
+        "elapsed time": 858.011582512001
+      },
+      {
+        "step": 3750,
+        "valid accuracy": 0.54,
+        "train loss": 0.6115566066503525,
+        "train samples": 15000,
+        "train time": 44.967253953713225,
+        "eval time": 13.822728877014015,
+        "tokens / sec": 4819.12905384576,
+        "mem allocated avg": 6807568994.304,
+        "mem reserved avg": 16073713778.688,
+        "elapsed time": 921.0013866620138
+      },
+      {
+        "step": 4000,
+        "valid accuracy": 0.38,
+        "train loss": 0.6212958168983459,
+        "train samples": 16000,
+        "train time": 43.98874276672723,
+        "eval time": 13.764712378993863,
+        "tokens / sec": 4646.029578153488,
+        "mem allocated avg": 6788096286.72,
+        "mem reserved avg": 15774408245.248,
+        "elapsed time": 983.0344113839965
+      },
+      {
+        "step": 4250,
+        "valid accuracy": 0.44,
+        "train loss": 0.6039810788631439,
+        "train samples": 17000,
+        "train time": 43.837045381893404,
+        "eval time": 10.33728496299591,
+        "tokens / sec": 4822.154370999484,
+        "mem allocated avg": 6799123298.304,
+        "mem reserved avg": 15757891076.096,
+        "elapsed time": 1041.2764754859963
+      },
+      {
+        "step": 4500,
+        "valid accuracy": 0.52,
+        "train loss": 0.6095903199911118,
+        "train samples": 18000,
+        "train time": 43.668030562083004,
+        "eval time": 7.992580848978832,
+        "tokens / sec": 4759.04219459919,
+        "mem allocated avg": 6794387621.888,
+        "mem reserved avg": 15730720374.784,
+        "elapsed time": 1097.2214920280094
+      },
+      {
+        "step": 4750,
+        "valid accuracy": 0.44,
+        "train loss": 0.6002804050445557,
+        "train samples": 19000,
+        "train time": 44.70480700198095,
+        "eval time": 8.954350112006068,
+        "tokens / sec": 4696.116907309257,
+        "mem allocated avg": 6796872839.168,
+        "mem reserved avg": 15848370601.984,
+        "elapsed time": 1155.0104628030094
+      },
+      {
+        "step": 5000,
+        "valid accuracy": 0.44,
+        "train loss": 0.6074805817604065,
+        "train samples": 20000,
+        "train time": 43.53095386302448,
+        "eval time": 9.616009549994487,
+        "tokens / sec": 4784.641307318437,
+        "mem allocated avg": 6793322434.56,
+        "mem reserved avg": 15480622415.872,
+        "elapsed time": 1212.3254529940023
+      },
+      {
+        "step": 5000,
+        "test accuracy": 0.4533737680060652,
+        "train loss": 0.6074805817604065,
+        "train samples": 20000,
+        "train total tokens": 4198051,
+        "forgetting": 0.12120223045349121
+      }
+    ]
+  },
+  "meta_info": {
+    "model_info": {
+      "sha": "13afe5124825b4f3751f836b40dafda64c1ed062",
+      "created_at": "2024-09-18T15:23:48+00:00"
+    },
+    "dataset_info": {
+      "metamath": {
+        "sha": "aa4f34d3d2d3231299b5b03d9b3e5a20da45aa18",
+        "created_at": "2023-09-21T17:22:46+00:00"
+      },
+      "gsm8k": {
+        "sha": "740312add88f781978c0658806c59bc2815b9866",
+        "created_at": "2022-04-12T10:22:10+00:00"
+      }
+    },
+    "package_info": {
+      "transformers-version": "5.12.1",
+      "transformers-commit-hash": null,
+      "peft-version": "0.20.1.dev0",
+      "peft-commit-hash": "c702de402330f48c0d8b1f260d59413aa3110b65",
+      "datasets-version": "4.8.4",
+      "datasets-commit-hash": null,
+      "bitsandbytes-version": "0.49.2",
+      "bitsandbytes-commit-hash": null,
+      "torch-version": "2.11.0+cu130",
+      "torch-commit-hash": null
+    },
+    "system_info": {
+      "system": "Linux",
+      "release": "7.0.0-1009-aws",
+      "version": "#9~24.04.1-Ubuntu SMP PREEMPT Mon Jul 13 22:43:41 UTC 2026",
+      "machine": "x86_64",
+      "processor": "x86_64",
+      "accelerator": "NVIDIA L40S"
+    },
+    "pytorch_info": "PyTorch built with:\n  - GCC 13.3\n  - C++ Version: 201703\n  - Intel(R) oneAPI Math Kernel Library Version 2024.2-Product Build 20240605 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v3.10.2 (Git Hash f1d471933dc852f956fd05389f9313c7148783d5)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 13.0\n  - NVCC architecture flags: -gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_120,code=sm_120\n  - CuDNN 90.7.1  (built against CUDA 12.8)\n    - Built with CuDNN 91.9\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, COMMIT_SHA=70d99e998b4955e0049d13a98d77ae1b14db1f45, CUDA_VERSION=13.0, CUDNN_VERSION=9.19.0, CXX_COMPILER=/opt/rh/gcc-toolset-13/root/usr/bin/c++, CXX_FLAGS= -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DLIBKINETO_NOXPUPTI=ON -DUSE_FBGEMM -DUSE_MSLK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -DC10_NODEPRECATED -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=range-loop-construct -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-unknown-pragmas -Wno-unused-parameter -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=old-style-cast -faligned-new -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-dangling-reference -Wno-error=dangling-reference -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, TORCH_VERSION=2.11.0, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, USE_XCCL=OFF, USE_XPU=OFF, \n"
+  }
+}

lokr--llama-3.2-3B-rank32-lr0.001--2026-08-05T16-41-19+00-00.json

@@ -0,0 +1,362 @@
+{
+  "run_info": {
+    "created_at": "2026-08-05T16:41:19+00:00",
+    "total_time": 1543.2893607050064,
+    "experiment_name": "lokr/llama-3.2-3B-rank32-lr0.001",
+    "peft_branch": "auto-research",
+    "train_config": {
+      "model_id": "meta-llama/Llama-3.2-3B",
+      "dtype": "bfloat16",
+      "max_seq_length": 768,
+      "batch_size": 4,
+      "batch_size_eval": 50,
+      "max_steps": 5000,
+      "eval_steps": 250,
+      "compile": false,
+      "use_gc": false,
+      "query_template": "Question: {query} Think step by step.\nAnswer:",
+      "seed": 0,
+      "grad_norm_clip": 1.0,
+      "optimizer_type": "AdamW",
+      "optimizer_kwargs": {
+        "lr": 0.001,
+        "weight_decay": 0.1
+      },
+      "lr_scheduler": "cosine",
+      "use_amp": false,
+      "autocast_adapter_dtype": true,
+      "generation_kwargs": {
+        "max_length": 800,
+        "max_new_tokens": 300
+      },
+      "attn_implementation": null,
+      "init_kv_cache_prefix": null
+    },
+    "peft_config": {
+      "task_type": null,
+      "peft_type": "LOKR",
+      "auto_mapping": null,
+      "peft_version": "0.20.1.dev0@UNKNOWN",
+      "base_model_name_or_path": "meta-llama/Llama-3.2-3B",
+      "revision": null,
+      "inference_mode": false,
+      "rank_pattern": {},
+      "alpha_pattern": {},
+      "r": 32,
+      "alpha": 64,
+      "rank_dropout": 0.0,
+      "module_dropout": 0.0,
+      "use_effective_conv2d": false,
+      "decompose_both": false,
+      "decompose_factor": -1,
+      "rank_dropout_scale": false,
+      "target_modules": [
+        "q_proj",
+        "v_proj"
+      ],
+      "exclude_modules": null,
+      "init_weights": true,
+      "layers_to_transform": null,
+      "layers_pattern": null,
+      "modules_to_save": null
+    },
+    "error_msg": ""
+  },
+  "train_info": {
+    "accelerator_memory_reserved_avg": 15793729726,
+    "accelerator_memory_max": 23559405568,
+    "accelerator_memory_reserved_99th": 21344813056,
+    "train_time": 1307.2431086740398,
+    "file_size": 1131984,
+    "num_trainable_params": 279552,
+    "num_total_params": 3213029376,
+    "status": "success",
+    "metrics": [
+      {
+        "step": 250,
+        "valid accuracy": 0.26,
+        "train loss": 0.9972863845825195,
+        "train samples": 1000,
+        "train time": 43.20129816728877,
+        "eval time": 13.620590639009606,
+        "tokens / sec": 4900.755509247862,
+        "mem allocated avg": 6785902575.616,
+        "mem reserved avg": 15916209274.88,
+        "elapsed time": 83.59115520500927
+      },
+      {
+        "step": 500,
+        "valid accuracy": 0.42,
+        "train loss": 0.7273374778032303,
+        "train samples": 2000,
+        "train time": 43.818635235045804,
+        "eval time": 13.440946741000516,
+        "tokens / sec": 4746.724741295621,
+        "mem allocated avg": 6778857619.456,
+        "mem reserved avg": 15578769129.472,
+        "elapsed time": 145.04160515699186
+      },
+      {
+        "step": 750,
+        "valid accuracy": 0.28,
+        "train loss": 0.6921107683181763,
+        "train samples": 3000,
+        "train time": 45.18761947474559,
+        "eval time": 11.329180876986356,
+        "tokens / sec": 4744.684550595196,
+        "mem allocated avg": 6789001691.136,
+        "mem reserved avg": 15775121276.928,
+        "elapsed time": 205.6760063529946
+      },
+      {
+        "step": 1000,
+        "valid accuracy": 0.4,
+        "train loss": 0.6745790269374847,
+        "train samples": 4000,
+        "train time": 44.184205388010014,
+        "eval time": 13.444282117998227,
+        "tokens / sec": 4715.169101050187,
+        "mem allocated avg": 6779753209.856,
+        "mem reserved avg": 15797661466.624,
+        "elapsed time": 267.54085831699194
+      },
+      {
+        "step": 1250,
+        "valid accuracy": 0.36,
+        "train loss": 0.6729275505542756,
+        "train samples": 5000,
+        "train time": 43.49395537111559,
+        "eval time": 13.46411624699249,
+        "tokens / sec": 4794.643260669975,
+        "mem allocated avg": 6780462469.12,
+        "mem reserved avg": 15846902595.584,
+        "elapsed time": 328.592146682
+      },
+      {
+        "step": 1500,
+        "valid accuracy": 0.34,
+        "train loss": 0.6668304501771927,
+        "train samples": 6000,
+        "train time": 44.26632965187309,
+        "eval time": 13.452810667018639,
+        "tokens / sec": 4728.89895426743,
+        "mem allocated avg": 6781956315.136,
+        "mem reserved avg": 15751717060.608,
+        "elapsed time": 390.52621169699705
+      },
+      {
+        "step": 1750,
+        "valid accuracy": 0.32,
+        "train loss": 0.6592434631586075,
+        "train samples": 7000,
+        "train time": 43.340643761039246,
+        "eval time": 13.473839386017062,
+        "tokens / sec": 4830.454322605105,
+        "mem allocated avg": 6783807119.36,
+        "mem reserved avg": 16151031578.624,
+        "elapsed time": 451.5246996629867
+      },
+      {
+        "step": 2000,
+        "valid accuracy": 0.36,
+        "train loss": 0.661458988070488,
+        "train samples": 8000,
+        "train time": 43.138098473020364,
+        "eval time": 7.384789164992981,
+        "tokens / sec": 4814.676755627006,
+        "mem allocated avg": 6779387142.144,
+        "mem reserved avg": 15765700870.144,
+        "elapsed time": 506.28373629099224
+      },
+      {
+        "step": 2250,
+        "valid accuracy": 0.4,
+        "train loss": 0.6549151700735092,
+        "train samples": 9000,
+        "train time": 45.04773759498494,
+        "eval time": 8.135625335009536,
+        "tokens / sec": 4771.560381845451,
+        "mem allocated avg": 6790785810.432,
+        "mem reserved avg": 16084132429.824,
+        "elapsed time": 563.5650068520044
+      },
+      {
+        "step": 2500,
+        "valid accuracy": 0.4,
+        "train loss": 0.6524245887994766,
+        "train samples": 10000,
+        "train time": 42.58069601622992,
+        "eval time": 8.95897142199101,
+        "tokens / sec": 4837.098010833225,
+        "mem allocated avg": 6777116250.112,
+        "mem reserved avg": 15533898465.28,
+        "elapsed time": 619.2634280459897
+      },
+      {
+        "step": 2750,
+        "valid accuracy": 0.44,
+        "train loss": 0.6446445236206054,
+        "train samples": 11000,
+        "train time": 43.804762633895734,
+        "eval time": 11.028623944002902,
+        "tokens / sec": 4836.939804258826,
+        "mem allocated avg": 6786684753.92,
+        "mem reserved avg": 15927106076.672,
+        "elapsed time": 678.1798055369873
+      },
+      {
+        "step": 3000,
+        "valid accuracy": 0.5,
+        "train loss": 0.636047769188881,
+        "train samples": 12000,
+        "train time": 44.29176679896773,
+        "eval time": 11.020784683001693,
+        "tokens / sec": 4712.63657075122,
+        "mem allocated avg": 6782002194.432,
+        "mem reserved avg": 15856759209.984,
+        "elapsed time": 737.7226473859919
+      },
+      {
+        "step": 3250,
+        "valid accuracy": 0.36,
+        "train loss": 0.6453784251213074,
+        "train samples": 13000,
+        "train time": 43.889022187067894,
+        "eval time": 13.485144512989791,
+        "tokens / sec": 4805.3246459007005,
+        "mem allocated avg": 6785020166.144,
+        "mem reserved avg": 15728002465.792,
+        "elapsed time": 799.1354676230112
+      },
+      {
+        "step": 3500,
+        "valid accuracy": 0.44,
+        "train loss": 0.6287569386959075,
+        "train samples": 14000,
+        "train time": 43.88188967798487,
+        "eval time": 13.49879609598429,
+        "tokens / sec": 4779.876198112534,
+        "mem allocated avg": 6781960970.24,
+        "mem reserved avg": 15699607027.712,
+        "elapsed time": 860.7000853430072
+      },
+      {
+        "step": 3750,
+        "valid accuracy": 0.46,
+        "train loss": 0.6270692372322082,
+        "train samples": 15000,
+        "train time": 44.66675748990383,
+        "eval time": 9.188075685990043,
+        "tokens / sec": 4851.549836564297,
+        "mem allocated avg": 6793733603.328,
+        "mem reserved avg": 16043883888.64,
+        "elapsed time": 918.7074330459873
+      },
+      {
+        "step": 4000,
+        "valid accuracy": 0.38,
+        "train loss": 0.6401721099615097,
+        "train samples": 16000,
+        "train time": 42.885482566198334,
+        "eval time": 13.488256345008267,
+        "tokens / sec": 4765.552064956443,
+        "mem allocated avg": 6774095478.784,
+        "mem reserved avg": 15747464036.352,
+        "elapsed time": 979.268849782995
+      },
+      {
+        "step": 4250,
+        "valid accuracy": 0.48,
+        "train loss": 0.6243705161809922,
+        "train samples": 17000,
+        "train time": 43.78779757398297,
+        "eval time": 13.449544289993355,
+        "tokens / sec": 4827.5778118970575,
+        "mem allocated avg": 6785076533.248,
+        "mem reserved avg": 15712022167.552,
+        "elapsed time": 1040.4841427920037
+      },
+      {
+        "step": 4500,
+        "valid accuracy": 0.46,
+        "train loss": 0.631668661236763,
+        "train samples": 18000,
+        "train time": 43.95947218083893,
+        "eval time": 13.468623130000196,
+        "tokens / sec": 4727.490792998734,
+        "mem allocated avg": 6780526002.176,
+        "mem reserved avg": 15685514166.272,
+        "elapsed time": 1102.1798365190043
+      },
+      {
+        "step": 4750,
+        "valid accuracy": 0.42,
+        "train loss": 0.6229247853755951,
+        "train samples": 19000,
+        "train time": 44.09635784616694,
+        "eval time": 8.945634042000165,
+        "tokens / sec": 4760.914738863153,
+        "mem allocated avg": 6783016738.816,
+        "mem reserved avg": 15821812269.056,
+        "elapsed time": 1159.333109217987
+      },
+      {
+        "step": 5000,
+        "valid accuracy": 0.46,
+        "train loss": 0.6298999850749969,
+        "train samples": 20000,
+        "train time": 43.606315371143864,
+        "eval time": 8.17935786198359,
+        "tokens / sec": 4776.372372379521,
+        "mem allocated avg": 6779397117.952,
+        "mem reserved avg": 15451279065.088,
+        "elapsed time": 1215.2514337110042
+      },
+      {
+        "step": 5000,
+        "test accuracy": 0.4116755117513268,
+        "train loss": 0.6298999850749969,
+        "train samples": 20000,
+        "train total tokens": 4198051,
+        "forgetting": 0.14964866638183594
+      }
+    ]
+  },
+  "meta_info": {
+    "model_info": {
+      "sha": "13afe5124825b4f3751f836b40dafda64c1ed062",
+      "created_at": "2024-09-18T15:23:48+00:00"
+    },
+    "dataset_info": {
+      "metamath": {
+        "sha": "aa4f34d3d2d3231299b5b03d9b3e5a20da45aa18",
+        "created_at": "2023-09-21T17:22:46+00:00"
+      },
+      "gsm8k": {
+        "sha": "740312add88f781978c0658806c59bc2815b9866",
+        "created_at": "2022-04-12T10:22:10+00:00"
+      }
+    },
+    "package_info": {
+      "transformers-version": "5.12.1",
+      "transformers-commit-hash": null,
+      "peft-version": "0.20.1.dev0",
+      "peft-commit-hash": "c702de402330f48c0d8b1f260d59413aa3110b65",
+      "datasets-version": "4.8.4",
+      "datasets-commit-hash": null,
+      "bitsandbytes-version": "0.49.2",
+      "bitsandbytes-commit-hash": null,
+      "torch-version": "2.11.0+cu130",
+      "torch-commit-hash": null
+    },
+    "system_info": {
+      "system": "Linux",
+      "release": "7.0.0-1009-aws",
+      "version": "#9~24.04.1-Ubuntu SMP PREEMPT Mon Jul 13 22:43:41 UTC 2026",
+      "machine": "x86_64",
+      "processor": "x86_64",
+      "accelerator": "NVIDIA L40S"
+    },
+    "pytorch_info": "PyTorch built with:\n  - GCC 13.3\n  - C++ Version: 201703\n  - Intel(R) oneAPI Math Kernel Library Version 2024.2-Product Build 20240605 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v3.10.2 (Git Hash f1d471933dc852f956fd05389f9313c7148783d5)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 13.0\n  - NVCC architecture flags: -gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_120,code=sm_120\n  - CuDNN 90.7.1  (built against CUDA 12.8)\n    - Built with CuDNN 91.9\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, COMMIT_SHA=70d99e998b4955e0049d13a98d77ae1b14db1f45, CUDA_VERSION=13.0, CUDNN_VERSION=9.19.0, CXX_COMPILER=/opt/rh/gcc-toolset-13/root/usr/bin/c++, CXX_FLAGS= -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DLIBKINETO_NOXPUPTI=ON -DUSE_FBGEMM -DUSE_MSLK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -DC10_NODEPRECATED -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=range-loop-construct -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-unknown-pragmas -Wno-unused-parameter -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=old-style-cast -faligned-new -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-dangling-reference -Wno-error=dangling-reference -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, TORCH_VERSION=2.11.0, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, USE_XCCL=OFF, USE_XPU=OFF, \n"
+  }
+}