Spaces:
Running on Zero
Running on Zero
Update app.py
Browse files
app.py
CHANGED
|
@@ -27,7 +27,7 @@ import gradio as gr
|
|
| 27 |
import pk_workflow as pk
|
| 28 |
|
| 29 |
MODEL_REPO = os.environ.get("H3_MODEL_REPO", "MiniMaxAI/MiniMax-H3")
|
| 30 |
-
CONDITIONER_SPACE = os.environ.get("H3_CONDITIONER", "
|
| 31 |
# `pack` places the transformer at startup, `lazy` moves everything on the first GPU call.
|
| 32 |
PLACEMENT = os.environ.get("H3_PLACEMENT", "pack").lower()
|
| 33 |
# cuDNN's fused attention is 10-20% faster than the SDPA default on this pool and needs nothing installed. It is
|
|
@@ -434,7 +434,7 @@ def get_duration(
|
|
| 434 |
return max(60, int((denoise + decode + post) * _MARGIN) + _PLACEMENT_ALLOWANCE)
|
| 435 |
|
| 436 |
|
| 437 |
-
@spaces.GPU(duration=
|
| 438 |
def _generate(
|
| 439 |
prompt_embeds,
|
| 440 |
text_token_tags,
|
|
|
|
| 27 |
import pk_workflow as pk
|
| 28 |
|
| 29 |
MODEL_REPO = os.environ.get("H3_MODEL_REPO", "MiniMaxAI/MiniMax-H3")
|
| 30 |
+
CONDITIONER_SPACE = os.environ.get("H3_CONDITIONER", "dagloop5/qwen3vl-conditioner")
|
| 31 |
# `pack` places the transformer at startup, `lazy` moves everything on the first GPU call.
|
| 32 |
PLACEMENT = os.environ.get("H3_PLACEMENT", "pack").lower()
|
| 33 |
# cuDNN's fused attention is 10-20% faster than the SDPA default on this pool and needs nothing installed. It is
|
|
|
|
| 434 |
return max(60, int((denoise + decode + post) * _MARGIN) + _PLACEMENT_ALLOWANCE)
|
| 435 |
|
| 436 |
|
| 437 |
+
@spaces.GPU(duration=125, size=GPU_SIZE)
|
| 438 |
def _generate(
|
| 439 |
prompt_embeds,
|
| 440 |
text_token_tags,
|