Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-20 01:31:13 +02:00
Code Issues Packages Projects Releases Wiki Activity

fix(ltx): wrap explicit encode_prompt in inference_context

Browse Source 
Upstream LTX2 encode_prompt is not @torch.no_grad-decorated; only __call__
is. Calling encode_prompt directly from ltx_process recorded the full
Gemma3 forward graph.
This commit is contained in:
CalamitousFelicitousness
2026-05-04 21:44:07 +01:00
parent 28fbf1f74d
commit a37443d8f1
1 changed files with 7 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/ltx/ltx_process.py
+7 -6
View File
@@ -326,12 +326,13 @@ def run_ltx(task_id,
extra_networks.activate(p, networks)
# Encode once and reuse across stages; encode_prompt short-circuits when
# embeds are passed to __call__. CPU park keeps them off GPU between stages.
prompt_embeds, prompt_attention_mask, negative_prompt_embeds, negative_prompt_attention_mask = shared.sd_model.encode_prompt(
prompt=prompt_final,
negative_prompt=negative_final,
do_classifier_free_guidance=True,
device=devices.device,
)
with devices.inference_context():
prompt_embeds, prompt_attention_mask, negative_prompt_embeds, negative_prompt_attention_mask = shared.sd_model.encode_prompt(
prompt=prompt_final,
negative_prompt=negative_final,
do_classifier_free_guidance=True,
device=devices.device,
)
prompt_embeds = prompt_embeds.cpu()
prompt_attention_mask = prompt_attention_mask.cpu() if prompt_attention_mask is not None else None
negative_prompt_embeds = negative_prompt_embeds.cpu() if negative_prompt_embeds is not None else None