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.

cleanup

HenryNdubuakud85195db77f4
diff --git a/src/data.py b/src/data.py
index a54e0fd..48d114c 100644
--- a/src/data.py
+++ b/src/data.py
@@ -730,11 +730,7 @@ def pack_sequences(cache_path, enc_vl, dec_in_vl, dec_tgt_vl, loss_vl):
enc_lens = _seq_lens(enc_vl)
dec_lens = _seq_lens(dec_in_vl)
- # Sort by encoder length descending — first-fit-decreasing
order = np.argsort(-enc_lens)
-
- # Vectorized bin packing: numpy arrays for bin capacities
- # Pre-allocate for worst case (every seq gets its own bin)
bin_enc_rem = np.empty(n, dtype=np.int32)
bin_dec_rem = np.empty(n, dtype=np.int32)
bin_contents = [[] for _ in range(n)]
@@ -744,7 +740,6 @@ def pack_sequences(cache_path, enc_vl, dec_in_vl, dec_tgt_vl, loss_vl):
el = int(enc_lens[idx])
dl = int(dec_lens[idx])
- # Vectorized search: find all bins where both enc and dec fit
if n_bins > 0:
fits = (bin_enc_rem[:n_bins] >= el) & (bin_dec_rem[:n_bins] >= dl)
candidates = np.flatnonzero(fits)
@@ -766,7 +761,6 @@ def pack_sequences(cache_path, enc_vl, dec_in_vl, dec_tgt_vl, loss_vl):
bin_dec_rem = bin_dec_rem[:n_bins]
bin_contents = bin_contents[:n_bins]
- # Fill packed arrays — parallelized with multiprocessing
packed_enc = np.zeros((n_bins, max_enc), dtype=np.int16)
packed_dec_in = np.zeros((n_bins, max_dec), dtype=np.int16)
packed_dec_tgt = np.zeros((n_bins, max_dec), dtype=np.int16)