Add files using upload-large-folder tool

LTA_openwebtext_dualt/logs/lta_lm1b_logisticnormal_linearmean_categorical_fullvocab_c1024_fullycoupled_flmpack_onehot_hardce_ddit_small_len128_gbs512_4gpu_1m_nw0.launcher.log

@@ -0,0 +1,127 @@
+[launch] method=logisticnormal_linearmean_categorical_fullvocab_c1024_fullycoupled host=di-20260411014000-djqhq time=2026-05-07T18:00:05+00:00
+[launch] cwd=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt
+[launch] run_name=lta_lm1b_logisticnormal_linearmean_categorical_fullvocab_c1024_fullycoupled_flmpack_onehot_hardce_ddit_small_len128_gbs512_4gpu_1m_nw0
+[launch] save_dir=runs/lta_lm1b_logisticnormal_linearmean_categorical_fullvocab_c1024_fullycoupled_flmpack_onehot_hardce_ddit_small_len128_gbs512_4gpu_1m_nw0
+[launch] log_file=logs/lta_lm1b_logisticnormal_linearmean_categorical_fullvocab_c1024_fullycoupled_flmpack_onehot_hardce_ddit_small_len128_gbs512_4gpu_1m_nw0.log
+
+*****************************************
+Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
+*****************************************
+NCCL version 2.25.1+cuda12.8
+{
+  "device": "cuda:0",
+  "rank": 0,
+  "world_size": 4,
+  "samples": "wrapped_stream",
+  "vocab_size": 30522,
+  "save_dir": "runs/lta_lm1b_logisticnormal_linearmean_categorical_fullvocab_c1024_fullycoupled_flmpack_onehot_hardce_ddit_small_len128_gbs512_4gpu_1m_nw0",
+  "batch_size": 64,
+  "grad_accum": 2,
+  "effective_batch_size": 512,
+  "global_batch_size": 512,
+  "lr_schedule": "constant_warmup",
+  "warmup_steps": 2500,
+  "adam_beta1": 0.9,
+  "adam_beta2": 0.999,
+  "adam_eps": 1e-08,
+  "model_type": "ddit",
+  "dual_t": true,
+  "corrupt_t_mode": "same",
+  "corrupt_min_t": 0.0,
+  "corrupt_max_t": 1.0,
+  "dirichlet_endpoint_mode": "categorical_dual_t",
+  "dirichlet_semantic_t_mode": "same",
+  "dirichlet_semantic_t_value": 0.0,
+  "categorical_wrong_from_full_vocab": true,
+  "simplex_bridge_sampler": "logistic_normal_linear_mean",
+  "logistic_normal_sigma_min": 0.18,
+  "logistic_normal_sigma_max": 2.2,
+  "logistic_normal_tau_min": 0.65,
+  "logistic_normal_tau_max": 1.15,
+  "torch_compile": false,
+  "compile_mode": "max-autotune",
+  "state_format": "prob",
+  "target_loss": "hard_ce",
+  "meanflow_weight": 0.0,
+  "bridge_noise_init": "logistic_normal",
+  "noise_sigma": -1.0,
+  "wrap": true,
+  "wrap_mode": "stream",
+  "wrap_record_buffer_size": 200,
+  "openwebtext_split": "all",
+  "detokenizer": "auto",
+  "resolved_detokenizer": "lm1b",
+  "num_workers": 0,
+  "latest_every": 1000,
+  "resume_path": ""
+}
+step=100 micro_steps=200 elapsed=33.7s lr=1.212000e-05 loss_all=10.1645 acc_all=0.5433 loss_corrupt=10.1710 acc_corrupt=0.3656 corrupt_frac=0.5489 loss=10.1710 loss_recon=10.1710 loss_meanflow=0.0000 mean_model_t=0.5004 mean_corrupt_t=0.5004 wrong_frac=0.4963 init_acc_corrupt=0.4923 init_gold_top10=0.4993 init_gold_top100=0.5039
+step=200 micro_steps=400 elapsed=32.1s lr=2.412000e-05 loss_all=8.9418 acc_all=0.1305 loss_corrupt=8.9447 acc_corrupt=0.0986 corrupt_frac=0.5503 loss=8.9447 loss_recon=8.9447 loss_meanflow=0.0000 mean_model_t=0.5021 mean_corrupt_t=0.5021 wrong_frac=0.4978 init_acc_corrupt=0.4912 init_gold_top10=0.4979 init_gold_top100=0.5023
+step=300 micro_steps=600 elapsed=32.1s lr=3.612000e-05 loss_all=6.9768 acc_all=0.1485 loss_corrupt=7.0881 acc_corrupt=0.1136 corrupt_frac=0.5541 loss=7.0881 loss_recon=7.0881 loss_meanflow=0.0000 mean_model_t=0.5008 mean_corrupt_t=0.5008 wrong_frac=0.4990 init_acc_corrupt=0.4898 init_gold_top10=0.4966 init_gold_top100=0.5011
+step=400 micro_steps=800 elapsed=32.2s lr=4.812000e-05 loss_all=3.6665 acc_all=0.5435 loss_corrupt=4.7788 acc_corrupt=0.3888 corrupt_frac=0.5525 loss=4.7788 loss_recon=4.7788 loss_meanflow=0.0000 mean_model_t=0.5025 mean_corrupt_t=0.5025 wrong_frac=0.4996 init_acc_corrupt=0.4892 init_gold_top10=0.4960 init_gold_top100=0.5005
+step=500 micro_steps=1000 elapsed=32.2s lr=6.012000e-05 loss_all=2.3262 acc_all=0.7071 loss_corrupt=3.7839 acc_corrupt=0.5148 corrupt_frac=0.5524 loss=3.7839 loss_recon=3.7839 loss_meanflow=0.0000 mean_model_t=0.4993 mean_corrupt_t=0.4993 wrong_frac=0.5015 init_acc_corrupt=0.4869 init_gold_top10=0.4939 init_gold_top100=0.4985
+step=600 micro_steps=1200 elapsed=32.2s lr=7.212000e-05 loss_all=2.0390 acc_all=0.7353 loss_corrupt=3.4621 acc_corrupt=0.5424 corrupt_frac=0.5500 loss=3.4621 loss_recon=3.4621 loss_meanflow=0.0000 mean_model_t=0.4995 mean_corrupt_t=0.4995 wrong_frac=0.5005 init_acc_corrupt=0.4882 init_gold_top10=0.4950 init_gold_top100=0.4995
+step=700 micro_steps=1400 elapsed=32.2s lr=8.412000e-05 loss_all=1.9037 acc_all=0.7466 loss_corrupt=3.2809 acc_corrupt=0.5560 corrupt_frac=0.5486 loss=3.2809 loss_recon=3.2809 loss_meanflow=0.0000 mean_model_t=0.4990 mean_corrupt_t=0.4990 wrong_frac=0.5002 init_acc_corrupt=0.4884 init_gold_top10=0.4953 init_gold_top100=0.4998
+step=800 micro_steps=1600 elapsed=32.1s lr=9.612000e-05 loss_all=1.8368 acc_all=0.7513 loss_corrupt=3.1706 acc_corrupt=0.5644 corrupt_frac=0.5505 loss=3.1706 loss_recon=3.1706 loss_meanflow=0.0000 mean_model_t=0.4990 mean_corrupt_t=0.4990 wrong_frac=0.5002 init_acc_corrupt=0.4884 init_gold_top10=0.4954 init_gold_top100=0.4998
+step=900 micro_steps=1800 elapsed=32.2s lr=1.081200e-04 loss_all=1.7957 acc_all=0.7535 loss_corrupt=3.1140 acc_corrupt=0.5668 corrupt_frac=0.5487 loss=3.1140 loss_recon=3.1140 loss_meanflow=0.0000 mean_model_t=0.4992 mean_corrupt_t=0.4992 wrong_frac=0.5027 init_acc_corrupt=0.4857 init_gold_top10=0.4927 init_gold_top100=0.4973
+step=1000 micro_steps=2000 elapsed=32.1s lr=1.201200e-04 loss_all=1.7723 acc_all=0.7540 loss_corrupt=3.0474 acc_corrupt=0.5718 corrupt_frac=0.5531 loss=3.0474 loss_recon=3.0474 loss_meanflow=0.0000 mean_model_t=0.4954 mean_corrupt_t=0.4954 wrong_frac=0.5041 init_acc_corrupt=0.4845 init_gold_top10=0.4913 init_gold_top100=0.4961
+step=1100 micro_steps=2200 elapsed=33.6s lr=1.321200e-04 loss_all=1.7107 acc_all=0.7603 loss_corrupt=2.9532 acc_corrupt=0.5815 corrupt_frac=0.5514 loss=2.9532 loss_recon=2.9532 loss_meanflow=0.0000 mean_model_t=0.5021 mean_corrupt_t=0.5021 wrong_frac=0.4960 init_acc_corrupt=0.4932 init_gold_top10=0.4997 init_gold_top100=0.5042
+step=1200 micro_steps=2400 elapsed=32.1s lr=1.441200e-04 loss_all=1.6886 acc_all=0.7618 loss_corrupt=2.9234 acc_corrupt=0.5834 corrupt_frac=0.5499 loss=2.9234 loss_recon=2.9234 loss_meanflow=0.0000 mean_model_t=0.5036 mean_corrupt_t=0.5036 wrong_frac=0.4975 init_acc_corrupt=0.4914 init_gold_top10=0.4980 init_gold_top100=0.5025
+step=1300 micro_steps=2600 elapsed=32.1s lr=1.561200e-04 loss_all=1.6636 acc_all=0.7640 loss_corrupt=2.8729 acc_corrupt=0.5880 corrupt_frac=0.5504 loss=2.8729 loss_recon=2.8729 loss_meanflow=0.0000 mean_model_t=0.5037 mean_corrupt_t=0.5037 wrong_frac=0.4960 init_acc_corrupt=0.4930 init_gold_top10=0.4997 init_gold_top100=0.5041
+step=1400 micro_steps=2800 elapsed=32.1s lr=1.681200e-04 loss_all=1.6454 acc_all=0.7659 loss_corrupt=2.8562 acc_corrupt=0.5891 corrupt_frac=0.5462 loss=2.8562 loss_recon=2.8562 loss_meanflow=0.0000 mean_model_t=0.5015 mean_corrupt_t=0.5015 wrong_frac=0.4978 init_acc_corrupt=0.4908 init_gold_top10=0.4978 init_gold_top100=0.5023
+step=1500 micro_steps=3000 elapsed=32.1s lr=1.801200e-04 loss_all=1.6487 acc_all=0.7642 loss_corrupt=2.8300 acc_corrupt=0.5910 corrupt_frac=0.5519 loss=2.8300 loss_recon=2.8300 loss_meanflow=0.0000 mean_model_t=0.5007 mean_corrupt_t=0.5007 wrong_frac=0.4967 init_acc_corrupt=0.4922 init_gold_top10=0.4990 init_gold_top100=0.5035
+step=1600 micro_steps=3200 elapsed=32.2s lr=1.921200e-04 loss_all=1.6357 acc_all=0.7651 loss_corrupt=2.8133 acc_corrupt=0.5920 corrupt_frac=0.5504 loss=2.8133 loss_recon=2.8133 loss_meanflow=0.0000 mean_model_t=0.5011 mean_corrupt_t=0.5011 wrong_frac=0.4992 init_acc_corrupt=0.4896 init_gold_top10=0.4963 init_gold_top100=0.5008
+step=1700 micro_steps=3400 elapsed=32.1s lr=2.041200e-04 loss_all=1.6167 acc_all=0.7668 loss_corrupt=2.7754 acc_corrupt=0.5956 corrupt_frac=0.5497 loss=2.7754 loss_recon=2.7754 loss_meanflow=0.0000 mean_model_t=0.5012 mean_corrupt_t=0.5012 wrong_frac=0.4969 init_acc_corrupt=0.4919 init_gold_top10=0.4987 init_gold_top100=0.5032
+step=1800 micro_steps=3600 elapsed=32.1s lr=2.161200e-04 loss_all=1.6118 acc_all=0.7670 loss_corrupt=2.7677 acc_corrupt=0.5959 corrupt_frac=0.5490 loss=2.7677 loss_recon=2.7677 loss_meanflow=0.0000 mean_model_t=0.4992 mean_corrupt_t=0.4992 wrong_frac=0.4991 init_acc_corrupt=0.4894 init_gold_top10=0.4963 init_gold_top100=0.5009
+step=1900 micro_steps=3800 elapsed=32.1s lr=2.281200e-04 loss_all=1.5905 acc_all=0.7690 loss_corrupt=2.7400 acc_corrupt=0.5984 corrupt_frac=0.5471 loss=2.7400 loss_recon=2.7400 loss_meanflow=0.0000 mean_model_t=0.4995 mean_corrupt_t=0.4995 wrong_frac=0.4992 init_acc_corrupt=0.4894 init_gold_top10=0.4963 init_gold_top100=0.5009
+step=2000 micro_steps=4000 elapsed=32.1s lr=2.401200e-04 loss_all=1.5999 acc_all=0.7667 loss_corrupt=2.7239 acc_corrupt=0.5995 corrupt_frac=0.5547 loss=2.7239 loss_recon=2.7239 loss_meanflow=0.0000 mean_model_t=0.4999 mean_corrupt_t=0.4999 wrong_frac=0.4999 init_acc_corrupt=0.4888 init_gold_top10=0.4956 init_gold_top100=0.5002
+step=2100 micro_steps=4200 elapsed=33.5s lr=2.521200e-04 loss_all=1.5770 acc_all=0.7691 loss_corrupt=2.7072 acc_corrupt=0.6004 corrupt_frac=0.5501 loss=2.7072 loss_recon=2.7072 loss_meanflow=0.0000 mean_model_t=0.4987 mean_corrupt_t=0.4987 wrong_frac=0.5014 init_acc_corrupt=0.4873 init_gold_top10=0.4942 init_gold_top100=0.4987
+step=2200 micro_steps=4400 elapsed=32.1s lr=2.641200e-04 loss_all=1.5405 acc_all=0.7737 loss_corrupt=2.6464 acc_corrupt=0.6083 corrupt_frac=0.5497 loss=2.6464 loss_recon=2.6464 loss_meanflow=0.0000 mean_model_t=0.5043 mean_corrupt_t=0.5043 wrong_frac=0.4950 init_acc_corrupt=0.4938 init_gold_top10=0.5006 init_gold_top100=0.5051
+step=2300 micro_steps=4600 elapsed=32.1s lr=2.761200e-04 loss_all=1.5372 acc_all=0.7734 loss_corrupt=2.6522 acc_corrupt=0.6061 corrupt_frac=0.5471 loss=2.6522 loss_recon=2.6522 loss_meanflow=0.0000 mean_model_t=0.5005 mean_corrupt_t=0.5005 wrong_frac=0.5012 init_acc_corrupt=0.4875 init_gold_top10=0.4944 init_gold_top100=0.4990
+step=2400 micro_steps=4800 elapsed=32.1s lr=2.881200e-04 loss_all=1.5456 acc_all=0.7718 loss_corrupt=2.6467 acc_corrupt=0.6063 corrupt_frac=0.5525 loss=2.6467 loss_recon=2.6467 loss_meanflow=0.0000 mean_model_t=0.4997 mean_corrupt_t=0.4997 wrong_frac=0.5011 init_acc_corrupt=0.4875 init_gold_top10=0.4944 init_gold_top100=0.4991
+step=2500 micro_steps=5000 elapsed=32.0s lr=3.000000e-04 loss_all=1.5094 acc_all=0.7760 loss_corrupt=2.6095 acc_corrupt=0.6098 corrupt_frac=0.5460 loss=2.6095 loss_recon=2.6095 loss_meanflow=0.0000 mean_model_t=0.4982 mean_corrupt_t=0.4982 wrong_frac=0.5001 init_acc_corrupt=0.4885 init_gold_top10=0.4954 init_gold_top100=0.4999
+step=2600 micro_steps=5200 elapsed=32.1s lr=3.000000e-04 loss_all=1.5078 acc_all=0.7758 loss_corrupt=2.5903 acc_corrupt=0.6120 corrupt_frac=0.5509 loss=2.5903 loss_recon=2.5903 loss_meanflow=0.0000 mean_model_t=0.4994 mean_corrupt_t=0.4994 wrong_frac=0.4994 init_acc_corrupt=0.4892 init_gold_top10=0.4961 init_gold_top100=0.5007
+step=2700 micro_steps=5400 elapsed=32.1s lr=3.000000e-04 loss_all=1.5007 acc_all=0.7764 loss_corrupt=2.5900 acc_corrupt=0.6113 corrupt_frac=0.5489 loss=2.5900 loss_recon=2.5900 loss_meanflow=0.0000 mean_model_t=0.5002 mean_corrupt_t=0.5002 wrong_frac=0.5004 init_acc_corrupt=0.4885 init_gold_top10=0.4952 init_gold_top100=0.4997
+step=2800 micro_steps=5600 elapsed=32.1s lr=3.000000e-04 loss_all=1.4679 acc_all=0.7805 loss_corrupt=2.5478 acc_corrupt=0.6162 corrupt_frac=0.5472 loss=2.5478 loss_recon=2.5478 loss_meanflow=0.0000 mean_model_t=0.5018 mean_corrupt_t=0.5018 wrong_frac=0.4977 init_acc_corrupt=0.4909 init_gold_top10=0.4978 init_gold_top100=0.5023
+step=2900 micro_steps=5800 elapsed=32.1s lr=3.000000e-04 loss_all=1.4539 acc_all=0.7822 loss_corrupt=2.5214 acc_corrupt=0.6193 corrupt_frac=0.5481 loss=2.5214 loss_recon=2.5214 loss_meanflow=0.0000 mean_model_t=0.5021 mean_corrupt_t=0.5021 wrong_frac=0.4967 init_acc_corrupt=0.4921 init_gold_top10=0.4989 init_gold_top100=0.5033
+step=3000 micro_steps=6000 elapsed=32.1s lr=3.000000e-04 loss_all=1.4724 acc_all=0.7793 loss_corrupt=2.5567 acc_corrupt=0.6138 corrupt_frac=0.5484 loss=2.5567 loss_recon=2.5567 loss_meanflow=0.0000 mean_model_t=0.4967 mean_corrupt_t=0.4967 wrong_frac=0.5032 init_acc_corrupt=0.4854 init_gold_top10=0.4922 init_gold_top100=0.4969
+step=3100 micro_steps=6200 elapsed=33.7s lr=3.000000e-04 loss_all=1.4427 acc_all=0.7830 loss_corrupt=2.5009 acc_corrupt=0.6208 corrupt_frac=0.5492 loss=2.5009 loss_recon=2.5009 loss_meanflow=0.0000 mean_model_t=0.5036 mean_corrupt_t=0.5036 wrong_frac=0.4967 init_acc_corrupt=0.4921 init_gold_top10=0.4988 init_gold_top100=0.5033
+step=3200 micro_steps=6400 elapsed=32.2s lr=3.000000e-04 loss_all=1.4411 acc_all=0.7829 loss_corrupt=2.5122 acc_corrupt=0.6186 corrupt_frac=0.5470 loss=2.5122 loss_recon=2.5122 loss_meanflow=0.0000 mean_model_t=0.5012 mean_corrupt_t=0.5012 wrong_frac=0.5008 init_acc_corrupt=0.4879 init_gold_top10=0.4948 init_gold_top100=0.4993
+step=3300 micro_steps=6600 elapsed=32.2s lr=3.000000e-04 loss_all=1.4424 acc_all=0.7825 loss_corrupt=2.5125 acc_corrupt=0.6182 corrupt_frac=0.5477 loss=2.5125 loss_recon=2.5125 loss_meanflow=0.0000 mean_model_t=0.4977 mean_corrupt_t=0.4977 wrong_frac=0.5028 init_acc_corrupt=0.4857 init_gold_top10=0.4926 init_gold_top100=0.4973
+step=3400 micro_steps=6800 elapsed=32.2s lr=3.000000e-04 loss_all=1.4111 acc_all=0.7869 loss_corrupt=2.4540 acc_corrupt=0.6265 corrupt_frac=0.5486 loss=2.4540 loss_recon=2.4540 loss_meanflow=0.0000 mean_model_t=0.5032 mean_corrupt_t=0.5032 wrong_frac=0.4946 init_acc_corrupt=0.4944 init_gold_top10=0.5011 init_gold_top100=0.5055
+step=3500 micro_steps=7000 elapsed=32.1s lr=3.000000e-04 loss_all=1.4202 acc_all=0.7849 loss_corrupt=2.4856 acc_corrupt=0.6206 corrupt_frac=0.5455 loss=2.4856 loss_recon=2.4856 loss_meanflow=0.0000 mean_model_t=0.4976 mean_corrupt_t=0.4976 wrong_frac=0.5027 init_acc_corrupt=0.4857 init_gold_top10=0.4927 init_gold_top100=0.4974
+step=3600 micro_steps=7200 elapsed=32.1s lr=3.000000e-04 loss_all=1.4081 acc_all=0.7865 loss_corrupt=2.4649 acc_corrupt=0.6233 corrupt_frac=0.5459 loss=2.4649 loss_recon=2.4649 loss_meanflow=0.0000 mean_model_t=0.4991 mean_corrupt_t=0.4991 wrong_frac=0.4999 init_acc_corrupt=0.4888 init_gold_top10=0.4955 init_gold_top100=0.5001
+W0507 18:19:39.536000 353386 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
+W0507 18:19:39.537000 353386 torch/distributed/elastic/multiprocessing/api.py:898] Sending process 353453 closing signal SIGTERM
+W0507 18:19:39.537000 353386 torch/distributed/elastic/multiprocessing/api.py:898] Sending process 353454 closing signal SIGTERM
+W0507 18:19:39.538000 353386 torch/distributed/elastic/multiprocessing/api.py:898] Sending process 353455 closing signal SIGTERM
+W0507 18:19:39.538000 353386 torch/distributed/elastic/multiprocessing/api.py:898] Sending process 353456 closing signal SIGTERM
+Traceback (most recent call last):
+  File "<frozen runpy>", line 198, in _run_module_as_main
+  File "<frozen runpy>", line 88, in _run_code
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 922, in <module>
+    main()
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
+    return f(*args, **kwargs)
+           ^^^^^^^^^^^^^^^^^^
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 918, in main
+    run(args)
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 909, in run
+    elastic_launch(
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
+    return launch_agent(self._config, self._entrypoint, list(args))
+           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
+    result = agent.run()
+             ^^^^^^^^^^^
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 137, in wrapper
+    result = f(*args, **kwargs)
+             ^^^^^^^^^^^^^^^^^^
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
+    result = self._invoke_run(role)
+             ^^^^^^^^^^^^^^^^^^^^^^
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
+    time.sleep(monitor_interval)
+  File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
+    raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
+torch.distributed.elastic.multiprocessing.api.SignalException: Process 353386 got signal: 15

LTA_openwebtext_dualt/logs/lta_owt_c1024_len1024_t0to1_lowk64plus_cleanbridge_buf1000_gbs128_4gpu_2k.nohup

@@ -0,0 +1,85 @@
+[launch] owt low-t-cut low-k 100-step pilot
+[launch] run_name=lta_owt_c1024_len1024_t0to1_lowk64plus_cleanbridge_buf1000_gbs128_4gpu_2k
+[launch] save_dir=runs/lta_owt_c1024_len1024_t0to1_lowk64plus_cleanbridge_buf1000_gbs128_4gpu_2k
+[launch] t=0.0..1.0 mask=0.01..1.0
+[launch] mixture original=0.0 lowk=1.0 all=0.0
+[launch] clean_state_mode=bridge
+
+*****************************************
+Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
+*****************************************
+NCCL version 2.25.1+cuda12.8
+{
+  "device": "cuda:0",
+  "rank": 0,
+  "world_size": 4,
+  "samples": "wrapped_stream_online_shuffle:1000",
+  "vocab_size": 50257,
+  "save_dir": "runs/lta_owt_c1024_len1024_t0to1_lowk64plus_cleanbridge_buf1000_gbs128_4gpu_2k",
+  "batch_size": 16,
+  "grad_accum": 2,
+  "effective_batch_size": 128,
+  "global_batch_size": 128,
+  "lr_schedule": "constant_warmup",
+  "warmup_steps": 100,
+  "adam_beta1": 0.9,
+  "adam_beta2": 0.999,
+  "adam_eps": 1e-08,
+  "model_type": "ddit",
+  "dual_t": true,
+  "corrupt_t_mode": "same",
+  "corrupt_min_t": 0.0,
+  "corrupt_max_t": 1.0,
+  "dirichlet_endpoint_mode": "categorical_dual_t",
+  "dirichlet_semantic_t_mode": "same",
+  "dirichlet_semantic_t_value": 0.0,
+  "categorical_wrong_from_full_vocab": true,
+  "simplex_bridge_sampler": "dirichlet",
+  "logistic_normal_sigma_min": 0.18,
+  "logistic_normal_sigma_max": 2.2,
+  "logistic_normal_tau_min": 0.65,
+  "logistic_normal_tau_max": 1.15,
+  "torch_compile": false,
+  "compile_mode": "max-autotune",
+  "state_format": "prob",
+  "target_loss": "hard_ce",
+  "meanflow_weight": 0.0,
+  "bridge_noise_init": "logistic_normal",
+  "noise_sigma": -1.0,
+  "wrap": true,
+  "wrap_mode": "stream",
+  "wrap_record_buffer_size": 200,
+  "owt_cached_chunks": false,
+  "owt_chunk_cache_dir": "",
+  "owt_chunk_cache_rebuild": false,
+  "owt_chunk_cache_write_batch": 4096,
+  "online_chunk_shuffle": true,
+  "online_chunk_shuffle_buffer": 1000,
+  "openwebtext_split": "train_minus_100k",
+  "detokenizer": "auto",
+  "resolved_detokenizer": null,
+  "num_workers": 0,
+  "latest_every": 500,
+  "resume_path": ""
+}
+step=100 micro_steps=200 elapsed=105.3s lr=3.000000e-04 loss_all=8.1603 acc_all=0.2428 loss_corrupt=8.2249 acc_corrupt=0.2239 corrupt_frac=0.8529 loss=8.2249 loss_recon=8.2249 loss_meanflow=0.0000 mean_model_t=0.5013 mean_corrupt_t=0.5013 wrong_frac=0.4990 init_acc_corrupt=0.4665 init_gold_top10=0.4953 init_gold_top100=0.5251
+step=200 micro_steps=400 elapsed=101.2s lr=3.000000e-04 loss_all=4.0894 acc_all=0.5048 loss_corrupt=4.4452 acc_corrupt=0.4609 corrupt_frac=0.8555 loss=4.4452 loss_recon=4.4452 loss_meanflow=0.0000 mean_model_t=0.4963 mean_corrupt_t=0.4963 wrong_frac=0.5040 init_acc_corrupt=0.4616 init_gold_top10=0.4904 init_gold_top100=0.5198
+step=300 micro_steps=600 elapsed=102.1s lr=3.000000e-04 loss_all=3.8334 acc_all=0.5283 loss_corrupt=4.1803 acc_corrupt=0.4854 corrupt_frac=0.8540 loss=4.1803 loss_recon=4.1803 loss_meanflow=0.0000 mean_model_t=0.5010 mean_corrupt_t=0.5010 wrong_frac=0.4986 init_acc_corrupt=0.4673 init_gold_top10=0.4959 init_gold_top100=0.5249
+step=400 micro_steps=800 elapsed=101.3s lr=3.000000e-04 loss_all=3.6010 acc_all=0.5477 loss_corrupt=3.9351 acc_corrupt=0.5057 corrupt_frac=0.8540 loss=3.9351 loss_recon=3.9351 loss_meanflow=0.0000 mean_model_t=0.5092 mean_corrupt_t=0.5092 wrong_frac=0.4904 init_acc_corrupt=0.4760 init_gold_top10=0.5043 init_gold_top100=0.5332
+step=500 micro_steps=1000 elapsed=101.6s lr=3.000000e-04 loss_all=3.5088 acc_all=0.5530 loss_corrupt=3.8326 acc_corrupt=0.5115 corrupt_frac=0.8551 loss=3.8326 loss_recon=3.8326 loss_meanflow=0.0000 mean_model_t=0.5078 mean_corrupt_t=0.5078 wrong_frac=0.4921 init_acc_corrupt=0.4738 init_gold_top10=0.5027 init_gold_top100=0.5309
+step=600 micro_steps=1200 elapsed=121.0s lr=3.000000e-04 loss_all=3.5477 acc_all=0.5448 loss_corrupt=3.8690 acc_corrupt=0.5034 corrupt_frac=0.8546 loss=3.8690 loss_recon=3.8690 loss_meanflow=0.0000 mean_model_t=0.4973 mean_corrupt_t=0.4973 wrong_frac=0.5028 init_acc_corrupt=0.4617 init_gold_top10=0.4912 init_gold_top100=0.5216
+step=700 micro_steps=1400 elapsed=151.2s lr=3.000000e-04 loss_all=3.4579 acc_all=0.5530 loss_corrupt=3.7778 acc_corrupt=0.5112 corrupt_frac=0.8540 loss=3.7778 loss_recon=3.7778 loss_meanflow=0.0000 mean_model_t=0.5022 mean_corrupt_t=0.5022 wrong_frac=0.4988 init_acc_corrupt=0.4661 init_gold_top10=0.4956 init_gold_top100=0.5251
+step=800 micro_steps=1600 elapsed=101.6s lr=3.000000e-04 loss_all=3.3878 acc_all=0.5601 loss_corrupt=3.6868 acc_corrupt=0.5205 corrupt_frac=0.8564 loss=3.6868 loss_recon=3.6868 loss_meanflow=0.0000 mean_model_t=0.5055 mean_corrupt_t=0.5055 wrong_frac=0.4933 init_acc_corrupt=0.4730 init_gold_top10=0.5013 init_gold_top100=0.5302
+step=900 micro_steps=1800 elapsed=101.6s lr=3.000000e-04 loss_all=3.4095 acc_all=0.5549 loss_corrupt=3.7220 acc_corrupt=0.5133 corrupt_frac=0.8515 loss=3.7220 loss_recon=3.7220 loss_meanflow=0.0000 mean_model_t=0.4971 mean_corrupt_t=0.4971 wrong_frac=0.5024 init_acc_corrupt=0.4637 init_gold_top10=0.4917 init_gold_top100=0.5223
+step=1000 micro_steps=2000 elapsed=150.6s lr=3.000000e-04 loss_all=3.4462 acc_all=0.5495 loss_corrupt=3.7466 acc_corrupt=0.5093 corrupt_frac=0.8552 loss=3.7466 loss_recon=3.7466 loss_meanflow=0.0000 mean_model_t=0.4921 mean_corrupt_t=0.4921 wrong_frac=0.5076 init_acc_corrupt=0.4575 init_gold_top10=0.4863 init_gold_top100=0.5181
+step=1100 micro_steps=2200 elapsed=137.7s lr=3.000000e-04 loss_all=3.3705 acc_all=0.5575 loss_corrupt=3.6763 acc_corrupt=0.5165 corrupt_frac=0.8529 loss=3.6763 loss_recon=3.6763 loss_meanflow=0.0000 mean_model_t=0.4963 mean_corrupt_t=0.4963 wrong_frac=0.5031 init_acc_corrupt=0.4619 init_gold_top10=0.4911 init_gold_top100=0.5209
+step=1200 micro_steps=2400 elapsed=101.7s lr=3.000000e-04 loss_all=3.3290 acc_all=0.5614 loss_corrupt=3.6456 acc_corrupt=0.5187 corrupt_frac=0.8546 loss=3.6456 loss_recon=3.6456 loss_meanflow=0.0000 mean_model_t=0.4982 mean_corrupt_t=0.4982 wrong_frac=0.5023 init_acc_corrupt=0.4630 init_gold_top10=0.4921 init_gold_top100=0.5221
+step=1300 micro_steps=2600 elapsed=101.5s lr=3.000000e-04 loss_all=3.3538 acc_all=0.5572 loss_corrupt=3.6600 acc_corrupt=0.5157 corrupt_frac=0.8554 loss=3.6600 loss_recon=3.6600 loss_meanflow=0.0000 mean_model_t=0.4943 mean_corrupt_t=0.4943 wrong_frac=0.5056 init_acc_corrupt=0.4591 init_gold_top10=0.4888 init_gold_top100=0.5192
+step=1400 micro_steps=2800 elapsed=101.7s lr=3.000000e-04 loss_all=3.3495 acc_all=0.5595 loss_corrupt=3.6461 acc_corrupt=0.5193 corrupt_frac=0.8568 loss=3.6461 loss_recon=3.6461 loss_meanflow=0.0000 mean_model_t=0.4980 mean_corrupt_t=0.4980 wrong_frac=0.5018 init_acc_corrupt=0.4625 init_gold_top10=0.4922 init_gold_top100=0.5239
+step=1500 micro_steps=3000 elapsed=102.7s lr=3.000000e-04 loss_all=3.2241 acc_all=0.5725 loss_corrupt=3.5293 acc_corrupt=0.5309 corrupt_frac=0.8553 loss=3.5293 loss_recon=3.5293 loss_meanflow=0.0000 mean_model_t=0.5072 mean_corrupt_t=0.5072 wrong_frac=0.4926 init_acc_corrupt=0.4732 init_gold_top10=0.5020 init_gold_top100=0.5305
+step=1600 micro_steps=3200 elapsed=152.8s lr=3.000000e-04 loss_all=3.2893 acc_all=0.5636 loss_corrupt=3.5949 acc_corrupt=0.5217 corrupt_frac=0.8537 loss=3.5949 loss_recon=3.5949 loss_meanflow=0.0000 mean_model_t=0.4976 mean_corrupt_t=0.4976 wrong_frac=0.5025 init_acc_corrupt=0.4617 init_gold_top10=0.4918 init_gold_top100=0.5222
+step=1700 micro_steps=3400 elapsed=190.0s lr=3.000000e-04 loss_all=3.2442 acc_all=0.5687 loss_corrupt=3.5504 acc_corrupt=0.5266 corrupt_frac=0.8535 loss=3.5504 loss_recon=3.5504 loss_meanflow=0.0000 mean_model_t=0.5021 mean_corrupt_t=0.5021 wrong_frac=0.4979 init_acc_corrupt=0.4670 init_gold_top10=0.4966 init_gold_top100=0.5258
+step=1800 micro_steps=3600 elapsed=103.1s lr=3.000000e-04 loss_all=3.2934 acc_all=0.5620 loss_corrupt=3.5990 acc_corrupt=0.5200 corrupt_frac=0.8535 loss=3.5990 loss_recon=3.5990 loss_meanflow=0.0000 mean_model_t=0.4942 mean_corrupt_t=0.4942 wrong_frac=0.5053 init_acc_corrupt=0.4599 init_gold_top10=0.4888 init_gold_top100=0.5191
+step=1900 micro_steps=3800 elapsed=102.9s lr=3.000000e-04 loss_all=3.2194 acc_all=0.5701 loss_corrupt=3.5194 acc_corrupt=0.5286 corrupt_frac=0.8545 loss=3.5194 loss_recon=3.5194 loss_meanflow=0.0000 mean_model_t=0.5006 mean_corrupt_t=0.5006 wrong_frac=0.4989 init_acc_corrupt=0.4664 init_gold_top10=0.4955 init_gold_top100=0.5250
+step=2000 micro_steps=4000 elapsed=103.0s lr=3.000000e-04 loss_all=3.2098 acc_all=0.5715 loss_corrupt=3.5113 acc_corrupt=0.5298 corrupt_frac=0.8547 loss=3.5113 loss_recon=3.5113 loss_meanflow=0.0000 mean_model_t=0.5035 mean_corrupt_t=0.5035 wrong_frac=0.4971 init_acc_corrupt=0.4693 init_gold_top10=0.4973 init_gold_top100=0.5262
+scripts/launch_lta_owt_c1024_fullycoupled_4gpu_len1024_lowtcut_lowk_100step.sh: line 137: --dirichlet_endpoint_mode: command not found

LTA_openwebtext_dualt/logs/lta_owt_gpt2cached_len1024_ddit768x12_elfopt_only_muon_ema_gbs512_8gpu_1m_20260513_032747.log

@@ -0,0 +1,92 @@
+{
+  "device": "cuda:0",
+  "rank": 0,
+  "world_size": 1,
+  "samples": "owt_cached_chunks:8734897",
+  "vocab_size": 50257,
+  "tokenizer_vocab_size": 50257,
+  "save_dir": "runs/lta_owt_gpt2cached_len1024_ddit768x12_elfopt_only_muon_ema_gbs512_8gpu_1m_20260513_032747",
+  "batch_size": 32,
+  "grad_accum": 16,
+  "effective_batch_size": 512,
+  "global_batch_size": 512,
+  "lr_schedule": "constant_warmup",
+  "optimizer": "muon",
+  "warmup_steps": 2000,
+  "min_lr": 0.0,
+  "weight_decay": 0.0,
+  "adamw_param_groups": "nanogpt",
+  "adam_beta1": 0.9,
+  "adam_beta2": 0.95,
+  "adam_eps": 1e-08,
+  "muon_momentum": 0.95,
+  "muon_ns_steps": 5,
+  "muon_update_scale": 1.0,
+  "ema_decay": 0.9999,
+  "ema_start_step": 0,
+  "model_type": "ddit",
+  "dual_t": true,
+  "corrupt_t_mode": "same",
+  "corrupt_min_t": 0.0,
+  "corrupt_max_t": 1.0,
+  "prefix_block_prob": 0.0,
+  "prefix_block_len": 128,
+  "dirichlet_endpoint_mode": "categorical_dual_t",
+  "dirichlet_semantic_t_mode": "same",
+  "dirichlet_semantic_t_value": 0.0,
+  "categorical_wrong_from_full_vocab": true,
+  "categorical_wrong_from_batch_valid_tokens": false,
+  "mask_mixture_original_prob": 0.0,
+  "mask_mixture_lowk_prob": 0.0,
+  "mask_mixture_lowcorrupt_prob": 0.0,
+  "mask_mixture_block_prob": 0.0,
+  "mask_mixture_all_prob": 0.0,
+  "mask_mixture_lowk_clean_tokens": "1,2,4,8,16,32,64",
+  "mask_mixture_lowcorrupt_tokens": "1,2,4,8,16,32,64",
+  "mask_mixture_block_tokens": "64,128",
+  "simplex_bridge_sampler": "dirichlet",
+  "logistic_normal_sigma_min": 0.18,
+  "logistic_normal_sigma_max": 2.2,
+  "logistic_normal_tau_min": 0.65,
+  "logistic_normal_tau_max": 1.15,
+  "torch_compile": false,
+  "compile_mode": "max-autotune",
+  "state_format": "prob",
+  "target_loss": "hard_ce",
+  "meanflow_weight": 0.0,
+  "bridge_noise_init": "logistic_normal",
+  "noise_sigma": -1.0,
+  "allow_tf32": true,
+  "activation_checkpointing": true,
+  "activation_checkpoint_interval": 1,
+  "ddp_static_graph": false,
+  "ddp_gradient_as_bucket_view": true,
+  "blocking_data_transfer": false,
+  "dataloader_prefetch_factor": 4,
+  "full_train_stats": false,
+  "wrap": true,
+  "wrap_mode": "stream",
+  "wrap_record_buffer_size": 200,
+  "owt_cached_chunks": true,
+  "owt_chunk_cache_dir": "/e2e-data/evad-tech-vla/wanghan58/data/small_benchmarks/langflow_2604_11748/openwebtext_lta_cached_chunks/gpt2_len1024_train_minus_100k",
+  "owt_chunk_cache_rebuild": false,
+  "owt_chunk_cache_write_batch": 4096,
+  "owt_exact_repeat_per_chunk": 0,
+  "online_chunk_shuffle": false,
+  "online_chunk_shuffle_buffer": 10000,
+  "openwebtext_split": "train_minus_100k",
+  "detokenizer": "auto",
+  "resolved_detokenizer": null,
+  "num_workers": 1,
+  "latest_every": 5000,
+  "resume_path": ""
+}
+step=50 micro_steps=800 elapsed=571.2s lr=5.100000e-05 loss_all=10.8125 acc_all=0.5605 loss_corrupt=10.8125 acc_corrupt=0.3880 corrupt_frac=0.5761 loss=10.8125 loss_recon=10.8125 loss_meanflow=0.0000 mean_model_t=0.4831 mean_corrupt_t=0.4831 mean_loss_t_weight=1.0000 prior_center_loss_beta=0.0000 wrong_frac=0.4951 init_acc_corrupt=0.4676 init_gold_top10=0.4988 init_gold_top100=0.5335
+step=100 micro_steps=1600 elapsed=570.4s lr=1.010000e-04 loss_all=10.7564 acc_all=0.5739 loss_corrupt=10.7781 acc_corrupt=0.3814 corrupt_frac=0.5033 loss=10.7781 loss_recon=10.7781 loss_meanflow=0.0000 mean_model_t=0.4987 mean_corrupt_t=0.4987 mean_loss_t_weight=1.0000 prior_center_loss_beta=0.0000 wrong_frac=0.4954 init_acc_corrupt=0.4742 init_gold_top10=0.4992 init_gold_top100=0.5288
+step=150 micro_steps=2400 elapsed=570.4s lr=1.510000e-04 loss_all=10.6733 acc_all=0.5642 loss_corrupt=10.7260 acc_corrupt=0.3500 corrupt_frac=0.5002 loss=10.7260 loss_recon=10.7260 loss_meanflow=0.0000 mean_model_t=0.5064 mean_corrupt_t=0.5064 mean_loss_t_weight=1.0000 prior_center_loss_beta=0.0000 wrong_frac=0.5343 init_acc_corrupt=0.4266 init_gold_top10=0.4587 init_gold_top100=0.4930
+step=200 micro_steps=3200 elapsed=570.5s lr=2.010000e-04 loss_all=10.5303 acc_all=0.5759 loss_corrupt=10.5970 acc_corrupt=0.4295 corrupt_frac=0.5151 loss=10.5970 loss_recon=10.5970 loss_meanflow=0.0000 mean_model_t=0.5509 mean_corrupt_t=0.5509 mean_loss_t_weight=1.0000 prior_center_loss_beta=0.0000 wrong_frac=0.4047 init_acc_corrupt=0.5650 init_gold_top10=0.5911 init_gold_top100=0.6175
+step=250 micro_steps=4000 elapsed=570.6s lr=2.510000e-04 loss_all=10.3466 acc_all=0.5599 loss_corrupt=10.4928 acc_corrupt=0.3590 corrupt_frac=0.4792 loss=10.4928 loss_recon=10.4928 loss_meanflow=0.0000 mean_model_t=0.5542 mean_corrupt_t=0.5542 mean_loss_t_weight=1.0000 prior_center_loss_beta=0.0000 wrong_frac=0.4766 init_acc_corrupt=0.4833 init_gold_top10=0.5175 init_gold_top100=0.5533
+step=300 micro_steps=4800 elapsed=570.4s lr=3.010000e-04 loss_all=10.1577 acc_all=0.5103 loss_corrupt=10.3788 acc_corrupt=0.3098 corrupt_frac=0.4810 loss=10.3788 loss_recon=10.3788 loss_meanflow=0.0000 mean_model_t=0.4728 mean_corrupt_t=0.4728 mean_loss_t_weight=1.0000 prior_center_loss_beta=0.0000 wrong_frac=0.5348 init_acc_corrupt=0.4165 init_gold_top10=0.4597 init_gold_top100=0.4903
+step=350 micro_steps=5600 elapsed=570.5s lr=3.510000e-04 loss_all=9.9247 acc_all=0.4639 loss_corrupt=10.2645 acc_corrupt=0.2476 corrupt_frac=0.5316 loss=10.2645 loss_recon=10.2645 loss_meanflow=0.0000 mean_model_t=0.4559 mean_corrupt_t=0.4559 mean_loss_t_weight=1.0000 prior_center_loss_beta=0.0000 wrong_frac=0.6194 init_acc_corrupt=0.3267 init_gold_top10=0.3700 init_gold_top100=0.4187
+step=400 micro_steps=6400 elapsed=571.0s lr=4.010000e-04 loss_all=9.6225 acc_all=0.4531 loss_corrupt=9.8933 acc_corrupt=0.3268 corrupt_frac=0.6250 loss=9.8933 loss_recon=9.8933 loss_meanflow=0.0000 mean_model_t=0.5057 mean_corrupt_t=0.5057 mean_loss_t_weight=1.0000 prior_center_loss_beta=0.0000 wrong_frac=0.4999 init_acc_corrupt=0.4570 init_gold_top10=0.4944 init_gold_top100=0.5211
+step=450 micro_steps=7200 elapsed=570.9s lr=4.510000e-04 loss_all=9.1788 acc_all=0.4718 loss_corrupt=9.5245 acc_corrupt=0.3493 corrupt_frac=0.6076 loss=9.5245 loss_recon=9.5245 loss_meanflow=0.0000 mean_model_t=0.5421 mean_corrupt_t=0.5421 mean_loss_t_weight=1.0000 prior_center_loss_beta=0.0000 wrong_frac=0.4807 init_acc_corrupt=0.4938 init_gold_top10=0.5117 init_gold_top100=0.5420

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/__init__.py

@@ -0,0 +1,204 @@
+# This file is auto-generated by `utils/generate_inference_types.py`.
+# Do not modify it manually.
+#
+# ruff: noqa: F401
+
+from .audio_classification import (
+    AudioClassificationInput,
+    AudioClassificationOutputElement,
+    AudioClassificationOutputTransform,
+    AudioClassificationParameters,
+)
+from .audio_to_audio import AudioToAudioInput, AudioToAudioOutputElement
+from .automatic_speech_recognition import (
+    AutomaticSpeechRecognitionEarlyStoppingEnum,
+    AutomaticSpeechRecognitionGenerationParameters,
+    AutomaticSpeechRecognitionInput,
+    AutomaticSpeechRecognitionOutput,
+    AutomaticSpeechRecognitionOutputChunk,
+    AutomaticSpeechRecognitionParameters,
+)
+from .base import BaseInferenceType
+from .chat_completion import (
+    ChatCompletionInput,
+    ChatCompletionInputFunctionDefinition,
+    ChatCompletionInputFunctionName,
+    ChatCompletionInputGrammarType,
+    ChatCompletionInputJSONSchema,
+    ChatCompletionInputMessage,
+    ChatCompletionInputMessageChunk,
+    ChatCompletionInputMessageChunkType,
+    ChatCompletionInputResponseFormatJSONObject,
+    ChatCompletionInputResponseFormatJSONSchema,
+    ChatCompletionInputResponseFormatText,
+    ChatCompletionInputStreamOptions,
+    ChatCompletionInputTool,
+    ChatCompletionInputToolCall,
+    ChatCompletionInputToolChoiceClass,
+    ChatCompletionInputToolChoiceEnum,
+    ChatCompletionInputURL,
+    ChatCompletionOutput,
+    ChatCompletionOutputComplete,
+    ChatCompletionOutputFunctionDefinition,
+    ChatCompletionOutputLogprob,
+    ChatCompletionOutputLogprobs,
+    ChatCompletionOutputMessage,
+    ChatCompletionOutputToolCall,
+    ChatCompletionOutputTopLogprob,
+    ChatCompletionOutputUsage,
+    ChatCompletionStreamOutput,
+    ChatCompletionStreamOutputChoice,
+    ChatCompletionStreamOutputDelta,
+    ChatCompletionStreamOutputDeltaToolCall,
+    ChatCompletionStreamOutputFunction,
+    ChatCompletionStreamOutputLogprob,
+    ChatCompletionStreamOutputLogprobs,
+    ChatCompletionStreamOutputTopLogprob,
+    ChatCompletionStreamOutputUsage,
+)
+from .depth_estimation import DepthEstimationInput, DepthEstimationOutput
+from .document_question_answering import (
+    DocumentQuestionAnsweringInput,
+    DocumentQuestionAnsweringInputData,
+    DocumentQuestionAnsweringOutputElement,
+    DocumentQuestionAnsweringParameters,
+)
+from .feature_extraction import FeatureExtractionInput, FeatureExtractionInputTruncationDirection
+from .fill_mask import FillMaskInput, FillMaskOutputElement, FillMaskParameters
+from .image_classification import (
+    ImageClassificationInput,
+    ImageClassificationOutputElement,
+    ImageClassificationOutputTransform,
+    ImageClassificationParameters,
+)
+from .image_segmentation import (
+    ImageSegmentationInput,
+    ImageSegmentationOutputElement,
+    ImageSegmentationParameters,
+    ImageSegmentationSubtask,
+)
+from .image_text_to_image import (
+    ImageTextToImageInput,
+    ImageTextToImageOutput,
+    ImageTextToImageParameters,
+    ImageTextToImageTargetSize,
+)
+from .image_text_to_video import (
+    ImageTextToVideoInput,
+    ImageTextToVideoOutput,
+    ImageTextToVideoParameters,
+    ImageTextToVideoTargetSize,
+)
+from .image_to_image import ImageToImageInput, ImageToImageOutput, ImageToImageParameters, ImageToImageTargetSize
+from .image_to_text import (
+    ImageToTextEarlyStoppingEnum,
+    ImageToTextGenerationParameters,
+    ImageToTextInput,
+    ImageToTextOutput,
+    ImageToTextParameters,
+)
+from .image_to_video import ImageToVideoInput, ImageToVideoOutput, ImageToVideoParameters, ImageToVideoTargetSize
+from .object_detection import (
+    ObjectDetectionBoundingBox,
+    ObjectDetectionInput,
+    ObjectDetectionOutputElement,
+    ObjectDetectionParameters,
+)
+from .question_answering import (
+    QuestionAnsweringInput,
+    QuestionAnsweringInputData,
+    QuestionAnsweringOutputElement,
+    QuestionAnsweringParameters,
+)
+from .sentence_similarity import SentenceSimilarityInput, SentenceSimilarityInputData
+from .summarization import (
+    SummarizationInput,
+    SummarizationOutput,
+    SummarizationParameters,
+    SummarizationTruncationStrategy,
+)
+from .table_question_answering import (
+    Padding,
+    TableQuestionAnsweringInput,
+    TableQuestionAnsweringInputData,
+    TableQuestionAnsweringOutputElement,
+    TableQuestionAnsweringParameters,
+)
+from .text2text_generation import (
+    Text2TextGenerationInput,
+    Text2TextGenerationOutput,
+    Text2TextGenerationParameters,
+    Text2TextGenerationTruncationStrategy,
+)
+from .text_classification import (
+    TextClassificationInput,
+    TextClassificationOutputElement,
+    TextClassificationOutputTransform,
+    TextClassificationParameters,
+)
+from .text_generation import (
+    TextGenerationInput,
+    TextGenerationInputGenerateParameters,
+    TextGenerationInputGrammarType,
+    TextGenerationOutput,
+    TextGenerationOutputBestOfSequence,
+    TextGenerationOutputDetails,
+    TextGenerationOutputFinishReason,
+    TextGenerationOutputPrefillToken,
+    TextGenerationOutputToken,
+    TextGenerationStreamOutput,
+    TextGenerationStreamOutputStreamDetails,
+    TextGenerationStreamOutputToken,
+    TypeEnum,
+)
+from .text_to_audio import (
+    TextToAudioEarlyStoppingEnum,
+    TextToAudioGenerationParameters,
+    TextToAudioInput,
+    TextToAudioOutput,
+    TextToAudioParameters,
+)
+from .text_to_image import TextToImageInput, TextToImageOutput, TextToImageParameters
+from .text_to_speech import (
+    TextToSpeechEarlyStoppingEnum,
+    TextToSpeechGenerationParameters,
+    TextToSpeechInput,
+    TextToSpeechOutput,
+    TextToSpeechParameters,
+)
+from .text_to_video import TextToVideoInput, TextToVideoOutput, TextToVideoParameters
+from .token_classification import (
+    TokenClassificationAggregationStrategy,
+    TokenClassificationInput,
+    TokenClassificationOutputElement,
+    TokenClassificationParameters,
+)
+from .translation import TranslationInput, TranslationOutput, TranslationParameters, TranslationTruncationStrategy
+from .video_classification import (
+    VideoClassificationInput,
+    VideoClassificationOutputElement,
+    VideoClassificationOutputTransform,
+    VideoClassificationParameters,
+)
+from .visual_question_answering import (
+    VisualQuestionAnsweringInput,
+    VisualQuestionAnsweringInputData,
+    VisualQuestionAnsweringOutputElement,
+    VisualQuestionAnsweringParameters,
+)
+from .zero_shot_classification import (
+    ZeroShotClassificationInput,
+    ZeroShotClassificationOutputElement,
+    ZeroShotClassificationParameters,
+)
+from .zero_shot_image_classification import (
+    ZeroShotImageClassificationInput,
+    ZeroShotImageClassificationOutputElement,
+    ZeroShotImageClassificationParameters,
+)
+from .zero_shot_object_detection import (
+    ZeroShotObjectDetectionBoundingBox,
+    ZeroShotObjectDetectionInput,
+    ZeroShotObjectDetectionOutputElement,
+    ZeroShotObjectDetectionParameters,
+)

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/audio_to_audio.py

@@ -0,0 +1,30 @@
+# Inference code generated from the JSON schema spec in @huggingface/tasks.
+#
+# See:
+#   - script: https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-codegen.ts
+#   - specs:  https://github.com/huggingface/huggingface.js/tree/main/packages/tasks/src/tasks.
+from typing import Any
+
+from .base import BaseInferenceType, dataclass_with_extra
+
+
+@dataclass_with_extra
+class AudioToAudioInput(BaseInferenceType):
+    """Inputs for Audio to Audio inference"""
+
+    inputs: Any
+    """The input audio data"""
+
+
+@dataclass_with_extra
+class AudioToAudioOutputElement(BaseInferenceType):
+    """Outputs of inference for the Audio To Audio task
+    A generated audio file with its label.
+    """
+
+    blob: Any
+    """The generated audio file."""
+    content_type: str
+    """The content type of audio file."""
+    label: str
+    """The label of the audio file."""

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/base.py

@@ -0,0 +1,167 @@
+# Copyright 2024 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Contains a base class for all inference types."""
+
+import inspect
+import json
+import types
+from dataclasses import asdict, dataclass
+from typing import Any, TypeVar, get_args
+
+from typing_extensions import dataclass_transform
+
+
+T = TypeVar("T", bound="BaseInferenceType")
+
+
+def _repr_with_extra(self):
+    fields = list(self.__dataclass_fields__.keys())
+    other_fields = list(k for k in self.__dict__ if k not in fields)
+    return f"{self.__class__.__name__}({', '.join(f'{k}={self.__dict__[k]!r}' for k in fields + other_fields)})"
+
+
+@dataclass_transform()
+def dataclass_with_extra(cls: type[T]) -> type[T]:
+    """Decorator to add a custom __repr__ method to a dataclass, showing all fields, including extra ones.
+
+    This decorator only works with dataclasses that inherit from `BaseInferenceType`.
+    """
+    cls = dataclass(cls)
+    cls.__repr__ = _repr_with_extra  # type: ignore[method-assign]
+    return cls
+
+
+@dataclass
+class BaseInferenceType(dict):
+    """Base class for all inference types.
+
+    Object is a dataclass and a dict for backward compatibility but plan is to remove the dict part in the future.
+
+    Handle parsing from dict, list and json strings in a permissive way to ensure future-compatibility (e.g. all fields
+    are made optional, and non-expected fields are added as dict attributes).
+    """
+
+    @classmethod
+    def parse_obj_as_list(cls: type[T], data: bytes | str | list | dict) -> list[T]:
+        """Alias to parse server response and return a single instance.
+
+        See `parse_obj` for more details.
+        """
+        output = cls.parse_obj(data)
+        if not isinstance(output, list):
+            raise ValueError(f"Invalid input data for {cls}. Expected a list, but got {type(output)}.")
+        return output
+
+    @classmethod
+    def parse_obj_as_instance(cls: type[T], data: bytes | str | list | dict) -> T:
+        """Alias to parse server response and return a single instance.
+
+        See `parse_obj` for more details.
+        """
+        output = cls.parse_obj(data)
+        if isinstance(output, list):
+            raise ValueError(f"Invalid input data for {cls}. Expected a single instance, but got a list.")
+        return output
+
+    @classmethod
+    def parse_obj(cls: type[T], data: bytes | str | list | dict) -> list[T] | T:
+        """Parse server response as a dataclass or list of dataclasses.
+
+        To enable future-compatibility, we want to handle cases where the server return more fields than expected.
+        In such cases, we don't want to raise an error but still create the dataclass object. Remaining fields are
+        added as dict attributes.
+        """
+        # Parse server response (from bytes)
+        if isinstance(data, bytes):
+            data = data.decode()
+        if isinstance(data, str):
+            data = json.loads(data)
+
+        # If a list, parse each item individually
+        if isinstance(data, list):
+            return [cls.parse_obj(d) for d in data]  # type: ignore
+
+        # At this point, we expect a dict
+        if not isinstance(data, dict):
+            raise ValueError(f"Invalid data type: {type(data)}")
+
+        init_values = {}
+        other_values = {}
+        for key, value in data.items():
+            key = normalize_key(key)
+            if key in cls.__dataclass_fields__ and cls.__dataclass_fields__[key].init:
+                if isinstance(value, dict) or isinstance(value, list):
+                    field_type = cls.__dataclass_fields__[key].type
+
+                    # if `field_type` is a `BaseInferenceType`, parse it
+                    if inspect.isclass(field_type) and issubclass(field_type, BaseInferenceType):
+                        value = field_type.parse_obj(value)
+
+                    # otherwise, recursively parse nested dataclasses (if possible)
+                    # `get_args` returns handle Union and Optional for us
+                    else:
+                        expected_types = get_args(field_type)
+                        for expected_type in expected_types:
+                            if (
+                                isinstance(expected_type, types.GenericAlias) and expected_type.__origin__ is list
+                            ) or getattr(expected_type, "_name", None) == "List":
+                                expected_type = get_args(expected_type)[
+                                    0
+                                ]  # assume same type for all items in the list
+                            if inspect.isclass(expected_type) and issubclass(expected_type, BaseInferenceType):
+                                value = expected_type.parse_obj(value)
+                                break
+                init_values[key] = value
+            else:
+                other_values[key] = value
+
+        # Make all missing fields default to None
+        # => ensure that dataclass initialization will never fail even if the server does not return all fields.
+        for key in cls.__dataclass_fields__:
+            if key not in init_values:
+                init_values[key] = None
+
+        # Initialize dataclass with expected values
+        item = cls(**init_values)
+
+        # Add remaining fields as dict attributes
+        item.update(other_values)
+
+        # Add remaining fields as extra dataclass fields.
+        # They won't be part of the dataclass fields but will be accessible as attributes.
+        # Use @dataclass_with_extra to show them in __repr__.
+        item.__dict__.update(other_values)
+        return item
+
+    def __post_init__(self):
+        self.update(asdict(self))
+
+    def __setitem__(self, __key: Any, __value: Any) -> None:
+        # Hacky way to keep dataclass values in sync when dict is updated
+        super().__setitem__(__key, __value)
+        if __key in self.__dataclass_fields__ and getattr(self, __key, None) != __value:
+            self.__setattr__(__key, __value)
+        return
+
+    def __setattr__(self, __name: str, __value: Any) -> None:
+        # Hacky way to keep dict values is sync when dataclass is updated
+        super().__setattr__(__name, __value)
+        if self.get(__name) != __value:
+            self[__name] = __value
+        return
+
+
+def normalize_key(key: str) -> str:
+    # e.g "content-type" -> "content_type", "Accept" -> "accept"
+    return key.replace("-", "_").replace(" ", "_").lower()

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/chat_completion.py

@@ -0,0 +1,347 @@
+# Inference code generated from the JSON schema spec in @huggingface/tasks.
+#
+# See:
+#   - script: https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-codegen.ts
+#   - specs:  https://github.com/huggingface/huggingface.js/tree/main/packages/tasks/src/tasks.
+from typing import Any, Literal, Union
+
+from .base import BaseInferenceType, dataclass_with_extra
+
+
+@dataclass_with_extra
+class ChatCompletionInputURL(BaseInferenceType):
+    url: str
+
+
+ChatCompletionInputMessageChunkType = Literal["text", "image_url"]
+
+
+@dataclass_with_extra
+class ChatCompletionInputMessageChunk(BaseInferenceType):
+    type: "ChatCompletionInputMessageChunkType"
+    image_url: ChatCompletionInputURL | None = None
+    text: str | None = None
+
+
+@dataclass_with_extra
+class ChatCompletionInputFunctionDefinition(BaseInferenceType):
+    name: str
+    parameters: Any
+    description: str | None = None
+
+
+@dataclass_with_extra
+class ChatCompletionInputToolCall(BaseInferenceType):
+    function: ChatCompletionInputFunctionDefinition
+    id: str
+    type: str
+
+
+@dataclass_with_extra
+class ChatCompletionInputMessage(BaseInferenceType):
+    role: str
+    content: list[ChatCompletionInputMessageChunk] | str | None = None
+    name: str | None = None
+    tool_calls: list[ChatCompletionInputToolCall] | None = None
+
+
+@dataclass_with_extra
+class ChatCompletionInputJSONSchema(BaseInferenceType):
+    name: str
+    """
+    The name of the response format.
+    """
+    description: str | None = None
+    """
+    A description of what the response format is for, used by the model to determine
+    how to respond in the format.
+    """
+    schema: dict[str, object] | None = None
+    """
+    The schema for the response format, described as a JSON Schema object. Learn how
+    to build JSON schemas [here](https://json-schema.org/).
+    """
+    strict: bool | None = None
+    """
+    Whether to enable strict schema adherence when generating the output. If set to
+    true, the model will always follow the exact schema defined in the `schema`
+    field.
+    """
+
+
+@dataclass_with_extra
+class ChatCompletionInputResponseFormatText(BaseInferenceType):
+    type: Literal["text"]
+
+
+@dataclass_with_extra
+class ChatCompletionInputResponseFormatJSONSchema(BaseInferenceType):
+    type: Literal["json_schema"]
+    json_schema: ChatCompletionInputJSONSchema
+
+
+@dataclass_with_extra
+class ChatCompletionInputResponseFormatJSONObject(BaseInferenceType):
+    type: Literal["json_object"]
+
+
+ChatCompletionInputGrammarType = Union[
+    ChatCompletionInputResponseFormatText,
+    ChatCompletionInputResponseFormatJSONSchema,
+    ChatCompletionInputResponseFormatJSONObject,
+]
+
+
+@dataclass_with_extra
+class ChatCompletionInputStreamOptions(BaseInferenceType):
+    include_usage: bool | None = None
+    """If set, an additional chunk will be streamed before the data: [DONE] message. The usage
+    field on this chunk shows the token usage statistics for the entire request, and the
+    choices field will always be an empty array. All other chunks will also include a usage
+    field, but with a null value.
+    """
+
+
+@dataclass_with_extra
+class ChatCompletionInputFunctionName(BaseInferenceType):
+    name: str
+
+
+@dataclass_with_extra
+class ChatCompletionInputToolChoiceClass(BaseInferenceType):
+    function: ChatCompletionInputFunctionName
+
+
+ChatCompletionInputToolChoiceEnum = Literal["auto", "none", "required"]
+
+
+@dataclass_with_extra
+class ChatCompletionInputTool(BaseInferenceType):
+    function: ChatCompletionInputFunctionDefinition
+    type: str
+
+
+@dataclass_with_extra
+class ChatCompletionInput(BaseInferenceType):
+    """Chat Completion Input.
+    Auto-generated from TGI specs.
+    For more details, check out
+    https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-tgi-import.ts.
+    """
+
+    messages: list[ChatCompletionInputMessage]
+    """A list of messages comprising the conversation so far."""
+    frequency_penalty: float | None = None
+    """Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing
+    frequency in the text so far,
+    decreasing the model's likelihood to repeat the same line verbatim.
+    """
+    logit_bias: list[float] | None = None
+    """UNUSED
+    Modify the likelihood of specified tokens appearing in the completion. Accepts a JSON
+    object that maps tokens
+    (specified by their token ID in the tokenizer) to an associated bias value from -100 to
+    100. Mathematically,
+    the bias is added to the logits generated by the model prior to sampling. The exact
+    effect will vary per model,
+    but values between -1 and 1 should decrease or increase likelihood of selection; values
+    like -100 or 100 should
+    result in a ban or exclusive selection of the relevant token.
+    """
+    logprobs: bool | None = None
+    """Whether to return log probabilities of the output tokens or not. If true, returns the log
+    probabilities of each
+    output token returned in the content of message.
+    """
+    max_tokens: int | None = None
+    """The maximum number of tokens that can be generated in the chat completion."""
+    model: str | None = None
+    """[UNUSED] ID of the model to use. See the model endpoint compatibility table for details
+    on which models work with the Chat API.
+    """
+    n: int | None = None
+    """UNUSED
+    How many chat completion choices to generate for each input message. Note that you will
+    be charged based on the
+    number of generated tokens across all of the choices. Keep n as 1 to minimize costs.
+    """
+    presence_penalty: float | None = None
+    """Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they
+    appear in the text so far,
+    increasing the model's likelihood to talk about new topics
+    """
+    response_format: ChatCompletionInputGrammarType | None = None
+    seed: int | None = None
+    stop: list[str] | None = None
+    """Up to 4 sequences where the API will stop generating further tokens."""
+    stream: bool | None = None
+    stream_options: ChatCompletionInputStreamOptions | None = None
+    temperature: float | None = None
+    """What sampling temperature to use, between 0 and 2. Higher values like 0.8 will make the
+    output more random, while
+    lower values like 0.2 will make it more focused and deterministic.
+    We generally recommend altering this or `top_p` but not both.
+    """
+    tool_choice: Union[ChatCompletionInputToolChoiceClass, "ChatCompletionInputToolChoiceEnum"] | None = None
+    tool_prompt: str | None = None
+    """A prompt to be appended before the tools"""
+    tools: list[ChatCompletionInputTool] | None = None
+    """A list of tools the model may call. Currently, only functions are supported as a tool.
+    Use this to provide a list of
+    functions the model may generate JSON inputs for.
+    """
+    top_logprobs: int | None = None
+    """An integer between 0 and 5 specifying the number of most likely tokens to return at each
+    token position, each with
+    an associated log probability. logprobs must be set to true if this parameter is used.
+    """
+    top_p: float | None = None
+    """An alternative to sampling with temperature, called nucleus sampling, where the model
+    considers the results of the
+    tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10%
+    probability mass are considered.
+    """
+
+
+@dataclass_with_extra
+class ChatCompletionOutputTopLogprob(BaseInferenceType):
+    logprob: float
+    token: str
+
+
+@dataclass_with_extra
+class ChatCompletionOutputLogprob(BaseInferenceType):
+    logprob: float
+    token: str
+    top_logprobs: list[ChatCompletionOutputTopLogprob]
+
+
+@dataclass_with_extra
+class ChatCompletionOutputLogprobs(BaseInferenceType):
+    content: list[ChatCompletionOutputLogprob]
+
+
+@dataclass_with_extra
+class ChatCompletionOutputFunctionDefinition(BaseInferenceType):
+    arguments: str
+    name: str
+    description: str | None = None
+
+
+@dataclass_with_extra
+class ChatCompletionOutputToolCall(BaseInferenceType):
+    function: ChatCompletionOutputFunctionDefinition
+    id: str
+    type: str
+
+
+@dataclass_with_extra
+class ChatCompletionOutputMessage(BaseInferenceType):
+    role: str
+    content: str | None = None
+    reasoning: str | None = None
+    tool_call_id: str | None = None
+    tool_calls: list[ChatCompletionOutputToolCall] | None = None
+
+
+@dataclass_with_extra
+class ChatCompletionOutputComplete(BaseInferenceType):
+    finish_reason: str
+    index: int
+    message: ChatCompletionOutputMessage
+    logprobs: ChatCompletionOutputLogprobs | None = None
+
+
+@dataclass_with_extra
+class ChatCompletionOutputUsage(BaseInferenceType):
+    completion_tokens: int
+    prompt_tokens: int
+    total_tokens: int
+
+
+@dataclass_with_extra
+class ChatCompletionOutput(BaseInferenceType):
+    """Chat Completion Output.
+    Auto-generated from TGI specs.
+    For more details, check out
+    https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-tgi-import.ts.
+    """
+
+    choices: list[ChatCompletionOutputComplete]
+    created: int
+    id: str
+    model: str
+    system_fingerprint: str
+    usage: ChatCompletionOutputUsage
+
+
+@dataclass_with_extra
+class ChatCompletionStreamOutputFunction(BaseInferenceType):
+    arguments: str
+    name: str | None = None
+
+
+@dataclass_with_extra
+class ChatCompletionStreamOutputDeltaToolCall(BaseInferenceType):
+    function: ChatCompletionStreamOutputFunction
+    id: str
+    index: int
+    type: str
+
+
+@dataclass_with_extra
+class ChatCompletionStreamOutputDelta(BaseInferenceType):
+    role: str
+    content: str | None = None
+    reasoning: str | None = None
+    tool_call_id: str | None = None
+    tool_calls: list[ChatCompletionStreamOutputDeltaToolCall] | None = None
+
+
+@dataclass_with_extra
+class ChatCompletionStreamOutputTopLogprob(BaseInferenceType):
+    logprob: float
+    token: str
+
+
+@dataclass_with_extra
+class ChatCompletionStreamOutputLogprob(BaseInferenceType):
+    logprob: float
+    token: str
+    top_logprobs: list[ChatCompletionStreamOutputTopLogprob]
+
+
+@dataclass_with_extra
+class ChatCompletionStreamOutputLogprobs(BaseInferenceType):
+    content: list[ChatCompletionStreamOutputLogprob]
+
+
+@dataclass_with_extra
+class ChatCompletionStreamOutputChoice(BaseInferenceType):
+    delta: ChatCompletionStreamOutputDelta
+    index: int
+    finish_reason: str | None = None
+    logprobs: ChatCompletionStreamOutputLogprobs | None = None
+
+
+@dataclass_with_extra
+class ChatCompletionStreamOutputUsage(BaseInferenceType):
+    completion_tokens: int
+    prompt_tokens: int
+    total_tokens: int
+
+
+@dataclass_with_extra
+class ChatCompletionStreamOutput(BaseInferenceType):
+    """Chat Completion Stream Output.
+    Auto-generated from TGI specs.
+    For more details, check out
+    https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-tgi-import.ts.
+    """
+
+    choices: list[ChatCompletionStreamOutputChoice]
+    created: int
+    id: str
+    model: str
+    system_fingerprint: str
+    usage: ChatCompletionStreamOutputUsage | None = None

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/feature_extraction.py

@@ -0,0 +1,36 @@
+# Inference code generated from the JSON schema spec in @huggingface/tasks.
+#
+# See:
+#   - script: https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-codegen.ts
+#   - specs:  https://github.com/huggingface/huggingface.js/tree/main/packages/tasks/src/tasks.
+from typing import Literal, Optional
+
+from .base import BaseInferenceType, dataclass_with_extra
+
+
+FeatureExtractionInputTruncationDirection = Literal["left", "right"]
+
+
+@dataclass_with_extra
+class FeatureExtractionInput(BaseInferenceType):
+    """Feature Extraction Input.
+    Auto-generated from TEI specs.
+    For more details, check out
+    https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-tei-import.ts.
+    """
+
+    inputs: list[str] | str
+    """The text or list of texts to embed."""
+    normalize: bool | None = None
+    prompt_name: str | None = None
+    """The name of the prompt that should be used by for encoding. If not set, no prompt
+    will be applied.
+    Must be a key in the `sentence-transformers` configuration `prompts` dictionary.
+    For example if ``prompt_name`` is "query" and the ``prompts`` is {"query": "query: ",
+    ...},
+    then the sentence "What is the capital of France?" will be encoded as
+    "query: What is the capital of France?" because the prompt text will be prepended before
+    any text to encode.
+    """
+    truncate: bool | None = None
+    truncation_direction: Optional["FeatureExtractionInputTruncationDirection"] = None

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/fill_mask.py

@@ -0,0 +1,47 @@
+# Inference code generated from the JSON schema spec in @huggingface/tasks.
+#
+# See:
+#   - script: https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-codegen.ts
+#   - specs:  https://github.com/huggingface/huggingface.js/tree/main/packages/tasks/src/tasks.
+from typing import Any
+
+from .base import BaseInferenceType, dataclass_with_extra
+
+
+@dataclass_with_extra
+class FillMaskParameters(BaseInferenceType):
+    """Additional inference parameters for Fill Mask"""
+
+    targets: list[str] | None = None
+    """When passed, the model will limit the scores to the passed targets instead of looking up
+    in the whole vocabulary. If the provided targets are not in the model vocab, they will be
+    tokenized and the first resulting token will be used (with a warning, and that might be
+    slower).
+    """
+    top_k: int | None = None
+    """When passed, overrides the number of predictions to return."""
+
+
+@dataclass_with_extra
+class FillMaskInput(BaseInferenceType):
+    """Inputs for Fill Mask inference"""
+
+    inputs: str
+    """The text with masked tokens"""
+    parameters: FillMaskParameters | None = None
+    """Additional inference parameters for Fill Mask"""
+
+
+@dataclass_with_extra
+class FillMaskOutputElement(BaseInferenceType):
+    """Outputs of inference for the Fill Mask task"""
+
+    score: float
+    """The corresponding probability"""
+    sequence: str
+    """The corresponding input with the mask token prediction."""
+    token: int
+    """The predicted token id (to replace the masked one)."""
+    token_str: Any
+    fill_mask_output_token_str: str | None = None
+    """The predicted token (to replace the masked one)."""

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/image_classification.py

@@ -0,0 +1,43 @@
+# Inference code generated from the JSON schema spec in @huggingface/tasks.
+#
+# See:
+#   - script: https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-codegen.ts
+#   - specs:  https://github.com/huggingface/huggingface.js/tree/main/packages/tasks/src/tasks.
+from typing import Literal, Optional
+
+from .base import BaseInferenceType, dataclass_with_extra
+
+
+ImageClassificationOutputTransform = Literal["sigmoid", "softmax", "none"]
+
+
+@dataclass_with_extra
+class ImageClassificationParameters(BaseInferenceType):
+    """Additional inference parameters for Image Classification"""
+
+    function_to_apply: Optional["ImageClassificationOutputTransform"] = None
+    """The function to apply to the model outputs in order to retrieve the scores."""
+    top_k: int | None = None
+    """When specified, limits the output to the top K most probable classes."""
+
+
+@dataclass_with_extra
+class ImageClassificationInput(BaseInferenceType):
+    """Inputs for Image Classification inference"""
+
+    inputs: str
+    """The input image data as a base64-encoded string. If no `parameters` are provided, you can
+    also provide the image data as a raw bytes payload.
+    """
+    parameters: ImageClassificationParameters | None = None
+    """Additional inference parameters for Image Classification"""
+
+
+@dataclass_with_extra
+class ImageClassificationOutputElement(BaseInferenceType):
+    """Outputs of inference for the Image Classification task"""
+
+    label: str
+    """The predicted class label."""
+    score: float
+    """The corresponding probability."""

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/image_text_to_image.py

@@ -0,0 +1,67 @@
+# Inference code generated from the JSON schema spec in @huggingface/tasks.
+#
+# See:
+#   - script: https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-codegen.ts
+#   - specs:  https://github.com/huggingface/huggingface.js/tree/main/packages/tasks/src/tasks.
+from typing import Any
+
+from .base import BaseInferenceType, dataclass_with_extra
+
+
+@dataclass_with_extra
+class ImageTextToImageTargetSize(BaseInferenceType):
+    """The size in pixels of the output image. This parameter is only supported by some
+    providers and for specific models. It will be ignored when unsupported.
+    """
+
+    height: int
+    width: int
+
+
+@dataclass_with_extra
+class ImageTextToImageParameters(BaseInferenceType):
+    """Additional inference parameters for Image Text To Image"""
+
+    guidance_scale: float | None = None
+    """For diffusion models. A higher guidance scale value encourages the model to generate
+    images closely linked to the text prompt at the expense of lower image quality.
+    """
+    negative_prompt: str | None = None
+    """One prompt to guide what NOT to include in image generation."""
+    num_inference_steps: int | None = None
+    """For diffusion models. The number of denoising steps. More denoising steps usually lead to
+    a higher quality image at the expense of slower inference.
+    """
+    prompt: str | None = None
+    """The text prompt to guide the image generation. Either this or inputs (image) must be
+    provided.
+    """
+    seed: int | None = None
+    """Seed for the random number generator."""
+    target_size: ImageTextToImageTargetSize | None = None
+    """The size in pixels of the output image. This parameter is only supported by some
+    providers and for specific models. It will be ignored when unsupported.
+    """
+
+
+@dataclass_with_extra
+class ImageTextToImageInput(BaseInferenceType):
+    """Inputs for Image Text To Image inference. Either inputs (image) or prompt (in parameters)
+    must be provided, or both.
+    """
+
+    inputs: str | None = None
+    """The input image data as a base64-encoded string. If no `parameters` are provided, you can
+    also provide the image data as a raw bytes payload. Either this or prompt must be
+    provided.
+    """
+    parameters: ImageTextToImageParameters | None = None
+    """Additional inference parameters for Image Text To Image"""
+
+
+@dataclass_with_extra
+class ImageTextToImageOutput(BaseInferenceType):
+    """Outputs of inference for the Image Text To Image task"""
+
+    image: Any
+    """The generated image returned as raw bytes in the payload."""

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/table_question_answering.py

@@ -0,0 +1,62 @@
+# Inference code generated from the JSON schema spec in @huggingface/tasks.
+#
+# See:
+#   - script: https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-codegen.ts
+#   - specs:  https://github.com/huggingface/huggingface.js/tree/main/packages/tasks/src/tasks.
+from typing import Literal, Optional
+
+from .base import BaseInferenceType, dataclass_with_extra
+
+
+@dataclass_with_extra
+class TableQuestionAnsweringInputData(BaseInferenceType):
+    """One (table, question) pair to answer"""
+
+    question: str
+    """The question to be answered about the table"""
+    table: dict[str, list[str]]
+    """The table to serve as context for the questions"""
+
+
+Padding = Literal["do_not_pad", "longest", "max_length"]
+
+
+@dataclass_with_extra
+class TableQuestionAnsweringParameters(BaseInferenceType):
+    """Additional inference parameters for Table Question Answering"""
+
+    padding: Optional["Padding"] = None
+    """Activates and controls padding."""
+    sequential: bool | None = None
+    """Whether to do inference sequentially or as a batch. Batching is faster, but models like
+    SQA require the inference to be done sequentially to extract relations within sequences,
+    given their conversational nature.
+    """
+    truncation: bool | None = None
+    """Activates and controls truncation."""
+
+
+@dataclass_with_extra
+class TableQuestionAnsweringInput(BaseInferenceType):
+    """Inputs for Table Question Answering inference"""
+
+    inputs: TableQuestionAnsweringInputData
+    """One (table, question) pair to answer"""
+    parameters: TableQuestionAnsweringParameters | None = None
+    """Additional inference parameters for Table Question Answering"""
+
+
+@dataclass_with_extra
+class TableQuestionAnsweringOutputElement(BaseInferenceType):
+    """Outputs of inference for the Table Question Answering task"""
+
+    answer: str
+    """The answer of the question given the table. If there is an aggregator, the answer will be
+    preceded by `AGGREGATOR >`.
+    """
+    cells: list[str]
+    """list of strings made up of the answer cell values."""
+    coordinates: list[list[int]]
+    """Coordinates of the cells of the answers."""
+    aggregator: str | None = None
+    """If the model has an aggregator, this returns the aggregator."""

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/text_classification.py

@@ -0,0 +1,41 @@
+# Inference code generated from the JSON schema spec in @huggingface/tasks.
+#
+# See:
+#   - script: https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-codegen.ts
+#   - specs:  https://github.com/huggingface/huggingface.js/tree/main/packages/tasks/src/tasks.
+from typing import Literal, Optional
+
+from .base import BaseInferenceType, dataclass_with_extra
+
+
+TextClassificationOutputTransform = Literal["sigmoid", "softmax", "none"]
+
+
+@dataclass_with_extra
+class TextClassificationParameters(BaseInferenceType):
+    """Additional inference parameters for Text Classification"""
+
+    function_to_apply: Optional["TextClassificationOutputTransform"] = None
+    """The function to apply to the model outputs in order to retrieve the scores."""
+    top_k: int | None = None
+    """When specified, limits the output to the top K most probable classes."""
+
+
+@dataclass_with_extra
+class TextClassificationInput(BaseInferenceType):
+    """Inputs for Text Classification inference"""
+
+    inputs: str
+    """The text to classify"""
+    parameters: TextClassificationParameters | None = None
+    """Additional inference parameters for Text Classification"""
+
+
+@dataclass_with_extra
+class TextClassificationOutputElement(BaseInferenceType):
+    """Outputs of inference for the Text Classification task"""
+
+    label: str
+    """The predicted class label."""
+    score: float
+    """The corresponding probability."""

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/text_to_video.py

@@ -0,0 +1,46 @@
+# Inference code generated from the JSON schema spec in @huggingface/tasks.
+#
+# See:
+#   - script: https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-codegen.ts
+#   - specs:  https://github.com/huggingface/huggingface.js/tree/main/packages/tasks/src/tasks.
+from typing import Any
+
+from .base import BaseInferenceType, dataclass_with_extra
+
+
+@dataclass_with_extra
+class TextToVideoParameters(BaseInferenceType):
+    """Additional inference parameters for Text To Video"""
+
+    guidance_scale: float | None = None
+    """A higher guidance scale value encourages the model to generate videos closely linked to
+    the text prompt, but values too high may cause saturation and other artifacts.
+    """
+    negative_prompt: list[str] | None = None
+    """One or several prompt to guide what NOT to include in video generation."""
+    num_frames: float | None = None
+    """The num_frames parameter determines how many video frames are generated."""
+    num_inference_steps: int | None = None
+    """The number of denoising steps. More denoising steps usually lead to a higher quality
+    video at the expense of slower inference.
+    """
+    seed: int | None = None
+    """Seed for the random number generator."""
+
+
+@dataclass_with_extra
+class TextToVideoInput(BaseInferenceType):
+    """Inputs for Text To Video inference"""
+
+    inputs: str
+    """The input text data (sometimes called "prompt")"""
+    parameters: TextToVideoParameters | None = None
+    """Additional inference parameters for Text To Video"""
+
+
+@dataclass_with_extra
+class TextToVideoOutput(BaseInferenceType):
+    """Outputs of inference for the Text To Video task"""
+
+    video: Any
+    """The generated video returned as raw bytes in the payload."""

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/video_classification.py

@@ -0,0 +1,45 @@
+# Inference code generated from the JSON schema spec in @huggingface/tasks.
+#
+# See:
+#   - script: https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-codegen.ts
+#   - specs:  https://github.com/huggingface/huggingface.js/tree/main/packages/tasks/src/tasks.
+from typing import Any, Literal, Optional
+
+from .base import BaseInferenceType, dataclass_with_extra
+
+
+VideoClassificationOutputTransform = Literal["sigmoid", "softmax", "none"]
+
+
+@dataclass_with_extra
+class VideoClassificationParameters(BaseInferenceType):
+    """Additional inference parameters for Video Classification"""
+
+    frame_sampling_rate: int | None = None
+    """The sampling rate used to select frames from the video."""
+    function_to_apply: Optional["VideoClassificationOutputTransform"] = None
+    """The function to apply to the model outputs in order to retrieve the scores."""
+    num_frames: int | None = None
+    """The number of sampled frames to consider for classification."""
+    top_k: int | None = None
+    """When specified, limits the output to the top K most probable classes."""
+
+
+@dataclass_with_extra
+class VideoClassificationInput(BaseInferenceType):
+    """Inputs for Video Classification inference"""
+
+    inputs: Any
+    """The input video data"""
+    parameters: VideoClassificationParameters | None = None
+    """Additional inference parameters for Video Classification"""
+
+
+@dataclass_with_extra
+class VideoClassificationOutputElement(BaseInferenceType):
+    """Outputs of inference for the Video Classification task"""
+
+    label: str
+    """The predicted class label."""
+    score: float
+    """The corresponding probability."""

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/visual_question_answering.py

@@ -0,0 +1,49 @@
+# Inference code generated from the JSON schema spec in @huggingface/tasks.
+#
+# See:
+#   - script: https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-codegen.ts
+#   - specs:  https://github.com/huggingface/huggingface.js/tree/main/packages/tasks/src/tasks.
+from typing import Any
+
+from .base import BaseInferenceType, dataclass_with_extra
+
+
+@dataclass_with_extra
+class VisualQuestionAnsweringInputData(BaseInferenceType):
+    """One (image, question) pair to answer"""
+
+    image: Any
+    """The image."""
+    question: str
+    """The question to answer based on the image."""
+
+
+@dataclass_with_extra
+class VisualQuestionAnsweringParameters(BaseInferenceType):
+    """Additional inference parameters for Visual Question Answering"""
+
+    top_k: int | None = None
+    """The number of answers to return (will be chosen by order of likelihood). Note that we
+    return less than topk answers if there are not enough options available within the
+    context.
+    """
+
+
+@dataclass_with_extra
+class VisualQuestionAnsweringInput(BaseInferenceType):
+    """Inputs for Visual Question Answering inference"""
+
+    inputs: VisualQuestionAnsweringInputData
+    """One (image, question) pair to answer"""
+    parameters: VisualQuestionAnsweringParameters | None = None
+    """Additional inference parameters for Visual Question Answering"""
+
+
+@dataclass_with_extra
+class VisualQuestionAnsweringOutputElement(BaseInferenceType):
+    """Outputs of inference for the Visual Question Answering task"""
+
+    score: float
+    """The associated score / probability"""
+    answer: str | None = None
+    """The answer to the question"""

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/huggingface_hub/inference/_generated/types/zero_shot_object_detection.py

@@ -0,0 +1,50 @@
+# Inference code generated from the JSON schema spec in @huggingface/tasks.
+#
+# See:
+#   - script: https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/scripts/inference-codegen.ts
+#   - specs:  https://github.com/huggingface/huggingface.js/tree/main/packages/tasks/src/tasks.
+from .base import BaseInferenceType, dataclass_with_extra
+
+
+@dataclass_with_extra
+class ZeroShotObjectDetectionParameters(BaseInferenceType):
+    """Additional inference parameters for Zero Shot Object Detection"""
+
+    candidate_labels: list[str]
+    """The candidate labels for this image"""
+
+
+@dataclass_with_extra
+class ZeroShotObjectDetectionInput(BaseInferenceType):
+    """Inputs for Zero Shot Object Detection inference"""
+
+    inputs: str
+    """The input image data as a base64-encoded string."""
+    parameters: ZeroShotObjectDetectionParameters
+    """Additional inference parameters for Zero Shot Object Detection"""
+
+
+@dataclass_with_extra
+class ZeroShotObjectDetectionBoundingBox(BaseInferenceType):
+    """The predicted bounding box. Coordinates are relative to the top left corner of the input
+    image.
+    """
+
+    xmax: int
+    xmin: int
+    ymax: int
+    ymin: int
+
+
+@dataclass_with_extra
+class ZeroShotObjectDetectionOutputElement(BaseInferenceType):
+    """Outputs of inference for the Zero Shot Object Detection task"""
+
+    box: ZeroShotObjectDetectionBoundingBox
+    """The predicted bounding box. Coordinates are relative to the top left corner of the input
+    image.
+    """
+    label: str
+    """A candidate label"""
+    score: float
+    """The associated score / probability"""