From 1b18454687409cdabfec4cd9a197f88fa996ae3c Mon Sep 17 00:00:00 2001
From: jmarin <javier.marin@satellogic.com>
Date: Fri, 23 Feb 2024 13:10:11 +0100
Subject: [PATCH 1/6] Clarify dynamic quantization behaviour

---
 recipes_source/quantization.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/recipes_source/quantization.rst b/recipes_source/quantization.rst
index a8719c19813..012e8c8b360 100644
--- a/recipes_source/quantization.rst
+++ b/recipes_source/quantization.rst
@@ -81,7 +81,7 @@ The full documentation of the `quantize_dynamic` API call is `here <https://pyto
 3. Post Training Static Quantization
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-This method converts both the weights and the activations to 8-bit integers beforehand so there won't be on-the-fly conversion on the activations during the inference, as the dynamic quantization does, hence improving the performance significantly.
+This method converts both the weights and the activations to 8-bit integers beforehand so there won’t be on-the-fly conversion on the activations during the inference, as the dynamic quantization does. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's performance.
 
 To apply static quantization on a model, run the following code:
 

From 2390287ce3e29b08d15d2d8658ce83e638ef37c4 Mon Sep 17 00:00:00 2001
From: jmarin <javier.marin@satellogic.com>
Date: Tue, 27 Feb 2024 17:35:55 +0100
Subject: [PATCH 2/6] Further elaborate on the root cause of model's
 degradation

---
 recipes_source/quantization.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/recipes_source/quantization.rst b/recipes_source/quantization.rst
index 012e8c8b360..14ab9cf7fbe 100644
--- a/recipes_source/quantization.rst
+++ b/recipes_source/quantization.rst
@@ -81,7 +81,7 @@ The full documentation of the `quantize_dynamic` API call is `here <https://pyto
 3. Post Training Static Quantization
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-This method converts both the weights and the activations to 8-bit integers beforehand so there won’t be on-the-fly conversion on the activations during the inference, as the dynamic quantization does. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's performance.
+This method converts both the weights and the activations to 8-bit integers beforehand so there won’t be on-the-fly conversion on the activations during the inference, as the dynamic quantization does. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's performance. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's performance. Converting weights and activation functions to 8-bit integers can slightly alter the network's behavior and activation responses, leading to a reduction in the model's original effectiveness.
 
 To apply static quantization on a model, run the following code:
 

From 95bdb46b15962f3600546d0b9b73a5230c2eddbd Mon Sep 17 00:00:00 2001
From: jmarin <javier.marin@satellogic.com>
Date: Tue, 27 Feb 2024 17:38:33 +0100
Subject: [PATCH 3/6] Change performance to accuracy as suggested.

---
 recipes_source/quantization.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/recipes_source/quantization.rst b/recipes_source/quantization.rst
index 14ab9cf7fbe..02e925c44af 100644
--- a/recipes_source/quantization.rst
+++ b/recipes_source/quantization.rst
@@ -81,7 +81,7 @@ The full documentation of the `quantize_dynamic` API call is `here <https://pyto
 3. Post Training Static Quantization
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-This method converts both the weights and the activations to 8-bit integers beforehand so there won’t be on-the-fly conversion on the activations during the inference, as the dynamic quantization does. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's performance. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's performance. Converting weights and activation functions to 8-bit integers can slightly alter the network's behavior and activation responses, leading to a reduction in the model's original effectiveness.
+This method converts both the weights and the activations to 8-bit integers beforehand so there won’t be on-the-fly conversion on the activations during the inference, as the dynamic quantization does. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's performance. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's accuracy. Converting weights and activation functions to 8-bit integers can slightly alter the network's behavior and activation responses, leading to a reduction in the model's original effectiveness.
 
 To apply static quantization on a model, run the following code:
 

From 0ce9191c2a1775281d83a1d604dedbb98dbc0eb8 Mon Sep 17 00:00:00 2001
From: jmarin <javier.marin@satellogic.com>
Date: Tue, 27 Feb 2024 17:50:43 +0100
Subject: [PATCH 4/6] Remove copy-paste mistake

---
 recipes_source/quantization.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/recipes_source/quantization.rst b/recipes_source/quantization.rst
index 02e925c44af..afccd56b918 100644
--- a/recipes_source/quantization.rst
+++ b/recipes_source/quantization.rst
@@ -81,7 +81,7 @@ The full documentation of the `quantize_dynamic` API call is `here <https://pyto
 3. Post Training Static Quantization
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-This method converts both the weights and the activations to 8-bit integers beforehand so there won’t be on-the-fly conversion on the activations during the inference, as the dynamic quantization does. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's performance. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's accuracy. Converting weights and activation functions to 8-bit integers can slightly alter the network's behavior and activation responses, leading to a reduction in the model's original effectiveness.
+This method converts both the weights and the activations to 8-bit integers beforehand so there won’t be on-the-fly conversion on the activations during the inference, as the dynamic quantization does. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's accuracy. Converting weights and activation functions to 8-bit integers can slightly alter the network's behavior and activation responses, leading to a reduction in the model's original effectiveness.
 
 To apply static quantization on a model, run the following code:
 

From 316377c3208a3f3182b233940420d941e4d1940b Mon Sep 17 00:00:00 2001
From: Javier Marin Tur <jmarin@csail.mit.edu>
Date: Tue, 27 Feb 2024 17:53:25 +0100
Subject: [PATCH 5/6] Update quantization.rst

Remove copy-paste mistake
---
 recipes_source/quantization.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/recipes_source/quantization.rst b/recipes_source/quantization.rst
index 02e925c44af..afccd56b918 100644
--- a/recipes_source/quantization.rst
+++ b/recipes_source/quantization.rst
@@ -81,7 +81,7 @@ The full documentation of the `quantize_dynamic` API call is `here <https://pyto
 3. Post Training Static Quantization
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-This method converts both the weights and the activations to 8-bit integers beforehand so there won’t be on-the-fly conversion on the activations during the inference, as the dynamic quantization does. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's performance. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's accuracy. Converting weights and activation functions to 8-bit integers can slightly alter the network's behavior and activation responses, leading to a reduction in the model's original effectiveness.
+This method converts both the weights and the activations to 8-bit integers beforehand so there won’t be on-the-fly conversion on the activations during the inference, as the dynamic quantization does. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's accuracy. Converting weights and activation functions to 8-bit integers can slightly alter the network's behavior and activation responses, leading to a reduction in the model's original effectiveness.
 
 To apply static quantization on a model, run the following code:
 

From 7d0df276eedf5dfdc24c07d6ff5703c465238965 Mon Sep 17 00:00:00 2001
From: jmarin <javier.marin@satellogic.com>
Date: Wed, 6 Mar 2024 23:05:45 +0100
Subject: [PATCH 6/6] Improve paragraph following suggestions

---
 recipes_source/quantization.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/recipes_source/quantization.rst b/recipes_source/quantization.rst
index afccd56b918..ac9cd48fe8c 100644
--- a/recipes_source/quantization.rst
+++ b/recipes_source/quantization.rst
@@ -81,7 +81,7 @@ The full documentation of the `quantize_dynamic` API call is `here <https://pyto
 3. Post Training Static Quantization
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-This method converts both the weights and the activations to 8-bit integers beforehand so there won’t be on-the-fly conversion on the activations during the inference, as the dynamic quantization does. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's accuracy. Converting weights and activation functions to 8-bit integers can slightly alter the network's behavior and activation responses, leading to a reduction in the model's original effectiveness.
+This method converts both the weights and the activations to 8-bit integers beforehand so there won’t be on-the-fly conversion on the activations during the inference, as the dynamic quantization does. While post-training static quantization can significantly enhance inference speed and reduce model size, this method may degrade the original model's accuracy more compared to post training dynamic quantization.
 
 To apply static quantization on a model, run the following code: