Tessa Han
th135
AI & ML interests
None yet
Recent Activity
updated a collection 1 day ago
klb updated a model 1 day ago
th135/OLMo-2-1B-Midtrain-50B-SAM-rho5e-2-metamathqa published a model 1 day ago
th135/OLMo-2-1B-Midtrain-50B-SAM-rho5e-2-metamathqaOrganizations
None yet
models pretrained on fw + finetuned on safety
-
th135/llama-0.5B-10BT-weightdecay0.0001-seed42-safetymixed300
0.5B • Updated • 4 -
th135/llama-0.5B-10BT-weightdecay0.001-seed42-safetymixed300
0.5B • Updated • 4 -
th135/llama-0.5B-10BT-weightdecay0.01-seed42-safetymixed300
0.5B • Updated • 5 -
th135/llama-0.5B-10BT-weightdecay0.1-seed42-safetymixed300
0.5B • Updated • 4
vary lr and wd during pt
models pretrained on fw + finetuned on CoT datasets
-
th135/llama-0.5B-10BT-weightdecay0.0001-seed42-metamathqa
0.5B • Updated • 4 -
th135/llama-0.5B-10BT-weightdecay0.001-seed42-metamathqa
0.5B • Updated • 4 -
th135/llama-0.5B-10BT-weightdecay0.01-seed42-metamathqa
0.5B • Updated • 4 -
th135/llama-0.5B-10BT-weightdecay0.1-seed42-metamathqa
0.5B • Updated • 3
sweep ft hyperparameters
-
th135/olmo-1B-30BT-weightdecay0.1-metamathqa-sweep-lr1.0e-5-bs8-wdft0.0
1B • Updated • 5 -
th135/olmo-1B-30BT-weightdecay0.1-metamathqa-sweep-lr1.0e-5-bs8-wdft0.1
1B • Updated • 7 -
th135/olmo-1B-30BT-weightdecay0.1-metamathqa-sweep-lr1.0e-5-bs8-wdft1.0
1B • Updated • 5 -
th135/olmo-1B-30BT-weightdecay0.1-metamathqa-sweep-lr1.0e-5-bs16-wdft0.1
1B • Updated • 5
models pretrained on fw + finetuned on language and general knowledge datasets
Model checkpoints for MedSafetyBench paper [NeurIPS 2024] (https://arxiv.org/abs/2403.03744)
pretrained models
sweep ft hyperparameters
-
th135/olmo-1B-30BT-weightdecay0.1-metamathqa-sweep-lr1.0e-5-bs8-wdft0.0
1B • Updated • 5 -
th135/olmo-1B-30BT-weightdecay0.1-metamathqa-sweep-lr1.0e-5-bs8-wdft0.1
1B • Updated • 7 -
th135/olmo-1B-30BT-weightdecay0.1-metamathqa-sweep-lr1.0e-5-bs8-wdft1.0
1B • Updated • 5 -
th135/olmo-1B-30BT-weightdecay0.1-metamathqa-sweep-lr1.0e-5-bs16-wdft0.1
1B • Updated • 5
models pretrained on fw + finetuned on safety
-
th135/llama-0.5B-10BT-weightdecay0.0001-seed42-safetymixed300
0.5B • Updated • 4 -
th135/llama-0.5B-10BT-weightdecay0.001-seed42-safetymixed300
0.5B • Updated • 4 -
th135/llama-0.5B-10BT-weightdecay0.01-seed42-safetymixed300
0.5B • Updated • 5 -
th135/llama-0.5B-10BT-weightdecay0.1-seed42-safetymixed300
0.5B • Updated • 4
models pretrained on fw + finetuned on language and general knowledge datasets
vary lr and wd during pt
Model checkpoints for MedSafetyBench paper [NeurIPS 2024] (https://arxiv.org/abs/2403.03744)
pretrained models
models pretrained on fw + finetuned on CoT datasets
-
th135/llama-0.5B-10BT-weightdecay0.0001-seed42-metamathqa
0.5B • Updated • 4 -
th135/llama-0.5B-10BT-weightdecay0.001-seed42-metamathqa
0.5B • Updated • 4 -
th135/llama-0.5B-10BT-weightdecay0.01-seed42-metamathqa
0.5B • Updated • 4 -
th135/llama-0.5B-10BT-weightdecay0.1-seed42-metamathqa
0.5B • Updated • 3