Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 01:04:32 +02:00
Code Issues Packages Projects Releases Wiki Activity

feat(vqa): persist thinking mode and improve reasoning output formatting

Browse Source 
- Add interrogate_vlm_thinking_mode setting to save checkbox state
- Update ui_caption to restore Thinking Mode preference on load
- Add blank line before 'Answer:' label for visual separation
- Remove '\n\n' replacement in clean() that stripped blank lines
- Fix Qwen reasoning detection when <think> tag is in prompt, not response
- Add reasoning icon to Moondream 2 and 3 model names
This commit is contained in:
CalamitousFelicitousness
2025-12-04 23:49:10 +00:00
parent a4b5e84a13
commit 2b6226b62b
3 changed files with 29 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/ui_caption.py
+12 -10
View File
@@ -30,7 +30,7 @@ def update_vlm_prompt_placeholder(question):
def update_vlm_params(*args):
vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking = args
vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking, vlm_thinking_mode = args
shared.opts.interrogate_vlm_max_length = int(vlm_max_tokens)
shared.opts.interrogate_vlm_num_beams = int(vlm_num_beams)
shared.opts.interrogate_vlm_temperature = float(vlm_temperature)
@@ -39,6 +39,7 @@ def update_vlm_params(*args):
shared.opts.interrogate_vlm_top_p = float(vlm_top_p)
shared.opts.interrogate_vlm_keep_prefill = bool(vlm_keep_prefill)
shared.opts.interrogate_vlm_keep_thinking = bool(vlm_keep_thinking)
shared.opts.interrogate_vlm_thinking_mode = bool(vlm_thinking_mode)
shared.opts.save(shared.config_filename)
@@ -87,20 +88,21 @@ def create_ui():
vlm_top_p = gr.Slider(label='Top-P', value=shared.opts.interrogate_vlm_top_p, minimum=0.0, maximum=1.0, step=0.01, elem_id='vlm_top_p')
with gr.Row():
vlm_do_sample = gr.Checkbox(label='Use sample', value=shared.opts.interrogate_vlm_do_sample, elem_id='vlm_do_sample')
vlm_thinking_mode = gr.Checkbox(label='Thinking Mode', value=False, elem_id='vlm_thinking_mode')
vlm_thinking_mode = gr.Checkbox(label='Thinking Mode', value=shared.opts.interrogate_vlm_thinking_mode, elem_id='vlm_thinking_mode')
with gr.Row():
vlm_keep_thinking = gr.Checkbox(label='Keep Thinking Trace', value=shared.opts.interrogate_vlm_keep_thinking, elem_id='vlm_keep_thinking')
vlm_keep_prefill = gr.Checkbox(label='Keep Prefill', value=shared.opts.interrogate_vlm_keep_prefill, elem_id='vlm_keep_prefill')
with gr.Row():
vlm_prefill = gr.Textbox(label='Prefill Text', value='', lines=1, elem_id='vlm_prefill', placeholder='Optional prefill text for model to continue from')
vlm_max_tokens.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking], outputs=[])
vlm_num_beams.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking], outputs=[])
vlm_temperature.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking], outputs=[])
vlm_do_sample.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking], outputs=[])
vlm_top_k.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking], outputs=[])
vlm_top_p.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking], outputs=[])
vlm_keep_prefill.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking], outputs=[])
vlm_keep_thinking.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking], outputs=[])
vlm_max_tokens.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking, vlm_thinking_mode], outputs=[])
vlm_num_beams.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking, vlm_thinking_mode], outputs=[])
vlm_temperature.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking, vlm_thinking_mode], outputs=[])
vlm_do_sample.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking, vlm_thinking_mode], outputs=[])
vlm_top_k.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking, vlm_thinking_mode], outputs=[])
vlm_top_p.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking, vlm_thinking_mode], outputs=[])
vlm_keep_prefill.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking, vlm_thinking_mode], outputs=[])
vlm_keep_thinking.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking, vlm_thinking_mode], outputs=[])
vlm_thinking_mode.change(fn=update_vlm_params, inputs=[vlm_max_tokens, vlm_num_beams, vlm_temperature, vlm_do_sample, vlm_top_k, vlm_top_p, vlm_keep_prefill, vlm_keep_thinking, vlm_thinking_mode], outputs=[])
with gr.Accordion(label='Batch caption', open=False, visible=True):
with gr.Row():
vlm_batch_files = gr.File(label="Files", show_label=True, file_count='multiple', file_types=['image'], interactive=True, height=100, elem_id='vlm_batch_files')