From 6d1f734d4215e48ccf69b7bda0388deb717a53f9 Mon Sep 17 00:00:00 2001
From: HU <franck.hu@ecl20.ec-lyon.fr>
Date: Mon, 20 Nov 2023 14:01:13 +0100
Subject: [PATCH] maj
---
TD2 Deep Learning.ipynb | 41 +++++++++++++++++-
...osch-South-Africa-closeup-1-DHD.jpg.icloud | Bin 0 -> 254 bytes
2 files changed, 39 insertions(+), 2 deletions(-)
create mode 100644 hymenoptera_data/test/ants/.ants-devouring-remains-of-large-dead-insect-on-red-tile-in-Stellenbosch-South-Africa-closeup-1-DHD.jpg.icloud
diff --git a/TD2 Deep Learning.ipynb b/TD2 Deep Learning.ipynb
index b7f5695..0d885ba 100644
--- a/TD2 Deep Learning.ipynb
+++ b/TD2 Deep Learning.ipynb
@@ -23769,12 +23769,20 @@
"print_size_of_model(model, \"fp32\")"
]
},
+ {
+ "cell_type": "markdown",
+ "id": "d1fc547c",
+ "metadata": {},
+ "source": [
+ "### La taille du modèle 1 initial est de 251.278 KB"
+ ]
+ },
{
"cell_type": "markdown",
"id": "05c4e9ad",
"metadata": {},
"source": [
- "Post training quantization example"
+ "### On va compresser ce modèle : "
]
},
{
@@ -23804,11 +23812,18 @@
"source": [
"import torch.quantization\n",
"\n",
- "\n",
"quantized_model = torch.quantization.quantize_dynamic(model, dtype=torch.qint8)\n",
"print_size_of_model(quantized_model, \"int8\")"
]
},
+ {
+ "cell_type": "markdown",
+ "id": "70bd3046",
+ "metadata": {},
+ "source": [
+ "### Le modèle compressé a une taille de 76.522 KB. Il y a donc eu une réduction de 174.756 KB."
+ ]
+ },
{
"cell_type": "code",
"execution_count": 19,
@@ -23897,6 +23912,20 @@
")"
]
},
+ {
+ "cell_type": "markdown",
+ "id": "50c95c3e",
+ "metadata": {},
+ "source": [
+ "### En terme de précision, elle est la même en pourcentage pour presque toutes les classes (en comparaison avec le modèle non compressé).\n",
+ "\n",
+ "#### Elle difére de 1% pour la classe horse.\n",
+ "\n",
+ "### Au global, la précision est la même.\n",
+ "\n",
+ "### Le modèle compressé conserve la précision du modèle tout en réduisant sa taille."
+ ]
+ },
{
"cell_type": "markdown",
"id": "7b108e17",
@@ -23913,6 +23942,14 @@
"Try training aware quantization to mitigate the impact on the accuracy (doc available here https://pytorch.org/docs/stable/quantization.html#torch.quantization.quantize_dynamic)"
]
},
+ {
+ "cell_type": "markdown",
+ "id": "a5ba7855",
+ "metadata": {},
+ "source": [
+ "### On va essayer le training aware quantization"
+ ]
+ },
{
"cell_type": "markdown",
"id": "201470f9",
diff --git a/hymenoptera_data/test/ants/.ants-devouring-remains-of-large-dead-insect-on-red-tile-in-Stellenbosch-South-Africa-closeup-1-DHD.jpg.icloud b/hymenoptera_data/test/ants/.ants-devouring-remains-of-large-dead-insect-on-red-tile-in-Stellenbosch-South-Africa-closeup-1-DHD.jpg.icloud
new file mode 100644
index 0000000000000000000000000000000000000000..6f5e2546ea17972a4437948bbb1ffd6a582d82a8
GIT binary patch
literal 254
zcmYc)$jK}&F)+By$i&RT$`<1n92(@~mzbOComv?$AOPmNW#*&?XI4RkB;Z0psm1xF
zMaiill?4!~w8XrUV%?O~vi#Db%)E5nqSV~P%)DaV{50L1#G>?6-IUbC6y40c;?(34
z-TXY=qSO@KlFXb`-ON1Q;F8pwoYcIe{Nm&c-QfJvk_=tPw4%)9MBU_^{NmKo0$oF0
y7Y`S`tb%kQ1_i;T@d7ef9TJq9UYe6w#KAA4@3nac0~jzef++?Db|?*_t^)weK~4w&
literal 0
HcmV?d00001
--
GitLab