Upload fine-tuned model - BLEU: 0.1636

.gitattributes

@@ -47,3 +47,5 @@ checkpoint-96/source.spm filter=lfs diff=lfs merge=lfs -text
 checkpoint-96/target.spm filter=lfs diff=lfs merge=lfs -text
 checkpoint-80/source.spm filter=lfs diff=lfs merge=lfs -text
 checkpoint-80/target.spm filter=lfs diff=lfs merge=lfs -text
+checkpoint-144/source.spm filter=lfs diff=lfs merge=lfs -text
+checkpoint-144/target.spm filter=lfs diff=lfs merge=lfs -text

README.md

@@ -40,13 +40,13 @@ The model was fine-tuned on a combination of:
 ## Performance
 
 ### Test Set (General Translation)
-- **BLEU:** 0.1720
-- **chrF:** 36.54
+- **BLEU:** 0.1636
+- **chrF:** 36.08
 - **Improvement over baseline:** +0.0%
 
 ### Domain Evaluation (Call Transcriptions)
 - **Domain BLEU:** 0.0000
-- **Domain chrF:** 1.88
+- **Domain chrF:** 2.12
 - **Domain COMET-QE:** 0.0000
 
 *Domain metrics are evaluated on real 10-minute call transcriptions from child helplines.*
@@ -88,7 +88,7 @@ print(translation)
 
 **Base Model:** Helsinki-NLP/opus-mt-mul-en  
 **Training Epochs:** 8  
-**Batch Size:** 330 (effective: 330)  
+**Batch Size:** 350 (effective: 350)  
 **Learning Rate:** 3e-05  
 **Optimizer:** AdamW  
 **Hardware:** NVIDIA GPU with FP16 mixed precision

baseline_results.json

@@ -1,11 +1,11 @@
 {
-  "eval_loss": 18.69961166381836,
-  "eval_model_preparation_time": 0.0025,
-  "eval_bleu": 0.03199810773966101,
-  "eval_chrf": 20.095068987858355,
-  "eval_runtime": 17.9508,
-  "eval_samples_per_second": 20.055,
-  "eval_steps_per_second": 0.111,
+  "eval_loss": 18.048294067382812,
+  "eval_model_preparation_time": 0.0023,
+  "eval_bleu": 0.03175826534675532,
+  "eval_chrf": 20.001735154905155,
+  "eval_runtime": 18.5174,
+  "eval_samples_per_second": 19.441,
+  "eval_steps_per_second": 0.108,
   "eval_comet": 0.0341746505332392,
   "eval_comet_std": 0.1127765594779826,
   "baseline_domain_bleu": 0.0,

checkpoint-144/config.json

@@ -0,0 +1,54 @@
+{
+  "activation_dropout": 0.0,
+  "activation_function": "swish",
+  "add_bias_logits": false,
+  "add_final_layer_norm": false,
+  "architectures": [
+    "MarianMTModel"
+  ],
+  "attention_dropout": 0.0,
+  "classif_dropout": 0.0,
+  "classifier_dropout": 0.0,
+  "d_model": 512,
+  "decoder_attention_heads": 8,
+  "decoder_ffn_dim": 2048,
+  "decoder_layerdrop": 0.0,
+  "decoder_layers": 6,
+  "decoder_start_token_id": 64171,
+  "decoder_vocab_size": 64172,
+  "dropout": 0.1,
+  "dtype": "float32",
+  "encoder_attention_heads": 8,
+  "encoder_ffn_dim": 2048,
+  "encoder_layerdrop": 0.0,
+  "encoder_layers": 6,
+  "eos_token_id": 0,
+  "extra_pos_embeddings": 64172,
+  "forced_eos_token_id": 0,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1",
+    "2": "LABEL_2"
+  },
+  "init_std": 0.02,
+  "is_encoder_decoder": true,
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1,
+    "LABEL_2": 2
+  },
+  "max_length": null,
+  "max_position_embeddings": 512,
+  "model_type": "marian",
+  "normalize_before": false,
+  "normalize_embedding": false,
+  "num_beams": null,
+  "num_hidden_layers": 6,
+  "pad_token_id": 64171,
+  "scale_embedding": true,
+  "share_encoder_decoder_embeddings": true,
+  "static_position_embeddings": true,
+  "transformers_version": "4.57.1",
+  "use_cache": true,
+  "vocab_size": 64172
+}

checkpoint-144/generation_config.json

@@ -0,0 +1,17 @@
+{
+  "bad_words_ids": [
+    [
+      64171
+    ]
+  ],
+  "decoder_start_token_id": 64171,
+  "eos_token_id": [
+    0
+  ],
+  "forced_eos_token_id": 0,
+  "max_length": 512,
+  "num_beams": 6,
+  "pad_token_id": 64171,
+  "renormalize_logits": true,
+  "transformers_version": "4.57.1"
+}

checkpoint-144/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e806cfcebe09031bb3997449d2294bc58d022486c0368850759b3bedf8cb904
+size 308263984

checkpoint-144/optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:29f78c47b18d5f95c267251341d9e5e8f56245d49d89db5f5677c796e8f78f32
+size 616171979

checkpoint-144/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66fb29b7a0256e5601dba02afcd343a88eeca85220cd4be6fe8007ff095049e7
+size 14645

checkpoint-144/scaler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f71bbc2be28fb2466cee77d17f19aa4d2f70898f072b16a7dedd850e3fb3f602
+size 1383

checkpoint-144/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:730b4f4d9246f798fbfc76673f473edf320c4b67bfb18d4d6e8aecae0caac18f
+size 1465

checkpoint-144/source.spm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c4a99ea3602b29fbf901ade8b93a45efa3d7c64eab8fc5fa812383efa327a87d
+size 706917

checkpoint-144/special_tokens_map.json

@@ -0,0 +1,5 @@
+{
+  "eos_token": "</s>",
+  "pad_token": "<pad>",
+  "unk_token": "<unk>"
+}

checkpoint-144/target.spm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6dce5fa58fcd7dde9e81e279b8c075bf42ee558278f73d6fb48e342029d7f19
+size 791194

checkpoint-144/tokenizer_config.json

@@ -0,0 +1,39 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "64171": {
+      "content": "<pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "extra_special_tokens": {},
+  "model_max_length": 512,
+  "pad_token": "<pad>",
+  "separate_vocabs": false,
+  "source_lang": "mul",
+  "sp_model_kwargs": {},
+  "target_lang": "eng",
+  "tokenizer_class": "MarianTokenizer",
+  "unk_token": "<unk>"
+}

checkpoint-144/trainer_state.json

@@ -0,0 +1,57 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.0,
+  "eval_steps": 1000,
+  "global_step": 144,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 2.7777777777777777,
+      "grad_norm": 2.7723793983459473,
+      "learning_rate": 2.5145069915713543e-05,
+      "loss": 6.7383,
+      "step": 50
+    },
+    {
+      "epoch": 5.555555555555555,
+      "grad_norm": 2.405052900314331,
+      "learning_rate": 8.141365150430423e-06,
+      "loss": 3.4434,
+      "step": 100
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 144,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 8,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "EarlyStoppingCallback": {
+      "args": {
+        "early_stopping_patience": 5,
+        "early_stopping_threshold": 0.001
+      },
+      "attributes": {
+        "early_stopping_patience_counter": 0
+      }
+    },
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1138408789377024.0,
+  "train_batch_size": 350,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-144/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dfd52f05cccd88a8ac6ef07201e53ca6487918fde26bf4b76757ae607de93a43
+size 5969

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d90c9b028acf77d356fb498689d67e8dd01efda8fa45b4a52202dbce37f7951
+oid sha256:4e806cfcebe09031bb3997449d2294bc58d022486c0368850759b3bedf8cb904
 size 308263984

test_results.json

@@ -1,16 +1,16 @@
 {
-  "eval_loss": 3.238176107406616,
-  "eval_model_preparation_time": 0.0025,
-  "eval_bleu": 0.17195884535965744,
-  "eval_chrf": 36.53906898689781,
-  "eval_runtime": 3.9316,
-  "eval_samples_per_second": 91.565,
-  "eval_steps_per_second": 0.509,
+  "eval_loss": 3.188666343688965,
+  "eval_model_preparation_time": 0.0023,
+  "eval_bleu": 0.1636321929455223,
+  "eval_chrf": 36.081372653845165,
+  "eval_runtime": 4.4962,
+  "eval_samples_per_second": 80.068,
+  "eval_steps_per_second": 0.445,
   "epoch": 8.0,
-  "eval_comet": 0.1170399222568853,
-  "eval_comet_std": 0.21053217439673014,
-  "domain_bleu": 5.3233561842199044e-17,
-  "domain_chrf": 1.8808933992523351,
-  "domain_comet": 4.171395121375099e-05,
-  "domain_comet_std": 3.3738691858045937e-07
+  "eval_comet": 0.11297648821764597,
+  "eval_comet_std": 0.2079275907563645,
+  "domain_bleu": 7.830089906681801e-16,
+  "domain_chrf": 2.119572055418669,
+  "domain_comet": 4.181272961432114e-05,
+  "domain_comet_std": 8.272744406669947e-08
 }

train_results.json

@@ -1,8 +1,8 @@
 {
-  "train_runtime": 61.5297,
-  "train_samples_per_second": 793.893,
-  "train_steps_per_second": 2.47,
+  "train_runtime": 63.5743,
+  "train_samples_per_second": 768.361,
+  "train_steps_per_second": 2.265,
   "total_flos": 1138408789377024.0,
-  "train_loss": 4.468797128451498,
+  "train_loss": 4.526931868659125,
   "epoch": 8.0
 }

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eaf461ec2530d2bc87154a757b8eee105da2d899e5550831a3cb068345247e5a
+oid sha256:dfd52f05cccd88a8ac6ef07201e53ca6487918fde26bf4b76757ae607de93a43
 size 5969

translation_config.json

@@ -30,7 +30,7 @@
   },
   "training_config": {
     "learning_rate": 3e-05,
-    "batch_size": 330,
+    "batch_size": 350,
     "gradient_accumulation_steps": 1,
     "num_epochs": 8,
     "max_length": 300,