Uploading results from cluster

.gitattributes

@@ -33,3 +33,9 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+eval/asvspoof19_eval/eval_test_epoch_best.json filter=lfs diff=lfs merge=lfs -text
+eval/asvspoof19_eval/eval_test_epoch_best_incremental.jsonl filter=lfs diff=lfs merge=lfs -text
+eval/itw/eval_test_epoch_best.json filter=lfs diff=lfs merge=lfs -text
+eval/itw/eval_test_epoch_best_incremental.jsonl filter=lfs diff=lfs merge=lfs -text
+eval/mlaad_en/eval_test_epoch_best.json filter=lfs diff=lfs merge=lfs -text
+eval/mlaad_en/eval_test_epoch_best_incremental.jsonl filter=lfs diff=lfs merge=lfs -text

eval/asvspoof19_eval/config.json

@@ -0,0 +1,39 @@
+{
+    "model": {
+        "llm_path": "/netscratch/fkallel/models/Qwen3.5-4B/",
+        "llm_type": "qwen3.5",
+        "audio_encoder_path": "/ds-slt/audio_weights/xlsr2_300m.pt",
+        "audio_encoder_type": "wav2vec2",
+        "audio_encoder_source": "fairseq",
+        "freeze_audio_encoder": false,
+        "lora": false,
+        "lora_rank": 64,
+        "lora_alpha": 16,
+        "lora_dropout": 0.05,
+        "special_tokens": []
+    },
+    "data": {
+        "train_path": [
+            "data/my_hir_sdd_binary_train_unique_2.json",
+            "data/asvspoof19_train.json"
+        ],
+        "val_path": [
+            "data/my_hir_sdd_binary_val_unique_2.json",
+            "data/asvspoof19_dev.json"
+        ],
+        "test_path": "data/my_hir_sdd_binary_test_unique_2.json",
+        "prompt_path": "prompts/train_prompt.json"
+    },
+    "train": {
+        "batch_size": 4,
+        "epochs": 5,
+        "lr": "1e-4",
+        "lr_adapter": "1e-5",
+        "lr_audio": "1e-6",
+        "lr_lora": "1e-5",
+        "num_workers": 4,
+        "seed": 240,
+        "grad_accum_steps": 1
+    },
+    "output_dir": "output_v6/qwen3.5_4B_no_lora_240_myhir_asvspoof19"
+}

eval/asvspoof19_eval/eval_test_epoch_best.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c66c86b0b7ec81f491f166ab2bd18cc899a2e47c5e9047bc013b9d481441bb96
+size 35761008

eval/asvspoof19_eval/eval_test_epoch_best_incremental.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:04beaf60ee22fd1bb4a859625263c498259ec86bc5d264c894706d0e6b9ef3f0
+size 32484104

eval/asvspoof19_eval/test.log

@@ -0,0 +1,44 @@
+08/04/2026 18:32:03 - INFO - Device: cuda
+08/04/2026 18:32:03 - INFO - Loading qwen3.5 Tokenizer & Model from /netscratch/fkallel/models/Qwen3.5-4B/
+08/04/2026 18:32:04 - INFO - Creating wav2vec2 encoder from /ds-slt/audio_weights/xlsr2_300m.pt
+08/04/2026 18:32:05 - INFO - Please install tensorboardX: pip install tensorboardX
+08/04/2026 18:32:13 - INFO - Loading full checkpoint from output_v6/qwen3.5_4B_no_lora_240/checkpoint_best.pt
+08/04/2026 18:56:45 - INFO - Device: cuda
+08/04/2026 18:56:45 - INFO - Loading qwen3.5 Tokenizer & Model from /netscratch/fkallel/models/Qwen3.5-4B/
+08/04/2026 18:56:46 - INFO - Creating wav2vec2 encoder from /ds-slt/audio_weights/xlsr2_300m.pt
+08/04/2026 18:56:47 - INFO - Please install tensorboardX: pip install tensorboardX
+08/04/2026 18:56:54 - INFO - Loading full checkpoint from output_v6/qwen3.5_4B_no_lora_240/checkpoint_best.pt
+08/04/2026 20:46:10 - INFO - Evaluation metrics: {
+  "loss": 0.2941332906781157,
+  "accuracy": 0.8955318163314008,
+  "balanced_accuracy": 0.5079210839342618,
+  "precision": 0.38337801608579086,
+  "recall": 0.019442556084296396,
+  "f1": 0.03700828157349896,
+  "precision_negative": 0.8982275908783021,
+  "recall_negative": 0.9963996117842272,
+  "f1_negative": 0.944770160153177,
+  "macro_f1": 0.490889220863338,
+  "tp": 143,
+  "tn": 63652,
+  "fp": 230,
+  "fn": 7212,
+  "total_samples": 71237
+}
+08/04/2026 21:15:23 - INFO - Evaluation metrics: {
+  "loss": 0.29413409428935977,
+  "accuracy": 0.8955177786824262,
+  "balanced_accuracy": 0.507853102968932,
+  "precision": 0.3817204301075269,
+  "recall": 0.01930659415363698,
+  "f1": 0.03675423838488417,
+  "precision_negative": 0.8982149156847526,
+  "recall_negative": 0.9963996117842272,
+  "f1_negative": 0.9447631487157413,
+  "macro_f1": 0.4907586935503127,
+  "tp": 142,
+  "tn": 63652,
+  "fp": 230,
+  "fn": 7213,
+  "total_samples": 71237
+}

eval/config.json

@@ -0,0 +1,33 @@
+{
+    "model": {
+        "llm_path": "/netscratch/fkallel/models/Qwen3.5-4B/",
+        "llm_type": "qwen3.5",
+        "audio_encoder_path": "/ds-slt/audio_weights/xlsr2_300m.pt",
+        "audio_encoder_type": "wav2vec2",
+        "audio_encoder_source": "fairseq",
+        "freeze_audio_encoder": false,
+        "lora": false,
+        "lora_rank": 64,
+        "lora_alpha": 16,
+        "lora_dropout": 0.05,
+        "special_tokens": []
+    },
+    "data": {
+        "train_path": "data/my_hir_sdd_binary_train_unique_2.json",
+        "val_path": "data/my_hir_sdd_binary_val_unique_2.json",
+        "test_path": "data/my_hir_sdd_binary_test_unique_2.json",
+        "prompt_path": "prompts/train_prompt.json"
+    },
+    "train": {
+        "batch_size": 4,
+        "epochs": 5,
+        "lr": "1e-4",
+        "lr_adapter": "1e-5",
+        "lr_audio": "1e-6",
+        "lr_lora": "1e-5",
+        "num_workers": 4,
+        "seed": 240,
+        "grad_accum_steps": 1
+    },
+    "output_dir": "output_v6/qwen3.5_4B_no_lora_240"
+}

eval/itw/config.json

@@ -0,0 +1,33 @@
+{
+    "model": {
+        "llm_path": "/netscratch/fkallel/models/Qwen3.5-4B/",
+        "llm_type": "qwen3.5",
+        "audio_encoder_path": "/ds-slt/audio_weights/xlsr2_300m.pt",
+        "audio_encoder_type": "wav2vec2",
+        "audio_encoder_source": "fairseq",
+        "freeze_audio_encoder": false,
+        "lora": false,
+        "lora_rank": 64,
+        "lora_alpha": 16,
+        "lora_dropout": 0.05,
+        "special_tokens": []
+    },
+    "data": {
+        "train_path": "data/asvspoof19_train.json",
+        "val_path": "data/asvspoof19_dev.json",
+        "test_path": "data/my_hir_sdd_binary_test_unique_2.json",
+        "prompt_path": "prompts/train_prompt.json"
+    },
+    "train": {
+        "batch_size": 4,
+        "epochs": 5,
+        "lr": "1e-4",
+        "lr_adapter": "1e-5",
+        "lr_audio": "1e-6",
+        "lr_lora": "1e-5",
+        "num_workers": 4,
+        "seed": 240,
+        "grad_accum_steps": 1
+    },
+    "output_dir": "output_v6/qwen3.5_4B_no_lora_240_myhir_asvspoof19"
+}

eval/itw/eval_test_epoch_best.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aec1823c4f3e13db9699406424dddbc54ad58ed18d862e994ba4c9910be8cacc
+size 14469006

eval/itw/eval_test_epoch_best_incremental.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7eaffb67cffda15d31b1df8eb91e44b369ac459482b14e08f608b0cf47475a4
+size 13007170

eval/itw/test.log

@@ -0,0 +1,44 @@
+08/04/2026 17:27:26 - INFO - Device: cuda
+08/04/2026 17:27:26 - INFO - Loading qwen3.5 Tokenizer & Model from /netscratch/fkallel/models/Qwen3.5-4B/
+08/04/2026 17:27:27 - INFO - Creating wav2vec2 encoder from /ds-slt/audio_weights/xlsr2_300m.pt
+08/04/2026 17:27:28 - INFO - Please install tensorboardX: pip install tensorboardX
+08/04/2026 17:27:35 - INFO - Loading full checkpoint from output_v6/qwen3.5_4B_no_lora_240/checkpoint_best.pt
+08/04/2026 17:49:25 - INFO - Device: cuda
+08/04/2026 17:49:25 - INFO - Loading qwen3.5 Tokenizer & Model from /netscratch/fkallel/models/Qwen3.5-4B/
+08/04/2026 17:49:26 - INFO - Creating wav2vec2 encoder from /ds-slt/audio_weights/xlsr2_300m.pt
+08/04/2026 17:49:27 - INFO - Please install tensorboardX: pip install tensorboardX
+08/04/2026 17:49:35 - INFO - Loading full checkpoint from output_v6/qwen3.5_4B_no_lora_240/checkpoint_best.pt
+08/04/2026 18:31:58 - INFO - Evaluation metrics: {
+  "loss": 0.9986640888436654,
+  "accuracy": 0.5610938040844583,
+  "balanced_accuracy": 0.643866926071187,
+  "precision": 0.9422143802382003,
+  "recall": 0.3209938386014126,
+  "f1": 0.4788521895082948,
+  "precision_negative": 0.45732244375050046,
+  "recall_negative": 0.9667400135409614,
+  "f1_negative": 0.6209164537696364,
+  "macro_f1": 0.5498843216389656,
+  "tp": 6408,
+  "tn": 11423,
+  "fp": 393,
+  "fn": 13555,
+  "total_samples": 31779
+}
+08/04/2026 18:56:40 - INFO - Evaluation metrics: {
+  "loss": 0.9986674823490951,
+  "accuracy": 0.5612196733692061,
+  "balanced_accuracy": 0.6439671114140714,
+  "precision": 0.9422483468038207,
+  "recall": 0.3211942092871813,
+  "f1": 0.4790794979079498,
+  "precision_negative": 0.45739569151917997,
+  "recall_negative": 0.9667400135409614,
+  "f1_negative": 0.6209839630334331,
+  "macro_f1": 0.5500317304706914,
+  "tp": 6412,
+  "tn": 11423,
+  "fp": 393,
+  "fn": 13551,
+  "total_samples": 31779
+}

eval/mlaad_en/config.json

@@ -0,0 +1,39 @@
+{
+    "model": {
+        "llm_path": "/netscratch/fkallel/models/Qwen3.5-4B/",
+        "llm_type": "qwen3.5",
+        "audio_encoder_path": "/ds-slt/audio_weights/xlsr2_300m.pt",
+        "audio_encoder_type": "wav2vec2",
+        "audio_encoder_source": "fairseq",
+        "freeze_audio_encoder": false,
+        "lora": false,
+        "lora_rank": 64,
+        "lora_alpha": 16,
+        "lora_dropout": 0.05,
+        "special_tokens": []
+    },
+    "data": {
+        "train_path": [
+            "data/my_hir_sdd_binary_train_unique_2.json",
+            "data/asvspoof19_train.json"
+        ],
+        "val_path": [
+            "data/my_hir_sdd_binary_val_unique_2.json",
+            "data/asvspoof19_dev.json"
+        ],
+        "test_path": "data/my_hir_sdd_binary_test_unique_2.json",
+        "prompt_path": "prompts/train_prompt.json"
+    },
+    "train": {
+        "batch_size": 4,
+        "epochs": 5,
+        "lr": "1e-4",
+        "lr_adapter": "1e-5",
+        "lr_audio": "1e-6",
+        "lr_lora": "1e-5",
+        "num_workers": 4,
+        "seed": 240,
+        "grad_accum_steps": 1
+    },
+    "output_dir": "output_v6/qwen3.5_4B_no_lora_240_myhir_asvspoof19"
+}

eval/mlaad_en/eval_test_epoch_best.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0dd4513f140ab7f1534ef36c366fa851a098eb63eb9eb2e42122bebef324e81
+size 22549888

eval/mlaad_en/eval_test_epoch_best_incremental.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd2fa97a0629f5f9e6895ffb03ea98e648e19780b3eab695f45dedd22845b0e0
+size 20801886

eval/mlaad_en/test.log

@@ -0,0 +1,44 @@
+08/04/2026 20:46:15 - INFO - Device: cuda
+08/04/2026 20:46:15 - INFO - Loading qwen3.5 Tokenizer & Model from /netscratch/fkallel/models/Qwen3.5-4B/
+08/04/2026 20:46:16 - INFO - Creating wav2vec2 encoder from /ds-slt/audio_weights/xlsr2_300m.pt
+08/04/2026 20:46:17 - INFO - Please install tensorboardX: pip install tensorboardX
+08/04/2026 20:46:25 - INFO - Loading full checkpoint from output_v6/qwen3.5_4B_no_lora_240/checkpoint_best.pt
+08/04/2026 21:15:28 - INFO - Device: cuda
+08/04/2026 21:15:28 - INFO - Loading qwen3.5 Tokenizer & Model from /netscratch/fkallel/models/Qwen3.5-4B/
+08/04/2026 21:15:29 - INFO - Creating wav2vec2 encoder from /ds-slt/audio_weights/xlsr2_300m.pt
+08/04/2026 21:15:30 - INFO - Please install tensorboardX: pip install tensorboardX
+08/04/2026 21:15:38 - INFO - Loading full checkpoint from output_v6/qwen3.5_4B_no_lora_240/checkpoint_best.pt
+08/04/2026 22:05:43 - INFO - Evaluation metrics: {
+  "loss": 0.01016120363089086,
+  "accuracy": 0.9944736842105263,
+  "balanced_accuracy": 0.9944736842105263,
+  "precision": 0.9969321908388872,
+  "recall": 0.992,
+  "f1": 0.9944599799504037,
+  "precision_negative": 0.9920393840997171,
+  "recall_negative": 0.9969473684210526,
+  "f1_negative": 0.9944873208379271,
+  "macro_f1": 0.9944736503941654,
+  "tp": 18848,
+  "tn": 18942,
+  "fp": 58,
+  "fn": 152,
+  "total_samples": 38000
+}
+08/04/2026 22:39:48 - INFO - Evaluation metrics: {
+  "loss": 0.01016135814392328,
+  "accuracy": 0.9944473684210526,
+  "balanced_accuracy": 0.9944473684210526,
+  "precision": 0.996932028563872,
+  "recall": 0.9919473684210526,
+  "f1": 0.9944334520511805,
+  "precision_negative": 0.991987431264729,
+  "recall_negative": 0.9969473684210526,
+  "f1_negative": 0.9944612153825961,
+  "macro_f1": 0.9944473337168883,
+  "tp": 18847,
+  "tn": 18942,
+  "fp": 58,
+  "fn": 153,
+  "total_samples": 38000
+}

eval/test.log

@@ -0,0 +1,19 @@
+08/04/2026 15:53:35 - INFO - Device: cuda
+08/04/2026 15:53:35 - INFO - Loading qwen3.5 Tokenizer & Model from /netscratch/fkallel/models/Qwen3.5-4B/
+08/04/2026 15:53:36 - INFO - Creating wav2vec2 encoder from /ds-slt/audio_weights/xlsr2_300m.pt
+08/04/2026 15:53:37 - INFO - Please install tensorboardX: pip install tensorboardX
+08/04/2026 15:53:44 - INFO - Loading full checkpoint from output_v6/qwen3.5_4B_no_lora_240/checkpoint_best.pt
+08/04/2026 16:13:17 - INFO - Evaluation metrics: {
+  "loss": 0.1341296887972373,
+  "accuracy": 0.9518874907475944,
+  "balanced_accuracy": 0.9531213155094851,
+  "precision": 0.9231570179092045,
+  "recall": 0.9809692409825183,
+  "f1": 0.9511854951185496,
+  "macro_f1" : 0.951877539,
+  "tp": 4433,
+  "tn": 4569,
+  "fp": 369,
+  "fn": 86,
+  "total_samples": 9457
+}