n0p/mlx-video
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix token handling in LTX-2 text encoder by directly appending response tokens to the generated tokens list, improving clarity and consistency in token generation.

Browse Source
This commit is contained in:
Prince Canuma
2026-03-18 13:50:33 +01:00
parent f5e311a77c
commit fea0f87df9
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
mlx_video/models/ltx_2/text_encoder.py
View File

@@ -1079,7 +1079,7 @@ class LTX2TextEncoder(nn.Module):
for i, response in enumerate(generator): for i, response in enumerate(generator):
next_token = mx.array([response.token]) next_token = mx.array([response.token])
input_ids = mx.concatenate([input_ids, next_token[None, :]], axis=1) input_ids = mx.concatenate([input_ids, next_token[None, :]], axis=1)
generated_tokens.append(next_token.squeeze()) generated_tokens.append(response.token)
generated_token_count += 1 generated_token_count += 1
progress.update(task, advance=1) progress.update(task, advance=1)