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.

training bugs

HenryNdubuaku5d123887c2d5
diff --git a/src/train.py b/src/train.py
index c7bb881..f4143eb 100644
--- a/src/train.py
+++ b/src/train.py
@@ -778,15 +778,17 @@ def train(args):
del _eval_params
- # --- tqdm postfix ---
+ # --- tqdm postfix (fixed keys to prevent bar flickering) ---
postfix = {
+ "speech_loss": f"{speech_loss_val:.4f}" if speech_loss_val is not None else "-",
"text_loss": f"{text_loss_val:.4f}",
"text_ppl": f"{last_val_ppl:.2f}" if last_val_ppl is not None else "?",
}
- if speech_loss_val is not None:
- postfix["speech_loss"] = f"{speech_loss_val:.4f}"
- if epoch == weight_prune_epoch and not gradual_sparsify_done:
- postfix["prune"] = f"{current_sparsity*100:.1f}%"
+ if sparsity_ratio > 0:
+ if epoch == weight_prune_epoch and not gradual_sparsify_done:
+ postfix["sparsification"] = f"{current_sparsity*100:.1f}%"
+ else:
+ postfix["sparsification"] = "done"
pbar.set_postfix(**postfix)
if use_wandb: