Skip to content

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.

updates

HenryNdubuakua4df722f6c50
diff --git a/src/train.py b/src/train.py
index bc39d68..c808c7d 100644
--- a/src/train.py
+++ b/src/train.py
@@ -783,9 +783,9 @@ def train(args):
cl_q_b, cl_t_b, cl_rngs, do_qat, do_cl, do_prune,
)
- text_loss_val = float(loss.addressable_shards[0].data)
+ text_loss_val = float(loss.addressable_shards[0].data[0])
text_losses.append(text_loss_val)
- step_grad_norm = float(grad_norm.addressable_shards[0].data)
+ step_grad_norm = float(grad_norm.addressable_shards[0].data[0])
global_step += 1
if epoch == weight_prune_epoch and not gradual_sparsify_done: