Fix compatibility issues and add colab notebook link (#18)

* Bug fixes * Add no of workers in sample run * Move import * Add workers and injections to the networks command * Add Colab notebook link in README --------- Co-authored-by: sharad339 <[email protected]>
ma3mool · Oct 22, 2024 · 6ecd77a · 6ecd77a
1 parent b9f53f1
commit 6ecd77a
Show file tree

Hide file tree

Showing 7 changed files with 28 additions and 21 deletions.
diff --git a/README.md b/README.md
@@ -25,6 +25,7 @@ GoldenEye is a functional simulator with fault injection capabilities for common
 ## Usage
 
 Take a look at our documentation [here](https://goldeneyedocs.readthedocs.io/en/stable/index.html).
+Check this [Colab notebook](https://colab.research.google.com/drive/1Om-Wg6wLOeaAKRWaDcpjoZZdF1LwXcCI) for a demo.
 
 ### Installing
 

diff --git a/scripts/accuracy_profile.sh b/scripts/accuracy_profile.sh
@@ -6,13 +6,14 @@ BATCH=${2} #128
 FORMAT=${3}    # simulated format
 BITWIDTH=${4}
 RADIX=${5}
+WORKERS=${6} # 16
 
 DATASET="IMAGENET"
 OUTPUT_PATH="../output/"
 SRC_PATH="../src/"
 LOG_PATH="./log/"
 SCRIPT1="../src/preprocess.py"
-SCRIPT2="../src/profile.py"
+SCRIPT2="../src/profile_model.py"
 SCRIPT3="../src/split_data.py"
 SCRIPT4="../src/injections.py"
 SCRIPT5="../src/postprocess.py"
@@ -23,7 +24,7 @@ PRECISION="FP32" # compute fabric
 QUANT="" # -q leave empty if you do not want quantization
 #BIT_FLIP="-e" # -e leave empty if you do not want bit flip model. NOTE: -q MUST BE ENABLED TOO WITH THIS
 TRAINSET="" # -r. leave empty if using testset
-WORKERS=16
+#WORKERS=16
 
 if [[ ${FORMAT} -eq "INT" ]]
 then

diff --git a/scripts/end_to_end.sh b/scripts/end_to_end.sh
@@ -8,13 +8,14 @@ FORMAT=${4}  #"fp_n"    # simulated format
 BITWIDTH=${5}   #32
 RADIX=${6}      #23
 INJECTIONS_LOC=${7}  # {0, no injection}. {1: value} or {2, META}
+WORKERS=${8} #16
 
 
 OUTPUT_PATH="../output/"
 SRC_PATH="../src/"
 LOG_PATH="./log/"
 SCRIPT1="../src/preprocess.py"
-SCRIPT2="../src/profile.py"
+SCRIPT2="../src/profile_model.py"
 SCRIPT3="../src/split_data.py"
 SCRIPT4="../src/injections.py"
 SCRIPT5="../src/postprocess.py"
@@ -25,9 +26,9 @@ BIAS="" # leave empty, or include the flag with the number: "-a -8" | ""
 QUANT="" # -q leave empty if you do not want quantization
 #BIT_FLIP="-e" # -e leave empty if you do not want bit flip model. NOTE: -q MUST BE ENABLED TOO WITH THIS
 TRAINSET="" # -r. leave empty if using testset
-WORKERS=16
+#WORKERS=16
 
-INJECTIONS=102400 #${BATCH}
+INJECTIONS=${9} #102400 #${BATCH}
 #INJECTIONS_LOC=1  # {0, no injection}. {1: value} or {2, META}
                   #OLD {2, INT value}, or {3, INT scaling}, or {4, block meta}, or {5, adaptive meta}
 
@@ -102,5 +103,5 @@ python3 ${SCRIPT4} -b ${BATCH} -n ${NETWORK} -d ${DATASET} -o ${OUTPUT_PATH} ${T
 
 # postprocessing
 echo -n "Postprocessing ... "
-python3 ${SCRIPT5} -b ${BATCH} -n ${NETWORK} -d ${DATASET} -o ${OUTPUT_PATH} ${TRAINSET} ${VERBOSE} ${DEBUG} -w ${WORKERS} -P ${PRECISION} -i ${INJECTIONS} -f ${FORMAT} -B ${BITWIDTH} -R ${RADIX} ${BIAS} ${QUANT}
+python3 ${SCRIPT5} -b ${BATCH} -n ${NETWORK} -d ${DATASET} -o ${OUTPUT_PATH} ${TRAINSET} ${VERBOSE} ${DEBUG} -w ${WORKERS} -P ${PRECISION} -i ${INJECTIONS} -f ${FORMAT} -B ${BITWIDTH} -R ${RADIX} -I ${INJECTIONS_LOC} ${BIAS} ${QUANT}
 echo -n "Done! "
diff --git a/scripts/networks.sh b/scripts/networks.sh
@@ -5,8 +5,10 @@ script="./end_to_end.sh"
 network=${1}
 dataset=${2}
 batchsize=${3}
+workers=${4:-8}
+injections=${5:-102400}
 
-time ${script} ${network} ${dataset} ${batchsize} fp32 32 23  1               #2&> ./log/alexnet.log
+time ${script} ${network} ${dataset} ${batchsize} fp32 32 23  1 ${workers} ${injections}              #2&> ./log/alexnet.log
 
 #time ${script} ${dataset} ${network} ${batchsize} block_fp 12 7  1               #2&> ./log/alexnet.log
 #time ${script} ${dataset} ${network} ${batchsize} adaptive_fp 8 3 1               #2&> ./log/alexnet.log

diff --git a/src/injections.py b/src/injections.py
@@ -170,7 +170,7 @@ def rand_neurons_batch(pfi_model, layer, shape, maxval, batchsize, function=-1):
             pbar.update(samples)
 
             # prep images
-            images, labels, img_ids, index = dataiter.next()
+            images, labels, img_ids, index = next(dataiter)
             if getCUDA_en():
                 labels = labels.cuda()
                 images = images.cuda()

diff --git a/src/profile.py → src/profile_model.py b/src/profile.py → src/profile_model.py
diff --git a/src/util.py b/src/util.py
@@ -10,7 +10,6 @@
 import timm
 import numpy as np
 from num_sys_class import *
-from othermodels import resnet, vgg, cifar10_nn
 
 '''
 Environment Variables
@@ -282,19 +281,22 @@ def getNetwork(networkName, DATASET):
 
     elif DATASET == 'CIFAR10' or DATASET == 'CIFAR100':
         if networkName == "resnet18":
-            MODEL = resnet.resnet18(pretrained=True)
-        elif networkName == "vgg19_bn":
-            MODEL = vgg.vgg19_bn(pretrained=True)
-        elif networkName == "cifar10_nn_baseline":
-            MODEL = cifar10_nn.baseline(pretrained=True, output_size=getNumClasses(DATASET))
-        elif networkName == "cifar10_nn_v1":
-            MODEL = cifar10_nn.v1(pretrained=True, output_size=getNumClasses(DATASET))
-        elif networkName == "cifar10_nn_v2":
-            MODEL = cifar10_nn.v2(pretrained=True, output_size=getNumClasses(DATASET))
-
-        # Error
+            MODEL = models.resnet18(pretrained=True, progress=True)
         else:
-            sys.exit("Network does not exist")
+            from othermodels import vgg, cifar10_nn
+
+            if networkName == "vgg19_bn":
+                MODEL = vgg.vgg19_bn(pretrained=True)
+            elif networkName == "cifar10_nn_baseline":
+                MODEL = cifar10_nn.baseline(pretrained=True, output_size=getNumClasses(DATASET))
+            elif networkName == "cifar10_nn_v1":
+                MODEL = cifar10_nn.v1(pretrained=True, output_size=getNumClasses(DATASET))
+            elif networkName == "cifar10_nn_v2":
+                MODEL = cifar10_nn.v2(pretrained=True, output_size=getNumClasses(DATASET))
+
+            # Error
+            else:
+                sys.exit("Network does not exist")
 
     # model upgrades
     if getCUDA_en():