issues
search
eth-easl
/
modyn
Modyn is a research-platform for training ML models on dynamic datasets.
MIT License
22
stars
3
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
feat: Add more powerful windowing strategies, warumup and dynamic thresholds
#564
robinholzi
opened
1 day ago
1
Enable finetuning on `RHO-LOSS` selection strategy when `Twin` strategy is used to produce the irreducible loss model
#563
XianzheMa
opened
1 day ago
3
fix: Adjust currently trained model to be defined for dataset end
#562
robinholzi
closed
1 day ago
0
fix: Fix minor issue in alibi detect config
#561
robinholzi
opened
1 day ago
2
Support threshold-based triggering for alibi-detect MMD
#560
MaxiBoether
closed
2 days ago
2
feat!: Batch evaluations in evaluator
#559
robinholzi
closed
2 days ago
7
Add rho training log to the pipeline file
#558
XianzheMa
closed
1 day ago
3
Windowing for data drift detection
#557
MaxiBoether
closed
2 days ago
3
Windowing for drift detection
#556
MaxiBoether
closed
2 days ago
0
feat: Create `DriftDetector` abstraction and introduce alibi-detect
#555
robinholzi
closed
2 days ago
3
Batch evaluation intervals into a single request and a single evaluation process
#554
XianzheMa
closed
2 days ago
8
Fix small issues in aggregation script
#553
MaxiBoether
closed
3 days ago
2
Twin RHO Model Step 2: split the training set and train the twin model
#552
XianzheMa
closed
4 days ago
3
feat: Create `DriftDetector` abstraction and introduce alibi-detect
#551
robinholzi
closed
2 days ago
3
Parameterize `drop_last`
#550
XianzheMa
closed
1 week ago
3
Add `cuda-cudart-dev` and `cuda-profiler-api` to environment
#549
MaxiBoether
closed
1 week ago
2
feat: Add unique sequence number for aggregation into supervisor
#548
robinholzi
closed
1 week ago
2
Twin RHO Model Step 1: create the Twin RHO Model
#547
XianzheMa
closed
4 days ago
2
feat: Refine triggering abstraction
#546
robinholzi
closed
1 week ago
2
fix: Allow to ignore certain assertion errors
#545
robinholzi
closed
1 week ago
2
Add `ratio_max` for presampling strategies
#544
MaxiBoether
closed
1 week ago
2
Weighted Optimization
#543
MaxiBoether
opened
1 week ago
1
BtS accumulation with non-multiples
#542
MaxiBoether
opened
1 week ago
0
Implement `ratio_max` scaling for downsamplers
#541
MaxiBoether
closed
1 week ago
2
Adding another way of calculating gradients in downsamplers
#540
XianzheMa
closed
1 week ago
3
Record training loss
#539
XianzheMa
closed
1 week ago
2
Missing commit SHA when running in Docker
#538
MaxiBoether
opened
2 weeks ago
0
feat: Log commit sha
#537
robinholzi
closed
2 weeks ago
2
Speeding up evaluations
#536
MaxiBoether
closed
2 days ago
0
fix: Adjust evaluation business logic
#535
robinholzi
closed
2 weeks ago
5
feat: More sophisticated evaluation logic
#534
robinholzi
closed
1 week ago
3
Fix batch number
#533
XianzheMa
closed
1 week ago
2
fix: Fix evaluation logic
#532
robinholzi
closed
2 weeks ago
2
Fix unfixed naming inconsistencies
#531
XianzheMa
closed
2 weeks ago
3
Fail explicitly in storage when asking for duplicate keys
#530
MaxiBoether
closed
2 weeks ago
2
Fix replacement option in loss and gradnorm downsampling
#529
MaxiBoether
closed
2 weeks ago
3
fix: Adjust active model definition
#528
robinholzi
closed
2 weeks ago
1
Fix recovery in datasets
#527
MaxiBoether
closed
2 weeks ago
3
Forward `shuffle` to trainer server and measure GPU
#526
XianzheMa
closed
2 weeks ago
2
Fix device for initial RHO-LOSS tensor
#525
MaxiBoether
closed
2 weeks ago
1
Multithreaded error handling in storage
#524
MaxiBoether
closed
2 weeks ago
2
refactor: Simplify yearbook data generation
#523
robinholzi
closed
2 weeks ago
2
Small stability and logging fixes
#522
MaxiBoether
closed
2 weeks ago
1
More robust gRPC connections, using `tenacity` where possible
#521
MaxiBoether
closed
2 weeks ago
5
Fix various issues in RS2
#520
MaxiBoether
closed
2 weeks ago
1
More exact logging in storage and returning errors
#519
MaxiBoether
closed
2 weeks ago
2
Fix metric in `UncertaintyDownsamplingStrategy`
#518
MaxiBoether
closed
2 weeks ago
2
Optionally hide evaluation progress
#517
MaxiBoether
closed
2 weeks ago
2
Add statistics for CGLM dataset
#516
MaxiBoether
closed
1 week ago
3
Fix inconsistency on passed number of samples in `batch-then-sample` mode
#515
XianzheMa
closed
2 weeks ago
3
Next