Initial mae d2go commit

Summary: Pull Request resolved: https://github.com/facebookresearch/d2go/pull/218

Reviewed By: newstzpz

Differential Revision: D35084145

fbshipit-source-id: f655b1a1cfa813f99e51f2d4527a6e480ef8608d

d2go/optimizer/build.py

@@ -251,9 +251,11 @@ def adamw(cfg, model: torch.nn.Module) -> torch.optim.Optimizer:
     Build an optimizer from config.
     """
     params = get_optimizer_param_groups(model, cfg)
-    return maybe_add_gradient_clipping(cfg, torch.optim.AdamW)(
-        params, cfg.SOLVER.BASE_LR
+
+    optim = maybe_add_gradient_clipping(cfg, torch.optim.AdamW)(
+        params, cfg.SOLVER.BASE_LR, betas=cfg.SOLVER.BETAS
     )
+    return optim
 
 
 @D2GO_OPTIM_MAPPER_REGISTRY.register()

d2go/utils/get_default_cfg.py

@@ -70,6 +70,9 @@ def get_default_cfg(_C):
     _C.SOLVER.LR_MULTIPLIER_OVERWRITE = []
     _C.SOLVER.WEIGHT_DECAY_EMBED = 0.0
 
+    # Betas are used in the AdamW optimizer
+    _C.SOLVER.BETAS = (0.9, 0.999)
+
     # RECOMPUTE_BOXES for LSJ Training
     _C.INPUT.RECOMPUTE_BOXES = False