Automation foundation model for tiny devices: 2-bit, 8-29 MB, tool calls, ASR, structured extraction and embeddings on phones, wearables, smart homes, robots, cars and microcontrollers.
bugfix
HenryNdubuakuea240109e1c0
diff --git a/src/train.py b/src/train.pyindex 1da0d7e..c61f852 100644--- a/src/train.py+++ b/src/train.py@@ -588,10 +588,10 @@ def train(args):del ema_unrprune_mask = jax_utils.replicate(mask)del mask- # Apply mask once via the fused step (reuses donated buffers)- state = state.replace(- params=jax.tree.map(lambda w, m: w * m, state.params, prune_mask))- ema_params = jax.tree.map(lambda w, m: w * m, ema_params, prune_mask)+ # Apply final mask to both params and EMA (EMA lags due to slow decay)+ state = state.replace(+ params=jax.tree.map(lambda w, m: w * m, state.params, prune_mask))+ ema_params = jax.tree.map(lambda w, m: w * m, ema_params, prune_mask)final_pruned = jax.tree.map(np.array, jax_utils.unreplicate(ema_params))total_p = sum(x.size for x in jax.tree.leaves(final_pruned))zero_p = sum(int(np.sum(np.abs(x) < 1e-6)) for x in jax.tree.leaves(final_pruned))