Skip to content

update configs for experiment run #126

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
houman1359 merged 1 commit into from
Jan 12, 2026
Merged

update configs for experiment run #126

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 1 addition & 2 deletions .gitignore
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,8 +2,7 @@
_arxiv/
# Ignore drafts by default, but keep the SCAR paper folder tracked (it has its own .gitignore).
drafts/
!drafts/LLM_prune/
!drafts/LLM_prune/**

checkpoints/
results/
logs/
Expand Down
25 changes: 15 additions & 10 deletions configs/vision_prune/mobilenetv2_cifar10_unified.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,18 @@ dataset:
batch_size: 128
num_workers: 4

# -----------------------------------------------------------------------------
# TRAINING (paper-quality CIFAR baselines)
# -----------------------------------------------------------------------------
training:
enabled: true
epochs: 50
learning_rate: 0.05
optimizer: "sgd"
scheduler: "cosine"
momentum: 0.9
weight_decay: 0.0005

# -----------------------------------------------------------------------------
# CALIBRATION
# -----------------------------------------------------------------------------
Expand Down Expand Up @@ -156,14 +168,6 @@ pruning:
- "redundancy_low" # Prune low redundancy (MI)
- "synergy_low" # Prune low synergy

# =========================================================================
# SINGLE METRICS - Prune HIGH (assumes high = unimportant)
# =========================================================================
- "rq_high" # Prune high RQ (TEST: is high RQ bad?)
- "redundancy_high" # Prune high redundancy (TEST: is high corr bad?)
- "synergy_high" # Prune high synergy
- "magnitude_high" # Prune high magnitude (inverse of standard)

# =========================================================================
# COMPOSITE COMBINATIONS
# =========================================================================
Expand Down Expand Up @@ -195,10 +199,11 @@ pruning:
- "composite_pos_red"

fine_tune:
enabled: false # Disabled to see pure pruning impact without recovery
epochs: 15 # MobileNet may need more fine-tuning
enabled: true # Enable recovery fine-tuning after pruning (standard for reporting)
epochs: 5
learning_rate: 0.0001
weight_decay: 0.00001
max_batches: 100

# -----------------------------------------------------------------------------
# EVALUATION (Enhanced for Vision)
Expand Down
27 changes: 17 additions & 10 deletions configs/vision_prune/resnet18_cifar10_unified.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,19 @@ dataset:
batch_size: 128
num_workers: 4

# -----------------------------------------------------------------------------
# TRAINING (paper-quality CIFAR baselines)
# -----------------------------------------------------------------------------
# NOTE: This trains/fine-tunes the model on CIFAR-10 before running the metric/cluster/pruning analyses.
training:
enabled: true
epochs: 50
learning_rate: 0.05
optimizer: "sgd"
scheduler: "cosine"
momentum: 0.9
weight_decay: 0.0005

# -----------------------------------------------------------------------------
# CALIBRATION
# -----------------------------------------------------------------------------
Expand Down Expand Up @@ -165,14 +178,6 @@ pruning:
- "redundancy_low" # Prune low redundancy (MI)
- "synergy_low" # Prune low synergy

# =========================================================================
# SINGLE METRICS - Prune HIGH (assumes high = unimportant)
# =========================================================================
- "rq_high" # Prune high RQ (TEST: is high RQ bad?)
- "redundancy_high" # Prune high redundancy (TEST: is high corr bad?)
- "synergy_high" # Prune high synergy
- "magnitude_high" # Prune high magnitude (inverse of standard)

# =========================================================================
# COMPOSITE COMBINATIONS
# =========================================================================
Expand Down Expand Up @@ -205,10 +210,12 @@ pruning:
- "composite_pos_red"

fine_tune:
enabled: false # Disabled to see pure pruning impact without recovery
epochs: 10
enabled: true # Enable recovery fine-tuning after pruning (standard for reporting)
epochs: 5
learning_rate: 0.0001
weight_decay: 0.0001
# Safety cap: limits fine-tune compute so the full method×ratio grid stays feasible on 1 GPU
max_batches: 100

# -----------------------------------------------------------------------------
# EVALUATION (Enhanced for Vision)
Expand Down
23 changes: 14 additions & 9 deletions configs/vision_prune/resnet50_imagenet100_unified.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,6 +45,17 @@ dataset:
image_size: 224
normalize: true

# -----------------------------------------------------------------------------
# TRAINING (required: classifier head is replaced for ImageNet-100)
# -----------------------------------------------------------------------------
training:
enabled: true
epochs: 30
learning_rate: 0.001
optimizer: "adam"
scheduler: "cosine"
weight_decay: 0.0001

# -----------------------------------------------------------------------------
# CALIBRATION
# -----------------------------------------------------------------------------
Expand Down Expand Up @@ -157,14 +168,6 @@ pruning:
- "redundancy_low" # Prune low redundancy (MI)
- "synergy_low" # Prune low synergy

# =========================================================================
# SINGLE METRICS - Prune HIGH (assumes high = unimportant)
# =========================================================================
- "rq_high" # Prune high RQ (TEST: is high RQ bad?)
- "redundancy_high" # Prune high redundancy (TEST: is high corr bad?)
- "synergy_high" # Prune high synergy
- "magnitude_high" # Prune high magnitude (inverse of standard)

# =========================================================================
# COMPOSITE COMBINATIONS
# =========================================================================
Expand Down Expand Up @@ -196,10 +199,12 @@ pruning:
- "composite_pos_red"

fine_tune:
enabled: false # Disabled to see pure pruning impact without recovery
enabled: true # Enable recovery fine-tuning after pruning (standard for reporting)
epochs: 5 # Fewer epochs for ImageNet
learning_rate: 0.00001
weight_decay: 0.0001
# Critical for feasibility: fine-tuning every (method,ratio) on ImageNet-100 otherwise explodes runtime.
max_batches: 10

# -----------------------------------------------------------------------------
# EVALUATION (Enhanced for Vision)
Expand Down
25 changes: 15 additions & 10 deletions configs/vision_prune/vgg16_cifar10_unified.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,6 +41,18 @@ dataset:
batch_size: 128
num_workers: 4

# -----------------------------------------------------------------------------
# TRAINING (paper-quality CIFAR baselines)
# -----------------------------------------------------------------------------
training:
enabled: true
epochs: 50
learning_rate: 0.05
optimizer: "sgd"
scheduler: "cosine"
momentum: 0.9
weight_decay: 0.0005

# -----------------------------------------------------------------------------
# CALIBRATION
# -----------------------------------------------------------------------------
Expand Down Expand Up @@ -153,14 +165,6 @@ pruning:
- "redundancy_low" # Prune low redundancy (MI)
- "synergy_low" # Prune low synergy

# =========================================================================
# SINGLE METRICS - Prune HIGH (assumes high = unimportant)
# =========================================================================
- "rq_high" # Prune high RQ (TEST: is high RQ bad?)
- "redundancy_high" # Prune high redundancy (TEST: is high corr bad?)
- "synergy_high" # Prune high synergy
- "magnitude_high" # Prune high magnitude (inverse of standard)

# =========================================================================
# COMPOSITE COMBINATIONS
# =========================================================================
Expand Down Expand Up @@ -192,10 +196,11 @@ pruning:
- "composite_pos_red"

fine_tune:
enabled: false # Disabled to see pure pruning impact without recovery
epochs: 10
enabled: true # Enable recovery fine-tuning after pruning (standard for reporting)
epochs: 5
learning_rate: 0.0001
weight_decay: 0.0001
max_batches: 100

# -----------------------------------------------------------------------------
# EVALUATION (Enhanced for Vision)
Expand Down
Loading
Loading