Skip to content

Implement model unloading and dtype fix #64

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
mykeehu wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Implement model unloading and dtype fix #64

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
0d44dd4
Implement model unloading and dtype fix
mykeehu Dec 29, 2025
b46c2da
Move checkbox before Caption
mykeehu Dec 30, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
28 changes: 23 additions & 5 deletions gradio-app/app.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -279,6 +279,17 @@ def show_global_error(message: str):

def hide_global_error():
return gr.update(value="", visible=False)

def unload_model():
global g_model, g_processor, g_quant
print("Unloading model and clearing memory...")
g_model = None
g_processor = None
g_quant = None
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()


def load_model(quant: str, status: gr.HTML | None = None):
"""Load the model and processor if not already loaded."""
Expand Down Expand Up @@ -313,7 +324,7 @@ def load_model(quant: str, status: gr.HTML | None = None):

try:
if quant == "bf16":
g_model = LlavaForConditionalGeneration.from_pretrained(MODEL_PATH, torch_dtype="bfloat16", device_map=0)
g_model = LlavaForConditionalGeneration.from_pretrained(MODEL_PATH, dtype="bfloat16", device_map=0)
assert isinstance(g_model, LlavaForConditionalGeneration), f"Expected LlavaForConditionalGeneration, got {type(g_model)}"
if _HAS_LIGER:
try:
Expand All @@ -338,7 +349,7 @@ def load_model(quant: str, status: gr.HTML | None = None):
else:
raise ValueError(f"Unknown quantization type: {quant}")

g_model = LlavaForConditionalGeneration.from_pretrained(MODEL_PATH, torch_dtype="auto", device_map=0, quantization_config=qnt_config)
g_model = LlavaForConditionalGeneration.from_pretrained(MODEL_PATH, dtype="auto", device_map=0, quantization_config=qnt_config)
assert isinstance(g_model, LlavaForConditionalGeneration), f"Expected LlavaForConditionalGeneration, got {type(g_model)}"

g_model.eval()
Expand Down Expand Up @@ -426,7 +437,7 @@ def print_system_info():


@torch.no_grad()
def chat_joycaption(input_image: Image.Image, prompt: str, temperature: float, top_p: float, max_new_tokens: int, quant: str) -> Generator[dict, None, None]:
def chat_joycaption(input_image: Image.Image, prompt: str, temperature: float, top_p: float, max_new_tokens: int, quant: str, unload_after: bool) -> Generator[dict, None, None]:
# Hide any previous global errors
yield {global_error: hide_global_error()}

Expand Down Expand Up @@ -491,6 +502,11 @@ def chat_joycaption(input_image: Image.Image, prompt: str, temperature: float, t

t.join()
yield {single_status_output: gr.update(value="Captioning complete!")}

if unload_after:
unload_model()
yield {single_status_output: format_info("Model unloaded from memory.")}

except Exception as e:
error_msg = f"Error during generation: {str(e)}"
print(error_msg)
Expand Down Expand Up @@ -752,8 +768,10 @@ def collate_fn(batch: list[tuple[Path, Image.Image, str, str] | None], *, proces
with gr.Column(scale=1):
initial_single_prompt = build_prompt(caption_type.value, caption_length.value, extra_options.value, name_input.value)
prompt_box_single = gr.Textbox(lines=4, label="Confirm or Edit Prompt", value=initial_single_prompt, interactive=True, elem_id="single_prompt_box")
unload_after_gen = gr.Checkbox(label="Unload model from RAM/VRAM after generation",value=False)
run_button_single = gr.Button("Caption", variant="primary")
output_caption_single = gr.Textbox(label="Generated Caption", lines=8, interactive=True, elem_id="single_output_box")


# Batch Processing Tab
with gr.TabItem("Batch Processing", id="batch_tab"):
Expand Down Expand Up @@ -810,7 +828,7 @@ def collate_fn(batch: list[tuple[Path, Image.Image, str, str] | None], *, proces
# Handle single image captioning
run_button_single.click(
chat_joycaption,
inputs=[input_image_single, prompt_box_single, temperature_slider, top_p_slider, max_tokens_slider, model_quantization],
inputs=[input_image_single, prompt_box_single, temperature_slider, top_p_slider, max_tokens_slider, model_quantization, unload_after_gen],
outputs=[single_status_output, output_caption_single, global_error],
)

Expand All @@ -824,4 +842,4 @@ def collate_fn(batch: list[tuple[Path, Image.Image, str, str] | None], *, proces

if __name__ == "__main__":
print_system_info()
demo.launch()
demo.launch()