update large encoders; slm and update c5 gpt4o

runs/api_models/gpt-4o/gpt-4o-2024-11-20-zero-shot-C5-essay_only/evaluation_results.csv

@@ -1,2 +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
-0.3188405797101449,60.048289746247356,0.5487540742298391,0.07971014492753625,0.2921402969790066,0.3188405797101449,0.29828469022017406,16,112,4,6,13,81,25,19,3,96,18,21,10,92,21,15,0,106,0,32,2,109,26,1,2025-07-02 21:08:07,gpt-4o-2024-11-20-zero-shot-C5-essay_only
+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
+0.3188405797101449,60.048289746247356,0.5487540742298391,0.07971014492753625,0.2921402969790066,0.3188405797101449,0.29828469022017406,16,112,4,6,13,81,25,19,3,96,18,21,10,92,21,15,0,106,0,32,2,109,26,1,2025-07-02 21:08:07,gpt-4o-2024-11-20-zero-shot-C5-essay_only

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C1-encoder_classification-C1-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only}/.hydra/config.yaml

@@ -20,12 +20,12 @@ post_training_results:
   model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
 experiments:
   model:
-    name: kamel-usp/jbcs2025_bertimbau-large-C1
+    name: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only
     type: encoder_classification
     num_labels: 6
     output_dir: ./results/bertimbau_large/C1
     logging_dir: ./logs/bertimbau_large/C1
-    best_model_dir: ./results/bertimbau_large/C1/best_model
+    best_model_dir: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only
   tokenizer:
     name: neuralmind/bert-large-portuguese-cased
   dataset:

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C3-encoder_classification-C3-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only}/.hydra/hydra.yaml

@@ -1,6 +1,6 @@
 hydra:
   run:
-    dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: inference_output/2025-07-10/01-10-43
   sweep:
     dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
     subdir: ${hydra.job.num}
@@ -110,13 +110,14 @@ hydra:
   output_subdir: .hydra
   overrides:
     hydra:
+    - hydra.run.dir=inference_output/2025-07-10/01-10-43
     - hydra.mode=RUN
     task:
-    - experiments=large_models/C3
+    - experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only
   job:
     name: run_inference_experiment
     chdir: null
-    override_dirname: experiments=large_models/C3
+    override_dirname: experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only
     id: ???
     num: ???
     config_name: config
@@ -141,9 +142,9 @@ hydra:
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /workspace/jbcs2025/outputs/2025-07-01/00-08-08
+    output_dir: /workspace/jbcs2025/inference_output/2025-07-10/01-10-43
     choices:
-      experiments: large_models/C3
+      experiments: temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only
       hydra/env: default
       hydra/callbacks: null
       hydra/job_logging: default

runs/large_models/bertimbau/jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C1-encoder_classification-C1-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only}/bootstrap_confidence_intervals.csv

@@ -1,2 +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_bertimbau-large-C1-encoder_classification-C1-essay_only,2025-07-01 00:02:02,0.7077306949563342,0.6143517863619989,0.7925777763902052,0.17822599002820627,0.5086103433798745,0.3864421058348054,0.6645304839656547,0.2780883781308493,0.7098050926433485,0.6316473666444833,0.7862820634470531,0.1546346968025698
+jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only,2025-07-10 01:10:49,0.6822787596623211,0.586354115876401,0.7722799106018048,0.18592579472540383,0.4971683868103125,0.37489665684586737,0.6515710555984839,0.27667439875261657,0.6920329870783886,0.612322638598863,0.767705078861486,0.15538244026262293

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C1-encoder_classification-C1-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only}/evaluation_results.csv

@@ -1,2 +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
-0.7028985507246377,24.55399256179405,0.7080553295362083,0.007246376811594235,0.4714726209463051,0.7028985507246377,0.7092465463174845,0,137,0,1,0,138,0,0,4,124,4,6,54,60,12,12,33,76,11,18,6,114,14,4,2025-07-01 00:02:02,jbcs2025_bertimbau-large-C1-encoder_classification-C1-essay_only
+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
+0.6884057971014492,25.02172968684897,0.6825694326120293,0.007246376811594235,0.4609419962901689,0.6884057971014492,0.6910750081919986,0,137,0,1,0,138,0,0,4,122,6,6,46,62,10,20,40,69,18,11,5,119,9,5,2025-07-10 01:10:49,jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C1-encoder_classification-C1-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only}/run_inference_experiment.log

@@ -1,5 +1,5 @@
-[2025-07-01 00:02:02,361][__main__][INFO] - Starting inference experiment
-[2025-07-01 00:02:02,362][__main__][INFO] - cache_dir: /tmp/
+[2025-07-10 01:10:49,396][__main__][INFO] - Starting inference experiment
+[2025-07-10 01:10:49,398][__main__][INFO] - cache_dir: /tmp/
 dataset:
   name: kamel-usp/aes_enem_dataset
   split: JBCS2025
@@ -21,12 +21,12 @@ post_training_results:
   model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
 experiments:
   model:
-    name: kamel-usp/jbcs2025_bertimbau-large-C1
+    name: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only
     type: encoder_classification
     num_labels: 6
     output_dir: ./results/bertimbau_large/C1
     logging_dir: ./logs/bertimbau_large/C1
-    best_model_dir: ./results/bertimbau_large/C1/best_model
+    best_model_dir: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only
   tokenizer:
     name: neuralmind/bert-large-portuguese-cased
   dataset:
@@ -41,9 +41,9 @@ experiments:
     gradient_accumulation_steps: 1
     gradient_checkpointing: false
 
-[2025-07-01 00:02:02,364][__main__][INFO] - Running inference with fine-tuned HF model
-[2025-07-01 00:02:08,354][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:02:08,355][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:10:49,400][__main__][INFO] - Running inference with fine-tuned HF model
+[2025-07-10 01:10:55,009][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:10:55,010][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -68,20 +68,14 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:02:10,698][transformers.tokenization_utils_base][INFO] - loading file vocab.txt from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/vocab.txt
-[2025-07-01 00:02:10,698][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at None
-[2025-07-01 00:02:10,698][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/added_tokens.json
-[2025-07-01 00:02:10,698][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/special_tokens_map.json
-[2025-07-01 00:02:10,698][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/tokenizer_config.json
-[2025-07-01 00:02:10,698][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
-[2025-07-01 00:02:10,698][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:02:10,699][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:10:55,328][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:10:55,328][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -106,14 +100,20 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:02:10,724][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:02:10,724][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:10:55,537][transformers.tokenization_utils_base][INFO] - loading file vocab.txt from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/vocab.txt
+[2025-07-10 01:10:55,537][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at None
+[2025-07-10 01:10:55,538][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/added_tokens.json
+[2025-07-10 01:10:55,538][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/special_tokens_map.json
+[2025-07-10 01:10:55,538][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/tokenizer_config.json
+[2025-07-10 01:10:55,538][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
+[2025-07-10 01:10:55,538][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:10:55,538][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -138,18 +138,73 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:02:10,741][__main__][INFO] - Tokenizer function parameters- Padding:max_length; Truncation: True; Use Full Context: False
-[2025-07-01 00:02:10,943][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bertimbau-large-C1
-[2025-07-01 00:02:10,943][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bertimbau-large-C1
-[2025-07-01 00:02:11,801][__main__][INFO] - Model need ≈ 2.62 GiB to run inference and 6.36 for training 
-[2025-07-01 00:02:12,612][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_bertimbau-large-C1/snapshots/3e737c8fdb77192423f85a28a47c007d664a9aab/config.json
-[2025-07-01 00:02:12,613][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:10:55,572][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:10:55,572][transformers.configuration_utils][INFO] - Model config BertConfig {
+  "architectures": [
+    "BertForMaskedLM"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "directionality": "bidi",
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "output_past": true,
+  "pad_token_id": 0,
+  "pooler_fc_size": 768,
+  "pooler_num_attention_heads": 12,
+  "pooler_num_fc_layers": 3,
+  "pooler_size_per_head": 128,
+  "pooler_type": "first_token_transform",
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 29794
+}
+
+[2025-07-10 01:10:55,590][__main__][INFO] - Tokenizer function parameters- Padding:longest; Truncation: True; Use Full Context: False
+[2025-07-10 01:10:56,012][__main__][INFO] - 
+Token statistics for 'train' split:
+[2025-07-10 01:10:56,012][__main__][INFO] -   Total examples: 500
+[2025-07-10 01:10:56,012][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:10:56,012][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:10:56,012][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:10:56,012][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:10:56,103][__main__][INFO] - 
+Token statistics for 'validation' split:
+[2025-07-10 01:10:56,103][__main__][INFO] -   Total examples: 132
+[2025-07-10 01:10:56,103][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:10:56,103][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:10:56,103][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:10:56,103][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:10:56,198][__main__][INFO] - 
+Token statistics for 'test' split:
+[2025-07-10 01:10:56,198][__main__][INFO] -   Total examples: 138
+[2025-07-10 01:10:56,198][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:10:56,198][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:10:56,198][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:10:56,198][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:10:56,198][__main__][INFO] - If token statistics are the same (max, avg, min) keep in mind that this is due to batched tokenization and padding.
+[2025-07-10 01:10:56,198][__main__][INFO] - Model max length: 512. If it is the same as stats, then there is a high chance that sequences are being truncated.
+[2025-07-10 01:10:56,198][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only
+[2025-07-10 01:10:56,199][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only
+[2025-07-10 01:10:57,324][__main__][INFO] - Model need ≈ 2.62 GiB to run inference and 6.36 for training 
+[2025-07-10 01:10:58,329][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only/snapshots/da30f459272b3b4c66170aa85fce0e1191eea445/config.json
+[2025-07-10 01:10:58,330][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForSequenceClassification"
   ],
@@ -190,37 +245,36 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "problem_type": "single_label_classification",
   "torch_dtype": "float32",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:02:52,491][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_bertimbau-large-C1/snapshots/3e737c8fdb77192423f85a28a47c007d664a9aab/model.safetensors
-[2025-07-01 00:02:52,492][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.float32 as defined in model's config object
-[2025-07-01 00:02:52,492][transformers.modeling_utils][INFO] - Instantiating BertForSequenceClassification model under default dtype torch.float32.
-[2025-07-01 00:02:53,107][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing BertForSequenceClassification.
+[2025-07-10 01:11:22,859][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only/snapshots/da30f459272b3b4c66170aa85fce0e1191eea445/model.safetensors
+[2025-07-10 01:11:22,862][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.float32 as defined in model's config object
+[2025-07-10 01:11:22,862][transformers.modeling_utils][INFO] - Instantiating BertForSequenceClassification model under default dtype torch.float32.
+[2025-07-10 01:11:23,611][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing BertForSequenceClassification.
 
-[2025-07-01 00:02:53,107][transformers.modeling_utils][INFO] - All the weights of BertForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_bertimbau-large-C1.
+[2025-07-10 01:11:23,612][transformers.modeling_utils][INFO] - All the weights of BertForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_bert-large-portuguese-cased-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 BertForSequenceClassification for predictions without further training.
-[2025-07-01 00:02:53,115][transformers.training_args][INFO] - PyTorch: setting up devices
-[2025-07-01 00:02:53,148][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 :-).
-[2025-07-01 00:02:53,153][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.
-[2025-07-01 00:02:53,170][transformers.trainer][INFO] - Using auto half precision backend
-[2025-07-01 00:02:57,479][__main__][INFO] - Running inference on test dataset
-[2025-07-01 00:02:57,480][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `BertForSequenceClassification.forward` and have been ignored: essay_text, reference, supporting_text, prompt, id_prompt, grades, id, essay_year. If essay_text, reference, supporting_text, prompt, id_prompt, grades, id, essay_year are not expected by `BertForSequenceClassification.forward`,  you can safely ignore this message.
-[2025-07-01 00:02:57,486][transformers.trainer][INFO] - 
+[2025-07-10 01:11:23,630][transformers.training_args][INFO] - PyTorch: setting up devices
+[2025-07-10 01:11:23,654][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 :-).
+[2025-07-10 01:11:23,667][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.
+[2025-07-10 01:11:23,697][transformers.trainer][INFO] - Using auto half precision backend
+[2025-07-10 01:11:27,018][__main__][INFO] - Running inference on test dataset
+[2025-07-10 01:11:27,019][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `BertForSequenceClassification.forward` and have been ignored: reference, id, grades, essay_text, id_prompt, supporting_text, prompt, essay_year. If reference, id, grades, essay_text, id_prompt, supporting_text, prompt, essay_year are not expected by `BertForSequenceClassification.forward`,  you can safely ignore this message.
+[2025-07-10 01:11:27,031][transformers.trainer][INFO] - 
 ***** Running Prediction *****
-[2025-07-01 00:02:57,486][transformers.trainer][INFO] -   Num examples = 138
-[2025-07-01 00:02:57,486][transformers.trainer][INFO] -   Batch size = 16
-[2025-07-01 00:02:57,969][__main__][INFO] - Inference results saved to jbcs2025_bertimbau-large-C1-encoder_classification-C1-essay_only_inference_results.jsonl
-[2025-07-01 00:02:57,975][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
-[2025-07-01 00:04:34,547][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
-[2025-07-01 00:04:34,547][__main__][INFO] - Bootstrap Confidence Intervals (95%):
-[2025-07-01 00:04:34,547][__main__][INFO] -   QWK: 0.7077 [0.6144, 0.7926]
-[2025-07-01 00:04:34,547][__main__][INFO] -   Macro_F1: 0.5086 [0.3864, 0.6645]
-[2025-07-01 00:04:34,547][__main__][INFO] -   Weighted_F1: 0.7098 [0.6316, 0.7863]
-[2025-07-01 00:04:34,547][__main__][INFO] - Inference results: {'accuracy': 0.7028985507246377, 'RMSE': 24.55399256179405, 'QWK': 0.7080553295362083, 'HDIV': 0.007246376811594235, 'Macro_F1': 0.4714726209463051, 'Micro_F1': 0.7028985507246377, 'Weighted_F1': 0.7092465463174845, '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(4), 'TN_2': np.int64(124), 'FP_2': np.int64(4), 'FN_2': np.int64(6), 'TP_3': np.int64(54), 'TN_3': np.int64(60), 'FP_3': np.int64(12), 'FN_3': np.int64(12), 'TP_4': np.int64(33), 'TN_4': np.int64(76), 'FP_4': np.int64(11), 'FN_4': np.int64(18), 'TP_5': np.int64(6), 'TN_5': np.int64(114), 'FP_5': np.int64(14), 'FN_5': np.int64(4)}
-[2025-07-01 00:04:34,547][__main__][INFO] - Inference experiment completed
+[2025-07-10 01:11:27,032][transformers.trainer][INFO] -   Num examples = 138
+[2025-07-10 01:11:27,032][transformers.trainer][INFO] -   Batch size = 16
+[2025-07-10 01:11:28,595][__main__][INFO] - Inference results saved to jbcs2025_bert-large-portuguese-cased-encoder_classification-C1-essay_only-encoder_classification-C1-essay_only_inference_results.jsonl
+[2025-07-10 01:11:28,596][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
+[2025-07-10 01:13:34,349][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
+[2025-07-10 01:13:34,350][__main__][INFO] - Bootstrap Confidence Intervals (95%):
+[2025-07-10 01:13:34,350][__main__][INFO] -   QWK: 0.6823 [0.5864, 0.7723]
+[2025-07-10 01:13:34,350][__main__][INFO] -   Macro_F1: 0.4972 [0.3749, 0.6516]
+[2025-07-10 01:13:34,350][__main__][INFO] -   Weighted_F1: 0.6920 [0.6123, 0.7677]
+[2025-07-10 01:13:34,350][__main__][INFO] - Inference results: {'accuracy': 0.6884057971014492, 'RMSE': 25.02172968684897, 'QWK': 0.6825694326120293, 'HDIV': 0.007246376811594235, 'Macro_F1': 0.4609419962901689, 'Micro_F1': 0.6884057971014492, 'Weighted_F1': 0.6910750081919986, '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(4), 'TN_2': np.int64(122), 'FP_2': np.int64(6), 'FN_2': np.int64(6), 'TP_3': np.int64(46), 'TN_3': np.int64(62), 'FP_3': np.int64(10), 'FN_3': np.int64(20), 'TP_4': np.int64(40), 'TN_4': np.int64(69), 'FP_4': np.int64(18), 'FN_4': np.int64(11), 'TP_5': np.int64(5), 'TN_5': np.int64(119), 'FP_5': np.int64(9), 'FN_5': np.int64(5)}
+[2025-07-10 01:13:34,354][__main__][INFO] - Inference experiment completed

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C2-encoder_classification-C2-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only}/.hydra/config.yaml

@@ -20,12 +20,12 @@ post_training_results:
   model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
 experiments:
   model:
-    name: kamel-usp/jbcs2025_bertimbau-large-C2
+    name: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only
     type: encoder_classification
     num_labels: 6
     output_dir: ./results/bertimbau_large/C2
     logging_dir: ./logs/bertimbau_large/C2
-    best_model_dir: ./results/bertimbau_large/C2/best_model
+    best_model_dir: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only
   tokenizer:
     name: neuralmind/bert-large-portuguese-cased
   dataset:

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C2-encoder_classification-C2-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only}/.hydra/hydra.yaml

@@ -1,6 +1,6 @@
 hydra:
   run:
-    dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: inference_output/2025-07-10/01-13-40
   sweep:
     dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
     subdir: ${hydra.job.num}
@@ -110,13 +110,14 @@ hydra:
   output_subdir: .hydra
   overrides:
     hydra:
+    - hydra.run.dir=inference_output/2025-07-10/01-13-40
     - hydra.mode=RUN
     task:
-    - experiments=large_models/C2
+    - experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only
   job:
     name: run_inference_experiment
     chdir: null
-    override_dirname: experiments=large_models/C2
+    override_dirname: experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only
     id: ???
     num: ???
     config_name: config
@@ -141,9 +142,9 @@ hydra:
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /workspace/jbcs2025/outputs/2025-07-01/00-04-40
+    output_dir: /workspace/jbcs2025/inference_output/2025-07-10/01-13-40
     choices:
-      experiments: large_models/C2
+      experiments: temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only
       hydra/env: default
       hydra/callbacks: null
       hydra/job_logging: default

runs/large_models/bertimbau/jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C4-encoder_classification-C4-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only}/bootstrap_confidence_intervals.csv

@@ -1,2 +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_bertimbau-large-C4-encoder_classification-C4-essay_only,2025-07-01 00:10:41,0.5689578701688437,0.46148695261109535,0.6680998143802181,0.2066128617691228,0.3225465526202049,0.23384768292840824,0.4348878912682667,0.20104020833985847,0.5689840307021065,0.4868628663380702,0.6499047041941646,0.16304183785609438
+jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only,2025-07-10 01:13:46,0.31345082186336776,0.14492296148559206,0.4709425779174953,0.32601961643190325,0.3376111572696462,0.25175523989642284,0.442912177116335,0.19115693721991217,0.4106064761329726,0.3275210469472954,0.49310656594887087,0.16558551900157548

runs/large_models/bertimbau/jbcs2025_bert-large-portuguese-cased-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
+0.39855072463768115,67.93005780671166,0.31696761677361585,0.1376811594202898,0.31891930577920913,0.39855072463768115,0.4108570406154947,0,137,0,1,16,73,30,19,3,114,19,2,16,79,8,35,10,102,10,16,10,102,16,10,2025-07-10 01:13:46,jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C2-encoder_classification-C2-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only}/run_inference_experiment.log

@@ -1,5 +1,5 @@
-[2025-07-01 00:04:40,281][__main__][INFO] - Starting inference experiment
-[2025-07-01 00:04:40,282][__main__][INFO] - cache_dir: /tmp/
+[2025-07-10 01:13:46,409][__main__][INFO] - Starting inference experiment
+[2025-07-10 01:13:46,411][__main__][INFO] - cache_dir: /tmp/
 dataset:
   name: kamel-usp/aes_enem_dataset
   split: JBCS2025
@@ -21,12 +21,12 @@ post_training_results:
   model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
 experiments:
   model:
-    name: kamel-usp/jbcs2025_bertimbau-large-C2
+    name: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only
     type: encoder_classification
     num_labels: 6
     output_dir: ./results/bertimbau_large/C2
     logging_dir: ./logs/bertimbau_large/C2
-    best_model_dir: ./results/bertimbau_large/C2/best_model
+    best_model_dir: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only
   tokenizer:
     name: neuralmind/bert-large-portuguese-cased
   dataset:
@@ -41,9 +41,9 @@ experiments:
     gradient_accumulation_steps: 1
     gradient_checkpointing: false
 
-[2025-07-01 00:04:40,284][__main__][INFO] - Running inference with fine-tuned HF model
-[2025-07-01 00:04:44,834][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:04:44,835][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:13:46,413][__main__][INFO] - Running inference with fine-tuned HF model
+[2025-07-10 01:13:51,408][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:13:51,409][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -68,20 +68,14 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:04:45,517][transformers.tokenization_utils_base][INFO] - loading file vocab.txt from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/vocab.txt
-[2025-07-01 00:04:45,517][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at None
-[2025-07-01 00:04:45,517][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/added_tokens.json
-[2025-07-01 00:04:45,517][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/special_tokens_map.json
-[2025-07-01 00:04:45,517][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/tokenizer_config.json
-[2025-07-01 00:04:45,517][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
-[2025-07-01 00:04:45,517][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:04:45,518][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:13:51,618][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:13:51,618][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -106,14 +100,20 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:04:45,543][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:04:45,544][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:13:51,813][transformers.tokenization_utils_base][INFO] - loading file vocab.txt from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/vocab.txt
+[2025-07-10 01:13:51,813][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at None
+[2025-07-10 01:13:51,813][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/added_tokens.json
+[2025-07-10 01:13:51,813][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/special_tokens_map.json
+[2025-07-10 01:13:51,813][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/tokenizer_config.json
+[2025-07-10 01:13:51,813][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
+[2025-07-10 01:13:51,814][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:13:51,814][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -138,18 +138,73 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:04:45,560][__main__][INFO] - Tokenizer function parameters- Padding:max_length; Truncation: True; Use Full Context: False
-[2025-07-01 00:04:45,768][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bertimbau-large-C2
-[2025-07-01 00:04:45,768][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bertimbau-large-C2
-[2025-07-01 00:04:46,679][__main__][INFO] - Model need ≈ 2.62 GiB to run inference and 6.36 for training 
-[2025-07-01 00:04:47,443][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_bertimbau-large-C2/snapshots/620fd996ab1895fef2aa7b5f2281eebcbe48864d/config.json
-[2025-07-01 00:04:47,444][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:13:51,844][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:13:51,845][transformers.configuration_utils][INFO] - Model config BertConfig {
+  "architectures": [
+    "BertForMaskedLM"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "directionality": "bidi",
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "output_past": true,
+  "pad_token_id": 0,
+  "pooler_fc_size": 768,
+  "pooler_num_attention_heads": 12,
+  "pooler_num_fc_layers": 3,
+  "pooler_size_per_head": 128,
+  "pooler_type": "first_token_transform",
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 29794
+}
+
+[2025-07-10 01:13:51,862][__main__][INFO] - Tokenizer function parameters- Padding:longest; Truncation: True; Use Full Context: False
+[2025-07-10 01:13:52,270][__main__][INFO] - 
+Token statistics for 'train' split:
+[2025-07-10 01:13:52,270][__main__][INFO] -   Total examples: 500
+[2025-07-10 01:13:52,270][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:13:52,270][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:13:52,270][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:13:52,270][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:13:52,360][__main__][INFO] - 
+Token statistics for 'validation' split:
+[2025-07-10 01:13:52,360][__main__][INFO] -   Total examples: 132
+[2025-07-10 01:13:52,360][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:13:52,360][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:13:52,360][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:13:52,360][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:13:52,454][__main__][INFO] - 
+Token statistics for 'test' split:
+[2025-07-10 01:13:52,454][__main__][INFO] -   Total examples: 138
+[2025-07-10 01:13:52,454][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:13:52,454][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:13:52,454][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:13:52,454][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:13:52,454][__main__][INFO] - If token statistics are the same (max, avg, min) keep in mind that this is due to batched tokenization and padding.
+[2025-07-10 01:13:52,454][__main__][INFO] - Model max length: 512. If it is the same as stats, then there is a high chance that sequences are being truncated.
+[2025-07-10 01:13:52,455][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only
+[2025-07-10 01:13:52,455][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only
+[2025-07-10 01:13:53,423][__main__][INFO] - Model need ≈ 2.62 GiB to run inference and 6.36 for training 
+[2025-07-10 01:13:54,248][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only/snapshots/a8253d03c01bf4be8629e8e2c587013f34834335/config.json
+[2025-07-10 01:13:54,248][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForSequenceClassification"
   ],
@@ -190,37 +245,36 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "problem_type": "single_label_classification",
   "torch_dtype": "float32",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:06:18,742][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_bertimbau-large-C2/snapshots/620fd996ab1895fef2aa7b5f2281eebcbe48864d/model.safetensors
-[2025-07-01 00:06:18,743][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.float32 as defined in model's config object
-[2025-07-01 00:06:18,743][transformers.modeling_utils][INFO] - Instantiating BertForSequenceClassification model under default dtype torch.float32.
-[2025-07-01 00:06:19,329][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing BertForSequenceClassification.
+[2025-07-10 01:14:18,652][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only/snapshots/a8253d03c01bf4be8629e8e2c587013f34834335/model.safetensors
+[2025-07-10 01:14:18,653][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.float32 as defined in model's config object
+[2025-07-10 01:14:18,654][transformers.modeling_utils][INFO] - Instantiating BertForSequenceClassification model under default dtype torch.float32.
+[2025-07-10 01:14:19,407][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing BertForSequenceClassification.
 
-[2025-07-01 00:06:19,330][transformers.modeling_utils][INFO] - All the weights of BertForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_bertimbau-large-C2.
+[2025-07-10 01:14:19,407][transformers.modeling_utils][INFO] - All the weights of BertForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_bert-large-portuguese-cased-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 BertForSequenceClassification for predictions without further training.
-[2025-07-01 00:06:19,337][transformers.training_args][INFO] - PyTorch: setting up devices
-[2025-07-01 00:06:19,391][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 :-).
-[2025-07-01 00:06:19,396][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.
-[2025-07-01 00:06:19,413][transformers.trainer][INFO] - Using auto half precision backend
-[2025-07-01 00:06:22,886][__main__][INFO] - Running inference on test dataset
-[2025-07-01 00:06:22,887][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `BertForSequenceClassification.forward` and have been ignored: supporting_text, reference, id_prompt, prompt, id, grades, essay_year, essay_text. If supporting_text, reference, id_prompt, prompt, id, grades, essay_year, essay_text are not expected by `BertForSequenceClassification.forward`,  you can safely ignore this message.
-[2025-07-01 00:06:22,893][transformers.trainer][INFO] - 
+[2025-07-10 01:14:19,425][transformers.training_args][INFO] - PyTorch: setting up devices
+[2025-07-10 01:14:19,450][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 :-).
+[2025-07-10 01:14:19,465][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.
+[2025-07-10 01:14:19,499][transformers.trainer][INFO] - Using auto half precision backend
+[2025-07-10 01:14:22,862][__main__][INFO] - Running inference on test dataset
+[2025-07-10 01:14:22,864][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `BertForSequenceClassification.forward` and have been ignored: prompt, reference, essay_text, essay_year, grades, id_prompt, id, supporting_text. If prompt, reference, essay_text, essay_year, grades, id_prompt, id, supporting_text are not expected by `BertForSequenceClassification.forward`,  you can safely ignore this message.
+[2025-07-10 01:14:22,875][transformers.trainer][INFO] - 
 ***** Running Prediction *****
-[2025-07-01 00:06:22,893][transformers.trainer][INFO] -   Num examples = 138
-[2025-07-01 00:06:22,893][transformers.trainer][INFO] -   Batch size = 16
-[2025-07-01 00:06:23,379][__main__][INFO] - Inference results saved to jbcs2025_bertimbau-large-C2-encoder_classification-C2-essay_only_inference_results.jsonl
-[2025-07-01 00:06:23,384][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
-[2025-07-01 00:08:02,421][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
-[2025-07-01 00:08:02,421][__main__][INFO] - Bootstrap Confidence Intervals (95%):
-[2025-07-01 00:08:02,421][__main__][INFO] -   QWK: 0.4215 [0.2689, 0.5652]
-[2025-07-01 00:08:02,421][__main__][INFO] -   Macro_F1: 0.2837 [0.2126, 0.3709]
-[2025-07-01 00:08:02,421][__main__][INFO] -   Weighted_F1: 0.3818 [0.2966, 0.4676]
-[2025-07-01 00:08:02,421][__main__][INFO] - Inference results: {'accuracy': 0.39855072463768115, 'RMSE': 62.22912315078803, 'QWK': 0.4242242542347474, 'HDIV': 0.13043478260869568, 'Macro_F1': 0.2673521459388278, 'Micro_F1': 0.39855072463768115, 'Weighted_F1': 0.38254856956135197, 'TP_0': np.int64(0), 'TN_0': np.int64(137), 'FP_0': np.int64(0), 'FN_0': np.int64(1), 'TP_1': np.int64(21), 'TN_1': np.int64(70), 'FP_1': np.int64(33), 'FN_1': np.int64(14), 'TP_2': np.int64(0), 'TN_2': np.int64(129), 'FP_2': np.int64(4), 'FN_2': np.int64(5), 'TP_3': np.int64(12), 'TN_3': np.int64(83), 'FP_3': np.int64(4), 'FN_3': np.int64(39), 'TP_4': np.int64(16), 'TN_4': np.int64(78), 'FP_4': np.int64(34), 'FN_4': np.int64(10), 'TP_5': np.int64(6), 'TN_5': np.int64(110), 'FP_5': np.int64(8), 'FN_5': np.int64(14)}
-[2025-07-01 00:08:02,421][__main__][INFO] - Inference experiment completed
+[2025-07-10 01:14:22,876][transformers.trainer][INFO] -   Num examples = 138
+[2025-07-10 01:14:22,876][transformers.trainer][INFO] -   Batch size = 16
+[2025-07-10 01:14:24,390][__main__][INFO] - Inference results saved to jbcs2025_bert-large-portuguese-cased-encoder_classification-C2-essay_only-encoder_classification-C2-essay_only_inference_results.jsonl
+[2025-07-10 01:14:24,391][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
+[2025-07-10 01:16:29,838][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
+[2025-07-10 01:16:29,841][__main__][INFO] - Bootstrap Confidence Intervals (95%):
+[2025-07-10 01:16:29,841][__main__][INFO] -   QWK: 0.3135 [0.1449, 0.4709]
+[2025-07-10 01:16:29,841][__main__][INFO] -   Macro_F1: 0.3376 [0.2518, 0.4429]
+[2025-07-10 01:16:29,841][__main__][INFO] -   Weighted_F1: 0.4106 [0.3275, 0.4931]
+[2025-07-10 01:16:29,841][__main__][INFO] - Inference results: {'accuracy': 0.39855072463768115, 'RMSE': 67.93005780671166, 'QWK': 0.31696761677361585, 'HDIV': 0.1376811594202898, 'Macro_F1': 0.31891930577920913, 'Micro_F1': 0.39855072463768115, 'Weighted_F1': 0.4108570406154947, 'TP_0': np.int64(0), 'TN_0': np.int64(137), 'FP_0': np.int64(0), 'FN_0': np.int64(1), 'TP_1': np.int64(16), 'TN_1': np.int64(73), 'FP_1': np.int64(30), 'FN_1': np.int64(19), 'TP_2': np.int64(3), 'TN_2': np.int64(114), 'FP_2': np.int64(19), 'FN_2': np.int64(2), 'TP_3': np.int64(16), 'TN_3': np.int64(79), 'FP_3': np.int64(8), 'FN_3': np.int64(35), 'TP_4': np.int64(10), 'TN_4': np.int64(102), 'FP_4': np.int64(10), 'FN_4': np.int64(16), 'TP_5': np.int64(10), 'TN_5': np.int64(102), 'FP_5': np.int64(16), 'FN_5': np.int64(10)}
+[2025-07-10 01:16:29,843][__main__][INFO] - Inference experiment completed

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C3-encoder_classification-C3-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only}/.hydra/config.yaml

@@ -20,12 +20,12 @@ post_training_results:
   model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
 experiments:
   model:
-    name: kamel-usp/jbcs2025_bertimbau-large-C3
+    name: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only
     type: encoder_classification
     num_labels: 6
     output_dir: ./results/bertimbau_large/C3
     logging_dir: ./logs/bertimbau_large/C3
-    best_model_dir: ./results/bertimbau_large/C3/best_model
+    best_model_dir: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only
   tokenizer:
     name: neuralmind/bert-large-portuguese-cased
   dataset:

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C1-encoder_classification-C1-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only}/.hydra/hydra.yaml

@@ -1,6 +1,6 @@
 hydra:
   run:
-    dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: inference_output/2025-07-10/01-16-36
   sweep:
     dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
     subdir: ${hydra.job.num}
@@ -110,13 +110,14 @@ hydra:
   output_subdir: .hydra
   overrides:
     hydra:
+    - hydra.run.dir=inference_output/2025-07-10/01-16-36
     - hydra.mode=RUN
     task:
-    - experiments=large_models/C1
+    - experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only
   job:
     name: run_inference_experiment
     chdir: null
-    override_dirname: experiments=large_models/C1
+    override_dirname: experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only
     id: ???
     num: ???
     config_name: config
@@ -141,9 +142,9 @@ hydra:
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /workspace/jbcs2025/outputs/2025-07-01/00-02-02
+    output_dir: /workspace/jbcs2025/inference_output/2025-07-10/01-16-36
     choices:
-      experiments: large_models/C1
+      experiments: temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only
       hydra/env: default
       hydra/callbacks: null
       hydra/job_logging: default

runs/large_models/bertimbau/jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only

runs/large_models/bertimbau/jbcs2025_bert-large-portuguese-cased-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_bert-large-portuguese-cased-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only,2025-07-10 01:16:42,0.24545954800981098,0.09445649614356424,0.39034083613362736,0.29588433999006314,0.18382071874415615,0.12381966213862766,0.2605668201553197,0.13674715801669202,0.2556045233592953,0.1820167330799211,0.33321252550062413,0.15119579242070302

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C2-encoder_classification-C2-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only}/evaluation_results.csv

@@ -1,2 +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
-0.39855072463768115,62.22912315078803,0.4242242542347474,0.13043478260869568,0.2673521459388278,0.39855072463768115,0.38254856956135197,0,137,0,1,21,70,33,14,0,129,4,5,12,83,4,39,16,78,34,10,6,110,8,14,2025-07-01 00:04:40,jbcs2025_bertimbau-large-C2-encoder_classification-C2-essay_only
+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
+0.2753623188405797,64.87446070815474,0.2477700693756193,0.13043478260869568,0.17399856052293136,0.2753623188405797,0.25553888306634265,0,137,0,1,9,66,43,20,0,118,2,18,9,74,19,36,19,71,29,19,1,124,7,6,2025-07-10 01:16:42,jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C3-encoder_classification-C3-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only}/run_inference_experiment.log

@@ -1,5 +1,5 @@
-[2025-07-01 00:08:08,174][__main__][INFO] - Starting inference experiment
-[2025-07-01 00:08:08,175][__main__][INFO] - cache_dir: /tmp/
+[2025-07-10 01:16:42,095][__main__][INFO] - Starting inference experiment
+[2025-07-10 01:16:42,097][__main__][INFO] - cache_dir: /tmp/
 dataset:
   name: kamel-usp/aes_enem_dataset
   split: JBCS2025
@@ -21,12 +21,12 @@ post_training_results:
   model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
 experiments:
   model:
-    name: kamel-usp/jbcs2025_bertimbau-large-C3
+    name: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only
     type: encoder_classification
     num_labels: 6
     output_dir: ./results/bertimbau_large/C3
     logging_dir: ./logs/bertimbau_large/C3
-    best_model_dir: ./results/bertimbau_large/C3/best_model
+    best_model_dir: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only
   tokenizer:
     name: neuralmind/bert-large-portuguese-cased
   dataset:
@@ -41,9 +41,9 @@ experiments:
     gradient_accumulation_steps: 1
     gradient_checkpointing: false
 
-[2025-07-01 00:08:08,177][__main__][INFO] - Running inference with fine-tuned HF model
-[2025-07-01 00:08:12,983][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:08:12,984][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:16:42,099][__main__][INFO] - Running inference with fine-tuned HF model
+[2025-07-10 01:16:46,973][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:16:46,974][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -68,20 +68,14 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:08:13,201][transformers.tokenization_utils_base][INFO] - loading file vocab.txt from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/vocab.txt
-[2025-07-01 00:08:13,201][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at None
-[2025-07-01 00:08:13,201][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/added_tokens.json
-[2025-07-01 00:08:13,201][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/special_tokens_map.json
-[2025-07-01 00:08:13,201][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/tokenizer_config.json
-[2025-07-01 00:08:13,201][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
-[2025-07-01 00:08:13,201][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:08:13,202][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:16:47,174][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:16:47,175][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -106,14 +100,20 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:08:13,228][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:08:13,229][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:16:47,374][transformers.tokenization_utils_base][INFO] - loading file vocab.txt from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/vocab.txt
+[2025-07-10 01:16:47,375][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at None
+[2025-07-10 01:16:47,375][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/added_tokens.json
+[2025-07-10 01:16:47,375][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/special_tokens_map.json
+[2025-07-10 01:16:47,375][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/tokenizer_config.json
+[2025-07-10 01:16:47,375][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
+[2025-07-10 01:16:47,375][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:16:47,376][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -138,18 +138,73 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:08:13,245][__main__][INFO] - Tokenizer function parameters- Padding:max_length; Truncation: True; Use Full Context: False
-[2025-07-01 00:08:13,452][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bertimbau-large-C3
-[2025-07-01 00:08:13,452][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bertimbau-large-C3
-[2025-07-01 00:08:14,514][__main__][INFO] - Model need ≈ 2.62 GiB to run inference and 6.36 for training 
-[2025-07-01 00:08:15,376][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_bertimbau-large-C3/snapshots/fc8d63cfeeb43af3963a4c9550e6c3c2e5276adf/config.json
-[2025-07-01 00:08:15,377][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:16:47,405][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:16:47,406][transformers.configuration_utils][INFO] - Model config BertConfig {
+  "architectures": [
+    "BertForMaskedLM"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "directionality": "bidi",
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "output_past": true,
+  "pad_token_id": 0,
+  "pooler_fc_size": 768,
+  "pooler_num_attention_heads": 12,
+  "pooler_num_fc_layers": 3,
+  "pooler_size_per_head": 128,
+  "pooler_type": "first_token_transform",
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 29794
+}
+
+[2025-07-10 01:16:47,422][__main__][INFO] - Tokenizer function parameters- Padding:longest; Truncation: True; Use Full Context: False
+[2025-07-10 01:16:47,831][__main__][INFO] - 
+Token statistics for 'train' split:
+[2025-07-10 01:16:47,831][__main__][INFO] -   Total examples: 500
+[2025-07-10 01:16:47,831][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:16:47,831][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:16:47,831][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:16:47,831][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:16:47,920][__main__][INFO] - 
+Token statistics for 'validation' split:
+[2025-07-10 01:16:47,921][__main__][INFO] -   Total examples: 132
+[2025-07-10 01:16:47,921][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:16:47,921][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:16:47,921][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:16:47,921][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:16:48,015][__main__][INFO] - 
+Token statistics for 'test' split:
+[2025-07-10 01:16:48,015][__main__][INFO] -   Total examples: 138
+[2025-07-10 01:16:48,015][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:16:48,015][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:16:48,015][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:16:48,015][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:16:48,015][__main__][INFO] - If token statistics are the same (max, avg, min) keep in mind that this is due to batched tokenization and padding.
+[2025-07-10 01:16:48,015][__main__][INFO] - Model max length: 512. If it is the same as stats, then there is a high chance that sequences are being truncated.
+[2025-07-10 01:16:48,016][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only
+[2025-07-10 01:16:48,016][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only
+[2025-07-10 01:16:48,940][__main__][INFO] - Model need ≈ 2.62 GiB to run inference and 6.36 for training 
+[2025-07-10 01:16:49,806][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only/snapshots/95506fbe9cc549ab2018d8ddd3a10010ee1c8d61/config.json
+[2025-07-10 01:16:49,807][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForSequenceClassification"
   ],
@@ -190,37 +245,36 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "problem_type": "single_label_classification",
   "torch_dtype": "float32",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:08:54,977][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_bertimbau-large-C3/snapshots/fc8d63cfeeb43af3963a4c9550e6c3c2e5276adf/model.safetensors
-[2025-07-01 00:08:54,978][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.float32 as defined in model's config object
-[2025-07-01 00:08:54,978][transformers.modeling_utils][INFO] - Instantiating BertForSequenceClassification model under default dtype torch.float32.
-[2025-07-01 00:08:55,595][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing BertForSequenceClassification.
+[2025-07-10 01:17:16,434][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only/snapshots/95506fbe9cc549ab2018d8ddd3a10010ee1c8d61/model.safetensors
+[2025-07-10 01:17:16,438][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.float32 as defined in model's config object
+[2025-07-10 01:17:16,438][transformers.modeling_utils][INFO] - Instantiating BertForSequenceClassification model under default dtype torch.float32.
+[2025-07-10 01:17:17,158][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing BertForSequenceClassification.
 
-[2025-07-01 00:08:55,596][transformers.modeling_utils][INFO] - All the weights of BertForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_bertimbau-large-C3.
+[2025-07-10 01:17:17,158][transformers.modeling_utils][INFO] - All the weights of BertForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_bert-large-portuguese-cased-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 BertForSequenceClassification for predictions without further training.
-[2025-07-01 00:08:55,603][transformers.training_args][INFO] - PyTorch: setting up devices
-[2025-07-01 00:08:55,640][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 :-).
-[2025-07-01 00:08:55,645][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.
-[2025-07-01 00:08:55,662][transformers.trainer][INFO] - Using auto half precision backend
-[2025-07-01 00:08:59,124][__main__][INFO] - Running inference on test dataset
-[2025-07-01 00:08:59,125][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `BertForSequenceClassification.forward` and have been ignored: grades, prompt, id, id_prompt, essay_year, supporting_text, reference, essay_text. If grades, prompt, id, id_prompt, essay_year, supporting_text, reference, essay_text are not expected by `BertForSequenceClassification.forward`,  you can safely ignore this message.
-[2025-07-01 00:08:59,133][transformers.trainer][INFO] - 
+[2025-07-10 01:17:17,177][transformers.training_args][INFO] - PyTorch: setting up devices
+[2025-07-10 01:17:17,200][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 :-).
+[2025-07-10 01:17:17,210][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.
+[2025-07-10 01:17:17,245][transformers.trainer][INFO] - Using auto half precision backend
+[2025-07-10 01:17:20,549][__main__][INFO] - Running inference on test dataset
+[2025-07-10 01:17:20,550][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `BertForSequenceClassification.forward` and have been ignored: grades, id, essay_year, supporting_text, reference, essay_text, id_prompt, prompt. If grades, id, essay_year, supporting_text, reference, essay_text, id_prompt, prompt are not expected by `BertForSequenceClassification.forward`,  you can safely ignore this message.
+[2025-07-10 01:17:20,563][transformers.trainer][INFO] - 
 ***** Running Prediction *****
-[2025-07-01 00:08:59,133][transformers.trainer][INFO] -   Num examples = 138
-[2025-07-01 00:08:59,133][transformers.trainer][INFO] -   Batch size = 16
-[2025-07-01 00:08:59,694][__main__][INFO] - Inference results saved to jbcs2025_bertimbau-large-C3-encoder_classification-C3-essay_only_inference_results.jsonl
-[2025-07-01 00:08:59,699][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
-[2025-07-01 00:10:36,121][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
-[2025-07-01 00:10:36,122][__main__][INFO] - Bootstrap Confidence Intervals (95%):
-[2025-07-01 00:10:36,122][__main__][INFO] -   QWK: 0.2668 [0.1309, 0.3967]
-[2025-07-01 00:10:36,122][__main__][INFO] -   Macro_F1: 0.2055 [0.1392, 0.2902]
-[2025-07-01 00:10:36,122][__main__][INFO] -   Weighted_F1: 0.2492 [0.1750, 0.3272]
-[2025-07-01 00:10:36,122][__main__][INFO] - Inference results: {'accuracy': 0.2898550724637681, 'RMSE': 51.07539184552491, 'QWK': 0.26937738246505727, 'HDIV': 0.021739130434782594, 'Macro_F1': 0.19411606228274925, 'Micro_F1': 0.2898550724637681, 'Weighted_F1': 0.24825898925023224, 'TP_0': np.int64(0), 'TN_0': np.int64(137), 'FP_0': np.int64(0), 'FN_0': np.int64(1), 'TP_1': np.int64(1), 'TN_1': np.int64(108), 'FP_1': np.int64(1), 'FN_1': np.int64(28), 'TP_2': np.int64(13), 'TN_2': np.int64(87), 'FP_2': np.int64(33), 'FN_2': np.int64(5), 'TP_3': np.int64(19), 'TN_3': np.int64(44), 'FP_3': np.int64(49), 'FN_3': np.int64(26), 'TP_4': np.int64(6), 'TN_4': np.int64(94), 'FP_4': np.int64(6), 'FN_4': np.int64(32), 'TP_5': np.int64(1), 'TN_5': np.int64(122), 'FP_5': np.int64(9), 'FN_5': np.int64(6)}
-[2025-07-01 00:10:36,122][__main__][INFO] - Inference experiment completed
+[2025-07-10 01:17:20,563][transformers.trainer][INFO] -   Num examples = 138
+[2025-07-10 01:17:20,563][transformers.trainer][INFO] -   Batch size = 16
+[2025-07-10 01:17:22,039][__main__][INFO] - Inference results saved to jbcs2025_bert-large-portuguese-cased-encoder_classification-C3-essay_only-encoder_classification-C3-essay_only_inference_results.jsonl
+[2025-07-10 01:17:22,040][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
+[2025-07-10 01:19:27,495][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
+[2025-07-10 01:19:27,497][__main__][INFO] - Bootstrap Confidence Intervals (95%):
+[2025-07-10 01:19:27,497][__main__][INFO] -   QWK: 0.2455 [0.0945, 0.3903]
+[2025-07-10 01:19:27,497][__main__][INFO] -   Macro_F1: 0.1838 [0.1238, 0.2606]
+[2025-07-10 01:19:27,497][__main__][INFO] -   Weighted_F1: 0.2556 [0.1820, 0.3332]
+[2025-07-10 01:19:27,498][__main__][INFO] - Inference results: {'accuracy': 0.2753623188405797, 'RMSE': 64.87446070815474, 'QWK': 0.2477700693756193, 'HDIV': 0.13043478260869568, 'Macro_F1': 0.17399856052293136, 'Micro_F1': 0.2753623188405797, 'Weighted_F1': 0.25553888306634265, '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(66), 'FP_1': np.int64(43), 'FN_1': np.int64(20), 'TP_2': np.int64(0), 'TN_2': np.int64(118), 'FP_2': np.int64(2), 'FN_2': np.int64(18), 'TP_3': np.int64(9), 'TN_3': np.int64(74), 'FP_3': np.int64(19), 'FN_3': np.int64(36), 'TP_4': np.int64(19), 'TN_4': np.int64(71), 'FP_4': np.int64(29), 'FN_4': np.int64(19), 'TP_5': np.int64(1), 'TN_5': np.int64(124), 'FP_5': np.int64(7), 'FN_5': np.int64(6)}
+[2025-07-10 01:19:27,503][__main__][INFO] - Inference experiment completed

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C4-encoder_classification-C4-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only}/.hydra/config.yaml

@@ -20,12 +20,12 @@ post_training_results:
   model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
 experiments:
   model:
-    name: kamel-usp/jbcs2025_bertimbau-large-C4
+    name: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only
     type: encoder_classification
     num_labels: 6
     output_dir: ./results/bertimbau_large/C4
     logging_dir: ./logs/bertimbau_large/C4
-    best_model_dir: ./results/bertimbau_large/C4/best_model
+    best_model_dir: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only
   tokenizer:
     name: neuralmind/bert-large-portuguese-cased
   dataset:

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C4-encoder_classification-C4-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only}/.hydra/hydra.yaml

@@ -1,6 +1,6 @@
 hydra:
   run:
-    dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: inference_output/2025-07-10/01-19-33
   sweep:
     dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
     subdir: ${hydra.job.num}
@@ -110,13 +110,14 @@ hydra:
   output_subdir: .hydra
   overrides:
     hydra:
+    - hydra.run.dir=inference_output/2025-07-10/01-19-33
     - hydra.mode=RUN
     task:
-    - experiments=large_models/C4
+    - experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only
   job:
     name: run_inference_experiment
     chdir: null
-    override_dirname: experiments=large_models/C4
+    override_dirname: experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only
     id: ???
     num: ???
     config_name: config
@@ -141,9 +142,9 @@ hydra:
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /workspace/jbcs2025/outputs/2025-07-01/00-10-41
+    output_dir: /workspace/jbcs2025/inference_output/2025-07-10/01-19-33
     choices:
-      experiments: large_models/C4
+      experiments: temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only
       hydra/env: default
       hydra/callbacks: null
       hydra/job_logging: default

runs/large_models/bertimbau/jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C2-encoder_classification-C2-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only}/bootstrap_confidence_intervals.csv

@@ -1,2 +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_bertimbau-large-C2-encoder_classification-C2-essay_only,2025-07-01 00:04:40,0.421524005294385,0.2688782617235809,0.5651890498796338,0.2963107881560529,0.28374338367400226,0.21261906130152639,0.3709122659559081,0.1582932046543817,0.38177503835634297,0.2966054262581051,0.4675676487920818,0.1709622225339767
+jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only,2025-07-10 01:19:39,0.5942508582936158,0.4700019118242705,0.7064182815355218,0.2364163697112513,0.4619540479839987,0.29942982127027257,0.6409612672040503,0.34153144593377777,0.6059987254912385,0.5248203752695526,0.6834775202623181,0.1586571449927655

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C3-encoder_classification-C3-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only}/evaluation_results.csv

@@ -1,2 +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
-0.2898550724637681,51.07539184552491,0.26937738246505727,0.021739130434782594,0.19411606228274925,0.2898550724637681,0.24825898925023224,0,137,0,1,1,108,1,28,13,87,33,5,19,44,49,26,6,94,6,32,1,122,9,6,2025-07-01 00:08:08,jbcs2025_bertimbau-large-C3-encoder_classification-C3-essay_only
+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
+0.5869565217391305,29.29114225312413,0.5985849056603774,0.007246376811594235,0.45502820381025977,0.5869565217391305,0.6030619269196934,0,137,0,1,1,136,1,0,5,114,15,4,42,52,10,34,29,69,23,17,4,125,8,1,2025-07-10 01:19:39,jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C4-encoder_classification-C4-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only}/run_inference_experiment.log

@@ -1,5 +1,5 @@
-[2025-07-01 00:10:41,832][__main__][INFO] - Starting inference experiment
-[2025-07-01 00:10:41,833][__main__][INFO] - cache_dir: /tmp/
+[2025-07-10 01:19:39,398][__main__][INFO] - Starting inference experiment
+[2025-07-10 01:19:39,399][__main__][INFO] - cache_dir: /tmp/
 dataset:
   name: kamel-usp/aes_enem_dataset
   split: JBCS2025
@@ -21,12 +21,12 @@ post_training_results:
   model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
 experiments:
   model:
-    name: kamel-usp/jbcs2025_bertimbau-large-C4
+    name: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only
     type: encoder_classification
     num_labels: 6
     output_dir: ./results/bertimbau_large/C4
     logging_dir: ./logs/bertimbau_large/C4
-    best_model_dir: ./results/bertimbau_large/C4/best_model
+    best_model_dir: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only
   tokenizer:
     name: neuralmind/bert-large-portuguese-cased
   dataset:
@@ -41,9 +41,9 @@ experiments:
     gradient_accumulation_steps: 1
     gradient_checkpointing: false
 
-[2025-07-01 00:10:41,835][__main__][INFO] - Running inference with fine-tuned HF model
-[2025-07-01 00:10:47,728][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:10:47,729][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:19:39,401][__main__][INFO] - Running inference with fine-tuned HF model
+[2025-07-10 01:19:44,024][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:19:44,025][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -68,20 +68,14 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:10:47,955][transformers.tokenization_utils_base][INFO] - loading file vocab.txt from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/vocab.txt
-[2025-07-01 00:10:47,955][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at None
-[2025-07-01 00:10:47,955][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/added_tokens.json
-[2025-07-01 00:10:47,955][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/special_tokens_map.json
-[2025-07-01 00:10:47,955][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/tokenizer_config.json
-[2025-07-01 00:10:47,955][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
-[2025-07-01 00:10:47,955][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:10:47,955][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:19:44,243][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:19:44,243][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -106,14 +100,20 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:10:47,981][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:10:47,982][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:19:44,443][transformers.tokenization_utils_base][INFO] - loading file vocab.txt from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/vocab.txt
+[2025-07-10 01:19:44,443][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at None
+[2025-07-10 01:19:44,443][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/added_tokens.json
+[2025-07-10 01:19:44,443][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/special_tokens_map.json
+[2025-07-10 01:19:44,443][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/tokenizer_config.json
+[2025-07-10 01:19:44,443][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
+[2025-07-10 01:19:44,443][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:19:44,444][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -138,18 +138,73 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:10:47,998][__main__][INFO] - Tokenizer function parameters- Padding:max_length; Truncation: True; Use Full Context: False
-[2025-07-01 00:10:48,205][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bertimbau-large-C4
-[2025-07-01 00:10:48,205][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bertimbau-large-C4
-[2025-07-01 00:10:49,076][__main__][INFO] - Model need ≈ 2.62 GiB to run inference and 6.36 for training 
-[2025-07-01 00:10:50,078][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_bertimbau-large-C4/snapshots/e9e2cf7e79031197ee29e0150459a3788e5e8d38/config.json
-[2025-07-01 00:10:50,078][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:19:44,474][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:19:44,475][transformers.configuration_utils][INFO] - Model config BertConfig {
+  "architectures": [
+    "BertForMaskedLM"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "directionality": "bidi",
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "output_past": true,
+  "pad_token_id": 0,
+  "pooler_fc_size": 768,
+  "pooler_num_attention_heads": 12,
+  "pooler_num_fc_layers": 3,
+  "pooler_size_per_head": 128,
+  "pooler_type": "first_token_transform",
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 29794
+}
+
+[2025-07-10 01:19:44,492][__main__][INFO] - Tokenizer function parameters- Padding:longest; Truncation: True; Use Full Context: False
+[2025-07-10 01:19:44,896][__main__][INFO] - 
+Token statistics for 'train' split:
+[2025-07-10 01:19:44,896][__main__][INFO] -   Total examples: 500
+[2025-07-10 01:19:44,897][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:19:44,897][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:19:44,897][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:19:44,897][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:19:44,985][__main__][INFO] - 
+Token statistics for 'validation' split:
+[2025-07-10 01:19:44,985][__main__][INFO] -   Total examples: 132
+[2025-07-10 01:19:44,985][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:19:44,985][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:19:44,986][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:19:44,986][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:19:45,083][__main__][INFO] - 
+Token statistics for 'test' split:
+[2025-07-10 01:19:45,083][__main__][INFO] -   Total examples: 138
+[2025-07-10 01:19:45,083][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:19:45,083][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:19:45,083][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:19:45,083][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:19:45,083][__main__][INFO] - If token statistics are the same (max, avg, min) keep in mind that this is due to batched tokenization and padding.
+[2025-07-10 01:19:45,083][__main__][INFO] - Model max length: 512. If it is the same as stats, then there is a high chance that sequences are being truncated.
+[2025-07-10 01:19:45,083][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only
+[2025-07-10 01:19:45,084][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only
+[2025-07-10 01:19:46,216][__main__][INFO] - Model need ≈ 2.62 GiB to run inference and 6.36 for training 
+[2025-07-10 01:19:47,074][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only/snapshots/0322eca1d32dfaf784bb8ad63fa85977ca109d48/config.json
+[2025-07-10 01:19:47,075][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForSequenceClassification"
   ],
@@ -190,37 +245,36 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "problem_type": "single_label_classification",
   "torch_dtype": "float32",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:11:28,782][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_bertimbau-large-C4/snapshots/e9e2cf7e79031197ee29e0150459a3788e5e8d38/model.safetensors
-[2025-07-01 00:11:28,783][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.float32 as defined in model's config object
-[2025-07-01 00:11:28,783][transformers.modeling_utils][INFO] - Instantiating BertForSequenceClassification model under default dtype torch.float32.
-[2025-07-01 00:11:29,434][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing BertForSequenceClassification.
+[2025-07-10 01:20:14,543][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only/snapshots/0322eca1d32dfaf784bb8ad63fa85977ca109d48/model.safetensors
+[2025-07-10 01:20:14,546][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.float32 as defined in model's config object
+[2025-07-10 01:20:14,546][transformers.modeling_utils][INFO] - Instantiating BertForSequenceClassification model under default dtype torch.float32.
+[2025-07-10 01:20:15,314][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing BertForSequenceClassification.
 
-[2025-07-01 00:11:29,434][transformers.modeling_utils][INFO] - All the weights of BertForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_bertimbau-large-C4.
+[2025-07-10 01:20:15,315][transformers.modeling_utils][INFO] - All the weights of BertForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_bert-large-portuguese-cased-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 BertForSequenceClassification for predictions without further training.
-[2025-07-01 00:11:29,442][transformers.training_args][INFO] - PyTorch: setting up devices
-[2025-07-01 00:11:29,480][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 :-).
-[2025-07-01 00:11:29,485][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.
-[2025-07-01 00:11:29,503][transformers.trainer][INFO] - Using auto half precision backend
-[2025-07-01 00:11:33,779][__main__][INFO] - Running inference on test dataset
-[2025-07-01 00:11:33,780][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `BertForSequenceClassification.forward` and have been ignored: prompt, essay_year, id, reference, supporting_text, essay_text, grades, id_prompt. If prompt, essay_year, id, reference, supporting_text, essay_text, grades, id_prompt are not expected by `BertForSequenceClassification.forward`,  you can safely ignore this message.
-[2025-07-01 00:11:33,787][transformers.trainer][INFO] - 
+[2025-07-10 01:20:15,333][transformers.training_args][INFO] - PyTorch: setting up devices
+[2025-07-10 01:20:15,364][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 :-).
+[2025-07-10 01:20:15,377][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.
+[2025-07-10 01:20:15,409][transformers.trainer][INFO] - Using auto half precision backend
+[2025-07-10 01:20:18,784][__main__][INFO] - Running inference on test dataset
+[2025-07-10 01:20:18,785][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `BertForSequenceClassification.forward` and have been ignored: reference, id_prompt, essay_year, prompt, essay_text, supporting_text, grades, id. If reference, id_prompt, essay_year, prompt, essay_text, supporting_text, grades, id are not expected by `BertForSequenceClassification.forward`,  you can safely ignore this message.
+[2025-07-10 01:20:18,793][transformers.trainer][INFO] - 
 ***** Running Prediction *****
-[2025-07-01 00:11:33,787][transformers.trainer][INFO] -   Num examples = 138
-[2025-07-01 00:11:33,787][transformers.trainer][INFO] -   Batch size = 16
-[2025-07-01 00:11:34,275][__main__][INFO] - Inference results saved to jbcs2025_bertimbau-large-C4-encoder_classification-C4-essay_only_inference_results.jsonl
-[2025-07-01 00:11:34,282][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
-[2025-07-01 00:13:11,390][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
-[2025-07-01 00:13:11,391][__main__][INFO] - Bootstrap Confidence Intervals (95%):
-[2025-07-01 00:13:11,391][__main__][INFO] -   QWK: 0.5690 [0.4615, 0.6681]
-[2025-07-01 00:13:11,391][__main__][INFO] -   Macro_F1: 0.3225 [0.2338, 0.4349]
-[2025-07-01 00:13:11,391][__main__][INFO] -   Weighted_F1: 0.5690 [0.4869, 0.6499]
-[2025-07-01 00:13:11,391][__main__][INFO] - Inference results: {'accuracy': 0.5434782608695652, 'RMSE': 30.45547950507524, 'QWK': 0.5718939041414612, 'HDIV': 0.007246376811594235, 'Macro_F1': 0.30149864910503205, 'Micro_F1': 0.5434782608695652, 'Weighted_F1': 0.5677143444488495, '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(135), 'FP_1': np.int64(2), 'FN_1': np.int64(1), 'TP_2': np.int64(6), 'TN_2': np.int64(109), 'FP_2': np.int64(20), 'FN_2': np.int64(3), 'TP_3': np.int64(40), 'TN_3': np.int64(50), 'FP_3': np.int64(12), 'FN_3': np.int64(36), 'TP_4': np.int64(27), 'TN_4': np.int64(71), 'FP_4': np.int64(21), 'FN_4': np.int64(19), 'TP_5': np.int64(2), 'TN_5': np.int64(125), 'FP_5': np.int64(8), 'FN_5': np.int64(3)}
-[2025-07-01 00:13:11,391][__main__][INFO] - Inference experiment completed
+[2025-07-10 01:20:18,793][transformers.trainer][INFO] -   Num examples = 138
+[2025-07-10 01:20:18,793][transformers.trainer][INFO] -   Batch size = 16
+[2025-07-10 01:20:20,322][__main__][INFO] - Inference results saved to jbcs2025_bert-large-portuguese-cased-encoder_classification-C4-essay_only-encoder_classification-C4-essay_only_inference_results.jsonl
+[2025-07-10 01:20:20,323][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
+[2025-07-10 01:22:24,410][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
+[2025-07-10 01:22:24,413][__main__][INFO] - Bootstrap Confidence Intervals (95%):
+[2025-07-10 01:22:24,413][__main__][INFO] -   QWK: 0.5943 [0.4700, 0.7064]
+[2025-07-10 01:22:24,413][__main__][INFO] -   Macro_F1: 0.4620 [0.2994, 0.6410]
+[2025-07-10 01:22:24,413][__main__][INFO] -   Weighted_F1: 0.6060 [0.5248, 0.6835]
+[2025-07-10 01:22:24,413][__main__][INFO] - Inference results: {'accuracy': 0.5869565217391305, 'RMSE': 29.29114225312413, 'QWK': 0.5985849056603774, 'HDIV': 0.007246376811594235, 'Macro_F1': 0.45502820381025977, 'Micro_F1': 0.5869565217391305, 'Weighted_F1': 0.6030619269196934, 'TP_0': np.int64(0), 'TN_0': np.int64(137), 'FP_0': np.int64(0), 'FN_0': np.int64(1), 'TP_1': np.int64(1), 'TN_1': np.int64(136), 'FP_1': np.int64(1), 'FN_1': np.int64(0), 'TP_2': np.int64(5), 'TN_2': np.int64(114), 'FP_2': np.int64(15), 'FN_2': np.int64(4), 'TP_3': np.int64(42), 'TN_3': np.int64(52), 'FP_3': np.int64(10), 'FN_3': np.int64(34), 'TP_4': np.int64(29), 'TN_4': np.int64(69), 'FP_4': np.int64(23), 'FN_4': np.int64(17), 'TP_5': np.int64(4), 'TN_5': np.int64(125), 'FP_5': np.int64(8), 'FN_5': np.int64(1)}
+[2025-07-10 01:22:24,419][__main__][INFO] - Inference experiment completed

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C5-encoder_classification-C5-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only}/.hydra/config.yaml

@@ -20,12 +20,12 @@ post_training_results:
   model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
 experiments:
   model:
-    name: kamel-usp/jbcs2025_bertimbau-large-C5
+    name: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only
     type: encoder_classification
     num_labels: 6
     output_dir: ./results/bertimbau_large/C5
     logging_dir: ./logs/bertimbau_large/C5
-    best_model_dir: ./results/bertimbau_large/C5/best_model
+    best_model_dir: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only
   tokenizer:
     name: neuralmind/bert-large-portuguese-cased
   dataset:

runs/large_models/bertimbau/jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only/.hydra/hydra.yaml

@@ -0,0 +1,157 @@
+hydra:
+  run:
+    dir: inference_output/2025-07-10/01-22-30
+  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/2025-07-10/01-22-30
+    - hydra.mode=RUN
+    task:
+    - experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only
+  job:
+    name: run_inference_experiment
+    chdir: null
+    override_dirname: experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-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/2025-07-10/01-22-30
+    choices:
+      experiments: temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-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/bertimbau/jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- experiments=temp_inference/kamel-usp_jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C3-encoder_classification-C3-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only}/bootstrap_confidence_intervals.csv

@@ -1,2 +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_bertimbau-large-C3-encoder_classification-C3-essay_only,2025-07-01 00:08:08,0.26675529638544276,0.13086754944731044,0.3967413784980497,0.26587382905073925,0.20553535282282293,0.13922610890459958,0.29020915603986924,0.15098304713526967,0.2492329798786624,0.17501884398554457,0.32721551607766175,0.15219667209211718
+jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only,2025-07-10 01:22:36,0.4570995991774188,0.3260729920596689,0.582680258986133,0.2566072669264641,0.3005407739499739,0.23044185263938957,0.3822276851313438,0.15178583249195424,0.35876023783511307,0.270756957437106,0.44869522570312514,0.17793826826601916

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C4-encoder_classification-C4-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only}/evaluation_results.csv

@@ -1,2 +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
-0.5434782608695652,30.45547950507524,0.5718939041414612,0.007246376811594235,0.30149864910503205,0.5434782608695652,0.5677143444488495,0,137,0,1,0,135,2,1,6,109,20,3,40,50,12,36,27,71,21,19,2,125,8,3,2025-07-01 00:10:41,jbcs2025_bertimbau-large-C4-encoder_classification-C4-essay_only
+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
+0.41304347826086957,59.75796593554388,0.45915406932356084,0.14492753623188404,0.3013676792556103,0.41304347826086957,0.3597310868375336,5,115,1,17,6,94,12,26,14,82,32,10,3,112,1,22,29,71,35,3,0,135,0,3,2025-07-10 01:22:36,jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only

runs/large_models/bertimbau/{jbcs2025_bertimbau-large-C5-encoder_classification-C5-essay_only → jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only}/run_inference_experiment.log

@@ -1,5 +1,5 @@
-[2025-07-01 00:13:17,175][__main__][INFO] - Starting inference experiment
-[2025-07-01 00:13:17,176][__main__][INFO] - cache_dir: /tmp/
+[2025-07-10 01:22:36,158][__main__][INFO] - Starting inference experiment
+[2025-07-10 01:22:36,160][__main__][INFO] - cache_dir: /tmp/
 dataset:
   name: kamel-usp/aes_enem_dataset
   split: JBCS2025
@@ -21,12 +21,12 @@ post_training_results:
   model_path: /workspace/jbcs2025/outputs/2025-03-24/20-42-59
 experiments:
   model:
-    name: kamel-usp/jbcs2025_bertimbau-large-C5
+    name: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only
     type: encoder_classification
     num_labels: 6
     output_dir: ./results/bertimbau_large/C5
     logging_dir: ./logs/bertimbau_large/C5
-    best_model_dir: ./results/bertimbau_large/C5/best_model
+    best_model_dir: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only
   tokenizer:
     name: neuralmind/bert-large-portuguese-cased
   dataset:
@@ -41,9 +41,9 @@ experiments:
     gradient_accumulation_steps: 1
     gradient_checkpointing: false
 
-[2025-07-01 00:13:17,178][__main__][INFO] - Running inference with fine-tuned HF model
-[2025-07-01 00:13:21,935][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:13:21,936][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:22:36,162][__main__][INFO] - Running inference with fine-tuned HF model
+[2025-07-10 01:22:40,872][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:22:40,873][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -68,20 +68,14 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:13:22,154][transformers.tokenization_utils_base][INFO] - loading file vocab.txt from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/vocab.txt
-[2025-07-01 00:13:22,154][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at None
-[2025-07-01 00:13:22,154][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/added_tokens.json
-[2025-07-01 00:13:22,154][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/special_tokens_map.json
-[2025-07-01 00:13:22,154][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/tokenizer_config.json
-[2025-07-01 00:13:22,154][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
-[2025-07-01 00:13:22,154][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:13:22,155][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:22:41,085][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:22:41,086][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -106,14 +100,20 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:13:22,182][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
-[2025-07-01 00:13:22,182][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:22:41,339][transformers.tokenization_utils_base][INFO] - loading file vocab.txt from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/vocab.txt
+[2025-07-10 01:22:41,339][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at None
+[2025-07-10 01:22:41,339][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/added_tokens.json
+[2025-07-10 01:22:41,339][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/special_tokens_map.json
+[2025-07-10 01:22:41,339][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/tokenizer_config.json
+[2025-07-10 01:22:41,339][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
+[2025-07-10 01:22:41,340][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:22:41,340][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForMaskedLM"
   ],
@@ -138,18 +138,73 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:13:22,200][__main__][INFO] - Tokenizer function parameters- Padding:max_length; Truncation: True; Use Full Context: False
-[2025-07-01 00:13:22,405][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bertimbau-large-C5
-[2025-07-01 00:13:22,405][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bertimbau-large-C5
-[2025-07-01 00:13:23,299][__main__][INFO] - Model need ≈ 2.62 GiB to run inference and 6.36 for training 
-[2025-07-01 00:13:24,131][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_bertimbau-large-C5/snapshots/0c5e7dc4b09aa42297d3fcb051dbd7433b740d47/config.json
-[2025-07-01 00:13:24,132][transformers.configuration_utils][INFO] - Model config BertConfig {
+[2025-07-10 01:22:41,370][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--neuralmind--bert-large-portuguese-cased/snapshots/aa302f6ea73b759f7df9cad58bd272127b67ec28/config.json
+[2025-07-10 01:22:41,371][transformers.configuration_utils][INFO] - Model config BertConfig {
+  "architectures": [
+    "BertForMaskedLM"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "directionality": "bidi",
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "output_past": true,
+  "pad_token_id": 0,
+  "pooler_fc_size": 768,
+  "pooler_num_attention_heads": 12,
+  "pooler_num_fc_layers": 3,
+  "pooler_size_per_head": 128,
+  "pooler_type": "first_token_transform",
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.53.1",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 29794
+}
+
+[2025-07-10 01:22:41,388][__main__][INFO] - Tokenizer function parameters- Padding:longest; Truncation: True; Use Full Context: False
+[2025-07-10 01:22:41,797][__main__][INFO] - 
+Token statistics for 'train' split:
+[2025-07-10 01:22:41,797][__main__][INFO] -   Total examples: 500
+[2025-07-10 01:22:41,797][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:22:41,797][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:22:41,797][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:22:41,797][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:22:41,888][__main__][INFO] - 
+Token statistics for 'validation' split:
+[2025-07-10 01:22:41,888][__main__][INFO] -   Total examples: 132
+[2025-07-10 01:22:41,888][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:22:41,888][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:22:41,888][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:22:41,888][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:22:41,983][__main__][INFO] - 
+Token statistics for 'test' split:
+[2025-07-10 01:22:41,983][__main__][INFO] -   Total examples: 138
+[2025-07-10 01:22:41,983][__main__][INFO] -   Min tokens: 512
+[2025-07-10 01:22:41,983][__main__][INFO] -   Max tokens: 512
+[2025-07-10 01:22:41,983][__main__][INFO] -   Avg tokens: 512.00
+[2025-07-10 01:22:41,983][__main__][INFO] -   Std tokens: 0.00
+[2025-07-10 01:22:41,983][__main__][INFO] - If token statistics are the same (max, avg, min) keep in mind that this is due to batched tokenization and padding.
+[2025-07-10 01:22:41,983][__main__][INFO] - Model max length: 512. If it is the same as stats, then there is a high chance that sequences are being truncated.
+[2025-07-10 01:22:41,983][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only
+[2025-07-10 01:22:41,983][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only
+[2025-07-10 01:22:43,230][__main__][INFO] - Model need ≈ 2.62 GiB to run inference and 6.36 for training 
+[2025-07-10 01:22:44,519][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--kamel-usp--jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only/snapshots/96dc152f3c31b3b49ac73d4d82b684487d6e2bf5/config.json
+[2025-07-10 01:22:44,521][transformers.configuration_utils][INFO] - Model config BertConfig {
   "architectures": [
     "BertForSequenceClassification"
   ],
@@ -190,37 +245,36 @@ experiments:
   "pooler_size_per_head": 128,
   "pooler_type": "first_token_transform",
   "position_embedding_type": "absolute",
-  "problem_type": "single_label_classification",
   "torch_dtype": "float32",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 29794
 }
 
-[2025-07-01 00:14:02,966][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_bertimbau-large-C5/snapshots/0c5e7dc4b09aa42297d3fcb051dbd7433b740d47/model.safetensors
-[2025-07-01 00:14:02,967][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.float32 as defined in model's config object
-[2025-07-01 00:14:02,967][transformers.modeling_utils][INFO] - Instantiating BertForSequenceClassification model under default dtype torch.float32.
-[2025-07-01 00:14:03,581][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing BertForSequenceClassification.
+[2025-07-10 01:23:10,193][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--kamel-usp--jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only/snapshots/96dc152f3c31b3b49ac73d4d82b684487d6e2bf5/model.safetensors
+[2025-07-10 01:23:10,194][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.float32 as defined in model's config object
+[2025-07-10 01:23:10,194][transformers.modeling_utils][INFO] - Instantiating BertForSequenceClassification model under default dtype torch.float32.
+[2025-07-10 01:23:10,938][transformers.modeling_utils][INFO] - All model checkpoint weights were used when initializing BertForSequenceClassification.
 
-[2025-07-01 00:14:03,582][transformers.modeling_utils][INFO] - All the weights of BertForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_bertimbau-large-C5.
+[2025-07-10 01:23:10,939][transformers.modeling_utils][INFO] - All the weights of BertForSequenceClassification were initialized from the model checkpoint at kamel-usp/jbcs2025_bert-large-portuguese-cased-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 BertForSequenceClassification for predictions without further training.
-[2025-07-01 00:14:03,589][transformers.training_args][INFO] - PyTorch: setting up devices
-[2025-07-01 00:14:03,639][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 :-).
-[2025-07-01 00:14:03,644][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.
-[2025-07-01 00:14:03,662][transformers.trainer][INFO] - Using auto half precision backend
-[2025-07-01 00:14:07,138][__main__][INFO] - Running inference on test dataset
-[2025-07-01 00:14:07,138][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `BertForSequenceClassification.forward` and have been ignored: essay_year, id, reference, grades, supporting_text, essay_text, id_prompt, prompt. If essay_year, id, reference, grades, supporting_text, essay_text, id_prompt, prompt are not expected by `BertForSequenceClassification.forward`,  you can safely ignore this message.
-[2025-07-01 00:14:07,145][transformers.trainer][INFO] - 
+[2025-07-10 01:23:10,952][transformers.training_args][INFO] - PyTorch: setting up devices
+[2025-07-10 01:23:10,975][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 :-).
+[2025-07-10 01:23:10,982][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.
+[2025-07-10 01:23:11,007][transformers.trainer][INFO] - Using auto half precision backend
+[2025-07-10 01:23:14,298][__main__][INFO] - Running inference on test dataset
+[2025-07-10 01:23:14,299][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `BertForSequenceClassification.forward` and have been ignored: id_prompt, prompt, reference, essay_year, grades, id, supporting_text, essay_text. If id_prompt, prompt, reference, essay_year, grades, id, supporting_text, essay_text are not expected by `BertForSequenceClassification.forward`,  you can safely ignore this message.
+[2025-07-10 01:23:14,307][transformers.trainer][INFO] - 
 ***** Running Prediction *****
-[2025-07-01 00:14:07,145][transformers.trainer][INFO] -   Num examples = 138
-[2025-07-01 00:14:07,145][transformers.trainer][INFO] -   Batch size = 16
-[2025-07-01 00:14:07,624][__main__][INFO] - Inference results saved to jbcs2025_bertimbau-large-C5-encoder_classification-C5-essay_only_inference_results.jsonl
-[2025-07-01 00:14:07,631][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
-[2025-07-01 00:15:43,889][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
-[2025-07-01 00:15:43,889][__main__][INFO] - Bootstrap Confidence Intervals (95%):
-[2025-07-01 00:15:43,889][__main__][INFO] -   QWK: 0.4766 [0.3498, 0.5988]
-[2025-07-01 00:15:43,889][__main__][INFO] -   Macro_F1: 0.3186 [0.2331, 0.4147]
-[2025-07-01 00:15:43,889][__main__][INFO] -   Weighted_F1: 0.3519 [0.2710, 0.4369]
-[2025-07-01 00:15:43,889][__main__][INFO] - Inference results: {'accuracy': 0.36231884057971014, 'RMSE': 61.10100926607787, 'QWK': 0.4785241515279538, 'HDIV': 0.14492753623188404, 'Macro_F1': 0.3255241258616379, 'Micro_F1': 0.36231884057971014, 'Weighted_F1': 0.3520852841017614, 'TP_0': np.int64(6), 'TN_0': np.int64(112), 'FP_0': np.int64(4), 'FN_0': np.int64(16), 'TP_1': np.int64(10), 'TN_1': np.int64(88), 'FP_1': np.int64(18), 'FN_1': np.int64(22), 'TP_2': np.int64(4), 'TN_2': np.int64(108), 'FP_2': np.int64(6), 'FN_2': np.int64(20), 'TP_3': np.int64(8), 'TN_3': np.int64(85), 'FP_3': np.int64(28), 'FN_3': np.int64(17), 'TP_4': np.int64(21), 'TN_4': np.int64(79), 'FP_4': np.int64(27), 'FN_4': np.int64(11), 'TP_5': np.int64(1), 'TN_5': np.int64(130), 'FP_5': np.int64(5), 'FN_5': np.int64(2)}
-[2025-07-01 00:15:43,890][__main__][INFO] - Inference experiment completed
+[2025-07-10 01:23:14,307][transformers.trainer][INFO] -   Num examples = 138
+[2025-07-10 01:23:14,307][transformers.trainer][INFO] -   Batch size = 16
+[2025-07-10 01:23:15,542][__main__][INFO] - Inference results saved to jbcs2025_bert-large-portuguese-cased-encoder_classification-C5-essay_only-encoder_classification-C5-essay_only_inference_results.jsonl
+[2025-07-10 01:23:15,543][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
+[2025-07-10 01:25:20,892][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
+[2025-07-10 01:25:20,892][__main__][INFO] - Bootstrap Confidence Intervals (95%):
+[2025-07-10 01:25:20,892][__main__][INFO] -   QWK: 0.4571 [0.3261, 0.5827]
+[2025-07-10 01:25:20,892][__main__][INFO] -   Macro_F1: 0.3005 [0.2304, 0.3822]
+[2025-07-10 01:25:20,892][__main__][INFO] -   Weighted_F1: 0.3588 [0.2708, 0.4487]
+[2025-07-10 01:25:20,892][__main__][INFO] - Inference results: {'accuracy': 0.41304347826086957, 'RMSE': 59.75796593554388, 'QWK': 0.45915406932356084, 'HDIV': 0.14492753623188404, 'Macro_F1': 0.3013676792556103, 'Micro_F1': 0.41304347826086957, 'Weighted_F1': 0.3597310868375336, 'TP_0': np.int64(5), 'TN_0': np.int64(115), 'FP_0': np.int64(1), 'FN_0': np.int64(17), 'TP_1': np.int64(6), 'TN_1': np.int64(94), 'FP_1': np.int64(12), 'FN_1': np.int64(26), 'TP_2': np.int64(14), 'TN_2': np.int64(82), 'FP_2': np.int64(32), 'FN_2': np.int64(10), 'TP_3': np.int64(3), 'TN_3': np.int64(112), 'FP_3': np.int64(1), 'FN_3': np.int64(22), 'TP_4': np.int64(29), 'TN_4': np.int64(71), 'FP_4': np.int64(35), 'FN_4': np.int64(3), 'TP_5': np.int64(0), 'TN_5': np.int64(135), 'FP_5': np.int64(0), 'FN_5': np.int64(3)}
+[2025-07-10 01:25:20,892][__main__][INFO] - Inference experiment completed

runs/large_models/bertimbau/jbcs2025_bertimbau-large-C1-encoder_classification-C1-essay_only/.hydra/overrides.yaml

@@ -1 +0,0 @@
-- experiments=large_models/C1

runs/large_models/bertimbau/jbcs2025_bertimbau-large-C2-encoder_classification-C2-essay_only/.hydra/overrides.yaml

@@ -1 +0,0 @@
-- experiments=large_models/C2

runs/large_models/bertimbau/jbcs2025_bertimbau-large-C3-encoder_classification-C3-essay_only/.hydra/overrides.yaml

@@ -1 +0,0 @@
-- experiments=large_models/C3

runs/large_models/bertimbau/jbcs2025_bertimbau-large-C4-encoder_classification-C4-essay_only/.hydra/overrides.yaml

@@ -1 +0,0 @@
-- experiments=large_models/C4

runs/large_models/bertimbau/jbcs2025_bertimbau-large-C5-encoder_classification-C5-essay_only/.hydra/hydra.yaml

@@ -1,156 +0,0 @@
-hydra:
-  run:
-    dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S}
-  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.mode=RUN
-    task:
-    - experiments=large_models/C5
-  job:
-    name: run_inference_experiment
-    chdir: null
-    override_dirname: experiments=large_models/C5
-    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/outputs/2025-07-01/00-13-17
-    choices:
-      experiments: large_models/C5
-      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/bertimbau/jbcs2025_bertimbau-large-C5-encoder_classification-C5-essay_only/.hydra/overrides.yaml

@@ -1 +0,0 @@
-- experiments=large_models/C5

runs/large_models/bertimbau/jbcs2025_bertimbau-large-C5-encoder_classification-C5-essay_only/bootstrap_confidence_intervals.csv

@@ -1,2 +0,0 @@
-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_bertimbau-large-C5-encoder_classification-C5-essay_only,2025-07-01 00:13:17,0.4765829843065011,0.34979869197784125,0.5987775413884535,0.24897884941061227,0.3185511976163596,0.23313502944056644,0.4147475443192737,0.18161251487870725,0.3518695561256887,0.2709850141597966,0.4369107158200968,0.16592570166030018

runs/large_models/bertimbau/jbcs2025_bertimbau-large-C5-encoder_classification-C5-essay_only/evaluation_results.csv

@@ -1,2 +0,0 @@
-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
-0.36231884057971014,61.10100926607787,0.4785241515279538,0.14492753623188404,0.3255241258616379,0.36231884057971014,0.3520852841017614,6,112,4,16,10,88,18,22,4,108,6,20,8,85,28,17,21,79,27,11,1,130,5,2,2025-07-01 00:13:17,jbcs2025_bertimbau-large-C5-encoder_classification-C5-essay_only

runs/slm_decoder_models/llama-3.1-8b/jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16-llama31_classification_lora-C1-essay_only-r16/.hydra/hydra.yaml

@@ -1,6 +1,6 @@
 hydra:
   run:
-    dir: inference_output/2025-07-06/16-22-15
+    dir: inference_output/2025-07-09/19-56-19
   sweep:
     dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
     subdir: ${hydra.job.num}
@@ -110,7 +110,7 @@ hydra:
   output_subdir: .hydra
   overrides:
     hydra:
-    - hydra.run.dir=inference_output/2025-07-06/16-22-15
+    - hydra.run.dir=inference_output/2025-07-09/19-56-19
     - hydra.mode=RUN
     task:
     - experiments=temp_inference/kamel-usp_jbcs2025_Llama-3_1-8B-llama31_classification_lora-C1-essay_only-r16
@@ -142,7 +142,7 @@ hydra:
     - path: ''
       schema: structured
       provider: schema
-    output_dir: /workspace/jbcs2025/inference_output/2025-07-06/16-22-15
+    output_dir: /workspace/jbcs2025/inference_output/2025-07-09/19-56-19
     choices:
       experiments: temp_inference/kamel-usp_jbcs2025_Llama-3_1-8B-llama31_classification_lora-C1-essay_only-r16
       hydra/env: default

runs/slm_decoder_models/llama-3.1-8b/jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16-llama31_classification_lora-C1-essay_only-r16/bootstrap_confidence_intervals.csv

@@ -1,2 +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_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16-llama31_classification_lora-C1-essay_only-r16,2025-07-06 16:22:20,0.6852010393656068,0.5881177044303095,0.7734829594694392,0.1853652550391297,0.5201141818478505,0.40078232785453927,0.6699223338679838,0.26914000601344457,0.6659186862378589,0.5849321458851546,0.7433844647084771,0.15845231882332245
+jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16-llama31_classification_lora-C1-essay_only-r16,2025-07-09 19:56:26,0.6852010393656068,0.5881177044303095,0.7734829594694392,0.1853652550391297,0.5201141818478505,0.40078232785453927,0.6699223338679838,0.26914000601344457,0.6659186862378589,0.5849321458851546,0.7433844647084771,0.15845231882332245

runs/slm_decoder_models/llama-3.1-8b/jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16-llama31_classification_lora-C1-essay_only-r16/evaluation_results.csv

@@ -1,2 +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
-0.6594202898550725,25.931906372573962,0.6867564182842831,0.007246376811594235,0.48154898864576284,0.6594202898550725,0.6655256851610287,0,137,0,1,0,138,0,0,7,123,5,3,50,58,14,16,28,73,14,23,6,114,14,4,2025-07-06 16:22:20,jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16-llama31_classification_lora-C1-essay_only-r16
+0.6594202898550725,25.931906372573962,0.6867564182842831,0.007246376811594235,0.48154898864576284,0.6594202898550725,0.6655256851610287,0,137,0,1,0,138,0,0,7,123,5,3,50,58,14,16,28,73,14,23,6,114,14,4,2025-07-09 19:56:26,jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16-llama31_classification_lora-C1-essay_only-r16

runs/slm_decoder_models/llama-3.1-8b/jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16-llama31_classification_lora-C1-essay_only-r16/run_inference_experiment.log

@@ -1,5 +1,5 @@
-[2025-07-06 16:22:20,211][__main__][INFO] - Starting inference experiment
-[2025-07-06 16:22:20,212][__main__][INFO] - cache_dir: /tmp/
+[2025-07-09 19:56:26,399][__main__][INFO] - Starting inference experiment
+[2025-07-09 19:56:26,400][__main__][INFO] - cache_dir: /tmp/
 dataset:
   name: kamel-usp/aes_enem_dataset
   split: JBCS2025
@@ -45,22 +45,45 @@ experiments:
     gradient_accumulation_steps: 2
     gradient_checkpointing: true
 
-[2025-07-06 16:22:20,214][__main__][INFO] - Running inference with fine-tuned HF model
-[2025-07-06 16:22:23,567][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at /tmp/models--meta-llama--Llama-3.1-8B/snapshots/d04e592bb4f6aa9cfee91e2e20afa771667e1d4b/tokenizer.json
-[2025-07-06 16:22:23,567][transformers.tokenization_utils_base][INFO] - loading file tokenizer.model from cache at None
-[2025-07-06 16:22:23,567][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at None
-[2025-07-06 16:22:23,567][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--meta-llama--Llama-3.1-8B/snapshots/d04e592bb4f6aa9cfee91e2e20afa771667e1d4b/special_tokens_map.json
-[2025-07-06 16:22:23,567][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--meta-llama--Llama-3.1-8B/snapshots/d04e592bb4f6aa9cfee91e2e20afa771667e1d4b/tokenizer_config.json
-[2025-07-06 16:22:23,567][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
-[2025-07-06 16:22:23,864][transformers.tokenization_utils_base][INFO] - Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
-[2025-07-06 16:22:23,871][__main__][INFO] - Tokenizer function parameters- Padding:longest; Truncation: False; Use Full Context: False
-[2025-07-06 16:22:24,163][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16
-[2025-07-06 16:22:24,164][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16
-[2025-07-06 16:22:25,971][__main__][INFO] - Model need ≈ 46.09 GiB to run inference and 136.77 for training 
-[2025-07-06 16:22:26,133][__main__][INFO] - Loading PEFT model configuration from kamel-usp/jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16
-[2025-07-06 16:22:26,133][__main__][INFO] - Base model name: meta-llama/Llama-3.1-8B
-[2025-07-06 16:22:26,273][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--meta-llama--Llama-3.1-8B/snapshots/d04e592bb4f6aa9cfee91e2e20afa771667e1d4b/config.json
-[2025-07-06 16:22:26,275][transformers.configuration_utils][INFO] - Model config LlamaConfig {
+[2025-07-09 19:56:26,404][__main__][INFO] - Running inference with fine-tuned HF model
+[2025-07-09 19:56:31,012][transformers.tokenization_utils_base][INFO] - loading file tokenizer.json from cache at /tmp/models--meta-llama--Llama-3.1-8B/snapshots/d04e592bb4f6aa9cfee91e2e20afa771667e1d4b/tokenizer.json
+[2025-07-09 19:56:31,013][transformers.tokenization_utils_base][INFO] - loading file tokenizer.model from cache at None
+[2025-07-09 19:56:31,013][transformers.tokenization_utils_base][INFO] - loading file added_tokens.json from cache at None
+[2025-07-09 19:56:31,013][transformers.tokenization_utils_base][INFO] - loading file special_tokens_map.json from cache at /tmp/models--meta-llama--Llama-3.1-8B/snapshots/d04e592bb4f6aa9cfee91e2e20afa771667e1d4b/special_tokens_map.json
+[2025-07-09 19:56:31,013][transformers.tokenization_utils_base][INFO] - loading file tokenizer_config.json from cache at /tmp/models--meta-llama--Llama-3.1-8B/snapshots/d04e592bb4f6aa9cfee91e2e20afa771667e1d4b/tokenizer_config.json
+[2025-07-09 19:56:31,013][transformers.tokenization_utils_base][INFO] - loading file chat_template.jinja from cache at None
+[2025-07-09 19:56:31,557][transformers.tokenization_utils_base][INFO] - Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
+[2025-07-09 19:56:31,566][__main__][INFO] - Tokenizer function parameters- Padding:longest; Truncation: False; Use Full Context: False
+[2025-07-09 19:56:33,275][__main__][INFO] - 
+Token statistics for 'train' split:
+[2025-07-09 19:56:33,275][__main__][INFO] -   Total examples: 500
+[2025-07-09 19:56:33,275][__main__][INFO] -   Min tokens: 2479
+[2025-07-09 19:56:33,275][__main__][INFO] -   Max tokens: 2479
+[2025-07-09 19:56:33,275][__main__][INFO] -   Avg tokens: 2479.00
+[2025-07-09 19:56:33,276][__main__][INFO] -   Std tokens: 0.00
+[2025-07-09 19:56:33,503][__main__][INFO] - 
+Token statistics for 'validation' split:
+[2025-07-09 19:56:33,504][__main__][INFO] -   Total examples: 132
+[2025-07-09 19:56:33,504][__main__][INFO] -   Min tokens: 2193
+[2025-07-09 19:56:33,504][__main__][INFO] -   Max tokens: 2193
+[2025-07-09 19:56:33,504][__main__][INFO] -   Avg tokens: 2193.00
+[2025-07-09 19:56:33,504][__main__][INFO] -   Std tokens: 0.00
+[2025-07-09 19:56:33,753][__main__][INFO] - 
+Token statistics for 'test' split:
+[2025-07-09 19:56:33,753][__main__][INFO] -   Total examples: 138
+[2025-07-09 19:56:33,753][__main__][INFO] -   Min tokens: 2254
+[2025-07-09 19:56:33,753][__main__][INFO] -   Max tokens: 2254
+[2025-07-09 19:56:33,753][__main__][INFO] -   Avg tokens: 2254.00
+[2025-07-09 19:56:33,753][__main__][INFO] -   Std tokens: 0.00
+[2025-07-09 19:56:33,753][__main__][INFO] - If token statistics are the same (max, avg, min) keep in mind that this is due to batched tokenization and padding.
+[2025-07-09 19:56:33,753][__main__][INFO] - Model max length: 131072. If it is the same as stats, then there is a high chance that sequences are being truncated.
+[2025-07-09 19:56:33,754][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16
+[2025-07-09 19:56:33,754][__main__][INFO] - Loading model from: kamel-usp/jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16
+[2025-07-09 19:56:36,618][__main__][INFO] - Model need ≈ 46.09 GiB to run inference and 136.77 for training 
+[2025-07-09 19:56:36,879][__main__][INFO] - Loading PEFT model configuration from kamel-usp/jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16
+[2025-07-09 19:56:36,879][__main__][INFO] - Base model name: meta-llama/Llama-3.1-8B
+[2025-07-09 19:56:37,086][transformers.configuration_utils][INFO] - loading configuration file config.json from cache at /tmp/models--meta-llama--Llama-3.1-8B/snapshots/d04e592bb4f6aa9cfee91e2e20afa771667e1d4b/config.json
+[2025-07-09 19:56:37,089][transformers.configuration_utils][INFO] - Model config LlamaConfig {
   "architectures": [
     "LlamaForCausalLM"
   ],
@@ -107,38 +130,38 @@ experiments:
   "rope_theta": 500000.0,
   "tie_word_embeddings": false,
   "torch_dtype": "bfloat16",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.53.1",
   "use_cache": true,
   "vocab_size": 128256
 }
 
-[2025-07-06 16:22:26,422][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--meta-llama--Llama-3.1-8B/snapshots/d04e592bb4f6aa9cfee91e2e20afa771667e1d4b/model.safetensors.index.json
-[2025-07-06 16:22:26,422][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.bfloat16 as defined in model's config object
-[2025-07-06 16:22:26,422][transformers.modeling_utils][INFO] - Instantiating LlamaForSequenceClassification model under default dtype torch.bfloat16.
-[2025-07-06 16:22:29,885][transformers.modeling_utils][INFO] - Some weights of the model checkpoint at meta-llama/Llama-3.1-8B were not used when initializing LlamaForSequenceClassification: ['lm_head.weight']
+[2025-07-09 19:56:37,312][transformers.modeling_utils][INFO] - loading weights file model.safetensors from cache at /tmp/models--meta-llama--Llama-3.1-8B/snapshots/d04e592bb4f6aa9cfee91e2e20afa771667e1d4b/model.safetensors.index.json
+[2025-07-09 19:56:37,313][transformers.modeling_utils][INFO] - Will use torch_dtype=torch.bfloat16 as defined in model's config object
+[2025-07-09 19:56:37,313][transformers.modeling_utils][INFO] - Instantiating LlamaForSequenceClassification model under default dtype torch.bfloat16.
+[2025-07-09 19:56:41,901][transformers.modeling_utils][INFO] - Some weights of the model checkpoint at meta-llama/Llama-3.1-8B were not used when initializing LlamaForSequenceClassification: ['lm_head.weight']
 - This IS expected if you are initializing LlamaForSequenceClassification from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
 - This IS NOT expected if you are initializing LlamaForSequenceClassification from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
-[2025-07-06 16:22:29,885][transformers.modeling_utils][WARNING] - Some weights of LlamaForSequenceClassification were not initialized from the model checkpoint at meta-llama/Llama-3.1-8B and are newly initialized: ['score.weight']
+[2025-07-09 19:56:41,902][transformers.modeling_utils][WARNING] - Some weights of LlamaForSequenceClassification were not initialized from the model checkpoint at meta-llama/Llama-3.1-8B and are newly initialized: ['score.weight']
 You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
-[2025-07-06 16:22:33,282][__main__][INFO] - Loaded fine-tuned PEFT model from kamel-usp/jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16
-[2025-07-06 16:22:33,285][__main__][INFO] - None
-[2025-07-06 16:22:33,288][transformers.training_args][INFO] - PyTorch: setting up devices
-[2025-07-06 16:22:33,342][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 :-).
-[2025-07-06 16:22:33,347][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.
-[2025-07-06 16:22:33,366][transformers.trainer][INFO] - Using auto half precision backend
-[2025-07-06 16:22:33,367][transformers.trainer][WARNING] - No label_names provided for model class `PeftModelForSequenceClassification`. Since `PeftModel` hides base models input arguments, if label_names is not given, label_names can't be set automatically within `Trainer`. Note that empty label_names list will be used instead.
-[2025-07-06 16:22:36,746][__main__][INFO] - Running inference on test dataset
-[2025-07-06 16:22:36,746][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `PeftModelForSequenceClassification.forward` and have been ignored: essay_year, reference, grades, id_prompt, essay_text, supporting_text, id, prompt. If essay_year, reference, grades, id_prompt, essay_text, supporting_text, id, prompt are not expected by `PeftModelForSequenceClassification.forward`,  you can safely ignore this message.
-[2025-07-06 16:22:36,762][transformers.trainer][INFO] - 
+[2025-07-09 19:56:48,788][__main__][INFO] - Loaded fine-tuned PEFT model from kamel-usp/jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16
+[2025-07-09 19:56:48,794][__main__][INFO] - None
+[2025-07-09 19:56:48,806][transformers.training_args][INFO] - PyTorch: setting up devices
+[2025-07-09 19:56:48,831][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 :-).
+[2025-07-09 19:56:48,845][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.
+[2025-07-09 19:56:48,882][transformers.trainer][INFO] - Using auto half precision backend
+[2025-07-09 19:56:48,883][transformers.trainer][WARNING] - No label_names provided for model class `PeftModelForSequenceClassification`. Since `PeftModel` hides base models input arguments, if label_names is not given, label_names can't be set automatically within `Trainer`. Note that empty label_names list will be used instead.
+[2025-07-09 19:56:52,225][__main__][INFO] - Running inference on test dataset
+[2025-07-09 19:56:52,226][transformers.trainer][INFO] - The following columns in the test set don't have a corresponding argument in `PeftModelForSequenceClassification.forward` and have been ignored: grades, prompt, essay_year, essay_text, supporting_text, id, reference, id_prompt. If grades, prompt, essay_year, essay_text, supporting_text, id, reference, id_prompt are not expected by `PeftModelForSequenceClassification.forward`,  you can safely ignore this message.
+[2025-07-09 19:56:52,252][transformers.trainer][INFO] - 
 ***** Running Prediction *****
-[2025-07-06 16:22:36,763][transformers.trainer][INFO] -   Num examples = 138
-[2025-07-06 16:22:36,763][transformers.trainer][INFO] -   Batch size = 4
-[2025-07-06 16:22:49,505][__main__][INFO] - Inference results saved to jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16-llama31_classification_lora-C1-essay_only-r16_inference_results.jsonl
-[2025-07-06 16:22:49,533][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
-[2025-07-06 16:24:20,979][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
-[2025-07-06 16:24:20,979][__main__][INFO] - Bootstrap Confidence Intervals (95%):
-[2025-07-06 16:24:20,979][__main__][INFO] -   QWK: 0.6852 [0.5881, 0.7735]
-[2025-07-06 16:24:20,980][__main__][INFO] -   Macro_F1: 0.5201 [0.4008, 0.6699]
-[2025-07-06 16:24:20,980][__main__][INFO] -   Weighted_F1: 0.6659 [0.5849, 0.7434]
-[2025-07-06 16:24:20,980][__main__][INFO] - Inference results: {'accuracy': 0.6594202898550725, 'RMSE': 25.931906372573962, 'QWK': 0.6867564182842831, 'HDIV': 0.007246376811594235, 'Macro_F1': 0.48154898864576284, 'Micro_F1': 0.6594202898550725, 'Weighted_F1': 0.6655256851610287, '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(7), 'TN_2': np.int64(123), 'FP_2': np.int64(5), 'FN_2': np.int64(3), 'TP_3': np.int64(50), 'TN_3': np.int64(58), 'FP_3': np.int64(14), 'FN_3': np.int64(16), 'TP_4': np.int64(28), 'TN_4': np.int64(73), 'FP_4': np.int64(14), 'FN_4': np.int64(23), 'TP_5': np.int64(6), 'TN_5': np.int64(114), 'FP_5': np.int64(14), 'FN_5': np.int64(4)}
-[2025-07-06 16:24:20,980][__main__][INFO] - Inference experiment completed
+[2025-07-09 19:56:52,252][transformers.trainer][INFO] -   Num examples = 138
+[2025-07-09 19:56:52,252][transformers.trainer][INFO] -   Batch size = 4
+[2025-07-09 19:57:52,735][__main__][INFO] - Inference results saved to jbcs2025_Llama-3.1-8B-llama31_classification_lora-C1-essay_only-r16-llama31_classification_lora-C1-essay_only-r16_inference_results.jsonl
+[2025-07-09 19:57:52,737][__main__][INFO] - Computing bootstrap confidence intervals for metrics: ['QWK', 'Macro_F1', 'Weighted_F1']
+[2025-07-09 20:00:05,966][__main__][INFO] - Bootstrap CI results saved to bootstrap_confidence_intervals.csv
+[2025-07-09 20:00:05,967][__main__][INFO] - Bootstrap Confidence Intervals (95%):
+[2025-07-09 20:00:05,967][__main__][INFO] -   QWK: 0.6852 [0.5881, 0.7735]
+[2025-07-09 20:00:05,967][__main__][INFO] -   Macro_F1: 0.5201 [0.4008, 0.6699]
+[2025-07-09 20:00:05,967][__main__][INFO] -   Weighted_F1: 0.6659 [0.5849, 0.7434]
+[2025-07-09 20:00:05,967][__main__][INFO] - Inference results: {'accuracy': 0.6594202898550725, 'RMSE': 25.931906372573962, 'QWK': 0.6867564182842831, 'HDIV': 0.007246376811594235, 'Macro_F1': 0.48154898864576284, 'Micro_F1': 0.6594202898550725, 'Weighted_F1': 0.6655256851610287, '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(7), 'TN_2': np.int64(123), 'FP_2': np.int64(5), 'FN_2': np.int64(3), 'TP_3': np.int64(50), 'TN_3': np.int64(58), 'FP_3': np.int64(14), 'FN_3': np.int64(16), 'TP_4': np.int64(28), 'TN_4': np.int64(73), 'FP_4': np.int64(14), 'FN_4': np.int64(23), 'TP_5': np.int64(6), 'TN_5': np.int64(114), 'FP_5': np.int64(14), 'FN_5': np.int64(4)}
+[2025-07-09 20:00:05,972][__main__][INFO] - Inference experiment completed