add albertina

bootstrap_confidence_intervals-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:74dcbf8cc574efe06dda34ff9b1ccb6d2a07e4f9cb21d0f66d1e4dc0e7e95f6b
-size 29581
+oid sha256:074e9f9d7094de3d194528f70edbd86e929dd3303cb3beb55fc4a4fec07f7d46
+size 28701

create_parquet_files.py

@@ -104,6 +104,8 @@ def simplify_experiment_name(name):
             name = name.replace('bert-base-multilingual-cased', 'mbert-base')
         elif 'bert-large-portuguese-cased' in name:
             name = name.replace('bert-large-portuguese-cased', 'bertimbau-large')
+        elif 'albertina-1b5-portuguese-ptbr-encoder' in name:
+            name = name.replace('albertina-1b5-portuguese-ptbr-encoder', 'albertina-1b5-ptbr')
         
         # Handle Llama variants
         elif 'Llama-3.1-8B-llama31_classification_lora' in name:

evaluation_results-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0fca8588cb67b7a43ce4d331b7c998443115ff62ed9e462adb1d33eb032426a8
-size 70007
+oid sha256:314a1e842e3c233857b23955ffe06d109c0b465905994de4ea68fcfe40a92c84
+size 68039

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only/.hydra/config.yaml

@@ -0,0 +1,41 @@
+cache_dir: /tmp/
+dataset:
+  name: kamel-usp/aes_enem_dataset
+  split: JBCS2025
+training_params:
+  seed: 42
+  num_train_epochs: 20
+  logging_steps: 100
+  metric_for_best_model: QWK
+  bf16: true
+bootstrap:
+  enabled: true
+  n_bootstrap: 10000
+  bootstrap_seed: 42
+  metrics:
+  - QWK
+  - Macro_F1
+  - Weighted_F1
+post_training_results:
+  model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
+experiments:
+  model:
+    name: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only
+    type: encoder_classification
+    num_labels: 6
+    output_dir: ./results/
+    logging_dir: ./logs/
+    best_model_dir: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only
+  tokenizer:
+    name: PORTULAN/albertina-1b5-portuguese-ptbr-encoder
+  dataset:
+    grade_index: 0
+    use_full_context: false
+  training_params:
+    weight_decay: 0.01
+    warmup_ratio: 0.1
+    learning_rate: 5.0e-05
+    train_batch_size: 4
+    eval_batch_size: 4
+    gradient_accumulation_steps: 4
+    gradient_checkpointing: false

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only/.hydra/hydra.yaml

@@ -0,0 +1,157 @@
+hydra:
+  run:
+    dir: inference_output/2025-07-11/01-30-40
+  sweep:
+    dir: 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.run.dir=inference_output/2025-07-11/01-30-40
+    - hydra.mode=RUN
+    task:
+    - experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only
+  job:
+    name: run_inference_experiment
+    chdir: null
+    override_dirname: experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only
+    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: '1.1'
+    cwd: /workspace/jbcs2025
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /workspace/jbcs2025/configs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/jbcs2025/inference_output/2025-07-11/01-30-40
+    choices:
+      experiments: temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only
+      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

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only/bootstrap_confidence_intervals.csv

@@ -0,0 +1,2 @@
+experiment_id,timestamp,QWK_mean,QWK_lower_95ci,QWK_upper_95ci,QWK_ci_width,Macro_F1_mean,Macro_F1_lower_95ci,Macro_F1_upper_95ci,Macro_F1_ci_width,Weighted_F1_mean,Weighted_F1_lower_95ci,Weighted_F1_upper_95ci,Weighted_F1_ci_width
+jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only,2025-07-11 01:30:46,0.6816712351479063,0.5762288119767628,0.7805907228920596,0.20436191091529676,0.536327190509169,0.41269287700049234,0.6944784058515344,0.2817855288510421,0.7122266522749188,0.6365110603189872,0.7850158989286448,0.14850483860965757

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only/evaluation_results.csv

@@ -0,0 +1,2 @@
+accuracy,RMSE,QWK,HDIV,Macro_F1,Micro_F1,Weighted_F1,TP_0,TN_0,FP_0,FN_0,TP_1,TN_1,FP_1,FN_1,TP_2,TN_2,FP_2,FN_2,TP_3,TN_3,FP_3,FN_3,TP_4,TN_4,FP_4,FN_4,TP_5,TN_5,FP_5,FN_5,timestamp,id
+0.7028985507246377,25.931906372573962,0.6826328310864394,0.007246376811594235,0.49688027796692447,0.7028985507246377,0.7116672747290037,0,137,0,1,0,138,0,0,6,122,6,4,45,65,7,21,40,71,16,11,6,116,12,4,2025-07-11 01:30:46,jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only/run_inference_experiment.log

@@ -0,0 +1,199 @@
+[2025-07-11 01:30:46,572][__main__][INFO] - Starting inference experiment
+[2025-07-11 01:30:46,574][__main__][INFO] - cache_dir: /tmp/
+dataset:
+  name: kamel-usp/aes_enem_dataset
+  split: JBCS2025
+training_params:
+  seed: 42
+  num_train_epochs: 20
+  logging_steps: 100
+  metric_for_best_model: QWK
+  bf16: true
+bootstrap:
+  enabled: true
+  n_bootstrap: 10000
+  bootstrap_seed: 42
+  metrics:
+  - QWK
+  - Macro_F1
+  - Weighted_F1
+post_training_results:
+  model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
+experiments:
+  model:
+    name: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only
+    type: encoder_classification
+    num_labels: 6
+    output_dir: ./results/
+    logging_dir: ./logs/
+    best_model_dir: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only
+  tokenizer:
+    name: PORTULAN/albertina-1b5-portuguese-ptbr-encoder
+  dataset:
+    grade_index: 0
+    use_full_context: false
+  training_params:
+    weight_decay: 0.01
+    warmup_ratio: 0.1
+    learning_rate: 5.0e-05
+    train_batch_size: 4
+    eval_batch_size: 4
+    gradient_accumulation_steps: 4
+    gradient_checkpointing: false
+
+[2025-07-11 01:30:46,576][__main__][INFO] - Running inference with fine-tuned HF model
+[2025-07-11 01:30:50,657][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/config.json
+[2025-07-11 01:30:50,660][transformers.configuration_utils][INFO] - Model config DebertaV2Config {
+  "architectures": [
+    "DebertaV2ForMaskedLM"
+  ],
+  "attention_head_size": 64,
+  "attention_probs_dropout_prob": 0.1,
+  "conv_act": "gelu",
+  "conv_kernel_size": 3,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 6144,
+  "layer_norm_eps": 1e-07,
+  "legacy": true,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "model_type": "deberta-v2",
+  "norm_rel_ebd": "layer_norm",
+  "num_attention_heads": 24,
+  "num_hidden_layers": 48,
+  "pad_token_id": 0,
+  "pooler_dropout": 0,
+  "pooler_hidden_act": "gelu",
+  "pooler_hidden_size": 1536,
+  "pos_att_type": [
+    "p2c",
+    "c2p"
+  ],
+  "position_biased_input": false,
+  "position_buckets": 256,
+  "relative_attention": true,
+  "share_att_key": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 0,
+  "vocab_size": 128100
+}
+
+[2025-07-11 01:30:51,082][transformers.tokenization_utils_base][INFO] - loading file spm.model from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/spm.model
+[2025-07-11 01:30:51,083][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/tokenizer.json
+[2025-07-11 01:30:51,083][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/added_tokens.json
+[2025-07-11 01:30:51,083][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/special_tokens_map.json
+[2025-07-11 01:30:51,083][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/tokenizer_config.json
+[2025-07-11 01:30:51,083][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
+[2025-07-11 01:30:51,312][__main__][INFO] - Tokenizer function parameters- Padding:longest; Truncation: True; Use Full Context: False
+[2025-07-11 01:30:51,723][__main__][INFO] - 
+Token statistics for 'train' split:
+[2025-07-11 01:30:51,723][__main__][INFO] -   Total examples: 500
+[2025-07-11 01:30:51,723][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:30:51,723][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:30:51,723][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:30:51,723][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:30:51,817][__main__][INFO] - 
+Token statistics for 'validation' split:
+[2025-07-11 01:30:51,817][__main__][INFO] -   Total examples: 132
+[2025-07-11 01:30:51,817][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:30:51,817][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:30:51,817][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:30:51,817][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:30:51,915][__main__][INFO] - 
+Token statistics for 'test' split:
+[2025-07-11 01:30:51,916][__main__][INFO] -   Total examples: 138
+[2025-07-11 01:30:51,916][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:30:51,916][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:30:51,916][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:30:51,916][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:30:51,916][__main__][INFO] - If token statistics are the same (max, avg, min) keep in mind that this is due to batched tokenization and padding.
+[2025-07-11 01:30:51,916][__main__][INFO] - Model max length: 512. If it is the same as stats, then there is a high chance that sequences are being truncated.
+[2025-07-11 01:30:51,916][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only
+[2025-07-11 01:30:51,916][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only
+[2025-07-11 01:30:53,129][__main__][INFO] - Model need ≈ 9.51 GiB to run inference and 27.02 for training 
+[2025-07-11 01:30:53,932][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only/snapshots/5ad996b4a4a15ce313490a7dd5f9001638938bae/config.json
+[2025-07-11 01:30:53,933][transformers.configuration_utils][INFO] - Model config DebertaV2Config {
+  "architectures": [
+    "DebertaV2ForSequenceClassification"
+  ],
+  "attention_head_size": 64,
+  "attention_probs_dropout_prob": 0.1,
+  "conv_act": "gelu",
+  "conv_kernel_size": 3,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1536,
+  "id2label": {
+    "0": 0,
+    "1": 40,
+    "2": 80,
+    "3": 120,
+    "4": 160,
+    "5": 200
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 6144,
+  "label2id": {
+    "0": 0,
+    "40": 1,
+    "80": 2,
+    "120": 3,
+    "160": 4,
+    "200": 5
+  },
+  "layer_norm_eps": 1e-07,
+  "legacy": true,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "model_type": "deberta-v2",
+  "norm_rel_ebd": "layer_norm",
+  "num_attention_heads": 24,
+  "num_hidden_layers": 48,
+  "pad_token_id": 0,
+  "pooler_dropout": 0,
+  "pooler_hidden_act": "gelu",
+  "pooler_hidden_size": 1536,
+  "pos_att_type": [
+    "p2c",
+    "c2p"
+  ],
+  "position_biased_input": false,
+  "position_buckets": 256,
+  "relative_attention": true,
+  "share_att_key": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 0,
+  "vocab_size": 128100
+}
+
+[2025-07-11 01:31:54,528][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only/snapshots/5ad996b4a4a15ce313490a7dd5f9001638938bae/model.safetensors
+[2025-07-11 01:31:54,532][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.bfloat16 as defined in model's config object
+[2025-07-11 01:31:54,532][transformers.modeling_utils][INFO] - Instantiating DebertaV2ForSequenceClassification model under default dtype torch.bfloat16.
+[2025-07-11 01:31:56,446][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing DebertaV2ForSequenceClassification.
+
+[2025-07-11 01:31:56,446][transformers.modeling_utils][INFO] - All the weights of DebertaV2ForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only.
+If your task is similar to the task the model of the checkpoint was trained on, you can already use DebertaV2ForSequenceClassification for predictions without further training.
+[2025-07-11 01:31:56,466][transformers.training_args][INFO] - PyTorch: setting up devices
+[2025-07-11 01:31:56,489][transformers.training_args][INFO] - The default value for the training argument `--report_to` will change in v5 (from all installed integrations to none). In v5, you will need to use `--report_to all` to get the same behavior as now. You should start updating your code and make this info disappear :-).
+[2025-07-11 01:31:56,497][transformers.trainer][INFO] - You have loaded a model on multiple GPUs. `is_model_parallel` attribute will be force-set to `True` to avoid any unexpected behavior such as device placement mismatching.
+[2025-07-11 01:31:56,522][transformers.trainer][INFO] - Using auto half precision backend
+[2025-07-11 01:31:59,885][__main__][INFO] - Running inference on test dataset
+[2025-07-11 01:31:59,887][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `DebertaV2ForSequenceClassification.forward` and have been ignored: essay_text, prompt, grades, supporting_text, id_prompt, essay_year, reference, id. If essay_text, prompt, grades, supporting_text, id_prompt, essay_year, reference, id are not expected by `DebertaV2ForSequenceClassification.forward`,  you can safely ignore this message.
+[2025-07-11 01:31:59,900][transformers.trainer][INFO] - 
+***** Running Prediction *****
+[2025-07-11 01:31:59,900][transformers.trainer][INFO] -   Num examples = 138
+[2025-07-11 01:31:59,900][transformers.trainer][INFO] -   Batch size = 4
+[2025-07-11 01:32:08,637][__main__][INFO] - Inference results saved to jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only_inference_results.jsonl
+[2025-07-11 01:32:08,638][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
+[2025-07-11 01:34:15,233][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
+[2025-07-11 01:34:15,233][__main__][INFO] - Bootstrap Confidence Intervals (95%):
+[2025-07-11 01:34:15,233][__main__][INFO] -   QWK: 0.6817 [0.5762, 0.7806]
+[2025-07-11 01:34:15,233][__main__][INFO] -   Macro_F1: 0.5363 [0.4127, 0.6945]
+[2025-07-11 01:34:15,233][__main__][INFO] -   Weighted_F1: 0.7122 [0.6365, 0.7850]
+[2025-07-11 01:34:15,233][__main__][INFO] - Inference results: {'accuracy': 0.7028985507246377, 'RMSE': 25.931906372573962, 'QWK': 0.6826328310864394, 'HDIV': 0.007246376811594235, 'Macro_F1': 0.49688027796692447, 'Micro_F1': 0.7028985507246377, 'Weighted_F1': 0.7116672747290037, 'TP_0': np.int64(0), 'TN_0': np.int64(137), 'FP_0': np.int64(0), 'FN_0': np.int64(1), 'TP_1': np.int64(0), 'TN_1': np.int64(138), 'FP_1': np.int64(0), 'FN_1': np.int64(0), 'TP_2': np.int64(6), 'TN_2': np.int64(122), 'FP_2': np.int64(6), 'FN_2': np.int64(4), 'TP_3': np.int64(45), 'TN_3': np.int64(65), 'FP_3': np.int64(7), 'FN_3': np.int64(21), 'TP_4': np.int64(40), 'TN_4': np.int64(71), 'FP_4': np.int64(16), 'FN_4': np.int64(11), 'TP_5': np.int64(6), 'TN_5': np.int64(116), 'FP_5': np.int64(12), 'FN_5': np.int64(4)}
+[2025-07-11 01:34:15,237][__main__][INFO] - Inference experiment completed

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only/.hydra/config.yaml

@@ -0,0 +1,41 @@
+cache_dir: /tmp/
+dataset:
+  name: kamel-usp/aes_enem_dataset
+  split: JBCS2025
+training_params:
+  seed: 42
+  num_train_epochs: 20
+  logging_steps: 100
+  metric_for_best_model: QWK
+  bf16: true
+bootstrap:
+  enabled: true
+  n_bootstrap: 10000
+  bootstrap_seed: 42
+  metrics:
+  - QWK
+  - Macro_F1
+  - Weighted_F1
+post_training_results:
+  model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
+experiments:
+  model:
+    name: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only
+    type: encoder_classification
+    num_labels: 6
+    output_dir: ./results/
+    logging_dir: ./logs/
+    best_model_dir: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only
+  tokenizer:
+    name: PORTULAN/albertina-1b5-portuguese-ptbr-encoder
+  dataset:
+    grade_index: 1
+    use_full_context: false
+  training_params:
+    weight_decay: 0.01
+    warmup_ratio: 0.1
+    learning_rate: 5.0e-05
+    train_batch_size: 4
+    eval_batch_size: 4
+    gradient_accumulation_steps: 4
+    gradient_checkpointing: false

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only/.hydra/hydra.yaml

@@ -0,0 +1,157 @@
+hydra:
+  run:
+    dir: inference_output/2025-07-11/01-34-20
+  sweep:
+    dir: 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.run.dir=inference_output/2025-07-11/01-34-20
+    - hydra.mode=RUN
+    task:
+    - experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only
+  job:
+    name: run_inference_experiment
+    chdir: null
+    override_dirname: experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only
+    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: '1.1'
+    cwd: /workspace/jbcs2025
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /workspace/jbcs2025/configs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/jbcs2025/inference_output/2025-07-11/01-34-20
+    choices:
+      experiments: temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only
+      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

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only/bootstrap_confidence_intervals.csv

@@ -0,0 +1,2 @@
+experiment_id,timestamp,QWK_mean,QWK_lower_95ci,QWK_upper_95ci,QWK_ci_width,Macro_F1_mean,Macro_F1_lower_95ci,Macro_F1_upper_95ci,Macro_F1_ci_width,Weighted_F1_mean,Weighted_F1_lower_95ci,Weighted_F1_upper_95ci,Weighted_F1_ci_width
+jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only,2025-07-11 01:34:26,0.2382531468171123,0.08558340882557128,0.386519313107836,0.3009359042822647,0.13898233480051359,0.08334403539825075,0.20313595816462865,0.1197919227663779,0.15418097819475352,0.0921235036938581,0.22086688751892972,0.1287433838250716

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only/evaluation_results.csv

@@ -0,0 +1,2 @@
+accuracy,RMSE,QWK,HDIV,Macro_F1,Micro_F1,Weighted_F1,TP_0,TN_0,FP_0,FN_0,TP_1,TN_1,FP_1,FN_1,TP_2,TN_2,FP_2,FN_2,TP_3,TN_3,FP_3,FN_3,TP_4,TN_4,FP_4,FN_4,TP_5,TN_5,FP_5,FN_5,timestamp,id
+0.15942028985507245,60.72031391337771,0.24034067303697548,0.1159420289855072,0.13165647872210337,0.15942028985507245,0.1542258389739044,0,137,0,1,0,99,4,35,3,74,59,2,6,73,14,45,9,79,33,17,4,112,6,16,2025-07-11 01:34:26,jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only/run_inference_experiment.log

@@ -0,0 +1,199 @@
+[2025-07-11 01:34:25,998][__main__][INFO] - Starting inference experiment
+[2025-07-11 01:34:25,999][__main__][INFO] - cache_dir: /tmp/
+dataset:
+  name: kamel-usp/aes_enem_dataset
+  split: JBCS2025
+training_params:
+  seed: 42
+  num_train_epochs: 20
+  logging_steps: 100
+  metric_for_best_model: QWK
+  bf16: true
+bootstrap:
+  enabled: true
+  n_bootstrap: 10000
+  bootstrap_seed: 42
+  metrics:
+  - QWK
+  - Macro_F1
+  - Weighted_F1
+post_training_results:
+  model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
+experiments:
+  model:
+    name: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only
+    type: encoder_classification
+    num_labels: 6
+    output_dir: ./results/
+    logging_dir: ./logs/
+    best_model_dir: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only
+  tokenizer:
+    name: PORTULAN/albertina-1b5-portuguese-ptbr-encoder
+  dataset:
+    grade_index: 1
+    use_full_context: false
+  training_params:
+    weight_decay: 0.01
+    warmup_ratio: 0.1
+    learning_rate: 5.0e-05
+    train_batch_size: 4
+    eval_batch_size: 4
+    gradient_accumulation_steps: 4
+    gradient_checkpointing: false
+
+[2025-07-11 01:34:26,001][__main__][INFO] - Running inference with fine-tuned HF model
+[2025-07-11 01:34:30,940][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/config.json
+[2025-07-11 01:34:30,943][transformers.configuration_utils][INFO] - Model config DebertaV2Config {
+  "architectures": [
+    "DebertaV2ForMaskedLM"
+  ],
+  "attention_head_size": 64,
+  "attention_probs_dropout_prob": 0.1,
+  "conv_act": "gelu",
+  "conv_kernel_size": 3,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 6144,
+  "layer_norm_eps": 1e-07,
+  "legacy": true,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "model_type": "deberta-v2",
+  "norm_rel_ebd": "layer_norm",
+  "num_attention_heads": 24,
+  "num_hidden_layers": 48,
+  "pad_token_id": 0,
+  "pooler_dropout": 0,
+  "pooler_hidden_act": "gelu",
+  "pooler_hidden_size": 1536,
+  "pos_att_type": [
+    "p2c",
+    "c2p"
+  ],
+  "position_biased_input": false,
+  "position_buckets": 256,
+  "relative_attention": true,
+  "share_att_key": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 0,
+  "vocab_size": 128100
+}
+
+[2025-07-11 01:34:31,458][transformers.tokenization_utils_base][INFO] - loading file spm.model from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/spm.model
+[2025-07-11 01:34:31,458][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/tokenizer.json
+[2025-07-11 01:34:31,458][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/added_tokens.json
+[2025-07-11 01:34:31,458][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/special_tokens_map.json
+[2025-07-11 01:34:31,458][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/tokenizer_config.json
+[2025-07-11 01:34:31,458][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
+[2025-07-11 01:34:31,689][__main__][INFO] - Tokenizer function parameters- Padding:longest; Truncation: True; Use Full Context: False
+[2025-07-11 01:34:32,088][__main__][INFO] - 
+Token statistics for 'train' split:
+[2025-07-11 01:34:32,088][__main__][INFO] -   Total examples: 500
+[2025-07-11 01:34:32,088][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:34:32,088][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:34:32,088][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:34:32,088][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:34:32,181][__main__][INFO] - 
+Token statistics for 'validation' split:
+[2025-07-11 01:34:32,181][__main__][INFO] -   Total examples: 132
+[2025-07-11 01:34:32,181][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:34:32,181][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:34:32,181][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:34:32,181][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:34:32,281][__main__][INFO] - 
+Token statistics for 'test' split:
+[2025-07-11 01:34:32,281][__main__][INFO] -   Total examples: 138
+[2025-07-11 01:34:32,281][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:34:32,281][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:34:32,281][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:34:32,281][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:34:32,281][__main__][INFO] - If token statistics are the same (max, avg, min) keep in mind that this is due to batched tokenization and padding.
+[2025-07-11 01:34:32,281][__main__][INFO] - Model max length: 512. If it is the same as stats, then there is a high chance that sequences are being truncated.
+[2025-07-11 01:34:32,281][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only
+[2025-07-11 01:34:32,282][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only
+[2025-07-11 01:34:33,362][__main__][INFO] - Model need ≈ 9.51 GiB to run inference and 27.02 for training 
+[2025-07-11 01:34:34,304][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only/snapshots/adb37bfa550aaf8233d141929271f33f9a7147f1/config.json
+[2025-07-11 01:34:34,304][transformers.configuration_utils][INFO] - Model config DebertaV2Config {
+  "architectures": [
+    "DebertaV2ForSequenceClassification"
+  ],
+  "attention_head_size": 64,
+  "attention_probs_dropout_prob": 0.1,
+  "conv_act": "gelu",
+  "conv_kernel_size": 3,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1536,
+  "id2label": {
+    "0": 0,
+    "1": 40,
+    "2": 80,
+    "3": 120,
+    "4": 160,
+    "5": 200
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 6144,
+  "label2id": {
+    "0": 0,
+    "40": 1,
+    "80": 2,
+    "120": 3,
+    "160": 4,
+    "200": 5
+  },
+  "layer_norm_eps": 1e-07,
+  "legacy": true,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "model_type": "deberta-v2",
+  "norm_rel_ebd": "layer_norm",
+  "num_attention_heads": 24,
+  "num_hidden_layers": 48,
+  "pad_token_id": 0,
+  "pooler_dropout": 0,
+  "pooler_hidden_act": "gelu",
+  "pooler_hidden_size": 1536,
+  "pos_att_type": [
+    "p2c",
+    "c2p"
+  ],
+  "position_biased_input": false,
+  "position_buckets": 256,
+  "relative_attention": true,
+  "share_att_key": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 0,
+  "vocab_size": 128100
+}
+
+[2025-07-11 01:35:30,309][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only/snapshots/adb37bfa550aaf8233d141929271f33f9a7147f1/model.safetensors
+[2025-07-11 01:35:30,313][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.bfloat16 as defined in model's config object
+[2025-07-11 01:35:30,313][transformers.modeling_utils][INFO] - Instantiating DebertaV2ForSequenceClassification model under default dtype torch.bfloat16.
+[2025-07-11 01:35:31,939][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing DebertaV2ForSequenceClassification.
+
+[2025-07-11 01:35:31,939][transformers.modeling_utils][INFO] - All the weights of DebertaV2ForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only.
+If your task is similar to the task the model of the checkpoint was trained on, you can already use DebertaV2ForSequenceClassification for predictions without further training.
+[2025-07-11 01:35:31,960][transformers.training_args][INFO] - PyTorch: setting up devices
+[2025-07-11 01:35:31,984][transformers.training_args][INFO] - The default value for the training argument `--report_to` will change in v5 (from all installed integrations to none). In v5, you will need to use `--report_to all` to get the same behavior as now. You should start updating your code and make this info disappear :-).
+[2025-07-11 01:35:31,991][transformers.trainer][INFO] - You have loaded a model on multiple GPUs. `is_model_parallel` attribute will be force-set to `True` to avoid any unexpected behavior such as device placement mismatching.
+[2025-07-11 01:35:32,016][transformers.trainer][INFO] - Using auto half precision backend
+[2025-07-11 01:35:35,310][__main__][INFO] - Running inference on test dataset
+[2025-07-11 01:35:35,311][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `DebertaV2ForSequenceClassification.forward` and have been ignored: essay_text, essay_year, id, id_prompt, reference, prompt, grades, supporting_text. If essay_text, essay_year, id, id_prompt, reference, prompt, grades, supporting_text are not expected by `DebertaV2ForSequenceClassification.forward`,  you can safely ignore this message.
+[2025-07-11 01:35:35,324][transformers.trainer][INFO] - 
+***** Running Prediction *****
+[2025-07-11 01:35:35,324][transformers.trainer][INFO] -   Num examples = 138
+[2025-07-11 01:35:35,324][transformers.trainer][INFO] -   Batch size = 4
+[2025-07-11 01:35:44,076][__main__][INFO] - Inference results saved to jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only_inference_results.jsonl
+[2025-07-11 01:35:44,077][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
+[2025-07-11 01:37:50,577][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
+[2025-07-11 01:37:50,578][__main__][INFO] - Bootstrap Confidence Intervals (95%):
+[2025-07-11 01:37:50,578][__main__][INFO] -   QWK: 0.2383 [0.0856, 0.3865]
+[2025-07-11 01:37:50,578][__main__][INFO] -   Macro_F1: 0.1390 [0.0833, 0.2031]
+[2025-07-11 01:37:50,578][__main__][INFO] -   Weighted_F1: 0.1542 [0.0921, 0.2209]
+[2025-07-11 01:37:50,578][__main__][INFO] - Inference results: {'accuracy': 0.15942028985507245, 'RMSE': 60.72031391337771, 'QWK': 0.24034067303697548, 'HDIV': 0.1159420289855072, 'Macro_F1': 0.13165647872210337, 'Micro_F1': 0.15942028985507245, 'Weighted_F1': 0.1542258389739044, 'TP_0': np.int64(0), 'TN_0': np.int64(137), 'FP_0': np.int64(0), 'FN_0': np.int64(1), 'TP_1': np.int64(0), 'TN_1': np.int64(99), 'FP_1': np.int64(4), 'FN_1': np.int64(35), 'TP_2': np.int64(3), 'TN_2': np.int64(74), 'FP_2': np.int64(59), 'FN_2': np.int64(2), 'TP_3': np.int64(6), 'TN_3': np.int64(73), 'FP_3': np.int64(14), 'FN_3': np.int64(45), 'TP_4': np.int64(9), 'TN_4': np.int64(79), 'FP_4': np.int64(33), 'FN_4': np.int64(17), 'TP_5': np.int64(4), 'TN_5': np.int64(112), 'FP_5': np.int64(6), 'FN_5': np.int64(16)}
+[2025-07-11 01:37:50,578][__main__][INFO] - Inference experiment completed

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only/.hydra/config.yaml

@@ -0,0 +1,41 @@
+cache_dir: /tmp/
+dataset:
+  name: kamel-usp/aes_enem_dataset
+  split: JBCS2025
+training_params:
+  seed: 42
+  num_train_epochs: 20
+  logging_steps: 100
+  metric_for_best_model: QWK
+  bf16: true
+bootstrap:
+  enabled: true
+  n_bootstrap: 10000
+  bootstrap_seed: 42
+  metrics:
+  - QWK
+  - Macro_F1
+  - Weighted_F1
+post_training_results:
+  model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
+experiments:
+  model:
+    name: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only
+    type: encoder_classification
+    num_labels: 6
+    output_dir: ./results/
+    logging_dir: ./logs/
+    best_model_dir: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only
+  tokenizer:
+    name: PORTULAN/albertina-1b5-portuguese-ptbr-encoder
+  dataset:
+    grade_index: 2
+    use_full_context: false
+  training_params:
+    weight_decay: 0.01
+    warmup_ratio: 0.1
+    learning_rate: 5.0e-05
+    train_batch_size: 4
+    eval_batch_size: 4
+    gradient_accumulation_steps: 4
+    gradient_checkpointing: false

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only/.hydra/hydra.yaml

@@ -0,0 +1,157 @@
+hydra:
+  run:
+    dir: inference_output/2025-07-11/01-37-55
+  sweep:
+    dir: 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.run.dir=inference_output/2025-07-11/01-37-55
+    - hydra.mode=RUN
+    task:
+    - experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only
+  job:
+    name: run_inference_experiment
+    chdir: null
+    override_dirname: experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only
+    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: '1.1'
+    cwd: /workspace/jbcs2025
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /workspace/jbcs2025/configs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/jbcs2025/inference_output/2025-07-11/01-37-55
+    choices:
+      experiments: temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only
+      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

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only/bootstrap_confidence_intervals.csv

@@ -0,0 +1,2 @@
+experiment_id,timestamp,QWK_mean,QWK_lower_95ci,QWK_upper_95ci,QWK_ci_width,Macro_F1_mean,Macro_F1_lower_95ci,Macro_F1_upper_95ci,Macro_F1_ci_width,Weighted_F1_mean,Weighted_F1_lower_95ci,Weighted_F1_upper_95ci,Weighted_F1_ci_width
+jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only,2025-07-11 01:38:01,0.2517666669121323,0.09322536648339473,0.40469018928519745,0.3114648228018027,0.22857107700198048,0.1640445655410444,0.305809643248977,0.14176507770793262,0.30780478274990675,0.23016566290720317,0.3851747099823908,0.15500904707518762

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only/evaluation_results.csv

@@ -0,0 +1,2 @@
+accuracy,RMSE,QWK,HDIV,Macro_F1,Micro_F1,Weighted_F1,TP_0,TN_0,FP_0,FN_0,TP_1,TN_1,FP_1,FN_1,TP_2,TN_2,FP_2,FN_2,TP_3,TN_3,FP_3,FN_3,TP_4,TN_4,FP_4,FN_4,TP_5,TN_5,FP_5,FN_5,timestamp,id
+0.3115942028985507,59.27093282765904,0.25441318069968977,0.09420289855072461,0.2158405391141964,0.3115942028985507,0.30799298925619395,0,137,0,1,9,78,31,20,7,103,17,11,14,67,26,31,13,81,19,25,0,129,2,7,2025-07-11 01:38:01,jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only/run_inference_experiment.log

@@ -0,0 +1,199 @@
+[2025-07-11 01:38:01,376][__main__][INFO] - Starting inference experiment
+[2025-07-11 01:38:01,377][__main__][INFO] - cache_dir: /tmp/
+dataset:
+  name: kamel-usp/aes_enem_dataset
+  split: JBCS2025
+training_params:
+  seed: 42
+  num_train_epochs: 20
+  logging_steps: 100
+  metric_for_best_model: QWK
+  bf16: true
+bootstrap:
+  enabled: true
+  n_bootstrap: 10000
+  bootstrap_seed: 42
+  metrics:
+  - QWK
+  - Macro_F1
+  - Weighted_F1
+post_training_results:
+  model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
+experiments:
+  model:
+    name: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only
+    type: encoder_classification
+    num_labels: 6
+    output_dir: ./results/
+    logging_dir: ./logs/
+    best_model_dir: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only
+  tokenizer:
+    name: PORTULAN/albertina-1b5-portuguese-ptbr-encoder
+  dataset:
+    grade_index: 2
+    use_full_context: false
+  training_params:
+    weight_decay: 0.01
+    warmup_ratio: 0.1
+    learning_rate: 5.0e-05
+    train_batch_size: 4
+    eval_batch_size: 4
+    gradient_accumulation_steps: 4
+    gradient_checkpointing: false
+
+[2025-07-11 01:38:01,380][__main__][INFO] - Running inference with fine-tuned HF model
+[2025-07-11 01:38:06,302][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/config.json
+[2025-07-11 01:38:06,305][transformers.configuration_utils][INFO] - Model config DebertaV2Config {
+  "architectures": [
+    "DebertaV2ForMaskedLM"
+  ],
+  "attention_head_size": 64,
+  "attention_probs_dropout_prob": 0.1,
+  "conv_act": "gelu",
+  "conv_kernel_size": 3,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 6144,
+  "layer_norm_eps": 1e-07,
+  "legacy": true,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "model_type": "deberta-v2",
+  "norm_rel_ebd": "layer_norm",
+  "num_attention_heads": 24,
+  "num_hidden_layers": 48,
+  "pad_token_id": 0,
+  "pooler_dropout": 0,
+  "pooler_hidden_act": "gelu",
+  "pooler_hidden_size": 1536,
+  "pos_att_type": [
+    "p2c",
+    "c2p"
+  ],
+  "position_biased_input": false,
+  "position_buckets": 256,
+  "relative_attention": true,
+  "share_att_key": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 0,
+  "vocab_size": 128100
+}
+
+[2025-07-11 01:38:06,748][transformers.tokenization_utils_base][INFO] - loading file spm.model from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/spm.model
+[2025-07-11 01:38:06,748][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/tokenizer.json
+[2025-07-11 01:38:06,748][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/added_tokens.json
+[2025-07-11 01:38:06,748][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/special_tokens_map.json
+[2025-07-11 01:38:06,748][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/tokenizer_config.json
+[2025-07-11 01:38:06,748][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
+[2025-07-11 01:38:06,967][__main__][INFO] - Tokenizer function parameters- Padding:longest; Truncation: True; Use Full Context: False
+[2025-07-11 01:38:07,435][__main__][INFO] - 
+Token statistics for 'train' split:
+[2025-07-11 01:38:07,435][__main__][INFO] -   Total examples: 500
+[2025-07-11 01:38:07,435][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:38:07,435][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:38:07,435][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:38:07,435][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:38:07,528][__main__][INFO] - 
+Token statistics for 'validation' split:
+[2025-07-11 01:38:07,528][__main__][INFO] -   Total examples: 132
+[2025-07-11 01:38:07,528][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:38:07,528][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:38:07,528][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:38:07,528][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:38:07,624][__main__][INFO] - 
+Token statistics for 'test' split:
+[2025-07-11 01:38:07,624][__main__][INFO] -   Total examples: 138
+[2025-07-11 01:38:07,624][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:38:07,624][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:38:07,624][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:38:07,624][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:38:07,625][__main__][INFO] - If token statistics are the same (max, avg, min) keep in mind that this is due to batched tokenization and padding.
+[2025-07-11 01:38:07,625][__main__][INFO] - Model max length: 512. If it is the same as stats, then there is a high chance that sequences are being truncated.
+[2025-07-11 01:38:07,625][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only
+[2025-07-11 01:38:07,625][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only
+[2025-07-11 01:38:08,597][__main__][INFO] - Model need ≈ 9.51 GiB to run inference and 27.02 for training 
+[2025-07-11 01:38:09,405][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only/snapshots/932551e130d45fd1107b0d1d9a6a645bfa6fb16a/config.json
+[2025-07-11 01:38:09,406][transformers.configuration_utils][INFO] - Model config DebertaV2Config {
+  "architectures": [
+    "DebertaV2ForSequenceClassification"
+  ],
+  "attention_head_size": 64,
+  "attention_probs_dropout_prob": 0.1,
+  "conv_act": "gelu",
+  "conv_kernel_size": 3,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1536,
+  "id2label": {
+    "0": 0,
+    "1": 40,
+    "2": 80,
+    "3": 120,
+    "4": 160,
+    "5": 200
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 6144,
+  "label2id": {
+    "0": 0,
+    "40": 1,
+    "80": 2,
+    "120": 3,
+    "160": 4,
+    "200": 5
+  },
+  "layer_norm_eps": 1e-07,
+  "legacy": true,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "model_type": "deberta-v2",
+  "norm_rel_ebd": "layer_norm",
+  "num_attention_heads": 24,
+  "num_hidden_layers": 48,
+  "pad_token_id": 0,
+  "pooler_dropout": 0,
+  "pooler_hidden_act": "gelu",
+  "pooler_hidden_size": 1536,
+  "pos_att_type": [
+    "p2c",
+    "c2p"
+  ],
+  "position_biased_input": false,
+  "position_buckets": 256,
+  "relative_attention": true,
+  "share_att_key": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 0,
+  "vocab_size": 128100
+}
+
+[2025-07-11 01:39:10,302][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only/snapshots/932551e130d45fd1107b0d1d9a6a645bfa6fb16a/model.safetensors
+[2025-07-11 01:39:10,305][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.bfloat16 as defined in model's config object
+[2025-07-11 01:39:10,305][transformers.modeling_utils][INFO] - Instantiating DebertaV2ForSequenceClassification model under default dtype torch.bfloat16.
+[2025-07-11 01:39:11,869][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing DebertaV2ForSequenceClassification.
+
+[2025-07-11 01:39:11,869][transformers.modeling_utils][INFO] - All the weights of DebertaV2ForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only.
+If your task is similar to the task the model of the checkpoint was trained on, you can already use DebertaV2ForSequenceClassification for predictions without further training.
+[2025-07-11 01:39:11,889][transformers.training_args][INFO] - PyTorch: setting up devices
+[2025-07-11 01:39:11,916][transformers.training_args][INFO] - The default value for the training argument `--report_to` will change in v5 (from all installed integrations to none). In v5, you will need to use `--report_to all` to get the same behavior as now. You should start updating your code and make this info disappear :-).
+[2025-07-11 01:39:11,924][transformers.trainer][INFO] - You have loaded a model on multiple GPUs. `is_model_parallel` attribute will be force-set to `True` to avoid any unexpected behavior such as device placement mismatching.
+[2025-07-11 01:39:11,949][transformers.trainer][INFO] - Using auto half precision backend
+[2025-07-11 01:39:15,275][__main__][INFO] - Running inference on test dataset
+[2025-07-11 01:39:15,276][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `DebertaV2ForSequenceClassification.forward` and have been ignored: id, essay_year, essay_text, id_prompt, grades, prompt, reference, supporting_text. If id, essay_year, essay_text, id_prompt, grades, prompt, reference, supporting_text are not expected by `DebertaV2ForSequenceClassification.forward`,  you can safely ignore this message.
+[2025-07-11 01:39:15,289][transformers.trainer][INFO] - 
+***** Running Prediction *****
+[2025-07-11 01:39:15,289][transformers.trainer][INFO] -   Num examples = 138
+[2025-07-11 01:39:15,289][transformers.trainer][INFO] -   Batch size = 4
+[2025-07-11 01:39:24,057][__main__][INFO] - Inference results saved to jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only_inference_results.jsonl
+[2025-07-11 01:39:24,058][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
+[2025-07-11 01:41:31,078][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
+[2025-07-11 01:41:31,079][__main__][INFO] - Bootstrap Confidence Intervals (95%):
+[2025-07-11 01:41:31,079][__main__][INFO] -   QWK: 0.2518 [0.0932, 0.4047]
+[2025-07-11 01:41:31,079][__main__][INFO] -   Macro_F1: 0.2286 [0.1640, 0.3058]
+[2025-07-11 01:41:31,079][__main__][INFO] -   Weighted_F1: 0.3078 [0.2302, 0.3852]
+[2025-07-11 01:41:31,079][__main__][INFO] - Inference results: {'accuracy': 0.3115942028985507, 'RMSE': 59.27093282765904, 'QWK': 0.25441318069968977, 'HDIV': 0.09420289855072461, 'Macro_F1': 0.2158405391141964, 'Micro_F1': 0.3115942028985507, 'Weighted_F1': 0.30799298925619395, 'TP_0': np.int64(0), 'TN_0': np.int64(137), 'FP_0': np.int64(0), 'FN_0': np.int64(1), 'TP_1': np.int64(9), 'TN_1': np.int64(78), 'FP_1': np.int64(31), 'FN_1': np.int64(20), 'TP_2': np.int64(7), 'TN_2': np.int64(103), 'FP_2': np.int64(17), 'FN_2': np.int64(11), 'TP_3': np.int64(14), 'TN_3': np.int64(67), 'FP_3': np.int64(26), 'FN_3': np.int64(31), 'TP_4': np.int64(13), 'TN_4': np.int64(81), 'FP_4': np.int64(19), 'FN_4': np.int64(25), 'TP_5': np.int64(0), 'TN_5': np.int64(129), 'FP_5': np.int64(2), 'FN_5': np.int64(7)}
+[2025-07-11 01:41:31,079][__main__][INFO] - Inference experiment completed

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only/.hydra/config.yaml

@@ -0,0 +1,41 @@
+cache_dir: /tmp/
+dataset:
+  name: kamel-usp/aes_enem_dataset
+  split: JBCS2025
+training_params:
+  seed: 42
+  num_train_epochs: 20
+  logging_steps: 100
+  metric_for_best_model: QWK
+  bf16: true
+bootstrap:
+  enabled: true
+  n_bootstrap: 10000
+  bootstrap_seed: 42
+  metrics:
+  - QWK
+  - Macro_F1
+  - Weighted_F1
+post_training_results:
+  model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
+experiments:
+  model:
+    name: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only
+    type: encoder_classification
+    num_labels: 6
+    output_dir: ./results/
+    logging_dir: ./logs/
+    best_model_dir: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only
+  tokenizer:
+    name: PORTULAN/albertina-1b5-portuguese-ptbr-encoder
+  dataset:
+    grade_index: 3
+    use_full_context: false
+  training_params:
+    weight_decay: 0.01
+    warmup_ratio: 0.1
+    learning_rate: 5.0e-05
+    train_batch_size: 4
+    eval_batch_size: 4
+    gradient_accumulation_steps: 4
+    gradient_checkpointing: false

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only/.hydra/hydra.yaml

@@ -0,0 +1,157 @@
+hydra:
+  run:
+    dir: inference_output/2025-07-11/01-41-35
+  sweep:
+    dir: 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.run.dir=inference_output/2025-07-11/01-41-35
+    - hydra.mode=RUN
+    task:
+    - experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only
+  job:
+    name: run_inference_experiment
+    chdir: null
+    override_dirname: experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only
+    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: '1.1'
+    cwd: /workspace/jbcs2025
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /workspace/jbcs2025/configs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/jbcs2025/inference_output/2025-07-11/01-41-35
+    choices:
+      experiments: temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only
+      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

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only/bootstrap_confidence_intervals.csv

@@ -0,0 +1,2 @@
+experiment_id,timestamp,QWK_mean,QWK_lower_95ci,QWK_upper_95ci,QWK_ci_width,Macro_F1_mean,Macro_F1_lower_95ci,Macro_F1_upper_95ci,Macro_F1_ci_width,Weighted_F1_mean,Weighted_F1_lower_95ci,Weighted_F1_upper_95ci,Weighted_F1_ci_width
+jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only,2025-07-11 01:41:41,0.5396531933398239,0.4204347375953364,0.6511327204944922,0.23069798289915583,0.34029020389270115,0.2300104953539315,0.4995002696501232,0.2694897742961917,0.5926251278756054,0.5065006889621307,0.6785371638323412,0.17203647487021045

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only/evaluation_results.csv

@@ -0,0 +1,2 @@
+accuracy,RMSE,QWK,HDIV,Macro_F1,Micro_F1,Weighted_F1,TP_0,TN_0,FP_0,FN_0,TP_1,TN_1,FP_1,FN_1,TP_2,TN_2,FP_2,FN_2,TP_3,TN_3,FP_3,FN_3,TP_4,TN_4,FP_4,FN_4,TP_5,TN_5,FP_5,FN_5,timestamp,id
+0.6014492753623188,27.662562992324986,0.543046357615894,0.007246376811594235,0.2974558904383466,0.6014492753623188,0.5920471313301213,0,137,0,1,0,137,0,1,3,122,7,6,60,36,26,16,18,78,14,28,2,125,8,3,2025-07-11 01:41:41,jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only/run_inference_experiment.log

@@ -0,0 +1,199 @@
+[2025-07-11 01:41:41,920][__main__][INFO] - Starting inference experiment
+[2025-07-11 01:41:41,921][__main__][INFO] - cache_dir: /tmp/
+dataset:
+  name: kamel-usp/aes_enem_dataset
+  split: JBCS2025
+training_params:
+  seed: 42
+  num_train_epochs: 20
+  logging_steps: 100
+  metric_for_best_model: QWK
+  bf16: true
+bootstrap:
+  enabled: true
+  n_bootstrap: 10000
+  bootstrap_seed: 42
+  metrics:
+  - QWK
+  - Macro_F1
+  - Weighted_F1
+post_training_results:
+  model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
+experiments:
+  model:
+    name: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only
+    type: encoder_classification
+    num_labels: 6
+    output_dir: ./results/
+    logging_dir: ./logs/
+    best_model_dir: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only
+  tokenizer:
+    name: PORTULAN/albertina-1b5-portuguese-ptbr-encoder
+  dataset:
+    grade_index: 3
+    use_full_context: false
+  training_params:
+    weight_decay: 0.01
+    warmup_ratio: 0.1
+    learning_rate: 5.0e-05
+    train_batch_size: 4
+    eval_batch_size: 4
+    gradient_accumulation_steps: 4
+    gradient_checkpointing: false
+
+[2025-07-11 01:41:41,924][__main__][INFO] - Running inference with fine-tuned HF model
+[2025-07-11 01:41:46,936][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/config.json
+[2025-07-11 01:41:46,940][transformers.configuration_utils][INFO] - Model config DebertaV2Config {
+  "architectures": [
+    "DebertaV2ForMaskedLM"
+  ],
+  "attention_head_size": 64,
+  "attention_probs_dropout_prob": 0.1,
+  "conv_act": "gelu",
+  "conv_kernel_size": 3,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 6144,
+  "layer_norm_eps": 1e-07,
+  "legacy": true,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "model_type": "deberta-v2",
+  "norm_rel_ebd": "layer_norm",
+  "num_attention_heads": 24,
+  "num_hidden_layers": 48,
+  "pad_token_id": 0,
+  "pooler_dropout": 0,
+  "pooler_hidden_act": "gelu",
+  "pooler_hidden_size": 1536,
+  "pos_att_type": [
+    "p2c",
+    "c2p"
+  ],
+  "position_biased_input": false,
+  "position_buckets": 256,
+  "relative_attention": true,
+  "share_att_key": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 0,
+  "vocab_size": 128100
+}
+
+[2025-07-11 01:41:47,798][transformers.tokenization_utils_base][INFO] - loading file spm.model from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/spm.model
+[2025-07-11 01:41:47,798][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/tokenizer.json
+[2025-07-11 01:41:47,798][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/added_tokens.json
+[2025-07-11 01:41:47,798][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/special_tokens_map.json
+[2025-07-11 01:41:47,798][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/tokenizer_config.json
+[2025-07-11 01:41:47,798][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
+[2025-07-11 01:41:48,068][__main__][INFO] - Tokenizer function parameters- Padding:longest; Truncation: True; Use Full Context: False
+[2025-07-11 01:41:48,553][__main__][INFO] - 
+Token statistics for 'train' split:
+[2025-07-11 01:41:48,554][__main__][INFO] -   Total examples: 500
+[2025-07-11 01:41:48,554][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:41:48,554][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:41:48,554][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:41:48,554][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:41:48,646][__main__][INFO] - 
+Token statistics for 'validation' split:
+[2025-07-11 01:41:48,646][__main__][INFO] -   Total examples: 132
+[2025-07-11 01:41:48,646][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:41:48,646][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:41:48,647][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:41:48,647][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:41:48,746][__main__][INFO] - 
+Token statistics for 'test' split:
+[2025-07-11 01:41:48,747][__main__][INFO] -   Total examples: 138
+[2025-07-11 01:41:48,747][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:41:48,747][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:41:48,747][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:41:48,747][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:41:48,747][__main__][INFO] - If token statistics are the same (max, avg, min) keep in mind that this is due to batched tokenization and padding.
+[2025-07-11 01:41:48,747][__main__][INFO] - Model max length: 512. If it is the same as stats, then there is a high chance that sequences are being truncated.
+[2025-07-11 01:41:48,747][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only
+[2025-07-11 01:41:48,747][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only
+[2025-07-11 01:41:49,700][__main__][INFO] - Model need ≈ 9.51 GiB to run inference and 27.02 for training 
+[2025-07-11 01:41:50,879][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only/snapshots/6873e152b4a5203d81466439ffd264a562da1f0b/config.json
+[2025-07-11 01:41:50,880][transformers.configuration_utils][INFO] - Model config DebertaV2Config {
+  "architectures": [
+    "DebertaV2ForSequenceClassification"
+  ],
+  "attention_head_size": 64,
+  "attention_probs_dropout_prob": 0.1,
+  "conv_act": "gelu",
+  "conv_kernel_size": 3,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1536,
+  "id2label": {
+    "0": 0,
+    "1": 40,
+    "2": 80,
+    "3": 120,
+    "4": 160,
+    "5": 200
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 6144,
+  "label2id": {
+    "0": 0,
+    "40": 1,
+    "80": 2,
+    "120": 3,
+    "160": 4,
+    "200": 5
+  },
+  "layer_norm_eps": 1e-07,
+  "legacy": true,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "model_type": "deberta-v2",
+  "norm_rel_ebd": "layer_norm",
+  "num_attention_heads": 24,
+  "num_hidden_layers": 48,
+  "pad_token_id": 0,
+  "pooler_dropout": 0,
+  "pooler_hidden_act": "gelu",
+  "pooler_hidden_size": 1536,
+  "pos_att_type": [
+    "p2c",
+    "c2p"
+  ],
+  "position_biased_input": false,
+  "position_buckets": 256,
+  "relative_attention": true,
+  "share_att_key": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 0,
+  "vocab_size": 128100
+}
+
+[2025-07-11 01:42:46,329][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only/snapshots/6873e152b4a5203d81466439ffd264a562da1f0b/model.safetensors
+[2025-07-11 01:42:46,334][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.bfloat16 as defined in model's config object
+[2025-07-11 01:42:46,334][transformers.modeling_utils][INFO] - Instantiating DebertaV2ForSequenceClassification model under default dtype torch.bfloat16.
+[2025-07-11 01:42:47,880][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing DebertaV2ForSequenceClassification.
+
+[2025-07-11 01:42:47,881][transformers.modeling_utils][INFO] - All the weights of DebertaV2ForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only.
+If your task is similar to the task the model of the checkpoint was trained on, you can already use DebertaV2ForSequenceClassification for predictions without further training.
+[2025-07-11 01:42:47,901][transformers.training_args][INFO] - PyTorch: setting up devices
+[2025-07-11 01:42:47,932][transformers.training_args][INFO] - The default value for the training argument `--report_to` will change in v5 (from all installed integrations to none). In v5, you will need to use `--report_to all` to get the same behavior as now. You should start updating your code and make this info disappear :-).
+[2025-07-11 01:42:47,939][transformers.trainer][INFO] - You have loaded a model on multiple GPUs. `is_model_parallel` attribute will be force-set to `True` to avoid any unexpected behavior such as device placement mismatching.
+[2025-07-11 01:42:47,964][transformers.trainer][INFO] - Using auto half precision backend
+[2025-07-11 01:42:51,283][__main__][INFO] - Running inference on test dataset
+[2025-07-11 01:42:51,285][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `DebertaV2ForSequenceClassification.forward` and have been ignored: grades, prompt, id, essay_text, id_prompt, supporting_text, essay_year, reference. If grades, prompt, id, essay_text, id_prompt, supporting_text, essay_year, reference are not expected by `DebertaV2ForSequenceClassification.forward`,  you can safely ignore this message.
+[2025-07-11 01:42:51,297][transformers.trainer][INFO] - 
+***** Running Prediction *****
+[2025-07-11 01:42:51,297][transformers.trainer][INFO] -   Num examples = 138
+[2025-07-11 01:42:51,297][transformers.trainer][INFO] -   Batch size = 4
+[2025-07-11 01:42:59,939][__main__][INFO] - Inference results saved to jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only_inference_results.jsonl
+[2025-07-11 01:42:59,940][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
+[2025-07-11 01:45:06,056][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
+[2025-07-11 01:45:06,057][__main__][INFO] - Bootstrap Confidence Intervals (95%):
+[2025-07-11 01:45:06,057][__main__][INFO] -   QWK: 0.5397 [0.4204, 0.6511]
+[2025-07-11 01:45:06,057][__main__][INFO] -   Macro_F1: 0.3403 [0.2300, 0.4995]
+[2025-07-11 01:45:06,057][__main__][INFO] -   Weighted_F1: 0.5926 [0.5065, 0.6785]
+[2025-07-11 01:45:06,057][__main__][INFO] - Inference results: {'accuracy': 0.6014492753623188, 'RMSE': 27.662562992324986, 'QWK': 0.543046357615894, 'HDIV': 0.007246376811594235, 'Macro_F1': 0.2974558904383466, 'Micro_F1': 0.6014492753623188, 'Weighted_F1': 0.5920471313301213, 'TP_0': np.int64(0), 'TN_0': np.int64(137), 'FP_0': np.int64(0), 'FN_0': np.int64(1), 'TP_1': np.int64(0), 'TN_1': np.int64(137), 'FP_1': np.int64(0), 'FN_1': np.int64(1), 'TP_2': np.int64(3), 'TN_2': np.int64(122), 'FP_2': np.int64(7), 'FN_2': np.int64(6), 'TP_3': np.int64(60), 'TN_3': np.int64(36), 'FP_3': np.int64(26), 'FN_3': np.int64(16), 'TP_4': np.int64(18), 'TN_4': np.int64(78), 'FP_4': np.int64(14), 'FN_4': np.int64(28), 'TP_5': np.int64(2), 'TN_5': np.int64(125), 'FP_5': np.int64(8), 'FN_5': np.int64(3)}
+[2025-07-11 01:45:06,058][__main__][INFO] - Inference experiment completed

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only/.hydra/config.yaml

@@ -0,0 +1,41 @@
+cache_dir: /tmp/
+dataset:
+  name: kamel-usp/aes_enem_dataset
+  split: JBCS2025
+training_params:
+  seed: 42
+  num_train_epochs: 20
+  logging_steps: 100
+  metric_for_best_model: QWK
+  bf16: true
+bootstrap:
+  enabled: true
+  n_bootstrap: 10000
+  bootstrap_seed: 42
+  metrics:
+  - QWK
+  - Macro_F1
+  - Weighted_F1
+post_training_results:
+  model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
+experiments:
+  model:
+    name: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only
+    type: encoder_classification
+    num_labels: 6
+    output_dir: ./results/
+    logging_dir: ./logs/
+    best_model_dir: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only
+  tokenizer:
+    name: PORTULAN/albertina-1b5-portuguese-ptbr-encoder
+  dataset:
+    grade_index: 4
+    use_full_context: false
+  training_params:
+    weight_decay: 0.01
+    warmup_ratio: 0.1
+    learning_rate: 5.0e-05
+    train_batch_size: 4
+    eval_batch_size: 4
+    gradient_accumulation_steps: 4
+    gradient_checkpointing: false

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only/.hydra/hydra.yaml

@@ -0,0 +1,157 @@
+hydra:
+  run:
+    dir: inference_output/2025-07-11/01-45-10
+  sweep:
+    dir: 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.run.dir=inference_output/2025-07-11/01-45-10
+    - hydra.mode=RUN
+    task:
+    - experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only
+  job:
+    name: run_inference_experiment
+    chdir: null
+    override_dirname: experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only
+    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: '1.1'
+    cwd: /workspace/jbcs2025
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /workspace/jbcs2025/configs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/jbcs2025/inference_output/2025-07-11/01-45-10
+    choices:
+      experiments: temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only
+      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

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- experiments=temp_inference/kamel-usp_jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only/bootstrap_confidence_intervals.csv

@@ -0,0 +1,2 @@
+experiment_id,timestamp,QWK_mean,QWK_lower_95ci,QWK_upper_95ci,QWK_ci_width,Macro_F1_mean,Macro_F1_lower_95ci,Macro_F1_upper_95ci,Macro_F1_ci_width,Weighted_F1_mean,Weighted_F1_lower_95ci,Weighted_F1_upper_95ci,Weighted_F1_ci_width
+jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only,2025-07-11 01:45:16,0.030023967736090135,-0.1487321238529284,0.2089807036607759,0.35771282751370426,0.15106212205759126,0.09700495581807897,0.20820622511097128,0.11120126929289231,0.1716628043898459,0.10670643853108219,0.24081069381270429,0.1341042552816221

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only/evaluation_results.csv

@@ -0,0 +1,2 @@
+accuracy,RMSE,QWK,HDIV,Macro_F1,Micro_F1,Weighted_F1,TP_0,TN_0,FP_0,FN_0,TP_1,TN_1,FP_1,FN_1,TP_2,TN_2,FP_2,FN_2,TP_3,TN_3,FP_3,FN_3,TP_4,TN_4,FP_4,FN_4,TP_5,TN_5,FP_5,FN_5,timestamp,id
+0.1956521739130435,76.7453912131782,0.028976674608011566,0.16666666666666663,0.1530071129978623,0.1956521739130435,0.1712828232958547,4,106,10,18,1,93,13,31,14,58,56,10,3,104,9,22,5,91,15,27,0,127,8,3,2025-07-11 01:45:16,jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only

runs/large_models/albertina/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only/run_inference_experiment.log

@@ -0,0 +1,199 @@
+[2025-07-11 01:45:16,521][__main__][INFO] - Starting inference experiment
+[2025-07-11 01:45:16,523][__main__][INFO] - cache_dir: /tmp/
+dataset:
+  name: kamel-usp/aes_enem_dataset
+  split: JBCS2025
+training_params:
+  seed: 42
+  num_train_epochs: 20
+  logging_steps: 100
+  metric_for_best_model: QWK
+  bf16: true
+bootstrap:
+  enabled: true
+  n_bootstrap: 10000
+  bootstrap_seed: 42
+  metrics:
+  - QWK
+  - Macro_F1
+  - Weighted_F1
+post_training_results:
+  model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
+experiments:
+  model:
+    name: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only
+    type: encoder_classification
+    num_labels: 6
+    output_dir: ./results/
+    logging_dir: ./logs/
+    best_model_dir: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only
+  tokenizer:
+    name: PORTULAN/albertina-1b5-portuguese-ptbr-encoder
+  dataset:
+    grade_index: 4
+    use_full_context: false
+  training_params:
+    weight_decay: 0.01
+    warmup_ratio: 0.1
+    learning_rate: 5.0e-05
+    train_batch_size: 4
+    eval_batch_size: 4
+    gradient_accumulation_steps: 4
+    gradient_checkpointing: false
+
+[2025-07-11 01:45:16,525][__main__][INFO] - Running inference with fine-tuned HF model
+[2025-07-11 01:45:21,093][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/config.json
+[2025-07-11 01:45:21,096][transformers.configuration_utils][INFO] - Model config DebertaV2Config {
+  "architectures": [
+    "DebertaV2ForMaskedLM"
+  ],
+  "attention_head_size": 64,
+  "attention_probs_dropout_prob": 0.1,
+  "conv_act": "gelu",
+  "conv_kernel_size": 3,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 6144,
+  "layer_norm_eps": 1e-07,
+  "legacy": true,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "model_type": "deberta-v2",
+  "norm_rel_ebd": "layer_norm",
+  "num_attention_heads": 24,
+  "num_hidden_layers": 48,
+  "pad_token_id": 0,
+  "pooler_dropout": 0,
+  "pooler_hidden_act": "gelu",
+  "pooler_hidden_size": 1536,
+  "pos_att_type": [
+    "p2c",
+    "c2p"
+  ],
+  "position_biased_input": false,
+  "position_buckets": 256,
+  "relative_attention": true,
+  "share_att_key": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 0,
+  "vocab_size": 128100
+}
+
+[2025-07-11 01:45:21,511][transformers.tokenization_utils_base][INFO] - loading file spm.model from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/spm.model
+[2025-07-11 01:45:21,511][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/tokenizer.json
+[2025-07-11 01:45:21,511][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/added_tokens.json
+[2025-07-11 01:45:21,511][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/special_tokens_map.json
+[2025-07-11 01:45:21,511][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--PORTULAN--albertina-1b5-portuguese-ptbr-encoder/snapshots/b22008e5096af9c398b75762d4e28e5008762916/tokenizer_config.json
+[2025-07-11 01:45:21,511][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
+[2025-07-11 01:45:21,789][__main__][INFO] - Tokenizer function parameters- Padding:longest; Truncation: True; Use Full Context: False
+[2025-07-11 01:45:22,298][__main__][INFO] - 
+Token statistics for 'train' split:
+[2025-07-11 01:45:22,298][__main__][INFO] -   Total examples: 500
+[2025-07-11 01:45:22,298][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:45:22,298][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:45:22,298][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:45:22,299][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:45:22,398][__main__][INFO] - 
+Token statistics for 'validation' split:
+[2025-07-11 01:45:22,398][__main__][INFO] -   Total examples: 132
+[2025-07-11 01:45:22,398][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:45:22,398][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:45:22,398][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:45:22,398][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:45:22,498][__main__][INFO] - 
+Token statistics for 'test' split:
+[2025-07-11 01:45:22,498][__main__][INFO] -   Total examples: 138
+[2025-07-11 01:45:22,498][__main__][INFO] -   Min tokens: 512
+[2025-07-11 01:45:22,498][__main__][INFO] -   Max tokens: 512
+[2025-07-11 01:45:22,498][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-11 01:45:22,498][__main__][INFO] -   Std tokens: 0.00
+[2025-07-11 01:45:22,498][__main__][INFO] - If token statistics are the same (max, avg, min) keep in mind that this is due to batched tokenization and padding.
+[2025-07-11 01:45:22,498][__main__][INFO] - Model max length: 512. If it is the same as stats, then there is a high chance that sequences are being truncated.
+[2025-07-11 01:45:22,499][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only
+[2025-07-11 01:45:22,499][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only
+[2025-07-11 01:45:23,468][__main__][INFO] - Model need ≈ 9.51 GiB to run inference and 27.02 for training 
+[2025-07-11 01:45:24,361][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only/snapshots/46296a27ad8fd570a2177c4bf8427be01ad50052/config.json
+[2025-07-11 01:45:24,362][transformers.configuration_utils][INFO] - Model config DebertaV2Config {
+  "architectures": [
+    "DebertaV2ForSequenceClassification"
+  ],
+  "attention_head_size": 64,
+  "attention_probs_dropout_prob": 0.1,
+  "conv_act": "gelu",
+  "conv_kernel_size": 3,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1536,
+  "id2label": {
+    "0": 0,
+    "1": 40,
+    "2": 80,
+    "3": 120,
+    "4": 160,
+    "5": 200
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 6144,
+  "label2id": {
+    "0": 0,
+    "40": 1,
+    "80": 2,
+    "120": 3,
+    "160": 4,
+    "200": 5
+  },
+  "layer_norm_eps": 1e-07,
+  "legacy": true,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "model_type": "deberta-v2",
+  "norm_rel_ebd": "layer_norm",
+  "num_attention_heads": 24,
+  "num_hidden_layers": 48,
+  "pad_token_id": 0,
+  "pooler_dropout": 0,
+  "pooler_hidden_act": "gelu",
+  "pooler_hidden_size": 1536,
+  "pos_att_type": [
+    "p2c",
+    "c2p"
+  ],
+  "position_biased_input": false,
+  "position_buckets": 256,
+  "relative_attention": true,
+  "share_att_key": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 0,
+  "vocab_size": 128100
+}
+
+[2025-07-11 01:46:19,417][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only/snapshots/46296a27ad8fd570a2177c4bf8427be01ad50052/model.safetensors
+[2025-07-11 01:46:19,420][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.bfloat16 as defined in model's config object
+[2025-07-11 01:46:19,420][transformers.modeling_utils][INFO] - Instantiating DebertaV2ForSequenceClassification model under default dtype torch.bfloat16.
+[2025-07-11 01:46:20,985][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing DebertaV2ForSequenceClassification.
+
+[2025-07-11 01:46:20,986][transformers.modeling_utils][INFO] - All the weights of DebertaV2ForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only.
+If your task is similar to the task the model of the checkpoint was trained on, you can already use DebertaV2ForSequenceClassification for predictions without further training.
+[2025-07-11 01:46:21,006][transformers.training_args][INFO] - PyTorch: setting up devices
+[2025-07-11 01:46:21,030][transformers.training_args][INFO] - The default value for the training argument `--report_to` will change in v5 (from all installed integrations to none). In v5, you will need to use `--report_to all` to get the same behavior as now. You should start updating your code and make this info disappear :-).
+[2025-07-11 01:46:21,037][transformers.trainer][INFO] - You have loaded a model on multiple GPUs. `is_model_parallel` attribute will be force-set to `True` to avoid any unexpected behavior such as device placement mismatching.
+[2025-07-11 01:46:21,062][transformers.trainer][INFO] - Using auto half precision backend
+[2025-07-11 01:46:24,415][__main__][INFO] - Running inference on test dataset
+[2025-07-11 01:46:24,417][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `DebertaV2ForSequenceClassification.forward` and have been ignored: reference, prompt, essay_year, id_prompt, essay_text, supporting_text, grades, id. If reference, prompt, essay_year, id_prompt, essay_text, supporting_text, grades, id are not expected by `DebertaV2ForSequenceClassification.forward`,  you can safely ignore this message.
+[2025-07-11 01:46:24,430][transformers.trainer][INFO] - 
+***** Running Prediction *****
+[2025-07-11 01:46:24,430][transformers.trainer][INFO] -   Num examples = 138
+[2025-07-11 01:46:24,430][transformers.trainer][INFO] -   Batch size = 4
+[2025-07-11 01:46:33,202][__main__][INFO] - Inference results saved to jbcs2025_albertina-1b5-portuguese-ptbr-encoder-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only_inference_results.jsonl
+[2025-07-11 01:46:33,203][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
+[2025-07-11 01:48:41,802][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
+[2025-07-11 01:48:41,802][__main__][INFO] - Bootstrap Confidence Intervals (95%):
+[2025-07-11 01:48:41,802][__main__][INFO] -   QWK: 0.0300 [-0.1487, 0.2090]
+[2025-07-11 01:48:41,802][__main__][INFO] -   Macro_F1: 0.1511 [0.0970, 0.2082]
+[2025-07-11 01:48:41,802][__main__][INFO] -   Weighted_F1: 0.1717 [0.1067, 0.2408]
+[2025-07-11 01:48:41,802][__main__][INFO] - Inference results: {'accuracy': 0.1956521739130435, 'RMSE': 76.7453912131782, 'QWK': 0.028976674608011566, 'HDIV': 0.16666666666666663, 'Macro_F1': 0.1530071129978623, 'Micro_F1': 0.1956521739130435, 'Weighted_F1': 0.1712828232958547, 'TP_0': np.int64(4), 'TN_0': np.int64(106), 'FP_0': np.int64(10), 'FN_0': np.int64(18), 'TP_1': np.int64(1), 'TN_1': np.int64(93), 'FP_1': np.int64(13), 'FN_1': np.int64(31), 'TP_2': np.int64(14), 'TN_2': np.int64(58), 'FP_2': np.int64(56), 'FN_2': np.int64(10), 'TP_3': np.int64(3), 'TN_3': np.int64(104), 'FP_3': np.int64(9), 'FN_3': np.int64(22), 'TP_4': np.int64(5), 'TN_4': np.int64(91), 'FP_4': np.int64(15), 'FN_4': np.int64(27), 'TP_5': np.int64(0), 'TN_5': np.int64(127), 'FP_5': np.int64(8), 'FN_5': np.int64(3)}
+[2025-07-11 01:48:41,803][__main__][INFO] - Inference experiment completed