update the model

README.md

@@ -22,16 +22,13 @@ model-index:
       name: Common Voice ar
       args: ar
     metrics:
-      - type: wer    # Required. Example: wer
-        value: 0.18  # Required. Example: 20.90
-        name: Test WER # Optional. Example: Test WER
+      - type: wer
+        value: 0.181
+        name: Test WER
         
-      - type: cer    # Required. Example: wer
-        value: 0.051  # Required. Example: 20.90
-        name: Test CER  # Optional. Example: Test WER
-        
-        WER: 0.18855042016806722
-CER: 0.05138746531806014
+      - type: cer
+        value: 0.049
+        name: Test CER
         
 ---
 <!-- This model card has been generated automatically according to the information the Trainer had access to. You
@@ -41,12 +38,16 @@ should probably proofread and complete it, then remove this comment. -->
 
 # نموذج **صوت سيناء** للتعرف على الأصوات العربية الفصحى و تحويلها إلى نصوص
 
-This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the common_voice 8 dataset.
+This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - AR dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.2141
+- Wer: 0.1808
 
 It achieves the following results on the evaluation set:
-- Loss: 0.22
-- Wer: 0.189
-- Cer:  0.051
+- eval_loss               =     0.2141
+- eval_samples            =      10388
+- eval_wer                =     0.181
+- eval_cer                =     0.049
 
 #### Evaluation Commands
 1. To evaluate on `mozilla-foundation/common_voice_8_0` with split `test`
@@ -95,11 +96,41 @@ The following hyperparameters were used during training:
 - train_batch_size: 32
 - eval_batch_size: 10
 - seed: 42
-- gradient_accumulation_steps: 2
-- total_train_batch_size: 128
+- distributed_type: multi-GPU
+- num_devices: 8
+- total_train_batch_size: 256
+- total_eval_batch_size: 80
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: linear
 - lr_scheduler_warmup_steps: 1000
-- num_epochs: 8.32
+- num_epochs: 10
 - mixed_precision_training: Native AMP
 
+
+### Training results
+
+| Training Loss | Epoch | Step  | Validation Loss | Wer    |
+|:-------------:|:-----:|:-----:|:---------------:|:------:|
+| 1.354         | 0.64  | 1000  | 0.4109          | 0.4493 |
+| 0.5886        | 1.28  | 2000  | 0.2798          | 0.3099 |
+| 0.4977        | 1.92  | 3000  | 0.2387          | 0.2673 |
+| 0.4253        | 2.56  | 4000  | 0.2266          | 0.2523 |
+| 0.3942        | 3.2   | 5000  | 0.2171          | 0.2437 |
+| 0.3619        | 3.84  | 6000  | 0.2076          | 0.2253 |
+| 0.3245        | 4.48  | 7000  | 0.2088          | 0.2186 |
+| 0.308         | 5.12  | 8000  | 0.2086          | 0.2206 |
+| 0.2881        | 5.76  | 9000  | 0.2089          | 0.2105 |
+| 0.2557        | 6.4   | 10000 | 0.2015          | 0.2004 |
+| 0.248         | 7.04  | 11000 | 0.2044          | 0.1953 |
+| 0.2251        | 7.68  | 12000 | 0.2058          | 0.1932 |
+| 0.2052        | 8.32  | 13000 | 0.2117          | 0.1878 |
+| 0.1976        | 8.96  | 14000 | 0.2104          | 0.1825 |
+| 0.1845        | 9.6   | 15000 | 0.2156          | 0.1821 |
+
+
+### Framework versions
+
+- Transformers 4.16.2
+- Pytorch 1.10.2+cu113
+- Datasets 1.18.3
+- Tokenizers 0.11.0

all_results.json

@@ -0,0 +1,14 @@
+{
+    "epoch": 10.0,
+    "eval_loss": 0.21412786841392517,
+    "eval_runtime": 70.9089,
+    "eval_samples": 10388,
+    "eval_samples_per_second": 146.498,
+    "eval_steps_per_second": 1.833,
+    "eval_wer": 0.18078979457836977,
+    "train_loss": 0.1316310991176183,
+    "train_runtime": 23113.6031,
+    "train_samples": 399991,
+    "train_samples_per_second": 173.054,
+    "train_steps_per_second": 0.676
+}

config.json

@@ -6,7 +6,7 @@
   "add_adapter": false,
   "apply_spec_augment": true,
   "architectures": [
-    "Wav2Vec2ForPreTraining"
+    "Wav2Vec2ForCTC"
   ],
   "attention_dropout": 0.0,
   "bos_token_id": 1,

eval_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 10.0,
+    "eval_loss": 0.21412786841392517,
+    "eval_runtime": 70.9089,
+    "eval_samples": 10388,
+    "eval_samples_per_second": 146.498,
+    "eval_steps_per_second": 1.833,
+    "eval_wer": 0.18078979457836977
+}

mozilla-foundation_common_voice_8_0_ar_test_eval_results.txt

@@ -1,2 +1,2 @@
-WER: 0.18855042016806722
-CER: 0.05138746531806014
+WER: 0.18172268907563024
+CER: 0.04875182561226061

optimizer.pt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:a2720401989ca4dbdc57ad7199cdc4116d02c36c374567e9a1295b01cd6c45b3
-size 625461417

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3299e75237418a55c403fc9c78801934d3057ff076e484879bc40bfd516f13ba
+oid sha256:588e6341d51008b353be1115b1e1e34d86bad4f676b32277cba57e5f7cff526a
 size 1262112241

rng_state_0.pth

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:39223200ea0fd8350b41518c6492d3d14d96d823ca3f6e9a374864389e6dbb1b
-size 14503

rng_state_1.pth

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:5091f67e4a1caff5227497b7c37712739cd8baf0783971573e46569913fdfff3
-size 14567

rng_state_2.pth

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:16de60f262d42506479df2d553aa073b64cf621ae9eebfc35fbc2b23021cee2d
-size 14503

rng_state_3.pth

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:a641931222bcf2b19d6a073bd70c446b59abafb3a4077c2d6a5aea1f4001e06a
-size 14503

rng_state_4.pth

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:da3f79cf3a4b4d9379e47a4353e6555235b22515af826d448849df6d7a5f04f0
-size 14567

rng_state_5.pth

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:16352761baa415d3af2ff3cea0ef555419a76b51449f27bcec56bcdba9a15ff5
-size 14567

rng_state_6.pth

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:8ea93d555065f78f93a91b315329c1c392289d935801de6b732e124d18b1586a
-size 14567

rng_state_7.pth

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:46629c3fa6cb14bd2274cb93b2b9a613c84794d3912a7fb43cf8af9b51524544
-size 14503

scaler.pt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:431319fd21daa87636d8253400f763a0a1a5400306ce1db9e67b38942c76551d
-size 559

scheduler.pt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:6f90f0da980bf8cd346163915335c996f7bbdbcfe93a2e624a8b02f902aa6d01
-size 623

train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 10.0,
+    "train_loss": 0.1316310991176183,
+    "train_runtime": 23113.6031,
+    "train_samples": 399991,
+    "train_samples_per_second": 173.054,
+    "train_steps_per_second": 0.676
+}

trainer_state.json

@@ -1,8 +1,8 @@
 {
-  "best_metric": 0.18776850201669637,
-  "best_model_checkpoint": "/workspace/cv-corpus-8.0-2022-01-19/output/checkpoint-13000",
-  "epoch": 8.317338451695457,
-  "global_step": 13000,
+  "best_metric": 0.4493387111903199,
+  "best_model_checkpoint": "/workspace/cv-corpus-8.0-2022-01-19/output/checkpoint-1000",
+  "epoch": 10.0,
+  "global_step": 15630,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -279,11 +279,68 @@
       "eval_steps_per_second": 1.836,
       "eval_wer": 0.18776850201669637,
       "step": 13000
+    },
+    {
+      "epoch": 8.64,
+      "learning_rate": 2.9241285030758714e-05,
+      "loss": 0.2026,
+      "step": 13500
+    },
+    {
+      "epoch": 8.96,
+      "learning_rate": 2.2406015037593985e-05,
+      "loss": 0.1976,
+      "step": 14000
+    },
+    {
+      "epoch": 8.96,
+      "eval_loss": 0.21043309569358826,
+      "eval_runtime": 71.1895,
+      "eval_samples_per_second": 145.92,
+      "eval_steps_per_second": 1.826,
+      "eval_wer": 0.18249695150548728,
+      "step": 14000
+    },
+    {
+      "epoch": 9.28,
+      "learning_rate": 1.5570745044429256e-05,
+      "loss": 0.1875,
+      "step": 14500
+    },
+    {
+      "epoch": 9.6,
+      "learning_rate": 8.735475051264526e-06,
+      "loss": 0.1845,
+      "step": 15000
+    },
+    {
+      "epoch": 9.6,
+      "eval_loss": 0.21563756465911865,
+      "eval_runtime": 71.0722,
+      "eval_samples_per_second": 146.161,
+      "eval_steps_per_second": 1.829,
+      "eval_wer": 0.18212175218084609,
+      "step": 15000
+    },
+    {
+      "epoch": 9.92,
+      "learning_rate": 1.9138755980861244e-06,
+      "loss": 0.1837,
+      "step": 15500
+    },
+    {
+      "epoch": 10.0,
+      "step": 15630,
+      "total_flos": 9.942412569719006e+20,
+      "train_loss": 0.1316310991176183,
+      "train_runtime": 23113.6031,
+      "train_samples_per_second": 173.054,
+      "train_steps_per_second": 0.676
     }
   ],
   "max_steps": 15630,
   "num_train_epochs": 10,
-  "total_flos": 8.268631785694088e+20,
+  "total_flos": 9.942412569719006e+20,
   "trial_name": null,
   "trial_params": null
 }