qwen3vl8b_full_v1: checkpoint-600

judge4/qwen3vl8b_full_v1/checkpoint-600/config.json

@@ -0,0 +1,66 @@
+{
+  "architectures": [
+    "Qwen3VLForConditionalGeneration"
+  ],
+  "dtype": "bfloat16",
+  "image_token_id": 151655,
+  "model_type": "qwen3_vl",
+  "text_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "bos_token_id": 151643,
+    "dtype": "bfloat16",
+    "eos_token_id": 151645,
+    "head_dim": 128,
+    "hidden_act": "silu",
+    "hidden_size": 4096,
+    "initializer_range": 0.02,
+    "intermediate_size": 12288,
+    "max_position_embeddings": 262144,
+    "model_type": "qwen3_vl_text",
+    "num_attention_heads": 32,
+    "num_hidden_layers": 36,
+    "num_key_value_heads": 8,
+    "pad_token_id": null,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "mrope_interleaved": true,
+      "mrope_section": [
+        24,
+        20,
+        20
+      ],
+      "rope_theta": 5000000,
+      "rope_type": "default"
+    },
+    "use_cache": true,
+    "vocab_size": 151936
+  },
+  "tie_word_embeddings": false,
+  "transformers_version": "5.10.2",
+  "use_cache": false,
+  "video_token_id": 151656,
+  "vision_config": {
+    "deepstack_visual_indexes": [
+      8,
+      16,
+      24
+    ],
+    "depth": 27,
+    "dtype": "bfloat16",
+    "hidden_act": "gelu_pytorch_tanh",
+    "hidden_size": 1152,
+    "in_channels": 3,
+    "initializer_range": 0.02,
+    "intermediate_size": 4304,
+    "model_type": "qwen3_vl_vision",
+    "num_heads": 16,
+    "num_position_embeddings": 2304,
+    "out_hidden_size": 4096,
+    "patch_size": 16,
+    "spatial_merge_size": 2,
+    "temporal_patch_size": 2
+  },
+  "vision_end_token_id": 151653,
+  "vision_start_token_id": 151652
+}

judge4/qwen3vl8b_full_v1/checkpoint-600/generation_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "repetition_penalty": 1.0,
+  "temperature": 0.7,
+  "top_k": 20,
+  "top_p": 0.8,
+  "transformers_version": "5.10.2"
+}

judge4/qwen3vl8b_full_v1/checkpoint-600/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96d2581260d403ff2f2d74da94f2fcaee18f61f3127eee2747387e41b6c1b289
+size 17534340584

judge4/qwen3vl8b_full_v1/checkpoint-600/trainer_state.json

@@ -0,0 +1,490 @@
+{
+  "best_global_step": 200,
+  "best_metric": 0.986340126194396,
+  "best_model_checkpoint": "/cache/hyeongjinkim/runs/yesno/qwen3vl8b/checkpoint-200",
+  "epoch": 0.32089851584436424,
+  "eval_steps": 200,
+  "global_step": 600,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.00534830859740607,
+      "grad_norm": 115.39302942622305,
+      "learning_rate": 1.5789473684210526e-06,
+      "loss": 0.6299215316772461,
+      "step": 10
+    },
+    {
+      "epoch": 0.01069661719481214,
+      "grad_norm": 39.09323838946906,
+      "learning_rate": 3.3333333333333333e-06,
+      "loss": 0.5217976570129395,
+      "step": 20
+    },
+    {
+      "epoch": 0.016044925792218213,
+      "grad_norm": 58.33327100367131,
+      "learning_rate": 5.087719298245615e-06,
+      "loss": 0.36051628589630125,
+      "step": 30
+    },
+    {
+      "epoch": 0.02139323438962428,
+      "grad_norm": 31.360935036252677,
+      "learning_rate": 6.842105263157896e-06,
+      "loss": 0.3351165771484375,
+      "step": 40
+    },
+    {
+      "epoch": 0.026741542987030353,
+      "grad_norm": 11.148106043175956,
+      "learning_rate": 8.596491228070176e-06,
+      "loss": 0.32320551872253417,
+      "step": 50
+    },
+    {
+      "epoch": 0.032089851584436425,
+      "grad_norm": 93.48741175980791,
+      "learning_rate": 9.999969973569808e-06,
+      "loss": 0.2841494083404541,
+      "step": 60
+    },
+    {
+      "epoch": 0.03743816018184249,
+      "grad_norm": 24.041199474493364,
+      "learning_rate": 9.99891908637917e-06,
+      "loss": 0.31807770729064944,
+      "step": 70
+    },
+    {
+      "epoch": 0.04278646877924856,
+      "grad_norm": 17.28784245820742,
+      "learning_rate": 9.996367238293978e-06,
+      "loss": 0.25719327926635743,
+      "step": 80
+    },
+    {
+      "epoch": 0.048134777376654635,
+      "grad_norm": 18.79954418521905,
+      "learning_rate": 9.992315195524705e-06,
+      "loss": 0.16733503341674805,
+      "step": 90
+    },
+    {
+      "epoch": 0.05348308597406071,
+      "grad_norm": 7.718337979076894,
+      "learning_rate": 9.98676417472592e-06,
+      "loss": 0.30640153884887694,
+      "step": 100
+    },
+    {
+      "epoch": 0.05883139457146677,
+      "grad_norm": 18.87648077862634,
+      "learning_rate": 9.979715842630971e-06,
+      "loss": 0.23051707744598388,
+      "step": 110
+    },
+    {
+      "epoch": 0.06417970316887285,
+      "grad_norm": 53.620896521195476,
+      "learning_rate": 9.971172315551539e-06,
+      "loss": 0.315609073638916,
+      "step": 120
+    },
+    {
+      "epoch": 0.06952801176627892,
+      "grad_norm": 6.175189141785783,
+      "learning_rate": 9.961136158742194e-06,
+      "loss": 0.2716816902160645,
+      "step": 130
+    },
+    {
+      "epoch": 0.07487632036368498,
+      "grad_norm": 3.6987055466078678,
+      "learning_rate": 9.94961038563017e-06,
+      "loss": 0.17460014820098876,
+      "step": 140
+    },
+    {
+      "epoch": 0.08022462896109106,
+      "grad_norm": 92.14918625907542,
+      "learning_rate": 9.936598456910553e-06,
+      "loss": 0.1968517541885376,
+      "step": 150
+    },
+    {
+      "epoch": 0.08557293755849713,
+      "grad_norm": 15.119460868723982,
+      "learning_rate": 9.922104279507188e-06,
+      "loss": 0.23519439697265626,
+      "step": 160
+    },
+    {
+      "epoch": 0.09092124615590319,
+      "grad_norm": 15.019433747757887,
+      "learning_rate": 9.90613220539959e-06,
+      "loss": 0.24371695518493652,
+      "step": 170
+    },
+    {
+      "epoch": 0.09626955475330927,
+      "grad_norm": 5.870677605808927,
+      "learning_rate": 9.888687030316244e-06,
+      "loss": 0.1655200958251953,
+      "step": 180
+    },
+    {
+      "epoch": 0.10161786335071533,
+      "grad_norm": 38.43177000299922,
+      "learning_rate": 9.869773992294645e-06,
+      "loss": 0.1303710460662842,
+      "step": 190
+    },
+    {
+      "epoch": 0.10696617194812141,
+      "grad_norm": 3.4771977179980036,
+      "learning_rate": 9.849398770108534e-06,
+      "loss": 0.1649288058280945,
+      "step": 200
+    },
+    {
+      "epoch": 0.10696617194812141,
+      "eval_auroc": 0.986340126194396,
+      "eval_loss": 0.1778445839881897,
+      "eval_runtime": 162.8402,
+      "eval_samples_per_second": 32.566,
+      "eval_steps_per_second": 2.039,
+      "step": 200
+    },
+    {
+      "epoch": 0.11231448054552748,
+      "grad_norm": 17.78250131729676,
+      "learning_rate": 9.82756748156282e-06,
+      "loss": 0.26313281059265137,
+      "step": 210
+    },
+    {
+      "epoch": 0.11766278914293354,
+      "grad_norm": 16.167615574624104,
+      "learning_rate": 9.804286681656661e-06,
+      "loss": 0.1850219964981079,
+      "step": 220
+    },
+    {
+      "epoch": 0.12301109774033962,
+      "grad_norm": 16.202819227774476,
+      "learning_rate": 9.779563360615269e-06,
+      "loss": 0.19695448875427246,
+      "step": 230
+    },
+    {
+      "epoch": 0.1283594063377457,
+      "grad_norm": 4.679400123941397,
+      "learning_rate": 9.753404941791063e-06,
+      "loss": 0.19695805311203002,
+      "step": 240
+    },
+    {
+      "epoch": 0.13370771493515177,
+      "grad_norm": 22.048169793689425,
+      "learning_rate": 9.725819279434736e-06,
+      "loss": 0.12451494932174682,
+      "step": 250
+    },
+    {
+      "epoch": 0.13905602353255783,
+      "grad_norm": 16.514362356407883,
+      "learning_rate": 9.696814656336974e-06,
+      "loss": 0.13828051090240479,
+      "step": 260
+    },
+    {
+      "epoch": 0.1444043321299639,
+      "grad_norm": 10.880812314019147,
+      "learning_rate": 9.666399781341487e-06,
+      "loss": 0.16824707984924317,
+      "step": 270
+    },
+    {
+      "epoch": 0.14975264072736996,
+      "grad_norm": 13.594908068539505,
+      "learning_rate": 9.63458378673011e-06,
+      "loss": 0.16994980573654175,
+      "step": 280
+    },
+    {
+      "epoch": 0.15510094932477603,
+      "grad_norm": 13.979685146852107,
+      "learning_rate": 9.60137622548078e-06,
+      "loss": 0.12250713109970093,
+      "step": 290
+    },
+    {
+      "epoch": 0.16044925792218212,
+      "grad_norm": 23.1890145595953,
+      "learning_rate": 9.566787068399182e-06,
+      "loss": 0.31440751552581786,
+      "step": 300
+    },
+    {
+      "epoch": 0.16579756651958819,
+      "grad_norm": 0.7897764763294275,
+      "learning_rate": 9.53082670112494e-06,
+      "loss": 0.15762845277786255,
+      "step": 310
+    },
+    {
+      "epoch": 0.17114587511699425,
+      "grad_norm": 2.8159576992499646,
+      "learning_rate": 9.493505921013264e-06,
+      "loss": 0.19779951572418214,
+      "step": 320
+    },
+    {
+      "epoch": 0.17649418371440032,
+      "grad_norm": 12.467613821043953,
+      "learning_rate": 9.454835933892961e-06,
+      "loss": 0.22400574684143065,
+      "step": 330
+    },
+    {
+      "epoch": 0.18184249231180638,
+      "grad_norm": 25.616160480376983,
+      "learning_rate": 9.414828350701808e-06,
+      "loss": 0.2321730375289917,
+      "step": 340
+    },
+    {
+      "epoch": 0.18719080090921247,
+      "grad_norm": 1.3969530235405139,
+      "learning_rate": 9.37349518400028e-06,
+      "loss": 0.1542872667312622,
+      "step": 350
+    },
+    {
+      "epoch": 0.19253910950661854,
+      "grad_norm": 15.068102705852885,
+      "learning_rate": 9.330848844364709e-06,
+      "loss": 0.17226810455322267,
+      "step": 360
+    },
+    {
+      "epoch": 0.1978874181040246,
+      "grad_norm": 5.7792217561834835,
+      "learning_rate": 9.2869021366609e-06,
+      "loss": 0.1545494318008423,
+      "step": 370
+    },
+    {
+      "epoch": 0.20323572670143067,
+      "grad_norm": 71.6843121958736,
+      "learning_rate": 9.241668256199392e-06,
+      "loss": 0.1899905562400818,
+      "step": 380
+    },
+    {
+      "epoch": 0.20858403529883673,
+      "grad_norm": 21.6463524086354,
+      "learning_rate": 9.195160784773471e-06,
+      "loss": 0.07489174604415894,
+      "step": 390
+    },
+    {
+      "epoch": 0.21393234389624283,
+      "grad_norm": 0.13359985882796035,
+      "learning_rate": 9.14739368658114e-06,
+      "loss": 0.09913319945335389,
+      "step": 400
+    },
+    {
+      "epoch": 0.21393234389624283,
+      "eval_auroc": 0.983682796757702,
+      "eval_loss": 0.2364487200975418,
+      "eval_runtime": 159.5655,
+      "eval_samples_per_second": 33.234,
+      "eval_steps_per_second": 2.081,
+      "step": 400
+    },
+    {
+      "epoch": 0.2192806524936489,
+      "grad_norm": 62.19974030049211,
+      "learning_rate": 9.098381304032267e-06,
+      "loss": 0.2020780086517334,
+      "step": 410
+    },
+    {
+      "epoch": 0.22462896109105496,
+      "grad_norm": 9.247102421418825,
+      "learning_rate": 9.048138353442173e-06,
+      "loss": 0.11548291444778443,
+      "step": 420
+    },
+    {
+      "epoch": 0.22997726968846102,
+      "grad_norm": 22.970689639607503,
+      "learning_rate": 8.996679920612964e-06,
+      "loss": 0.11144022941589356,
+      "step": 430
+    },
+    {
+      "epoch": 0.2353255782858671,
+      "grad_norm": 1.4437964908827596,
+      "learning_rate": 8.944021456303898e-06,
+      "loss": 0.10462523698806762,
+      "step": 440
+    },
+    {
+      "epoch": 0.24067388688327315,
+      "grad_norm": 1.4696765282633264,
+      "learning_rate": 8.890178771592198e-06,
+      "loss": 0.16833703517913817,
+      "step": 450
+    },
+    {
+      "epoch": 0.24602219548067925,
+      "grad_norm": 10.334755827866527,
+      "learning_rate": 8.835168033125657e-06,
+      "loss": 0.15125012397766113,
+      "step": 460
+    },
+    {
+      "epoch": 0.2513705040780853,
+      "grad_norm": 1.1380018464880775,
+      "learning_rate": 8.779005758268486e-06,
+      "loss": 0.10879876613616943,
+      "step": 470
+    },
+    {
+      "epoch": 0.2567188126754914,
+      "grad_norm": 27.554855927355863,
+      "learning_rate": 8.721708810141862e-06,
+      "loss": 0.10096737146377563,
+      "step": 480
+    },
+    {
+      "epoch": 0.26206712127289744,
+      "grad_norm": 10.67291054191312,
+      "learning_rate": 8.663294392560658e-06,
+      "loss": 0.13645027875900267,
+      "step": 490
+    },
+    {
+      "epoch": 0.26741542987030353,
+      "grad_norm": 11.160169426656719,
+      "learning_rate": 8.603780044867866e-06,
+      "loss": 0.13505263328552247,
+      "step": 500
+    },
+    {
+      "epoch": 0.27276373846770957,
+      "grad_norm": 1.7311663016149148,
+      "learning_rate": 8.543183636668302e-06,
+      "loss": 0.09905856847763062,
+      "step": 510
+    },
+    {
+      "epoch": 0.27811204706511566,
+      "grad_norm": 12.491616112911325,
+      "learning_rate": 8.481523362463111e-06,
+      "loss": 0.13770177364349365,
+      "step": 520
+    },
+    {
+      "epoch": 0.2834603556625217,
+      "grad_norm": 17.56462162498995,
+      "learning_rate": 8.418817736186751e-06,
+      "loss": 0.06779674887657165,
+      "step": 530
+    },
+    {
+      "epoch": 0.2888086642599278,
+      "grad_norm": 14.317316633123855,
+      "learning_rate": 8.355085585648054e-06,
+      "loss": 0.1754995107650757,
+      "step": 540
+    },
+    {
+      "epoch": 0.2941569728573339,
+      "grad_norm": 12.736983924949136,
+      "learning_rate": 8.290346046877033e-06,
+      "loss": 0.05573989748954773,
+      "step": 550
+    },
+    {
+      "epoch": 0.2995052814547399,
+      "grad_norm": 0.6150496518993013,
+      "learning_rate": 8.224618558379166e-06,
+      "loss": 0.05098409652709961,
+      "step": 560
+    },
+    {
+      "epoch": 0.304853590052146,
+      "grad_norm": 0.06977215369259143,
+      "learning_rate": 8.157922855298834e-06,
+      "loss": 0.06332494020462036,
+      "step": 570
+    },
+    {
+      "epoch": 0.31020189864955205,
+      "grad_norm": 5.211076096702457,
+      "learning_rate": 8.090278963493714e-06,
+      "loss": 0.1354996919631958,
+      "step": 580
+    },
+    {
+      "epoch": 0.31555020724695815,
+      "grad_norm": 81.93895874357261,
+      "learning_rate": 8.021707193521865e-06,
+      "loss": 0.05751166939735412,
+      "step": 590
+    },
+    {
+      "epoch": 0.32089851584436424,
+      "grad_norm": 1.8276016342124084,
+      "learning_rate": 7.952228134543357e-06,
+      "loss": 0.20583314895629884,
+      "step": 600
+    },
+    {
+      "epoch": 0.32089851584436424,
+      "eval_auroc": 0.9816760537862258,
+      "eval_loss": 0.2224961817264557,
+      "eval_runtime": 164.6281,
+      "eval_samples_per_second": 32.212,
+      "eval_steps_per_second": 2.017,
+      "step": 600
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 1870,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 200,
+  "stateful_callbacks": {
+    "EarlyStoppingCallback": {
+      "args": {
+        "early_stopping_patience": 5,
+        "early_stopping_threshold": 0.0
+      },
+      "attributes": {
+        "early_stopping_patience_counter": 2
+      }
+    },
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 28208346169344.0,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}

judge4/qwen3vl8b_full_v1/checkpoint-600/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5a6f4bcee28aa57305c334450d07345e21a06095708b28432288e4392d63148
+size 6673

judge4/qwen3vl8b_full_v1/checkpoint-600/zero_to_torch.py

@@ -0,0 +1,54 @@
+#!/usr/bin/env python
+
+# SPDX-License-Identifier: Apache-2.0
+# DeepSpeed Team
+
+import argparse
+
+if __package__:
+    from . import zero_to_fp32
+else:
+    import zero_to_fp32
+
+
+def main(args=None):
+    parser = argparse.ArgumentParser(
+        description="Convert a DeepSpeed ZeRO checkpoint to float32, float16, or bfloat16 PyTorch weights.")
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument("output_dir", type=str, help="directory for the converted PyTorch state_dict files")
+    parser.add_argument("--dtype",
+                        type=str,
+                        choices=sorted(zero_to_fp32.OUTPUT_DTYPE_NAMES),
+                        required=True,
+                        help="output tensor dtype")
+    parser.add_argument("--max_shard_size",
+                        type=str,
+                        default="5GB",
+                        help="maximum size of each checkpoint shard, such as 5GB or 500MB")
+    parser.add_argument("--safe_serialization",
+                        default=False,
+                        action='store_true',
+                        help="save with safetensors instead of PyTorch pickle serialization")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier, e.g., global_step1")
+    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug output")
+    parsed_args = parser.parse_args(args)
+
+    zero_to_fp32.debug = parsed_args.debug
+    zero_to_fp32.convert_zero_checkpoint_to_state_dict(parsed_args.checkpoint_dir,
+                                                       parsed_args.output_dir,
+                                                       dtype=parsed_args.dtype,
+                                                       max_shard_size=parsed_args.max_shard_size,
+                                                       safe_serialization=parsed_args.safe_serialization,
+                                                       tag=parsed_args.tag,
+                                                       exclude_frozen_parameters=parsed_args.exclude_frozen_parameters)
+
+
+if __name__ == "__main__":
+    main()