diff --git a/.mergify.yml b/.mergify.yml
index 5a1e1ff349e0a56e831a4cba8138af04c3033065..b11fd5c1f6613ea50a40dc4329781a797bca38a0 100644
--- a/.mergify.yml
+++ b/.mergify.yml
@@ -39,6 +39,18 @@ pull_request_rules:
     actions:
       label:
         remove: ["conflicts"]
+  - name: "auto add label=enhancement"
+    conditions:
+      - files~=^deepspeech/
+    actions:
+      label:
+        add: ["enhancement"]
+  - name: "auto add label=Example"
+    conditions:
+      - files~=^examples/
+    actions:
+      label:
+        add: ["Example"]
   - name: "auto add label=README"
     conditions:
       - files~=README.md
diff --git a/examples/aishell/s1/conf/conformer.yaml b/examples/aishell/s1/conf/conformer.yaml
index 40b04ed7bdd1afa4c648a3dae3fb360d122f0d8f..10c3a2822a37b30dad57dae58d534ca8638730e9 100644
--- a/examples/aishell/s1/conf/conformer.yaml
+++ b/examples/aishell/s1/conf/conformer.yaml
@@ -74,13 +74,13 @@ model:
 
 
 training:
-  n_epoch: 300
+  n_epoch: 240 
   accum_grad: 2
   global_grad_clip: 5.0
   optim: adam
   optim_conf:
     lr: 0.002
-    weight_decay: 1e-06
+    weight_decay: 1e-6
   scheduler: warmuplr     # pytorch v1.1.0+ required
   scheduler_conf:
     warmup_steps: 25000
@@ -99,7 +99,7 @@ decoding:
   cutoff_prob: 1.0
   cutoff_top_n: 0
   num_proc_bsearch: 8
-  ctc_weight: 0.0 # ctc weight for attention rescoring decode mode.
+  ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
   decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
       # <0: for decoding, use full chunk.
       # >0: for decoding, use fixed chunk size as set.
diff --git a/examples/librispeech/s1/conf/chunk_confermer.yaml b/examples/librispeech/s1/conf/chunk_confermer.yaml
index 3ee31e1b26e73e6dab2f5aced80569ed13a00127..0d07d0a7e82cb156f27b83a5f493eaadad3243af 100644
--- a/examples/librispeech/s1/conf/chunk_confermer.yaml
+++ b/examples/librispeech/s1/conf/chunk_confermer.yaml
@@ -104,7 +104,7 @@ decoding:
   cutoff_prob: 1.0
   cutoff_top_n: 0
   num_proc_bsearch: 8
-  ctc_weight: 0.0 # ctc weight for attention rescoring decode mode.
+  ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
   decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
       # <0: for decoding, use full chunk.
       # >0: for decoding, use fixed chunk size as set.
diff --git a/examples/librispeech/s1/conf/chunk_transformer.yaml b/examples/librispeech/s1/conf/chunk_transformer.yaml
index 265e6e0b674ddbde24d446816cf147035f3e73a0..3939ffc688e1de5dc66606328e48e2d69459b0b6 100644
--- a/examples/librispeech/s1/conf/chunk_transformer.yaml
+++ b/examples/librispeech/s1/conf/chunk_transformer.yaml
@@ -97,7 +97,7 @@ decoding:
   cutoff_prob: 1.0
   cutoff_top_n: 0
   num_proc_bsearch: 8
-  ctc_weight: 0.0 # ctc weight for attention rescoring decode mode.
+  ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
   decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
       # <0: for decoding, use full chunk.
       # >0: for decoding, use fixed chunk size as set.
diff --git a/examples/librispeech/s1/conf/conformer.yaml b/examples/librispeech/s1/conf/conformer.yaml
index 1981b946fd4e0503f83d09e514d894a9d70060b9..f89f82655fa79b2e3a8edc7721382abb94ae9926 100644
--- a/examples/librispeech/s1/conf/conformer.yaml
+++ b/examples/librispeech/s1/conf/conformer.yaml
@@ -8,7 +8,7 @@ data:
   spm_model_prefix: 'data/bpe_unigram_5000'
   mean_std_filepath: ""
   augmentation_config: conf/augmentation.json
-  batch_size: 64
+  batch_size: 16
   min_input_len: 0.5  # seconds
   max_input_len: 20.0 # seconds
   min_output_len: 0.0 # tokens
@@ -76,7 +76,7 @@ model:
 
 training:
   n_epoch: 120
-  accum_grad: 2
+  accum_grad: 8
   global_grad_clip: 5.0
   optim: adam
   optim_conf:
@@ -100,7 +100,7 @@ decoding:
   cutoff_prob: 1.0
   cutoff_top_n: 0
   num_proc_bsearch: 8
-  ctc_weight: 0.0 # ctc weight for attention rescoring decode mode.
+  ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
   decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
       # <0: for decoding, use full chunk.
       # >0: for decoding, use fixed chunk size as set.
diff --git a/examples/librispeech/s1/conf/transformer.yaml b/examples/librispeech/s1/conf/transformer.yaml
index 8ef9e12f1183550507e711625e99ae28acd2f6ea..9014e5b8443b67c4c0b436092ab21c03dd1d0086 100644
--- a/examples/librispeech/s1/conf/transformer.yaml
+++ b/examples/librispeech/s1/conf/transformer.yaml
@@ -95,7 +95,7 @@ decoding:
   cutoff_prob: 1.0
   cutoff_top_n: 0
   num_proc_bsearch: 8
-  ctc_weight: 0.0 # ctc weight for attention rescoring decode mode.
+  ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
   decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
       # <0: for decoding, use full chunk.
       # >0: for decoding, use fixed chunk size as set.
diff --git a/examples/tiny/s1/conf/chunk_confermer.yaml b/examples/tiny/s1/conf/chunk_confermer.yaml
index bd4279e2be5adc71937afb524c5739c689bae074..79006626408823732ba74838ebece5927b6a88f0 100644
--- a/examples/tiny/s1/conf/chunk_confermer.yaml
+++ b/examples/tiny/s1/conf/chunk_confermer.yaml
@@ -104,7 +104,7 @@ decoding:
   cutoff_prob: 1.0
   cutoff_top_n: 0
   num_proc_bsearch: 8
-  ctc_weight: 0.0 # ctc weight for attention rescoring decode mode.
+  ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
   decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
       # <0: for decoding, use full chunk.
       # >0: for decoding, use fixed chunk size as set.
diff --git a/examples/tiny/s1/conf/chunk_transformer.yaml b/examples/tiny/s1/conf/chunk_transformer.yaml
index ba60c273564540a627e7da3e9f4502ed4f2bb28a..aa2b145a681dff821d4695f96be8aef35d674a5e 100644
--- a/examples/tiny/s1/conf/chunk_transformer.yaml
+++ b/examples/tiny/s1/conf/chunk_transformer.yaml
@@ -97,7 +97,7 @@ decoding:
   cutoff_prob: 1.0
   cutoff_top_n: 0
   num_proc_bsearch: 8
-  ctc_weight: 0.0 # ctc weight for attention rescoring decode mode.
+  ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
   decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
       # <0: for decoding, use full chunk.
       # >0: for decoding, use fixed chunk size as set.
diff --git a/examples/tiny/s1/conf/conformer.yaml b/examples/tiny/s1/conf/conformer.yaml
index 83f4f5af46100735f5fc78761c4cc21901fe15e7..3813daa04a516c143d7a545cd28999518fecf2d8 100644
--- a/examples/tiny/s1/conf/conformer.yaml
+++ b/examples/tiny/s1/conf/conformer.yaml
@@ -100,7 +100,7 @@ decoding:
   cutoff_prob: 1.0
   cutoff_top_n: 0
   num_proc_bsearch: 8
-  ctc_weight: 0.0 # ctc weight for attention rescoring decode mode.
+  ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
   decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
       # <0: for decoding, use full chunk.
       # >0: for decoding, use fixed chunk size as set.
diff --git a/examples/tiny/s1/conf/transformer.yaml b/examples/tiny/s1/conf/transformer.yaml
index 3f3170bdfab088a85b7d365834e8f659c42f31bd..35c11731cc3acbc17fe4cf4c410f731b3b384e3c 100644
--- a/examples/tiny/s1/conf/transformer.yaml
+++ b/examples/tiny/s1/conf/transformer.yaml
@@ -95,7 +95,7 @@ decoding:
   cutoff_prob: 1.0
   cutoff_top_n: 0
   num_proc_bsearch: 8
-  ctc_weight: 0.0 # ctc weight for attention rescoring decode mode.
+  ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
   decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
       # <0: for decoding, use full chunk.
       # >0: for decoding, use fixed chunk size as set.