kingkenche commited on
Commit
9ef76aa
·
verified ·
1 Parent(s): 21f3837

Upload run_q4_output.log with huggingface_hub

Browse files
Files changed (1) hide show
  1. run_q4_output.log +55 -0
run_q4_output.log ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ /home/m25csa028/miniconda3/lib/python3.13/site-packages/torch/cuda/__init__.py:180: UserWarning: CUDA initialization: The NVIDIA driver on your system is too old (found version 12040). Please update your GPU driver by downloading and installing a new version from the URL: http://www.nvidia.com/Download/index.aspx Alternatively, go to: https://pytorch.org to install a PyTorch version that has been compiled with your version of the CUDA driver. (Triggered internally at /pytorch/c10/cuda/CUDAFunctions.cpp:119.)
2
+ return torch._C._cuda_getDeviceCount() > 0
3
+ ============================================================
4
+ Q4: ECAPA-TDNN Speaker Verification — Full Pipeline
5
+ ============================================================
6
+
7
+ [TASK 1] Loading pre-trained ECAPA-TDNN model...
8
+ Baseline GFLOPs (FP32): 11.3189
9
+
10
+ Loading SUPERB SI test split (streaming)...
11
+ Evaluating Baseline Accuracy (cosine-sim nearest neighbour)...
12
+ Baseline Top-1 Identification Accuracy: 100.00%
13
+
14
+ [TASK 2] Applying Post-Training Quantization (INT8 Dynamic)...
15
+ PTQ (INT8) Effective GFLOPs: 2.8297
16
+ GFLOPs Reduction: 8.4892 (75.0%)
17
+
18
+ [TASK 3] Evaluating PTQ model on test set...
19
+ PTQ Accuracy: 100.00%
20
+ Accuracy Change vs Baseline: +0.00% (maintained/increased)
21
+
22
+ [TASK 4] Running Optuna QAT hyperparameter search (4 trials)...
23
+ Trial 0: lr=1.33e-04, wd=7.11e-04, bs=8 → Adapter Acc=2.50%
24
+ Trial 1: lr=2.94e-05, wd=1.49e-06, bs=8 → Adapter Acc=0.42%
25
+ Trial 2: lr=1.15e-05, wd=8.12e-04, bs=8 → Adapter Acc=0.42%
26
+ Trial 3: lr=3.55e-05, wd=8.18e-06, bs=8 → Adapter Acc=0.00%
27
+
28
+ Best Hyperparameters: {'lr': 0.0001329291894316216, 'weight_decay': 0.0007114476009343421, 'batch_size': 8}
29
+ Best QAT Adapter Accuracy (val-trained): 2.50%
30
+ QAT Inference GFLOPs: 2.8297
31
+
32
+ [TASK 5] Final Trade-off Analysis...
33
+ Baseline Accuracy: 100.00%
34
+ Final QAT Model Accuracy: 100.00%
35
+ Total Perf Difference: +0.00% (improvement)
36
+ GFLOPs Saved permanently: 8.4892 GFLOPs
37
+
38
+ Results saved to q4_results.json
39
+
40
+ ============================================================
41
+ ANSWERS FOR SUBMISSION BLANKS:
42
+ ============================================================
43
+ Task 1: Baseline Accuracy = 100.00%
44
+ Task 1: Baseline GFLOPs = 11.3189 GFLOPs
45
+ Task 2: PTQ (INT8) GFLOPs = 2.8297 GFLOPs
46
+ Task 2: GFLOPs Impact = 8.4892 GFLOPs reduction (75.0%)
47
+ Task 3: PTQ Accuracy = 100.00% (Accuracy maintained/increased)
48
+ Task 4: Best Hyperparams = lr=1.33e-04, weight_decay=7.11e-04, batch_size=8
49
+ Task 4: Best QAT Accuracy = 2.50%
50
+ Task 4: QAT GFLOPs = 2.8297 GFLOPs
51
+ Task 5: Total Perf Diff = 0.00% absolute Accuracy difference
52
+ Task 5: GFLOPs Saved = 8.4892 GFLOPs permanently saved
53
+ ============================================================
54
+ '[Errno 9] Bad file descriptor' thrown while requesting GET https://huggingface.co/datasets/s3prl/superb/resolve/4b5ccb1d4776ce2af83f86a175137663835f0b1c/si/validation-00000-of-00001.parquet
55
+ Retrying in 1s [Retry 1/5].