Add files using upload-large-folder tool

.gitattributes

@@ -54,3 +54,5 @@ routing_rerun/N_8.0/wandb/run-20260524_111710-wj0poul7/run-wj0poul7.wandb filter
 routing_rerun/N_2.5/wandb/run-20260523_172819-hcp9axm4/run-hcp9axm4.wandb filter=lfs diff=lfs merge=lfs -text
 routing_rerun/N_4.0/wandb/run-20260524_000855-j30yot8x/run-j30yot8x.wandb filter=lfs diff=lfs merge=lfs -text
 routing_rerun/N_6.0/wandb/run-20260524_055222-tjovd8fs/run-tjovd8fs.wandb filter=lfs diff=lfs merge=lfs -text
+hnet_train_rerun/21-40-57/wandb/run-20260524_214058-3nxcy117/run-3nxcy117.wandb filter=lfs diff=lfs merge=lfs -text
+hnet_train_rerun/21-40-57/eval_results/predictions_initial_checkpoint.txt filter=lfs diff=lfs merge=lfs -text

hnet_train_rerun/21-40-57/.hydra/config.yaml

@@ -0,0 +1,55 @@
+model:
+  config_path: ${oc.env:PROJECT_ROOT}/hnet_project/configs/hnet_2stage_XL_code.json
+  checkpoint_path: ${oc.env:PROJECT_ROOT}/hnet_project/checkpoints/hnet_2stage_XL_code.pt
+training:
+  epochs: 3
+  batch_size: 8
+  eval_batch_size: 24
+  gradient_accumulation_steps: 4
+  lr: 0.0001
+  weight_decay: 0.1
+  betas:
+  - 0.9
+  - 0.95
+  eps: 1.0e-08
+  lr_scheduler: wsd
+  warmup_ratio: 0.1
+  decay_ratio: 0.2
+  warmup_steps: 100
+  min_lr_ratio: 0.1
+  lr_multiplier:
+  - 2.0
+  - 1.5
+  - 1.0
+  load_balancing_weight: 0.01
+  load_balancing_N: 4.0
+  max_grad_norm: 1.0
+  use_amp: true
+  resume: false
+  resume_checkpoint: null
+  warmup_model: true
+data:
+  path: ${oc.env:PROJECT_ROOT}/code_completion_exp/datasets/data_V4_full
+  max_context_len: 4096
+  max_target_len: 256
+  num_workers: 0
+  pin_memory: true
+  max_train_samples: null
+  max_val_samples: null
+logging:
+  log_interval: 10
+  save_interval: 3000
+  eval_interval: 2000
+  save_every_epoch: false
+tracking:
+  enabled: true
+  backend: wandb
+  project: hnet_rerun
+  run_name: hnet_train
+  entity: null
+  base_url: https://wandb.platun0v.ru
+  local_dir: ${paths.output_dir}
+paths:
+  output_dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S}
+seed: 42
+device: cuda

hnet_train_rerun/21-40-57/.hydra/hydra.yaml

@@ -0,0 +1,160 @@
+hydra:
+  run:
+    dir: ${paths.output_dir}
+  sweep:
+    dir: outputs/multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task: []
+  job:
+    name: train
+    chdir: false
+    override_dirname: ''
+    id: ???
+    num: ???
+    config_name: config
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /workspace/byte-llms-code/code_completion_exp/train_hnet
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /workspace/byte-llms-code/code_completion_exp/train_hnet/configs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/byte-llms-code/code_completion_exp/train_hnet/outputs/2026-05-24/21-40-57
+    choices:
+      paths: default
+      tracking: wandb
+      logging: default
+      data: default
+      training: default
+      model: hnet_xl_code
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

hnet_train_rerun/21-40-57/.hydra/overrides.yaml

@@ -0,0 +1 @@
+[]

hnet_train_rerun/21-40-57/eval_results/.ipynb_checkpoints/metrics_checkpoint_latest-checkpoint.txt

@@ -0,0 +1,17 @@
+Checkpoint: checkpoint_latest.pt
+================================================================================
+
+exact_match: 0.360191463901077
+token_accuracy: 0.47747439249588164
+bleu: 19.726981132809208
+bpb: 1.3918926184016422
+num_samples: 35098
+gen_wall_time_s: 6738.63248833298
+gen_samples_per_s: 5.208475170706725
+gen_time_per_sample_ms: 191.99477144945524
+gen_chars_per_s: 151.54513942822675
+gen_batch_mean_ms: 3071.391289121686
+gen_batch_p50_ms: 2684.8264584841672
+gen_batch_p95_ms: 7208.787969502737
+gen_batch_max_ms: 12667.352943972219
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/.ipynb_checkpoints/metrics_checkpoint_step_27000-checkpoint.txt

@@ -0,0 +1,17 @@
+Checkpoint: checkpoint_step_27000.pt
+================================================================================
+
+exact_match: 0.36064733033221263
+token_accuracy: 0.4773234898368026
+bleu: 19.665145938230435
+bpb: 1.3918926184016422
+num_samples: 35098
+gen_wall_time_s: 6899.235165899969
+gen_samples_per_s: 5.087230563392696
+gen_time_per_sample_ms: 196.57060704028632
+gen_chars_per_s: 148.02785750045956
+gen_batch_mean_ms: 3144.5921448951544
+gen_batch_p50_ms: 2730.471632006811
+gen_batch_p95_ms: 7366.315296370885
+gen_batch_max_ms: 12734.633811982349
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/.ipynb_checkpoints/metrics_checkpoint_step_3000-checkpoint.txt

@@ -0,0 +1,17 @@
+Checkpoint: checkpoint_step_3000.pt
+================================================================================
+
+exact_match: 0.36452219499686594
+token_accuracy: 0.4787851175154458
+bleu: 20.091589434631988
+bpb: 1.3343184788590108
+num_samples: 35098
+gen_wall_time_s: 6209.05275386211
+gen_samples_per_s: 5.652714092043846
+gen_time_per_sample_ms: 176.90616997726679
+gen_chars_per_s: 156.16971516256734
+gen_batch_mean_ms: 2830.014928834143
+gen_batch_p50_ms: 2555.72874849895
+gen_batch_p95_ms: 6390.771690043039
+gen_batch_max_ms: 9818.601018982008
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/.ipynb_checkpoints/metrics_model_best-checkpoint.txt

@@ -0,0 +1,17 @@
+Checkpoint: model_best.pt
+================================================================================
+
+exact_match: 0.3734400820559576
+token_accuracy: 0.4819705178766448
+bleu: 20.14870477803318
+bpb: 1.3264302330352213
+num_samples: 35098
+gen_wall_time_s: 6241.421260637813
+gen_samples_per_s: 5.623398667439622
+gen_time_per_sample_ms: 177.82840220633122
+gen_chars_per_s: 156.0005260565445
+gen_batch_mean_ms: 2844.76812244203
+gen_batch_p50_ms: 2527.919982007006
+gen_batch_p95_ms: 6982.412858091992
+gen_batch_max_ms: 10548.128249007277
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/eval_config.yaml

@@ -0,0 +1,29 @@
+model:
+  config_path: /workspace/byte-llms-code/hnet_project/configs/hnet_2stage_XL_code.json
+  checkpoint_path: /workspace/byte-llms-code/hnet_project/checkpoints/hnet_2stage_XL_code.pt
+data:
+  path: /workspace/byte-llms-code/code_completion_exp/datasets/data_V4_full
+  max_context_len: 4096
+  max_target_len: 256
+  num_workers: 0
+  pin_memory: true
+  max_train_samples: null
+  max_val_samples: null
+paths:
+  checkpoints_dir: outputs/2026-05-24/21-40-57
+  initial_checkpoint: auto
+  output_dir: outputs/2026-05-24/21-40-57/eval_results
+evaluation:
+  batch_size: 16
+  max_samples: null
+  compute_bpb: true
+  bleu_tokenize: none
+  use_amp: true
+  save_predictions: true
+generation:
+  max_length: 256
+  temperature: 0.1
+  top_k: 0
+  top_p: 1.0
+seed: 42
+device: cuda

hnet_train_rerun/21-40-57/eval_results/metrics_checkpoint_latest.txt

@@ -0,0 +1,17 @@
+Checkpoint: checkpoint_latest.pt
+================================================================================
+
+exact_match: 0.360191463901077
+token_accuracy: 0.47747439249588164
+bleu: 19.726981132809208
+bpb: 1.3918926184016422
+num_samples: 35098
+gen_wall_time_s: 6738.63248833298
+gen_samples_per_s: 5.208475170706725
+gen_time_per_sample_ms: 191.99477144945524
+gen_chars_per_s: 151.54513942822675
+gen_batch_mean_ms: 3071.391289121686
+gen_batch_p50_ms: 2684.8264584841672
+gen_batch_p95_ms: 7208.787969502737
+gen_batch_max_ms: 12667.352943972219
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/metrics_checkpoint_step_12000.txt

@@ -0,0 +1,17 @@
+Checkpoint: checkpoint_step_12000.pt
+================================================================================
+
+exact_match: 0.3741523733546071
+token_accuracy: 0.48474383789958975
+bleu: 20.866813431088616
+bpb: 1.335828999773514
+num_samples: 35098
+gen_wall_time_s: 6534.712194961612
+gen_samples_per_s: 5.37100930429059
+gen_time_per_sample_ms: 186.18474542599614
+gen_chars_per_s: 152.32575977370146
+gen_batch_mean_ms: 2978.44676160511
+gen_batch_p50_ms: 2646.6345914814156
+gen_batch_p95_ms: 7165.557659140904
+gen_batch_max_ms: 10396.650541981217
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/metrics_checkpoint_step_15000.txt

@@ -0,0 +1,17 @@
+Checkpoint: checkpoint_step_15000.pt
+================================================================================
+
+exact_match: 0.374294831614337
+token_accuracy: 0.48255671666768235
+bleu: 20.82318093842354
+bpb: 1.3339916854689853
+num_samples: 35098
+gen_wall_time_s: 6575.688067426148
+gen_samples_per_s: 5.3375402908578105
+gen_time_per_sample_ms: 187.35221572243853
+gen_chars_per_s: 149.28567017386504
+gen_batch_mean_ms: 2997.1230936308793
+gen_batch_p50_ms: 2574.5291139755864
+gen_batch_p95_ms: 7272.849028115161
+gen_batch_max_ms: 10864.763604011387
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/metrics_checkpoint_step_18000.txt

@@ -0,0 +1,17 @@
+Checkpoint: checkpoint_step_18000.pt
+================================================================================
+
+exact_match: 0.3744087982221209
+token_accuracy: 0.4840512333874068
+bleu: 21.172392577542894
+bpb: 1.3298998572417937
+num_samples: 35098
+gen_wall_time_s: 6599.062013073475
+gen_samples_per_s: 5.318634668149347
+gen_time_per_sample_ms: 188.0181780464264
+gen_chars_per_s: 149.91266910965837
+gen_batch_mean_ms: 3007.7766695868163
+gen_batch_p50_ms: 2595.0252109905705
+gen_batch_p95_ms: 7258.518203350831
+gen_batch_max_ms: 9577.27041101316
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/metrics_checkpoint_step_21000.txt

@@ -0,0 +1,17 @@
+Checkpoint: checkpoint_step_21000.pt
+================================================================================
+
+exact_match: 0.3584819647843182
+token_accuracy: 0.47367861022520286
+bleu: 19.489348058114224
+bpb: 1.4021340103652056
+num_samples: 35098
+gen_wall_time_s: 6887.526788596122
+gen_samples_per_s: 5.095878546434516
+gen_time_per_sample_ms: 196.23701602929287
+gen_chars_per_s: 147.68871776792565
+gen_batch_mean_ms: 3139.255601000967
+gen_batch_p50_ms: 2698.8774175115395
+gen_batch_p95_ms: 7320.19043576729
+gen_batch_max_ms: 12885.085268993862
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/metrics_checkpoint_step_24000.txt

@@ -0,0 +1,17 @@
+Checkpoint: checkpoint_step_24000.pt
+================================================================================
+
+exact_match: 0.3584819647843182
+token_accuracy: 0.4737385843589394
+bleu: 19.275525429511855
+bpb: 1.390391807400738
+num_samples: 35098
+gen_wall_time_s: 6441.2251053551445
+gen_samples_per_s: 5.44896342325003
+gen_time_per_sample_ms: 183.52114380748603
+gen_chars_per_s: 154.68423843340665
+gen_batch_mean_ms: 2935.836419943092
+gen_batch_p50_ms: 2630.608787992969
+gen_batch_p95_ms: 6906.270672706885
+gen_batch_max_ms: 12607.093763013836
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/metrics_checkpoint_step_27000.txt

@@ -0,0 +1,17 @@
+Checkpoint: checkpoint_step_27000.pt
+================================================================================
+
+exact_match: 0.36064733033221263
+token_accuracy: 0.4773234898368026
+bleu: 19.665145938230435
+bpb: 1.3918926184016422
+num_samples: 35098
+gen_wall_time_s: 6899.235165899969
+gen_samples_per_s: 5.087230563392696
+gen_time_per_sample_ms: 196.57060704028632
+gen_chars_per_s: 148.02785750045956
+gen_batch_mean_ms: 3144.5921448951544
+gen_batch_p50_ms: 2730.471632006811
+gen_batch_p95_ms: 7366.315296370885
+gen_batch_max_ms: 12734.633811982349
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/metrics_checkpoint_step_3000.txt

@@ -0,0 +1,17 @@
+Checkpoint: checkpoint_step_3000.pt
+================================================================================
+
+exact_match: 0.36452219499686594
+token_accuracy: 0.4787851175154458
+bleu: 20.091589434631988
+bpb: 1.3343184788590108
+num_samples: 35098
+gen_wall_time_s: 6209.05275386211
+gen_samples_per_s: 5.652714092043846
+gen_time_per_sample_ms: 176.90616997726679
+gen_chars_per_s: 156.16971516256734
+gen_batch_mean_ms: 2830.014928834143
+gen_batch_p50_ms: 2555.72874849895
+gen_batch_p95_ms: 6390.771690043039
+gen_batch_max_ms: 9818.601018982008
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/metrics_checkpoint_step_6000.txt

@@ -0,0 +1,17 @@
+Checkpoint: checkpoint_step_6000.pt
+================================================================================
+
+exact_match: 0.37295572397287596
+token_accuracy: 0.4793123094974845
+bleu: 20.604051625274728
+bpb: 1.3291212170475468
+num_samples: 35098
+gen_wall_time_s: 6369.97152560309
+gen_samples_per_s: 5.509914739638813
+gen_time_per_sample_ms: 181.49101161328537
+gen_chars_per_s: 152.10209278106132
+gen_batch_mean_ms: 2903.3598567014997
+gen_batch_p50_ms: 2524.245548993349
+gen_batch_p95_ms: 7080.392475571716
+gen_batch_max_ms: 9737.063668959308
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/metrics_checkpoint_step_9000.txt

@@ -0,0 +1,17 @@
+Checkpoint: checkpoint_step_9000.pt
+================================================================================
+
+exact_match: 0.37175907459114477
+token_accuracy: 0.48313711151029376
+bleu: 20.088935445782226
+bpb: 1.3244859882122255
+num_samples: 35098
+gen_wall_time_s: 6331.07599858538
+gen_samples_per_s: 5.543765389618182
+gen_time_per_sample_ms: 180.38281379524133
+gen_chars_per_s: 155.2655504719327
+gen_batch_mean_ms: 2885.631722235816
+gen_batch_p50_ms: 2578.9438789943233
+gen_batch_p95_ms: 6757.976525239064
+gen_batch_max_ms: 10634.354435955174
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/metrics_initial_checkpoint.txt

@@ -0,0 +1,17 @@
+Checkpoint: hnet_2stage_XL_code.pt
+================================================================================
+
+exact_match: 0.0
+token_accuracy: 0.47427544958835494
+bleu: 1.153252081120271
+bpb: 1.8748032956517047
+num_samples: 35098
+gen_wall_time_s: 28995.99217226659
+gen_samples_per_s: 1.2104431464693841
+gen_time_per_sample_ms: 826.1437168005751
+gen_chars_per_s: 308.8898957755469
+gen_batch_mean_ms: 13216.040187906377
+gen_batch_p50_ms: 13210.045973974047
+gen_batch_p95_ms: 13516.947992166388
+gen_batch_max_ms: 34002.42500199238
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/metrics_model_best.txt

@@ -0,0 +1,17 @@
+Checkpoint: model_best.pt
+================================================================================
+
+exact_match: 0.3734400820559576
+token_accuracy: 0.4819705178766448
+bleu: 20.14870477803318
+bpb: 1.3264302330352213
+num_samples: 35098
+gen_wall_time_s: 6241.421260637813
+gen_samples_per_s: 5.623398667439622
+gen_time_per_sample_ms: 177.82840220633122
+gen_chars_per_s: 156.0005260565445
+gen_batch_mean_ms: 2844.76812244203
+gen_batch_p50_ms: 2527.919982007006
+gen_batch_p95_ms: 6982.412858091992
+gen_batch_max_ms: 10548.128249007277
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/metrics_model_final.txt

@@ -0,0 +1,17 @@
+Checkpoint: model_final.pt
+================================================================================
+
+exact_match: 0.3617015214542139
+token_accuracy: 0.4779309697720693
+bleu: 19.715273255435186
+bpb: 1.392667822078947
+num_samples: 35098
+gen_wall_time_s: 6712.276489993092
+gen_samples_per_s: 5.228926438344038
+gen_time_per_sample_ms: 191.2438455180663
+gen_chars_per_s: 151.35192978337005
+gen_batch_mean_ms: 3059.3785277999505
+gen_batch_p50_ms: 2652.1117465163115
+gen_batch_p95_ms: 7237.394248164492
+gen_batch_max_ms: 12913.40813797433
+gen_num_batches: 2194

hnet_train_rerun/21-40-57/eval_results/predictions_initial_checkpoint.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ab78a591f30ed69ed1b65b60d94f0483695a8d7ed94fe1716ae5f4d755fe002
+size 17000214

hnet_train_rerun/21-40-57/eval_results/summary.txt

@@ -0,0 +1,18 @@
+EVALUATION SUMMARY
+==================================================================================================
+
+Checkpoint                      Exact Match    Token Acc     BLEU          BPB  ms/sample   samp/s
+--------------------------------------------------------------------------------------------------
+initial_checkpoint                    0.00%       47.43%     1.15         1.87      826.1     1.21
+checkpoint_latest                    36.02%       47.75%    19.73         1.39      192.0     5.21
+checkpoint_step_12000                37.42%       48.47%    20.87         1.34      186.2     5.37
+checkpoint_step_15000                37.43%       48.26%    20.82         1.33      187.4     5.34
+checkpoint_step_18000                37.44%       48.41%    21.17         1.33      188.0     5.32
+checkpoint_step_21000                35.85%       47.37%    19.49         1.40      196.2     5.10
+checkpoint_step_24000                35.85%       47.37%    19.28         1.39      183.5     5.45
+checkpoint_step_27000                36.06%       47.73%    19.67         1.39      196.6     5.09
+checkpoint_step_3000                 36.45%       47.88%    20.09         1.33      176.9     5.65
+checkpoint_step_6000                 37.30%       47.93%    20.60         1.33      181.5     5.51
+checkpoint_step_9000                 37.18%       48.31%    20.09         1.32      180.4     5.54
+model_best                           37.34%       48.20%    20.15         1.33      177.8     5.62
+model_final                          36.17%       47.79%    19.72         1.39      191.2     5.23

hnet_train_rerun/21-40-57/model_best.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf1579f27c0c45243bada34e891daf065c4d2a608dd49f80f8b9e1a0addd7479
+size 3315165139

hnet_train_rerun/21-40-57/model_final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:293b790edc4e3ba0f5c7ae637cea1e677509621fd740389cf3f37b59b6baaf0b
+size 3315165484

hnet_train_rerun/21-40-57/wandb/debug-internal.log

@@ -0,0 +1,13 @@
+{"time":"2026-05-24T21:40:59.042920084Z","level":"INFO","msg":"stream: starting","core version":"0.24.0"}
+{"time":"2026-05-24T21:40:59.47364371Z","level":"INFO","msg":"stream: created new stream","id":"3nxcy117"}
+{"time":"2026-05-24T21:40:59.473752145Z","level":"INFO","msg":"handler: started","stream_id":"3nxcy117"}
+{"time":"2026-05-24T21:40:59.473916706Z","level":"INFO","msg":"stream: started","id":"3nxcy117"}
+{"time":"2026-05-24T21:40:59.473946539Z","level":"INFO","msg":"sender: started","stream_id":"3nxcy117"}
+{"time":"2026-05-24T21:40:59.473947844Z","level":"INFO","msg":"writer: started","stream_id":"3nxcy117"}
+{"time":"2026-05-24T21:40:59.608257445Z","level":"ERROR","msg":"git repo not found","error":"repository does not exist"}
+{"time":"2026-05-25T09:11:44.25084313Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2026-05-25T09:11:44.382625783Z","level":"INFO","msg":"handler: operation stats","stats":{}}
+{"time":"2026-05-25T09:11:44.387865173Z","level":"INFO","msg":"stream: closing","id":"3nxcy117"}
+{"time":"2026-05-25T09:11:44.387891658Z","level":"INFO","msg":"handler: closed","stream_id":"3nxcy117"}
+{"time":"2026-05-25T09:11:44.388086651Z","level":"INFO","msg":"sender: closed","stream_id":"3nxcy117"}
+{"time":"2026-05-25T09:11:44.388103059Z","level":"INFO","msg":"stream: closed","id":"3nxcy117"}

hnet_train_rerun/21-40-57/wandb/debug.log

@@ -0,0 +1,24 @@
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_setup.py:_flush():81] Current SDK version is 0.24.0
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_setup.py:_flush():81] Configure stats pid to 83034
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_setup.py:_flush():81] Loading settings from environment variables
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_init.py:setup_run_log_directory():717] Logging user logs to outputs/2026-05-24/21-40-57/wandb/run-20260524_214058-3nxcy117/logs/debug.log
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_init.py:setup_run_log_directory():718] Logging internal logs to outputs/2026-05-24/21-40-57/wandb/run-20260524_214058-3nxcy117/logs/debug-internal.log
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_init.py:init():844] calling init triggers
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_init.py:init():849] wandb.init called with sweep_config: {}
+config: {'model': {'config_path': '/workspace/byte-llms-code/hnet_project/configs/hnet_2stage_XL_code.json', 'checkpoint_path': '/workspace/byte-llms-code/hnet_project/checkpoints/hnet_2stage_XL_code.pt'}, 'training': {'epochs': 3, 'batch_size': 8, 'eval_batch_size': 24, 'gradient_accumulation_steps': 4, 'lr': 0.0001, 'weight_decay': 0.1, 'betas': [0.9, 0.95], 'eps': 1e-08, 'lr_scheduler': 'wsd', 'warmup_ratio': 0.1, 'decay_ratio': 0.2, 'warmup_steps': 100, 'min_lr_ratio': 0.1, 'lr_multiplier': [2.0, 1.5, 1.0], 'load_balancing_weight': 0.01, 'load_balancing_N': 4.0, 'max_grad_norm': 1.0, 'use_amp': True, 'resume': False, 'resume_checkpoint': None, 'warmup_model': True}, 'data': {'path': '/workspace/byte-llms-code/code_completion_exp/datasets/data_V4_full', 'max_context_len': 4096, 'max_target_len': 256, 'num_workers': 0, 'pin_memory': True, 'max_train_samples': None, 'max_val_samples': None}, 'logging': {'log_interval': 10, 'save_interval': 3000, 'eval_interval': 2000, 'save_every_epoch': False}, 'tracking': {'enabled': True, 'backend': 'wandb', 'project': 'hnet_rerun', 'run_name': 'hnet_train', 'entity': None, 'base_url': 'https://wandb.platun0v.ru', 'local_dir': 'outputs/2026-05-24/21-40-57'}, 'paths': {'output_dir': 'outputs/2026-05-24/21-40-57'}, 'seed': 42, 'device': 'cuda', '_wandb': {'code_path': 'code/code_completion_exp/train_hnet/train.py'}}
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_init.py:init():892] starting backend
+2026-05-24 21:40:59,018 INFO    MainThread:83034 [wandb_init.py:init():895] sending inform_init request
+2026-05-24 21:40:59,039 INFO    MainThread:83034 [wandb_init.py:init():903] backend started and connected
+2026-05-24 21:40:59,047 INFO    MainThread:83034 [wandb_init.py:init():973] updated telemetry
+2026-05-24 21:40:59,077 INFO    MainThread:83034 [wandb_init.py:init():997] communicating run to backend with 90.0 second timeout
+2026-05-24 21:40:59,606 INFO    MainThread:83034 [wandb_init.py:init():1044] starting run threads in backend
+2026-05-24 21:40:59,860 INFO    MainThread:83034 [wandb_run.py:_console_start():2529] atexit reg
+2026-05-24 21:40:59,860 INFO    MainThread:83034 [wandb_run.py:_redirect():2377] redirect: wrap_raw
+2026-05-24 21:40:59,860 INFO    MainThread:83034 [wandb_run.py:_redirect():2446] Wrapping output streams.
+2026-05-24 21:40:59,861 INFO    MainThread:83034 [wandb_run.py:_redirect():2469] Redirects installed.
+2026-05-24 21:40:59,865 INFO    MainThread:83034 [wandb_init.py:init():1084] run started, returning control to user process
+2026-05-25 09:11:42,798 INFO    MainThread:83034 [wandb_run.py:_finish():2295] finishing run nikita/hnet_rerun/3nxcy117
+2026-05-25 09:11:42,798 INFO    MainThread:83034 [wandb_run.py:_atexit_cleanup():2494] got exitcode: 0
+2026-05-25 09:11:42,799 INFO    MainThread:83034 [wandb_run.py:_restore():2476] restore
+2026-05-25 09:11:42,799 INFO    MainThread:83034 [wandb_run.py:_restore():2482] restore done
+2026-05-25 09:11:44,386 INFO    MainThread:83034 [wandb_run.py:_footer_sync_info():3870] logging synced files

hnet_train_rerun/21-40-57/wandb/run-20260524_214058-3nxcy117/files/code/code_completion_exp/train_hnet/train.py

@@ -0,0 +1,284 @@
+"""
+Training Pipeline для HNet модели на задаче Code Completion.
+
+Конфигурация через Hydra + OmegaConf, логирование в Trackio.
+Поддержка DDP через Accelerate для multi-GPU тренировки.
+
+Использование:
+    # Базовый запуск (single GPU)
+    python train.py
+
+    # Multi-GPU с Accelerate
+    accelerate launch train.py
+
+    # Multi-GPU с указанием количества GPU
+    accelerate launch --num_processes=4 train.py
+
+    # Переопределение параметров через CLI
+    python train.py training.lr=1e-4 training.epochs=5
+
+    # Выбор другого конфига модели
+    python train.py model=hnet_small
+
+    # Multirun (sweep)
+    python train.py --multirun training.lr=1e-4,3e-4,1e-3
+
+    # Без логирования
+    python train.py tracking.enabled=false
+"""
+
+import os
+import math
+from pathlib import Path
+
+import torch
+import hydra
+from hydra.core.hydra_config import HydraConfig
+from omegaconf import DictConfig, OmegaConf
+from accelerate import Accelerator
+from accelerate.utils import set_seed as accelerate_set_seed
+
+# HNet imports
+from hnet.load_utils import load_from_pretrained, load_from_config
+from hnet.utils.tokenizers import ByteTokenizer
+from hnet.utils.train import group_params
+
+# Ensure repo root is on sys.path (needed when running from subdirectory)
+import sys
+sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
+
+# Shared training library
+from training_lib.utils import log_message
+from training_lib.checkpointing import save_checkpoint, load_checkpoint
+from training_lib.schedulers import get_lr_scheduler
+from training_lib.tracking import init_tracking, finish_tracking
+from training_lib.hnet.train_loop import train_epoch
+from training_lib.hnet.data import create_dataloaders
+
+
+@hydra.main(version_base=None, config_path="configs", config_name="config")
+def main(cfg: DictConfig):
+    """Глав��ая функция тренировки с поддержкой DDP чере�� Accelerate."""
+
+    # === Accelerator Setup ===
+    mixed_precision = "bf16" if cfg.training.use_amp else "no"
+
+    accelerator = Accelerator(
+        mixed_precision=mixed_precision,
+        gradient_accumulation_steps=cfg.training.gradient_accumulation_steps,
+    )
+
+    # === Setup ===
+    accelerate_set_seed(cfg.seed)
+
+    if cfg.paths.output_dir is None:
+        cfg.paths.output_dir = HydraConfig.get().runtime.output_dir
+
+    OmegaConf.resolve(cfg)
+
+    log_message(
+        f"CUDA_VISIBLE_DEVICES: {os.environ.get('CUDA_VISIBLE_DEVICES', 'not set')}",
+        cfg,
+        accelerator,
+    )
+    log_message(f"Number of processes: {accelerator.num_processes}", cfg, accelerator)
+    log_message(f"Process index: {accelerator.process_index}", cfg, accelerator)
+    log_message(f"Mixed precision: {mixed_precision}", cfg, accelerator)
+
+    log_message("=" * 60, cfg, accelerator)
+    log_message(
+        "HNet Training Pipeline (Hydra + Trackio + Accelerate)", cfg, accelerator
+    )
+    log_message("=" * 60, cfg, accelerator)
+    log_message(f"Config:\n{OmegaConf.to_yaml(cfg)}", cfg, accelerator)
+
+    # === Experiment Tracking Init ===
+    init_tracking(cfg, accelerator)
+
+    # === Tokenizer ===
+    log_message("Initializing tokenizer...", cfg, accelerator)
+    tokenizer = ByteTokenizer()
+
+    # === Model ===
+    log_message("Loading model...", cfg, accelerator)
+    if cfg.model.checkpoint_path:
+        model = load_from_pretrained(
+            model_path=cfg.model.checkpoint_path,
+            model_config_path=cfg.model.config_path,
+        )
+        log_message(f"Loaded pretrained: {cfg.model.checkpoint_path}", cfg, accelerator)
+    else:
+        model = load_from_config(
+            model_config_path=cfg.model.config_path,
+            device="cpu",
+        )
+        model.init_weights()
+        log_message("Initialized from scratch", cfg, accelerator)
+
+    model.train()
+
+    # LR multiplier для разны�� стадий (до prepare!)
+    lr_multiplier = list(cfg.training.lr_multiplier)
+    model.apply_lr_multiplier(lr_multiplier)
+    log_message(f"Applied LR multipliers: {lr_multiplier}", cfg, accelerator)
+
+    # Warmup для Triton kernels
+    if cfg.training.warmup_model:
+        log_message("Warming up model...", cfg, accelerator)
+        model = model.to(accelerator.device)
+        model.warmup(verbose=accelerator.is_main_process)
+
+    # Log model info
+    total_params = sum(p.numel() for p in model.parameters())
+    trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
+    log_message(f"Total params: {total_params:,}", cfg, accelerator)
+    log_message(f"Trainable params: {trainable_params:,}", cfg, accelerator)
+
+    # === Data ===
+    log_message("Creating dataloaders...", cfg, accelerator)
+    dataloaders = create_dataloaders(cfg, tokenizer)
+
+    train_dataloader = dataloaders["train"]
+    val_dataloader = dataloaders.get("validation", None)
+
+    log_message(
+        f"Train dataset size: {len(train_dataloader.dataset)}", cfg, accelerator
+    )
+    log_message(
+        f"Train batches per epoch (before DDP split): {len(train_dataloader)}",
+        cfg,
+        accelerator,
+    )
+
+    if val_dataloader:
+        log_message(
+            f"Validation dataset size: {len(val_dataloader.dataset)}", cfg, accelerator
+        )
+        log_message(f"Validation batches: {len(val_dataloader)}", cfg, accelerator)
+    else:
+        log_message("No validation dataset found", cfg, accelerator)
+
+    # === Optimizer ===
+    log_message("Creating optimizer...", cfg, accelerator)
+    param_groups = group_params(model)
+
+    for group in param_groups:
+        if "lr" not in group:
+            group["lr"] = cfg.training.lr
+        else:
+            group["lr"] = cfg.training.lr * group.get("lr_multiplier", 1.0)
+        if "weight_decay" not in group:
+            group["weight_decay"] = cfg.training.weight_decay
+
+    optimizer = torch.optim.AdamW(
+        param_groups,
+        lr=cfg.training.lr,
+        betas=tuple(cfg.training.betas),
+        eps=cfg.training.eps,
+    )
+
+    # === Scheduler ===
+    steps_per_epoch = math.ceil(len(train_dataloader) / accelerator.num_processes)
+    total_steps = (
+        cfg.training.epochs
+        * steps_per_epoch
+        // cfg.training.gradient_accumulation_steps
+    )
+    scheduler = get_lr_scheduler(optimizer, cfg, total_steps)
+
+    log_message(
+        f"Total steps: {total_steps}, Steps per epoch: {steps_per_epoch}",
+        cfg,
+        accelerator,
+    )
+
+    # === Accelerate Prepare ===
+    log_message(
+        "Preparing model, optimizer, and dataloaders with Accelerate...",
+        cfg,
+        accelerator,
+    )
+
+    if val_dataloader is not None:
+        model, optimizer, train_dataloader, val_dataloader, scheduler = (
+            accelerator.prepare(
+                model, optimizer, train_dataloader, val_dataloader, scheduler
+            )
+        )
+    else:
+        model, optimizer, train_dataloader, scheduler = accelerator.prepare(
+            model, optimizer, train_dataloader, scheduler
+        )
+
+    log_message(
+        f"Train batches per epoch (after DDP split): {len(train_dataloader)}",
+        cfg,
+        accelerator,
+    )
+
+    # === Resume ===
+    global_step = 0
+    start_epoch = 1
+
+    if cfg.training.resume and cfg.training.resume_checkpoint:
+        global_step, start_epoch = load_checkpoint(
+            model,
+            optimizer,
+            scheduler,
+            cfg.training.resume_checkpoint,
+            cfg,
+            accelerator,
+        )
+        start_epoch += 1
+
+    # === Training Loop ===
+    log_message("Starting training...", cfg, accelerator)
+
+    best_val_loss = float("inf")
+
+    try:
+        for epoch in range(start_epoch, cfg.training.epochs + 1):
+            log_message(f"\n{'=' * 60}", cfg, accelerator)
+            log_message(f"EPOCH {epoch}/{cfg.training.epochs}", cfg, accelerator)
+            log_message(f"{'=' * 60}", cfg, accelerator)
+
+            global_step, best_val_loss = train_epoch(
+                model=model,
+                dataloader=train_dataloader,
+                optimizer=optimizer,
+                scheduler=scheduler,
+                cfg=cfg,
+                epoch=epoch,
+                global_step=global_step,
+                accelerator=accelerator,
+                val_dataloader=val_dataloader,
+                best_val_loss=best_val_loss,
+            )
+
+            if cfg.logging.save_every_epoch:
+                save_checkpoint(
+                    model, optimizer, scheduler, global_step, epoch, cfg, accelerator
+                )
+
+    except KeyboardInterrupt:
+        log_message("Training interrupted by user", cfg, accelerator)
+        save_checkpoint(
+            model, optimizer, scheduler, global_step, epoch, cfg, accelerator
+        )
+
+    # === Final Save ===
+    log_message("\nTraining completed!", cfg, accelerator)
+
+    if accelerator.is_main_process:
+        final_model_path = Path(cfg.paths.output_dir) / "model_final.pt"
+        unwrapped_model = accelerator.unwrap_model(model)
+        torch.save(unwrapped_model.state_dict(), final_model_path)
+        log_message(f"Final model: {final_model_path}", cfg, accelerator)
+
+    accelerator.wait_for_everyone()
+    accelerator.end_training()
+    finish_tracking()
+
+
+if __name__ == "__main__":
+    main()

hnet_train_rerun/21-40-57/wandb/run-20260524_214058-3nxcy117/files/config.yaml

@@ -0,0 +1,129 @@
+_wandb:
+    value:
+        cli_version: 0.24.0
+        code_path: code/code_completion_exp/train_hnet/train.py
+        e:
+            8tbm13mz63ggpgppnq4dj4hkvc961ndv:
+                codePath: code_completion_exp/train_hnet/train.py
+                codePathLocal: train.py
+                cpu_count: 36
+                cpu_count_logical: 72
+                cudaVersion: "13.2"
+                disk:
+                    /:
+                        total: "234075717632"
+                        used: "19460378624"
+                email: nikita@local.ru
+                executable: /venv/bytellm/bin/python
+                git:
+                    commit: b57d04cd9f885094fe832cedecaaa9b128098ca3
+                    remote: https://github.com/naryst/byte-llms-code.git
+                gpu: NVIDIA A100 80GB PCIe
+                gpu_count: 1
+                gpu_nvidia:
+                    - architecture: Ampere
+                      cudaCores: 6912
+                      memoryTotal: "85899345920"
+                      name: NVIDIA A100 80GB PCIe
+                      uuid: GPU-ec9764c4-224a-9508-002f-de7762d78498
+                host: 1a023b226280
+                memory:
+                    total: "404274495488"
+                os: Linux-6.8.0-117-generic-x86_64-with-glibc2.39
+                program: /workspace/byte-llms-code/code_completion_exp/train_hnet/train.py
+                python: CPython 3.12.0
+                root: outputs/2026-05-24/21-40-57
+                startedAt: "2026-05-24T21:40:58.723386Z"
+                writerId: 8tbm13mz63ggpgppnq4dj4hkvc961ndv
+        m: []
+        python_version: 3.12.0
+        t:
+            "1":
+                - 1
+                - 11
+                - 49
+                - 50
+                - 51
+                - 71
+                - 105
+            "2":
+                - 1
+                - 11
+                - 49
+                - 50
+                - 51
+                - 71
+                - 105
+            "3":
+                - 2
+                - 13
+                - 16
+                - 61
+            "4": 3.12.0
+            "5": 0.24.0
+            "6": 4.57.6
+            "12": 0.24.0
+            "13": linux-x86_64
+data:
+    value:
+        max_context_len: 4096
+        max_target_len: 256
+        max_train_samples: null
+        max_val_samples: null
+        num_workers: 0
+        path: /workspace/byte-llms-code/code_completion_exp/datasets/data_V4_full
+        pin_memory: true
+device:
+    value: cuda
+logging:
+    value:
+        eval_interval: 2000
+        log_interval: 10
+        save_every_epoch: false
+        save_interval: 3000
+model:
+    value:
+        checkpoint_path: /workspace/byte-llms-code/hnet_project/checkpoints/hnet_2stage_XL_code.pt
+        config_path: /workspace/byte-llms-code/hnet_project/configs/hnet_2stage_XL_code.json
+paths:
+    value:
+        output_dir: outputs/2026-05-24/21-40-57
+seed:
+    value: 42
+tracking:
+    value:
+        backend: wandb
+        base_url: https://wandb.platun0v.ru
+        enabled: true
+        entity: null
+        local_dir: outputs/2026-05-24/21-40-57
+        project: hnet_rerun
+        run_name: hnet_train
+training:
+    value:
+        batch_size: 8
+        betas:
+            - 0.9
+            - 0.95
+        decay_ratio: 0.2
+        epochs: 3
+        eps: 1e-08
+        eval_batch_size: 24
+        gradient_accumulation_steps: 4
+        load_balancing_N: 4
+        load_balancing_weight: 0.01
+        lr: 0.0001
+        lr_multiplier:
+            - 2
+            - 1.5
+            - 1
+        lr_scheduler: wsd
+        max_grad_norm: 1
+        min_lr_ratio: 0.1
+        resume: false
+        resume_checkpoint: null
+        use_amp: true
+        warmup_model: true
+        warmup_ratio: 0.1
+        warmup_steps: 100
+        weight_decay: 0.1

hnet_train_rerun/21-40-57/wandb/run-20260524_214058-3nxcy117/files/requirements.txt

@@ -0,0 +1,245 @@
+setuptools==78.1.1
+wheel==0.45.1
+pip==25.2
+webencodings==0.5.1
+triton==3.2.0
+pytz==2025.2
+pydub==0.25.1
+pure_eval==0.2.3
+ptyprocess==0.7.0
+nvidia-ml-py==13.590.48
+nvidia-cusparselt-cu12==0.6.2
+mpmath==1.3.0
+ipython-genutils==0.2.0
+fastjsonschema==2.21.2
+brotli==1.2.0
+antlr4-python3-runtime==4.9.3
+xxhash==3.6.0
+widgetsnbextension==4.0.14
+websocket-client==1.9.0
+webcolors==24.11.1
+wcwidth==0.2.14
+urllib3==2.5.0
+uri-template==1.3.0
+tzdata==2025.2
+typing_extensions==4.15.0
+types-python-dateutil==2.9.0.20251008
+traitlets==5.14.3
+tqdm==4.67.1
+tornado==6.5.2
+tomlkit==0.13.3
+tinycss2==1.4.0
+tabulate==0.9.0
+sympy==1.13.1
+soupsieve==2.8
+sniffio==1.3.1
+smmap==5.0.2
+six==1.17.0
+shellingham==1.5.4
+Send2Trash==1.8.3
+semantic-version==2.10.0
+safetensors==0.6.2
+rpds-py==0.27.1
+rfc3986-validator==0.1.1
+regex==2025.9.18
+pyzmq==27.1.0
+PyYAML==6.0.3
+python-multipart==0.0.22
+python-json-logger==4.0.0
+python-dotenv==1.2.1
+pyparsing==3.2.5
+PyJWT==2.8.0
+Pygments==2.19.2
+pycparser==2.23
+pyarrow==22.0.0
+psutil==7.1.0
+protobuf==6.33.4
+propcache==0.4.1
+prometheus_client==0.23.1
+portalocker==3.2.0
+platformdirs==4.5.0
+pillow==11.3.0
+pexpect==4.9.0
+pathspec==1.0.4
+parso==0.8.5
+pandocfilters==1.5.1
+packaging==25.0
+orjson==3.11.6
+opt_einsum==3.4.0
+nvidia-nvtx-cu12==12.4.127
+nvidia-nvjitlink-cu12==12.4.127
+nvidia-nccl-cu12==2.21.5
+nvidia-curand-cu12==10.3.5.147
+nvidia-cufile-cu12==1.13.1.3
+nvidia-cufft-cu12==11.2.1.3
+nvidia-cuda-runtime-cu12==12.4.127
+nvidia-cuda-nvrtc-cu12==12.4.127
+nvidia-cuda-cupti-cu12==12.4.127
+nvidia-cublas-cu12==12.4.5.8
+numpy==2.3.3
+ninja==1.13.0
+networkx==3.5
+nest-asyncio==1.6.0
+narwhals==2.15.0
+mypy_extensions==1.1.0
+multidict==6.7.0
+mistune==3.1.4
+mdurl==0.1.2
+MarkupSafe==3.0.3
+lxml==6.0.2
+librt==0.8.0
+lark==1.3.0
+kiwisolver==1.4.9
+jupyterlab_widgets==3.0.15
+jupyterlab_pygments==0.3.0
+jsonpointer==3.0.0
+json5==0.12.1
+itsdangerous==2.2.0
+idna==3.10
+hf-xet==1.1.10
+h11==0.16.0
+groovy==0.1.2
+fsspec==2025.9.0
+frozenlist==1.8.0
+fqdn==1.5.1
+fonttools==4.60.1
+filelock==3.19.1
+ffmpy==1.0.0
+executing==2.2.1
+einops==0.8.1
+dill==0.4.0
+defusedxml==0.7.1
+decorator==5.2.1
+debugpy==1.8.17
+dacite==1.9.2
+cycler==0.12.1
+comm==0.2.3
+colorama==0.4.6
+click==8.3.1
+charset-normalizer==3.4.3
+certifi==2025.10.5
+bleach==6.2.0
+babel==2.17.0
+attrs==25.4.0
+async-lru==2.0.5
+asttokens==3.0.0
+annotated-types==0.7.0
+annotated-doc==0.0.4
+aiohappyeyeballs==2.6.1
+aiofiles==24.1.0
+yarl==1.22.0
+uvicorn==0.40.0
+typing-inspection==0.4.2
+terminado==0.18.1
+stack-data==0.6.3
+sentry-sdk==2.50.0
+scipy==1.17.0
+sacrebleu==2.6.0
+rfc3987-syntax==1.1.0
+rfc3339-validator==0.1.4
+requests==2.32.5
+reportlab==4.4.9
+referencing==0.36.2
+python-dateutil==2.9.0.post0
+pydantic_core==2.41.5
+prompt_toolkit==3.0.52
+plotly==6.5.2
+pathlib2==2.3.7.post1
+orderedmultidict==1.0.2
+optree==0.17.0
+omegaconf==2.3.0
+nvidia-cusparse-cu12==12.3.1.170
+nvidia-cudnn-cu12==9.1.0.70
+mypy==1.19.1
+multiprocess==0.70.16
+matplotlib-inline==0.1.7
+markdown-it-py==4.0.0
+jupyter_core==5.8.1
+Jinja2==3.1.6
+jedi==0.19.2
+ipython_pygments_lexers==1.1.1
+httpcore==1.0.9
+gitdb==4.0.12
+ftfy==6.3.1
+contourpy==1.3.3
+cffi==2.0.0
+beautifulsoup4==4.14.2
+anyio==4.11.0
+aiosignal==1.4.0
+starlette==0.50.0
+rich==14.2.0
+pydantic==2.12.5
+pandas==2.3.3
+nvidia-cusolver-cu12==11.6.1.9
+matplotlib==3.10.7
+jupyter_server_terminals==0.5.3
+jupyter_client==8.6.3
+jsonschema-specifications==2025.9.1
+ipython==9.6.0
+hydra-core==1.3.2
+huggingface-hub==0.35.3
+httpx==0.28.1
+GitPython==3.1.46
+furl==2.1.4
+cryptography==46.0.4
+arrow==1.3.0
+argon2-cffi-bindings==25.1.0
+aiohttp==3.13.1
+wandb==0.24.0
+typer==0.21.1
+torch==2.6.0
+tokenizers==0.22.1
+seaborn==0.13.2
+safehttpx==0.1.7
+jsonschema==4.25.1
+joypy==0.2.6
+isoduration==20.11.0
+ipywidgets==8.1.7
+ipykernel==6.30.1
+gradio_client==2.0.3
+fastapi==0.128.0
+Authlib==1.6.6
+argon2-cffi==25.1.0
+transformers==4.57.6
+nbformat==5.10.4
+mlstm_kernels==2.0.2
+jupyter-console==6.6.3
+gradio==6.5.1
+datasets==4.3.0
+clearml==1.16.4
+accelerate==1.10.1
+xlstm==2.0.4
+nbclient==0.10.2
+jupyter-events==0.12.0
+trackio==0.15.0
+nbconvert==7.16.6
+jupyter_server==2.17.0
+notebook_shim==0.2.4
+jupyterlab_server==2.27.3
+jupyter-lsp==2.3.0
+nbclassic==1.3.3
+jupyterlab==4.4.9
+notebook==7.4.7
+jupyter_contrib_core==0.4.2
+jupyter==1.1.1
+jupyter_nbextensions_configurator==0.6.4
+causal-conv1d==1.5.0.post8
+flash_attn==2.7.4.post1
+mamba-ssm==2.2.4
+hnet==0.0.1
+autocommand==2.2.2
+backports.tarfile==1.2.0
+importlib_metadata==8.0.0
+inflect==7.3.1
+jaraco.collections==5.1.0
+jaraco.context==5.3.0
+jaraco.functools==4.0.1
+jaraco.text==3.12.1
+more-itertools==10.3.0
+packaging==24.2
+platformdirs==4.2.2
+tomli==2.0.1
+typeguard==4.3.0
+typing_extensions==4.12.2
+wheel==0.45.1
+zipp==3.19.2

hnet_train_rerun/21-40-57/wandb/run-20260524_214058-3nxcy117/files/wandb-metadata.json

@@ -0,0 +1,40 @@
+{
+  "os": "Linux-6.8.0-117-generic-x86_64-with-glibc2.39",
+  "python": "CPython 3.12.0",
+  "startedAt": "2026-05-24T21:40:58.723386Z",
+  "program": "/workspace/byte-llms-code/code_completion_exp/train_hnet/train.py",
+  "codePath": "code_completion_exp/train_hnet/train.py",
+  "codePathLocal": "train.py",
+  "git": {
+    "remote": "https://github.com/naryst/byte-llms-code.git",
+    "commit": "b57d04cd9f885094fe832cedecaaa9b128098ca3"
+  },
+  "email": "nikita@local.ru",
+  "root": "outputs/2026-05-24/21-40-57",
+  "host": "1a023b226280",
+  "executable": "/venv/bytellm/bin/python",
+  "cpu_count": 36,
+  "cpu_count_logical": 72,
+  "gpu": "NVIDIA A100 80GB PCIe",
+  "gpu_count": 1,
+  "disk": {
+    "/": {
+      "total": "234075717632",
+      "used": "19460378624"
+    }
+  },
+  "memory": {
+    "total": "404274495488"
+  },
+  "gpu_nvidia": [
+    {
+      "name": "NVIDIA A100 80GB PCIe",
+      "memoryTotal": "85899345920",
+      "cudaCores": 6912,
+      "architecture": "Ampere",
+      "uuid": "GPU-ec9764c4-224a-9508-002f-de7762d78498"
+    }
+  ],
+  "cudaVersion": "13.2",
+  "writerId": "8tbm13mz63ggpgppnq4dj4hkvc961ndv"
+}

hnet_train_rerun/21-40-57/wandb/run-20260524_214058-3nxcy117/files/wandb-summary.json

@@ -0,0 +1 @@
+{"_wandb":{"runtime":41443},"train/hard_boundary_ratio_stage1":0.41428831398941984,"train/step_time":1.216536521911621,"_runtime":41443,"best/val_loss":0.2925644644940842,"train/chunk_len_stage0":2.823436495191354,"train/hard_boundary_ratio_stage0":0.35573476731596815,"train/loss_avg":0.15543903209881482,"train/lb_loss":1.1170897483825684,"val/perplexity":1.3941280151478466,"train/lm_loss":0.10374006628990173,"epoch/soft_boundary_ratio_stage0":0.3510054961482771,"train/soft_boundary_ratio_stage1":0.38348458492953624,"best/step":8000,"train/loss":0.2474285177886486,"_timestamp":1.7797002983859267e+09,"val/lm_loss":0.3314347568349802,"epoch/chunk_len_stage1":2.4207175499344795,"val/time":391.4486954212189,"epoch/hard_boundary_ratio_stage0":0.35573612602838667,"epoch/chunk_len_stage0":2.8234248996329034,"epoch/loss":0.15544442606447362,"epoch/lb_loss":1.0893232608403356,"epoch/time":13811.206544399261,"train/lr":1.000002524293968e-05,"epoch/hard_boundary_ratio_stage1":0.41428892502981024,"train/soft_boundary_ratio_stage0":0.35100414221000376,"epoch/soft_boundary_ratio_stage1":0.38348499507634093,"_step":29660,"train/epoch":3,"epoch/lm_loss":0.14452074673337414,"val/loss":0.34228874700274164,"val/lb_loss":1.0853990667925593,"train/chunk_len_stage1":2.4207217612011815,"best/val_perplexity":1.3263933377204589}

hnet_train_rerun/21-40-57/wandb/run-20260524_214058-3nxcy117/logs/debug-core.log

@@ -0,0 +1,16 @@
+{"time":"2026-05-24T21:40:58.831451528Z","level":"INFO","msg":"main: starting server","port-filename":"/tmp/tmpphm5sb3s/port-83034.txt","pid":83034,"log-level":0,"disable-analytics":false,"shutdown-on-parent-exit":false,"enable-dcgm-profiling":false}
+{"time":"2026-05-24T21:40:58.832303978Z","level":"INFO","msg":"server: will exit if parent process dies","ppid":83034}
+{"time":"2026-05-24T21:40:58.832293139Z","level":"INFO","msg":"server: accepting connections","addr":{"Name":"/tmp/wandb-83034-83044-3980445079/socket","Net":"unix"}}
+{"time":"2026-05-24T21:40:59.018472912Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"1(@)"}
+{"time":"2026-05-24T21:40:59.042712675Z","level":"INFO","msg":"handleInformInit: received","streamId":"3nxcy117","id":"1(@)"}
+{"time":"2026-05-24T21:40:59.473935735Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"3nxcy117","id":"1(@)"}
+{"time":"2026-05-25T09:11:44.387806371Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"3nxcy117","id":"1(@)"}
+{"time":"2026-05-25T09:11:44.402875303Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"3nxcy117","id":"1(@)"}
+{"time":"2026-05-25T09:11:44.423772595Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"}
+{"time":"2026-05-25T09:11:44.423812263Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"}
+{"time":"2026-05-25T09:11:44.423823672Z","level":"INFO","msg":"server is shutting down"}
+{"time":"2026-05-25T09:11:44.423847378Z","level":"INFO","msg":"connection: closing","id":"1(@)"}
+{"time":"2026-05-25T09:11:44.423931379Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-83034-83044-3980445079/socket","Net":"unix"}}
+{"time":"2026-05-25T09:11:44.423990926Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"}
+{"time":"2026-05-25T09:11:44.424021181Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"}
+{"time":"2026-05-25T09:11:44.424042817Z","level":"INFO","msg":"server is closed"}

hnet_train_rerun/21-40-57/wandb/run-20260524_214058-3nxcy117/logs/debug-internal.log

@@ -0,0 +1,13 @@
+{"time":"2026-05-24T21:40:59.042920084Z","level":"INFO","msg":"stream: starting","core version":"0.24.0"}
+{"time":"2026-05-24T21:40:59.47364371Z","level":"INFO","msg":"stream: created new stream","id":"3nxcy117"}
+{"time":"2026-05-24T21:40:59.473752145Z","level":"INFO","msg":"handler: started","stream_id":"3nxcy117"}
+{"time":"2026-05-24T21:40:59.473916706Z","level":"INFO","msg":"stream: started","id":"3nxcy117"}
+{"time":"2026-05-24T21:40:59.473946539Z","level":"INFO","msg":"sender: started","stream_id":"3nxcy117"}
+{"time":"2026-05-24T21:40:59.473947844Z","level":"INFO","msg":"writer: started","stream_id":"3nxcy117"}
+{"time":"2026-05-24T21:40:59.608257445Z","level":"ERROR","msg":"git repo not found","error":"repository does not exist"}
+{"time":"2026-05-25T09:11:44.25084313Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2026-05-25T09:11:44.382625783Z","level":"INFO","msg":"handler: operation stats","stats":{}}
+{"time":"2026-05-25T09:11:44.387865173Z","level":"INFO","msg":"stream: closing","id":"3nxcy117"}
+{"time":"2026-05-25T09:11:44.387891658Z","level":"INFO","msg":"handler: closed","stream_id":"3nxcy117"}
+{"time":"2026-05-25T09:11:44.388086651Z","level":"INFO","msg":"sender: closed","stream_id":"3nxcy117"}
+{"time":"2026-05-25T09:11:44.388103059Z","level":"INFO","msg":"stream: closed","id":"3nxcy117"}

hnet_train_rerun/21-40-57/wandb/run-20260524_214058-3nxcy117/logs/debug.log

@@ -0,0 +1,24 @@
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_setup.py:_flush():81] Current SDK version is 0.24.0
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_setup.py:_flush():81] Configure stats pid to 83034
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_setup.py:_flush():81] Loading settings from environment variables
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_init.py:setup_run_log_directory():717] Logging user logs to outputs/2026-05-24/21-40-57/wandb/run-20260524_214058-3nxcy117/logs/debug.log
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_init.py:setup_run_log_directory():718] Logging internal logs to outputs/2026-05-24/21-40-57/wandb/run-20260524_214058-3nxcy117/logs/debug-internal.log
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_init.py:init():844] calling init triggers
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_init.py:init():849] wandb.init called with sweep_config: {}
+config: {'model': {'config_path': '/workspace/byte-llms-code/hnet_project/configs/hnet_2stage_XL_code.json', 'checkpoint_path': '/workspace/byte-llms-code/hnet_project/checkpoints/hnet_2stage_XL_code.pt'}, 'training': {'epochs': 3, 'batch_size': 8, 'eval_batch_size': 24, 'gradient_accumulation_steps': 4, 'lr': 0.0001, 'weight_decay': 0.1, 'betas': [0.9, 0.95], 'eps': 1e-08, 'lr_scheduler': 'wsd', 'warmup_ratio': 0.1, 'decay_ratio': 0.2, 'warmup_steps': 100, 'min_lr_ratio': 0.1, 'lr_multiplier': [2.0, 1.5, 1.0], 'load_balancing_weight': 0.01, 'load_balancing_N': 4.0, 'max_grad_norm': 1.0, 'use_amp': True, 'resume': False, 'resume_checkpoint': None, 'warmup_model': True}, 'data': {'path': '/workspace/byte-llms-code/code_completion_exp/datasets/data_V4_full', 'max_context_len': 4096, 'max_target_len': 256, 'num_workers': 0, 'pin_memory': True, 'max_train_samples': None, 'max_val_samples': None}, 'logging': {'log_interval': 10, 'save_interval': 3000, 'eval_interval': 2000, 'save_every_epoch': False}, 'tracking': {'enabled': True, 'backend': 'wandb', 'project': 'hnet_rerun', 'run_name': 'hnet_train', 'entity': None, 'base_url': 'https://wandb.platun0v.ru', 'local_dir': 'outputs/2026-05-24/21-40-57'}, 'paths': {'output_dir': 'outputs/2026-05-24/21-40-57'}, 'seed': 42, 'device': 'cuda', '_wandb': {'code_path': 'code/code_completion_exp/train_hnet/train.py'}}
+2026-05-24 21:40:58,725 INFO    MainThread:83034 [wandb_init.py:init():892] starting backend
+2026-05-24 21:40:59,018 INFO    MainThread:83034 [wandb_init.py:init():895] sending inform_init request
+2026-05-24 21:40:59,039 INFO    MainThread:83034 [wandb_init.py:init():903] backend started and connected
+2026-05-24 21:40:59,047 INFO    MainThread:83034 [wandb_init.py:init():973] updated telemetry
+2026-05-24 21:40:59,077 INFO    MainThread:83034 [wandb_init.py:init():997] communicating run to backend with 90.0 second timeout
+2026-05-24 21:40:59,606 INFO    MainThread:83034 [wandb_init.py:init():1044] starting run threads in backend
+2026-05-24 21:40:59,860 INFO    MainThread:83034 [wandb_run.py:_console_start():2529] atexit reg
+2026-05-24 21:40:59,860 INFO    MainThread:83034 [wandb_run.py:_redirect():2377] redirect: wrap_raw
+2026-05-24 21:40:59,860 INFO    MainThread:83034 [wandb_run.py:_redirect():2446] Wrapping output streams.
+2026-05-24 21:40:59,861 INFO    MainThread:83034 [wandb_run.py:_redirect():2469] Redirects installed.
+2026-05-24 21:40:59,865 INFO    MainThread:83034 [wandb_init.py:init():1084] run started, returning control to user process
+2026-05-25 09:11:42,798 INFO    MainThread:83034 [wandb_run.py:_finish():2295] finishing run nikita/hnet_rerun/3nxcy117
+2026-05-25 09:11:42,798 INFO    MainThread:83034 [wandb_run.py:_atexit_cleanup():2494] got exitcode: 0
+2026-05-25 09:11:42,799 INFO    MainThread:83034 [wandb_run.py:_restore():2476] restore
+2026-05-25 09:11:42,799 INFO    MainThread:83034 [wandb_run.py:_restore():2482] restore done
+2026-05-25 09:11:44,386 INFO    MainThread:83034 [wandb_run.py:_footer_sync_info():3870] logging synced files