Redesign demo UI, add five curated examples, trim README

- New compact layout: hero, inputs left, poster + downloads right, example gallery
- Generation chains ZeroGPU steps automatically; downloads only appear when ready
- Five examples (NeurIPS 2022/2024, ICML 2022 x2, ICLR 2025) from the released checkpoint
- Remove placeholder example; LFS for png/pdf/zip; shorter README

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

.gitattributes

@@ -1,36 +1,3 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
-examples/gaussian_privacy/poster.png filter=lfs diff=lfs merge=lfs -text
+*.pdf filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,3 @@
+__pycache__/
+_stub/
+*.bak

README.md

@@ -9,52 +9,21 @@ python_version: 3.12
 app_file: app.py
 pinned: false
 startup_duration_timeout: 1h
-short_description: Turn a research paper into an editable academic poster
+short_description: Turn a research paper PDF into an editable academic poster
 ---
 
-# RAP — Paper to Poster
+# RAP · Paper to Poster
 
-Upload a paper PDF and a design instruction, then download an editable HTML
-poster, a PNG preview, and a ZIP of the selected figure/table/equation assets.
+Upload a research paper PDF and get an editable academic poster. A 27B vision-language model
+(Qwen3.8-27B, SFT + GRPO on real conference posters) reads the paper, selects figures and tables,
+and writes the poster as standalone HTML; the Space renders it to PNG.
 
-Model: RL checkpoint-240, BF16, **no quantization**, no thinking; total context
-is limited to 16,384 tokens. The model revision is pinned in `app.py`.
-The private model repository is accessed through a server-side HF_TOKEN Secret;
-it is never included in the app source or browser output.
+- Model: RL checkpoint-240, BF16, no quantization, 16K context.
+- Generation runs in ZeroGPU steps of up to 120 s; a full poster usually takes 3–6 steps and the
+  demo chains them automatically. Sign in to Hugging Face to use your GPU quota.
+- The examples on the page were generated by the same checkpoint with the default instruction and settings.
+- Research demo: scanned PDFs are not supported, uploads are temporary, and every statement,
+  number and caption should be checked before a poster is used.
 
-## How to use
-
-1. Upload a text-based research PDF (at most 40 MB).
-2. Set your instruction, canvas, columns and accent color. An optional advanced
-   layout specification overrides the automatic one.
-3. Click **Analyze PDF**. The training pipeline's LayoutLMv3 checkpoint extracts
-   assets from up to 12 pages; section-aware budgeting preserves method/results.
-4. Click **Generate / Continue**. Long outputs are generated over multiple
-   ZeroGPU requests; partial previews are explicitly labeled incomplete.
-5. Download the completed poster and check its scientific content and layout.
-
-A saved, actually generated example is available in the page without using GPU
-quota. It is clearly labeled as a saved output, not an interactive generation.
-
-## Runtime and limitations
-
-- ZeroGPU xlarge (96 GB) is requested per generation; no dedicated paid GPU.
-- Visitors use their own Hugging Face quota. A complete long poster can require
-  more time than a free visitor's daily quota. This is not an unlimited service.
-- Sign in before live generation. Anonymous quota is too small for the current
-  120-second xlarge reservation. Viewing the saved example needs no GPU quota.
-- Each request reserves 120 seconds; generation stops after approximately 100
-  seconds and can resume without adding chat turns or changing the prompt.
-- The first PDF extraction after a cold start installs an isolated CPU parser
-  environment, which may take several minutes. Subsequent PDFs reuse it.
-- The parser uses the original detector, but a new paper's automatically
-  selected assets/layout plan differ from the GT-derived test-set specifications.
-  This public demo is not an exact reproduction of the reported benchmark.
-- Scanned PDFs are not supported. Do not upload confidential papers.
-- Generated HTML is sanitized; scripts/external resources are disabled for
-  rendering. Scientific statements may still be wrong: review all outputs.
-- Files are temporary and may disappear on restart. Model weights are unchanged.
-
-The original detector and copied training preprocessing modules are described
-in [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md); upstream license text is
-included in [layout_vendor/LICENSE.md](layout_vendor/LICENSE.md).
+Third-party components (PDF-Extract-Kit LayoutLMv3 detector and the training preprocessing
+modules) are listed in [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md).

demo_ui.py

@@ -1,4 +1,4 @@
-"""Public PDF -> assets -> model HTML -> rendered poster workflow."""
+"""RAP demo UI: PDF -> assets -> model HTML -> rendered poster.  Pipeline unchanged; UI redesigned 2026-10-15."""
 import json
 import re
 import tempfile
@@ -7,153 +7,189 @@ from pathlib import Path
 import gradio as gr
 import torch
 import spaces
-from pdf_pipeline import prepare,chat,CANVASES,DEFAULT_INSTRUCTION
+from pdf_pipeline import prepare, chat, CANVASES, DEFAULT_INSTRUCTION
 from paper_sections import budget_body
 from poster_render import render
 
-ROOT=Path(__file__).resolve().parent
-ARTIFACTS=Path(tempfile.mkdtemp(prefix='rap_demo_'))
-WINDOW=16384
+ROOT = Path(__file__).resolve().parent
+ARTIFACTS = Path(tempfile.mkdtemp(prefix='rap_demo_'))
+WINDOW = 16384
+MAX_CHUNKS = 8
 
-def build(model,processor):
-    def analyze(pdf,instruction,canvas,accent,columns,custom_spec,progress=gr.Progress()):
-        if not pdf:raise gr.Error('Upload a research-paper PDF first.')
-        progress(.05,desc='Extracting the first 12 pages with LayoutLMv3 (CPU)…')
-        job=Path(tempfile.mkdtemp(prefix='paper_',dir=ARTIFACTS))
-        try:
-            p=prepare(pdf,job,instruction,canvas,accent,columns,custom_spec)
-            progress(.85,desc='Fitting paper text and visual assets into the 16K context…')
-            budget=22000
-            while True:
-                inputs=processor.apply_chat_template(chat(p),tokenize=True,add_generation_prompt=True,
-                    enable_thinking=False,return_dict=True,return_tensors='pt')
-                n=inputs['input_ids'].shape[1]
-                if n<=10000:break
-                if budget<=3500:
-                    raise ValueError('The paper and visual assets exceed the demo input budget.')
-                budget=max(3500,int(budget*.8))
-                p['body'],_=budget_body(p['sections'],total=budget,abstract_missing=True)
-            p['input_tokens']=n;p['continuation_ids']=[];p['raw']='';p['done']=False
-            summary=f'**Ready:** {p["title"]}\n\n{len(p["assets"])} assets · {n:,} input tokens · layout specification included.'
-            if p['stats'].get('total',0)<1000:
-                summary+='\n\nWarning: only a small amount of paper text was extracted.'
-            return p,summary,[(a['thumb'],f'{a["id"]} · page {a["page"]}') for a in p['assets']],p['spec'],None,None,None,''
-        except (ValueError,RuntimeError) as e:raise gr.Error(str(e)) from None
+CSS = '''
+.gradio-container {max-width: 1360px !important; margin: 0 auto;}
+#hero {padding: 26px 30px 22px; border-radius: 16px; margin-bottom: 6px;
+       background: linear-gradient(120deg, #1e1b4b 0%, #4c3d99 60%, #6d5bd0 100%); color: #fff;}
+#hero h1 {margin: 0 0 6px; font-size: 30px; letter-spacing: .2px; color: #fff;}
+#hero p {margin: 0; color: #e6e2ff; font-size: 15px;}
+#hero a {color: #fff; text-decoration: underline;}
+.badge {display: inline-block; margin: 10px 8px 0 0; padding: 3px 10px; border-radius: 999px;
+        background: rgba(255,255,255,.14); font-size: 12px; color: #fff;}
+#status {min-height: 24px; font-size: 14px;}
+#poster img {border-radius: 10px; box-shadow: 0 10px 30px rgba(20,16,60,.12);}
+#examples .thumbnail-item {border-radius: 10px;}
+footer {display: none !important;}
+'''
+
+def load_examples():
+    """examples/<name>/meta.json + paper.pdf + poster.png (+ poster.html, poster_bundle.zip)."""
+    out = []
+    for d in sorted((ROOT / 'examples').glob('*/')):
+        meta = d / 'meta.json'
+        if not meta.exists() or not (d / 'poster.png').exists() or not (d / 'paper.pdf').exists():
+            continue
+        m = json.loads(meta.read_text(encoding='utf-8'))
+        out.append({'dir': d, 'title': m.get('title', d.name), 'short': m.get('short', m.get('title', d.name))[:40], 'venue': m.get('venue', ''), 'instruction': m.get('instruction', DEFAULT_INSTRUCTION),
+                    'canvas': m.get('canvas', 'Landscape · 3:2'), 'columns': m.get('columns', 3), 'accent': m.get('accent', '#584287')})
+    return out
+
+def fileout(path):
+    return gr.update(value=str(path), visible=True) if path else gr.update(value=None, visible=False)
 
-    @spaces.GPU(size='xlarge',duration=120)
+def build(model, processor):
+    def analyze(pdf, instruction, canvas, accent, columns, custom_spec):
+        job = Path(tempfile.mkdtemp(prefix='paper_', dir=ARTIFACTS))
+        p = prepare(pdf, job, instruction, canvas, accent, columns, custom_spec)
+        budget = 22000
+        while True:
+            inputs = processor.apply_chat_template(chat(p), tokenize=True, add_generation_prompt=True,
+                                                   enable_thinking=False, return_dict=True, return_tensors='pt')
+            n = inputs['input_ids'].shape[1]
+            if n <= 10000:
+                break
+            if budget <= 3500:
+                raise ValueError('The paper and its figures exceed the demo input budget.')
+            budget = max(3500, int(budget * .8))
+            p['body'], _ = budget_body(p['sections'], total=budget, abstract_missing=True)
+        p['input_tokens'] = n; p['continuation_ids'] = []; p['raw'] = ''; p['done'] = False
+        return p
+
+    @spaces.GPU(size='xlarge', duration=120)
     def gpu_chunk(p):
-        inputs=processor.apply_chat_template(chat(p),tokenize=True,add_generation_prompt=True,
-            enable_thinking=False,return_dict=True,return_tensors='pt')
-        prefix=p.get('continuation_ids',[])
+        inputs = processor.apply_chat_template(chat(p), tokenize=True, add_generation_prompt=True,
+                                               enable_thinking=False, return_dict=True, return_tensors='pt')
+        prefix = p.get('continuation_ids', [])
         if prefix:
-            inputs['input_ids']=torch.cat([inputs['input_ids'],torch.tensor([prefix],dtype=torch.long)],dim=1)
-            inputs['attention_mask']=torch.ones_like(inputs['input_ids'])
-        length=int(inputs['input_ids'].shape[1])
-        remaining=WINDOW-length
-        if remaining<=0:raise ValueError('The 16K context budget has been exhausted.')
-        inputs=inputs.to('cuda')
-        start=time.monotonic()
+            inputs['input_ids'] = torch.cat([inputs['input_ids'], torch.tensor([prefix], dtype=torch.long)], dim=1)
+            inputs['attention_mask'] = torch.ones_like(inputs['input_ids'])
+        length = int(inputs['input_ids'].shape[1])
+        remaining = WINDOW - length
+        if remaining <= 0:
+            raise ValueError('The 16K context budget has been exhausted.')
+        inputs = inputs.to('cuda')
+        start = time.monotonic()
         with torch.inference_mode():
-            output=model.generate(**inputs,max_new_tokens=min(8192,remaining),max_time=100,
-                do_sample=False,use_cache=True,temperature=None,top_p=None,top_k=None,repetition_penalty=1.1)
-        new_ids=output[0,length:].detach().cpu().tolist()
-        all_ids=prefix+new_ids
-        text=processor.decode(all_ids,skip_special_tokens=True)
-        eos=model.generation_config.eos_token_id
-        eos=eos if isinstance(eos,list) else [eos]
-        complete=bool(re.search(r'</html\s*>',text,re.I))
-        stopped=complete or bool(new_ids and new_ids[-1] in eos) or length+len(new_ids)>=WINDOW
-        result={'ids':all_ids,'raw':text,'done':stopped,'complete':complete,'seconds':time.monotonic()-start,
-                'new_tokens':len(new_ids),'gpu':torch.cuda.get_device_name(),
-                'total_tokens':p['input_tokens']+len(all_ids)}
-        del inputs,output
+            output = model.generate(**inputs, max_new_tokens=min(8192, remaining), max_time=100, do_sample=False,
+                                    use_cache=True, temperature=None, top_p=None, top_k=None, repetition_penalty=1.1)
+        new_ids = output[0, length:].detach().cpu().tolist()
+        all_ids = prefix + new_ids
+        text = processor.decode(all_ids, skip_special_tokens=True)
+        eos = model.generation_config.eos_token_id
+        eos = eos if isinstance(eos, list) else [eos]
+        complete = bool(re.search(r'</html\s*>', text, re.I))
+        stopped = complete or bool(new_ids and new_ids[-1] in eos) or length + len(new_ids) >= WINDOW
+        result = {'ids': all_ids, 'raw': text, 'done': stopped, 'complete': complete, 'seconds': time.monotonic() - start,
+                  'new_tokens': len(new_ids), 'total_tokens': p['input_tokens'] + len(all_ids)}
+        del inputs, output
         return result
 
-    def generate(p,progress=gr.Progress()):
-        if not p:raise gr.Error('Click Analyze PDF first.')
-        if p.get('done'):raise gr.Error('Generation has ended. Analyze again to start a new poster.')
-        progress(.1,desc='Waiting for ZeroGPU xlarge, then generating HTML…')
-        result=gpu_chunk(p)
-        p=dict(p,continuation_ids=result['ids'],raw=result['raw'],done=result['done'])
-        rate=result['new_tokens']/max(result['seconds'],.01)
-        label=('Complete' if result['complete'] else 'Stopped — incomplete HTML; analyze again to retry'
-               if p['done'] else 'Partial — click Generate / Continue to resume')
-        status=(f'**{label}** · '
-                f'{len(result["ids"]):,} generated tokens · {result["total_tokens"]:,} / 16,384 total · '
-                f'{rate:.1f} tokens/s\n\nGPU: {result["gpu"]} · BF16 · checkpoint-240')
-        progress(.9,desc='Resolving assets and rendering the poster…')
+    def generate_loop(p, gallery, progress):
+        """ZeroGPU chunks until the HTML is complete; yields (state, status, png, html, zip, gallery, spec, raw, continue-button)."""
+        tokens = ''
+        for k in range(1, MAX_CHUNKS + 1):
+            progress(min(.15 + .1 * k, .9), desc=f'Generating poster HTML (step {k})…')
+            try:
+                r = gpu_chunk(p)
+            except Exception as e:                                         # noqa: BLE001  (quota exhausted, timeout, …)
+                yield (p, f"Paused after step {k - 1}: {str(e)[:200]}", None, None, None, gallery, p['spec'], p['raw'],
+                       gr.update(visible=bool(p.get('continuation_ids'))))
+                return
+            p = dict(p, continuation_ids=r['ids'], raw=r['raw'], done=r['done'])
+            tokens = f"{len(r['ids']):,} tokens · {r['new_tokens'] / max(r['seconds'], .01):.0f} tok/s"
+            if r['complete'] or p['done']:
+                break
+            yield (p, f"Step {k} done · {tokens} · continuing…", None, None, None, gallery, p['spec'], p['raw'], gr.update(visible=False))
+        progress(.95, desc='Rendering…')
         try:
-            png,html,archive,metadata=render(p['raw'],p)
-            if metadata['unresolved_assets']:status+='\n\nWarning: some generated asset IDs were not available.'
-            if not metadata['complete_html']:status+='\n\nThis preview is incomplete, not a finished poster.'
-            return p,status,png,html,archive,p['raw']
-        except ValueError as e:
-            return p,status+'\n\n'+str(e),None,None,None,p['raw']
-        except Exception as e:
-            print('POSTER_RENDER_ERROR',type(e).__name__,flush=True)
-            saved=Path(p['job'])/'output/poster.html'
-            return p,status+'\n\nPreview rendering failed; the generated HTML is preserved.',None,\
-                str(saved) if saved.exists() else None,None,p['raw']
+            png, html, archive, meta = render(p['raw'], p)
+        except Exception as e:                                             # noqa: BLE001
+            yield (p, f"The model stopped before a complete poster ({str(e)[:160]}).", None, None, None, gallery, p['spec'], p['raw'],
+                   gr.update(visible=not p.get('done')))
+            return
+        note = '' if meta['complete_html'] else ' · output incomplete'
+        yield (dict(p, done=True), f"Done · {tokens}{note}", png, fileout(html), fileout(archive), gallery, p['spec'], p['raw'], gr.update(visible=False))
 
-    css='''#hero {padding:20px 24px;border-radius:18px;background:linear-gradient(115deg,#241541,#534180);color:white}
-    #hero h1,#hero p {color:white} .gradio-container {max-width:1400px!important}'''
-    with gr.Blocks(title='RAP · Paper to Poster',css=css,delete_cache=(3600,86400),
-                   theme=gr.themes.Soft(primary_hue='violet')) as demo:
-        gr.Markdown('# RAP · Paper to Poster\nTurn a research paper into an editable academic poster.',elem_id='hero')
-        gr.Markdown('**RL checkpoint-240 · BF16, no quantization · no thinking · 16K context**\n\n'
-            'Upload a text-based PDF (≤40 MB). We analyze its first 12 pages, extract figures/tables/equations, '
-            'then generate poster HTML. Sign in to Hugging Face before generating. '
-            '**ZeroGPU xlarge uses 2× your personal GPU quota. A full poster usually needs several requests '
-            'and can exceed a free account’s daily allowance.** Partial previews are not finished posters. '
-            'The saved example below can be viewed without GPU quota.')
-        state=gr.State()
-        with gr.Row():
-            with gr.Column(scale=4):
-                pdf=gr.File(label='Research paper PDF',file_types=['.pdf'],type='filepath')
-                instruction=gr.Textbox(label='Your design instruction',value=DEFAULT_INSTRUCTION,lines=5)
+    def run(pdf, instruction, canvas, accent, columns, custom_spec, progress=gr.Progress()):
+        """One click: parse the PDF, then generate until the poster is complete."""
+        if not pdf:
+            raise gr.Error('Upload a paper PDF first.')
+        progress(.05, desc='Reading the paper (LayoutLMv3 on CPU)…')
+        try:
+            p = analyze(pdf, instruction, canvas, accent, columns, custom_spec)
+        except (ValueError, RuntimeError) as e:
+            raise gr.Error(str(e)) from None
+        gallery = [(a['thumb'], f"{a['id']} · page {a['page']}") for a in p['assets']]
+        yield (p, f"Parsed **{p['title'][:90]}** · {len(p['assets'])} figures/tables found · generating…",
+               None, None, None, gallery, p['spec'], '', gr.update(visible=False))
+        yield from generate_loop(p, gallery, progress)
+
+    def cont(p, progress=gr.Progress()):
+        if not p or p.get('done'):
+            raise gr.Error('Nothing to continue. Click Generate poster.')
+        gallery = [(a['thumb'], f"{a['id']} · page {a['page']}") for a in p['assets']]
+        yield from generate_loop(p, gallery, progress)
+
+    examples = load_examples()
+    def pick_example(evt: gr.SelectData):
+        ex = examples[evt.index]; d = ex['dir']
+        html = d / 'poster.html'; archive = d / 'poster_bundle.zip'
+        return (str(d / 'paper.pdf'), ex['instruction'], ex['canvas'], ex['accent'], ex['columns'],
+                f"Example · {ex['title']}" + (f" ({ex['venue']})" if ex['venue'] else ''),
+                str(d / 'poster.png'), fileout(html if html.exists() else None), fileout(archive if archive.exists() else None))
+
+    with gr.Blocks(title='RAP · Paper to Poster', css=CSS, delete_cache=(3600, 86400),
+                   theme=gr.themes.Soft(primary_hue='violet', neutral_hue='slate', radius_size='lg')) as demo:
+        gr.HTML('<div id="hero"><h1>RAP · Paper to Poster</h1>'
+                '<p>Upload a research paper and get an editable academic poster: a 27B vision-language model '
+                'reads the paper, picks figures and tables, and writes the poster as standalone HTML.</p>'
+                '<span class="badge">Qwen3.8-27B · SFT + GRPO</span><span class="badge">BF16 · ZeroGPU</span>'
+                '<span class="badge">Output: HTML + PNG</span></div>')
+        with gr.Row(equal_height=False):
+            with gr.Column(scale=5, min_width=360):
+                pdf = gr.File(label='Paper PDF', file_types=['.pdf'], type='filepath', height=110)
+                instruction = gr.Textbox(label='Instruction', value=DEFAULT_INSTRUCTION, lines=3)
                 with gr.Row():
-                    canvas=gr.Dropdown(list(CANVASES),value='Landscape · 3:2',label='Canvas')
-                    columns=gr.Slider(1,4,value=3,step=1,label='Columns')
-                accent=gr.ColorPicker(value='#584287',label='Accent color')
-                with gr.Accordion('Advanced: custom layout specification',open=False):
-                    custom_spec=gr.Textbox(label='Optional — leave blank for an automatic layout plan',lines=5)
-                analyze_btn=gr.Button('1 · Analyze PDF',variant='secondary')
-                generate_btn=gr.Button('2 · Generate / Continue',variant='primary')
+                    canvas = gr.Dropdown(list(CANVASES), value='Landscape · 3:2', label='Canvas', scale=2)
+                    columns = gr.Slider(1, 4, value=3, step=1, label='Columns', scale=2)
+                    accent = gr.ColorPicker(value='#584287', label='Accent', scale=1)
+                with gr.Accordion('Custom layout specification (optional)', open=False):
+                    custom_spec = gr.Textbox(show_label=False, placeholder='Leave empty to let the demo plan the layout.', lines=4)
+                run_btn = gr.Button('Generate poster', variant='primary', size='lg')
+                cont_btn = gr.Button('Continue generation', variant='secondary', visible=False)
+                gr.Markdown('<small>Generation runs in ZeroGPU steps of up to 120 s; a full poster usually takes 3–6 steps. '
+                            'Sign in to Hugging Face to use your GPU quota.</small>')
             with gr.Column(scale=7):
-                status=gr.Markdown('Upload a PDF to begin.')
-                poster=gr.Image(label='Rendered poster',type='filepath',interactive=False,height=600)
-                with gr.Row():
-                    html_download=gr.File(label='Standalone HTML',interactive=False)
-                    bundle=gr.File(label='Poster + selected assets ZIP',interactive=False)
-        with gr.Accordion('Extracted paper assets and layout plan',open=False):
-            gallery=gr.Gallery(label='Assets offered to the model',columns=4,height=350)
-            spec_view=gr.Textbox(label='Layout specification used for generation',lines=6,interactive=False)
-        with gr.Accordion('Model-generated HTML',open=False):
-            code=gr.Code(language='html',label='Model output (not executed in this page)',interactive=False)
-        example=ROOT/'examples/gaussian_privacy'
-        if (example/'poster.png').exists():
-            with gr.Accordion('View a completed example — no GPU quota required',open=False):
-                gr.Markdown('**Saved output, not a live generation.** This poster was generated on this '
-                    'Space using RL checkpoint-240 in BF16 from '
-                    '[Gaussian Membership Inference Privacy](https://arxiv.org/abs/2306.07273). '
-                    'Its figures are extracted from that paper. Please check scientific details before reuse.')
-                gr.Image(value=str(example/'poster.png'),label='Completed model-generated example',
-                         interactive=False,height=600)
+                status = gr.Markdown('Upload a PDF or pick an example below.', elem_id='status')
+                poster = gr.Image(label='Poster', type='filepath', interactive=False, height=560, elem_id='poster', show_download_button=True)
                 with gr.Row():
-                    gr.File(value=str(example/'poster.html'),label='Example HTML',interactive=False)
-                    gr.File(value=str(example/'poster_bundle.zip'),label='Example poster and assets',interactive=False)
-        gr.Markdown('Research demo: verify scientific statements, captions and layout before publication. '
-            'Uploaded papers are processed on the Space; do not upload confidential material. '
-            'Outputs are temporary and may disappear after a restart. '
-            '[Source and third-party notices](https://huggingface.co/spaces/VincentHancoder/RAP/tree/main).')
-        analyze_btn.click(analyze,[pdf,instruction,canvas,accent,columns,custom_spec],
-            [state,status,gallery,spec_view,poster,html_download,bundle,code],api_name='analyze',concurrency_limit=1)
-        generate_btn.click(generate,[state],[state,status,poster,html_download,bundle,code],
-            api_name='generate',concurrency_limit=1)
-        def reset_inputs():
-            return None,'Inputs changed — click Analyze PDF again.',None,None,None,'',[],''
-        for component in (pdf,instruction,canvas,accent,columns,custom_spec):
-            component.change(reset_inputs,None,
-                [state,status,poster,html_download,bundle,code,gallery,spec_view],
-                queue=False,api_name=False)
-    return demo,ARTIFACTS
+                    html_download = gr.File(label='Poster HTML', interactive=False, visible=False)
+                    bundle = gr.File(label='HTML + figures (ZIP)', interactive=False, visible=False)
+        if examples:
+            gr.Markdown('### Examples')
+            ex_gallery = gr.Gallery(value=[(str(e['dir'] / 'poster.png'), e['short']) for e in examples], columns=5, rows=1,
+                                    height=230, object_fit='contain', show_label=False, elem_id='examples', allow_preview=False)
+        with gr.Accordion('Details: extracted figures, layout plan, raw HTML', open=False):
+            gallery = gr.Gallery(label='Figures and tables offered to the model', columns=6, height=220)
+            spec_view = gr.Textbox(label='Layout specification', lines=4, interactive=False)
+            code = gr.Code(language='html', label='Model output', interactive=False)
+        with gr.Accordion('Notes', open=False):
+            gr.Markdown('Research demo. Scanned PDFs are not supported; uploads are processed on this Space and are temporary. '
+                        'Check every statement, number and caption before using a poster. '
+                        '[Source and third-party notices](https://huggingface.co/spaces/VincentHancoder/RAP/tree/main).')
+        state = gr.State()
+        outs = [state, status, poster, html_download, bundle, gallery, spec_view, code, cont_btn]
+        run_btn.click(run, [pdf, instruction, canvas, accent, columns, custom_spec], outs, api_name='generate', concurrency_limit=1)
+        cont_btn.click(cont, [state], outs, api_name='continue', concurrency_limit=1)
+        if examples:
+            ex_gallery.select(pick_example, None, [pdf, instruction, canvas, accent, columns, status, poster, html_download, bundle], api_name=False)
+    return demo, ARTIFACTS

examples/iclr_2025_5006/meta.json

@@ -0,0 +1,10 @@
+{
+ "title": "Matcha: Mitigating Graph Structure Shifts with Test-Time Adaptation",
+ "venue": "ICLR 2025",
+ "paper_id": "ICLR_2025_5006",
+ "short": "Matcha · ICLR 2025",
+ "instruction": "Create a clear academic poster with a strong title, a short motivation, the main method, key quantitative results, and takeaways. Prioritize the most informative figures. Use concise text, readable fonts, and generous spacing. Do not invent results.",
+ "canvas": "Landscape · 3:2",
+ "columns": 3,
+ "accent": "#584287"
+}

examples/iclr_2025_5006/paper.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b602b23c819046ab875dca439c16e512d799c1c89f3340e5bf11e24c51723198
+size 3939092

examples/icml_2022_16583/meta.json

@@ -0,0 +1,10 @@
+{
+ "title": "Adaptive Second Order Coresets for Data-efficient Machine Learning",
+ "venue": "ICML 2022",
+ "paper_id": "ICML_2022_16583",
+ "short": "AdaCore · ICML 2022",
+ "instruction": "Create a clear academic poster with a strong title, a short motivation, the main method, key quantitative results, and takeaways. Prioritize the most informative figures. Use concise text, readable fonts, and generous spacing. Do not invent results.",
+ "canvas": "Landscape · 3:2",
+ "columns": 3,
+ "accent": "#584287"
+}

examples/icml_2022_16583/paper.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce8ae89247f386c4fa0f7bc72a70ab0eda1cf3cd0ba3e4371f8922c3355ea01d
+size 1586267

examples/icml_2022_18241/meta.json

@@ -0,0 +1,10 @@
+{
+ "title": "Discovering Generalizable Spatial Goal Representations via Graph-based Active Reward Learning",
+ "venue": "ICML 2022",
+ "paper_id": "ICML_2022_18241",
+ "short": "GEM · ICML 2022",
+ "instruction": "Create a clear academic poster with a strong title, a short motivation, the main method, key quantitative results, and takeaways. Prioritize the most informative figures. Use concise text, readable fonts, and generous spacing. Do not invent results.",
+ "canvas": "Landscape · 3:2",
+ "columns": 3,
+ "accent": "#584287"
+}

examples/icml_2022_18241/paper.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3ca777da75c8b2f5e263786baec579cb877036fe5c0067d39aed04f8a921b60
+size 1344934

examples/neurips_2022_63389/meta.json

@@ -0,0 +1,10 @@
+{
+ "title": "Optimized Global Perturbation Attacks For Brain Tumour ROI Extraction From Binary Classification Models",
+ "venue": "NeurIPS 2022",
+ "paper_id": "NeurIPS_2022_63389",
+ "short": "Global Perturbation · NeurIPS 2022",
+ "instruction": "Create a clear academic poster with a strong title, a short motivation, the main method, key quantitative results, and takeaways. Prioritize the most informative figures. Use concise text, readable fonts, and generous spacing. Do not invent results.",
+ "canvas": "Landscape · 3:2",
+ "columns": 3,
+ "accent": "#584287"
+}

examples/neurips_2022_63389/paper.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21c8594705d06c371d6c320bf1d720bba3fcb4d375930ff57a88ed148bc53b07
+size 267114

examples/neurips_2024_4032/meta.json

@@ -0,0 +1,10 @@
+{
+ "title": "DiffuLT: Diffusion for Long-tail Recognition Without External Knowledge",
+ "venue": "NeurIPS 2024",
+ "paper_id": "NeurIPS_2024_4032",
+ "short": "DiffuLT · NeurIPS 2024",
+ "instruction": "Create a clear academic poster with a strong title, a short motivation, the main method, key quantitative results, and takeaways. Prioritize the most informative figures. Use concise text, readable fonts, and generous spacing. Do not invent results.",
+ "canvas": "Landscape · 3:2",
+ "columns": 3,
+ "accent": "#584287"
+}

examples/neurips_2024_4032/paper.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25e258b685407444c4f181dc019427dc2ad6d5590eab72af7ebddc203f1e1da0
+size 2809890