Sync backend files from GitHub 66452e1

Deployed-From: 66452e148f1e95cfb40220989f5b1cf28967c829

.github/workflows/sync-to-hf.yml

@@ -1,20 +1,99 @@
 name: Sync to Hugging Face Hub
 
+# GitHub main is the source of truth for both Spaces:
+# - mppolak/madtools (madtools.org) runs the website and gets the repo as-is.
+# - AccuratePaperExtraction/MADTOOL only displays the backend files: it gets a
+#   snapshot without index.html, with deploy/backend-space-README.md as README.md.
+# The two jobs are independent, so one failing doesn't block the other.
+
 on:
   push:
     branches: [ main ]
   workflow_dispatch:
 
 jobs:
-  sync:
+  website:
+    name: sync website (mppolak/madtools)
     runs-on: ubuntu-latest
+    env:
+      SPACE: mppolak/madtools
     steps:
       - uses: actions/checkout@v6
         with:
           fetch-depth: 0
           lfs: true
+
+      - name: Block deploy if the Space has commits GitHub doesn't
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN_MADTOOLS }}
+        run: |
+          git remote add hf https://user:$HF_TOKEN@huggingface.co/spaces/$SPACE
+          git fetch hf main
+          AHEAD=$(git rev-list --count HEAD..hf/main)
+          if [ "$AHEAD" -gt 0 ]; then
+            echo "::error::The Space $SPACE has $AHEAD commit(s) not on GitHub. Deploy blocked to avoid discarding them."
+            echo "Commits only on the Space:"
+            git log --format='  %h  %an  %s' HEAD..hf/main
+            echo ""
+            echo "To reconcile, from a local clone of this repo:"
+            echo "  git fetch https://huggingface.co/spaces/$SPACE main"
+            echo "  git merge FETCH_HEAD   # review the diff, resolve, or revert what you don't want"
+            echo "  git push origin main   # re-runs this workflow, which will then fast-forward the Space"
+            exit 1
+          fi
+          echo "Space $SPACE is in sync with GitHub; proceeding."
+
       - name: Push to HF Space
+        run: git push hf HEAD:main
+
+  backend-files:
+    name: sync backend files (AccuratePaperExtraction/MADTOOL)
+    runs-on: ubuntu-latest
+    env:
+      SPACE: AccuratePaperExtraction/MADTOOL
+    steps:
+      - uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
+          lfs: true
+
+      - name: Block deploy if the Space was edited directly
         env:
           HF_TOKEN: ${{ secrets.HF_TOKEN }}
         run: |
-          git push --force https://user:$HF_TOKEN@huggingface.co/spaces/AccuratePaperExtraction/MADTOOL main
+          git remote add hf https://user:$HF_TOKEN@huggingface.co/spaces/$SPACE
+          git fetch hf main
+          # The Space's history holds snapshots this job made (marked with a
+          # Deployed-From trailer); any other commit GitHub lacks is a manual edit.
+          MANUAL=$(git log --format='%h %(trailers:key=Deployed-From,valueonly,separator=)' HEAD..hf/main | awk 'NF==1 {print $1}')
+          if [ -n "$MANUAL" ]; then
+            echo "::error::The Space $SPACE has commits made directly on it. Deploy blocked to avoid discarding them."
+            for c in $MANUAL; do git log -1 --format='  %h  %an  %s' "$c"; done
+            echo ""
+            echo "To reconcile, from a local clone of this repo, bring any wanted changes into GitHub"
+            echo "(the README lives in deploy/backend-space-README.md), then record the edits as merged:"
+            echo "  git fetch https://huggingface.co/spaces/$SPACE main"
+            echo "  git merge -s ours FETCH_HEAD"
+            echo "  git push origin main"
+            exit 1
+          fi
+          echo "Space $SPACE has no direct edits; proceeding."
+
+      - name: Build backend-files snapshot and push
+        run: |
+          export GIT_INDEX_FILE="$RUNNER_TEMP/backend-files.index"
+          git read-tree HEAD
+          git rm --cached --quiet --ignore-unmatch index.html deploy/backend-space-README.md
+          git update-index --add --cacheinfo 100644,"$(git hash-object -w deploy/backend-space-README.md)",README.md
+          TREE=$(git write-tree)
+          unset GIT_INDEX_FILE
+          if [ "$TREE" = "$(git rev-parse 'hf/main^{tree}')" ]; then
+            echo "Space $SPACE already has these files; nothing to push."
+            exit 0
+          fi
+          COMMIT=$(git -c user.name="github-actions[bot]" \
+                       -c user.email="41898282+github-actions[bot]@users.noreply.github.com" \
+                       commit-tree "$TREE" -p hf/main \
+                       -m "Sync backend files from GitHub ${GITHUB_SHA::7}" \
+                       -m "Deployed-From: $GITHUB_SHA")
+          git push hf "$COMMIT:refs/heads/main"

main.py

@@ -717,21 +717,50 @@ def provider_display_name(provider: str) -> str:
     return {"openai": "OpenAI", "gemini": "Gemini"}.get(provider, provider or "Unknown")
 
 
+# Substrings of OpenAI model ids that are not text-chat models (same list
+# PlotExtract uses, plus completions-only "instruct" models). "search" also
+# covers the tool-only deep-research models.
+OPENAI_NON_CHAT_MARKERS = (
+    "audio",
+    "dall-e",
+    "embedding",
+    "gpt-image",
+    "instruct",
+    "moderation",
+    "realtime",
+    "search",
+    "transcribe",
+    "tts",
+    "whisper",
+)
+
+
+def is_openai_chat_model(model_id: str) -> bool:
+    mid = model_id.lower()
+    is_family = "gpt" in mid or re.match(r"^o\d+(?:-|$)", mid)  # GPT and o-series reasoning models
+    return bool(is_family) and not any(marker in mid for marker in OPENAI_NON_CHAT_MARKERS)
+
+
 def list_openai_chat_models(api_key: str) -> List[Dict[str, str]]:
     if OpenAIClient is None:
         raise RuntimeError("openai SDK is not installed.")
     client = OpenAIClient(api_key=api_key)
     page = client.models.list()
-    out: List[Dict[str, str]] = []
+    out: List[Dict[str, Any]] = []
     seen = set()
     for item in getattr(page, "data", []) or []:
         model_id = getattr(item, "id", None) or (item.get("id") if isinstance(item, dict) else None)
         if not model_id or model_id in seen:
             continue
-        if "gpt" not in model_id.lower():
+        if not is_openai_chat_model(model_id):
             continue
         seen.add(model_id)
-        out.append({"id": model_id, "display_name": model_id})
+        created = getattr(item, "created", None) or (item.get("created") if isinstance(item, dict) else None)
+        out.append({"id": model_id, "display_name": model_id, "created": created or 0})
+    # Newest first by release timestamp
+    out.sort(key=lambda m: (m["created"], m["id"]), reverse=True)
+    for m in out:
+        m.pop("created")
     return out
 
 
@@ -748,6 +777,45 @@ def _fetch_gemini_model_list(api_key: str) -> List[Dict[str, Any]]:
     return payload.get("models", []) or []
 
 
+# Substrings of Gemini model ids that support generateContent but are not
+# text-chat models (speech, image generation, live audio, tool-only agents).
+GEMINI_NON_CHAT_MARKERS = (
+    "aqa",
+    "audio",
+    "computer-use",
+    "embedding",
+    "image",
+    "live",
+    "robotics",
+    "tts",
+)
+
+
+def gemini_sort_key(model_id: str) -> Tuple:
+    """Newest-first ordering for Gemini ids, which carry no release date: the
+    version number tracks release order, then pro > flash > flash-lite, then
+    stable before preview/experimental. "-latest" aliases float to the top.
+    """
+    mid = model_id.lower()
+    version = re.search(r"-(\d+(?:\.\d+)?)(?:-|$)", mid)
+    if "lite" in mid:
+        tier = 1
+    elif "flash" in mid:
+        tier = 2
+    elif "pro" in mid:
+        tier = 3
+    else:
+        tier = 0
+    return (
+        "latest" in mid,
+        mid.startswith("gemini"),
+        float(version.group(1)) if version else 0.0,
+        tier,
+        not any(tag in mid for tag in ("preview", "exp")),
+        mid,
+    )
+
+
 def list_gemini_models(api_key: str) -> List[Dict[str, str]]:
     """List Gemini chat models via the public REST API to avoid pinning conflicts
     between google-generativeai and langchain-google-genai.
@@ -762,9 +830,12 @@ def list_gemini_models(api_key: str) -> List[Dict[str, str]]:
         model_id = raw_id.split("/", 1)[-1] if raw_id else ""
         if not model_id or model_id in seen:
             continue
+        if any(marker in model_id.lower() for marker in GEMINI_NON_CHAT_MARKERS):
+            continue
         seen.add(model_id)
         display = item.get("displayName") or model_id
         out.append({"id": model_id, "display_name": str(display)})
+    out.sort(key=lambda m: gemini_sort_key(m["id"]), reverse=True)
     return out
 
 
@@ -808,32 +879,10 @@ def fetch_models_for_provider(provider: str, api_key: str) -> List[Dict[str, str
     return []
 
 
-def _date_key(text: str) -> int:
-    value = (text or "").lower()
-    compact = re.search(r"(20\d{2})(\d{2})(\d{2})", value)
-    if compact:
-        return int("".join(compact.groups()))
-    dashed = re.search(r"(20\d{2})[-_](\d{2})[-_](\d{2})", value)
-    if dashed:
-        return int("".join(dashed.groups()))
-    return -1
-
-
 def choose_default_model(models: List[Dict[str, str]], provider: str) -> str:
-    if not models:
-        return ""
-    def priority(m):
-        mid = (m.get("id") or "").lower()
-        return (1 if "latest" in mid else 0, _date_key(mid), mid)
-    if provider == "openai":
-        gpt5 = [m for m in models if "gpt-5" in (m.get("id") or "").lower()]
-        if gpt5:
-            return max(gpt5, key=priority)["id"]
-    if provider == "gemini":
-        pros = [m for m in models if "pro" in (m.get("id") or "").lower()]
-        if pros:
-            return max(pros, key=priority)["id"]
-    return max(models, key=priority)["id"]
+    # The provider lists are already sorted newest-first, so the default is
+    # simply the top entry.
+    return models[0]["id"] if models else ""
 
 class AsyncEmbeddingAdapter:
     def __init__(self, embedder):

plotextract/app.py

@@ -458,51 +458,50 @@ def fetch_all_gemini_models(client):
     return [model for model in models if is_gemini_vision_model(model)]
 
 
+def gemini_sort_key(model):
+    # Gemini lists carry no release date: the version number tracks release
+    # order, then pro > flash > flash-lite, then stable before preview.
+    # "-latest" aliases float to the top.
+    model_id = (model.get("id") or "").lower()
+    version = re.search(r"-(\d+(?:\.\d+)?)(?:-|$)", model_id)
+    if "lite" in model_id:
+        tier = 1
+    elif "flash" in model_id:
+        tier = 2
+    elif "pro" in model_id:
+        tier = 3
+    else:
+        tier = 0
+    return (
+        "latest" in model_id,
+        float(version.group(1)) if version else 0.0,
+        tier,
+        not any(tag in model_id for tag in ["preview", "exp"]),
+        model_id,
+    )
+
+
+def sort_models_newest_first(models, provider):
+    if provider == "gemini":
+        return sorted(models, key=gemini_sort_key, reverse=True)
+    # Anthropic and OpenAI report a creation time (UTC ISO strings sort correctly)
+    return sorted(models, key=lambda model: (model.get("created_at") or "", model.get("id") or ""), reverse=True)
+
+
 def fetch_all_models(client, provider):
     if provider == "anthropic":
-        return fetch_all_anthropic_models(client)
+        return sort_models_newest_first(fetch_all_anthropic_models(client), provider)
     if provider == "openai":
-        return fetch_all_openai_models(client)
+        return sort_models_newest_first(fetch_all_openai_models(client), provider)
     if provider == "gemini":
-        return fetch_all_gemini_models(client)
+        return sort_models_newest_first(fetch_all_gemini_models(client), provider)
     return []
 
 
-def extract_date_key(text):
-    value = (text or "").lower()
-    compact = re.search(r"(20\d{2})(\d{2})(\d{2})", value)
-    if compact:
-        return int("".join(compact.groups()))
-    dashed = re.search(r"(20\d{2})[-_](\d{2})[-_](\d{2})", value)
-    if dashed:
-        return int("".join(dashed.groups()))
-    return -1
-
-
-def model_priority(model):
-    model_id = (model.get("id") or "").lower()
-    is_latest_alias = 1 if "latest" in model_id else 0
-    date_key = max(extract_date_key(model_id), extract_date_key(model.get("created_at") or ""))
-    created_key = model.get("created_at") or ""
-    return (is_latest_alias, date_key, created_key, model_id)
-
-
 def choose_default_model(models, provider):
-    if provider == "anthropic":
-        sonnets = [model for model in models if "sonnet" in (model.get("id") or "").lower()]
-        if sonnets:
-            return max(sonnets, key=model_priority)["id"]
-    if provider == "openai":
-        gpt_models = [model for model in models if "gpt" in (model.get("id") or "").lower()]
-        if gpt_models:
-            return max(gpt_models, key=model_priority)["id"]
-    if provider == "gemini":
-        pro_models = [model for model in models if "pro" in (model.get("id") or "").lower()]
-        if pro_models:
-            return max(pro_models, key=model_priority)["id"]
-    if models:
-        return max(models, key=model_priority)["id"]
-    return ""
+    # fetch_all_models returns lists sorted newest-first, so the default is
+    # simply the top entry.
+    return models[0]["id"] if models else ""
 
 
 def media_type_for_path(image_path):

plotextract/static/app.js

@@ -22,6 +22,9 @@ const copyCodeBtn = document.getElementById("copyCode");
 const toggleKeyBtn = document.getElementById("toggleKey");
 const zipDownload = document.getElementById("zipDownload");
 const resultCards = document.getElementById("resultCards");
+const infoBtn = document.getElementById("infoBtn");
+const infoModal = document.getElementById("infoModal");
+const infoCloseBtn = document.getElementById("infoCloseBtn");
 const APP_BASE = document.body?.dataset?.appBase || "";
 const appUrl = (path) => `${APP_BASE}${path}`;
 
@@ -33,6 +36,7 @@ const MODEL_PLACEHOLDER_ERROR = "Could not load models";
 
 let currentFile = null;
 let isZipUpload = false;
+let currentPreviewObjectUrl = "";
 let pollTimer = null;
 let loadedModelsKey = "";
 let modelsLoaded = false;
@@ -276,24 +280,33 @@ const setErrorDetails = (text) => {
   errorDetails.style.display = "block";
 };
 
+const releasePreviewObjectUrl = () => {
+  if (!currentPreviewObjectUrl) return;
+  URL.revokeObjectURL(currentPreviewObjectUrl);
+  currentPreviewObjectUrl = "";
+};
+
 const setPreview = (file) => {
+  releasePreviewObjectUrl();
   if (!file) {
     preview.innerHTML = "<p class=\"hint\">Preview will appear here.</p>";
     return;
   }
   const url = URL.createObjectURL(file);
+  currentPreviewObjectUrl = url;
   preview.innerHTML = "";
   const img = document.createElement("img");
   img.src = url;
-  img.onload = () => URL.revokeObjectURL(url);
   preview.appendChild(img);
 };
 
 const setPreviewMessage = (text) => {
+  releasePreviewObjectUrl();
   preview.innerHTML = `<p class="hint">${text}</p>`;
 };
 
 const setZipPreview = (fileName) => {
+  releasePreviewObjectUrl();
   preview.innerHTML = `
     <div class="zip-preview">
       <svg viewBox="0 0 24 24" aria-hidden="true">
@@ -2626,6 +2639,32 @@ setModelPlaceholder(MODEL_PLACEHOLDER_NO_KEY);
 
 improveManualBtn?.addEventListener("click", startManualImprove);
 
+const openInfoModal = () => {
+  if (!infoModal) return;
+  infoModal.classList.remove("hidden");
+  infoCloseBtn?.focus();
+};
+
+const closeInfoModal = () => {
+  if (!infoModal) return;
+  infoModal.classList.add("hidden");
+  infoBtn?.focus();
+};
+
+infoBtn?.addEventListener("click", openInfoModal);
+infoCloseBtn?.addEventListener("click", closeInfoModal);
+infoModal?.addEventListener("click", (event) => {
+  if (event.target === infoModal) {
+    closeInfoModal();
+  }
+});
+
+document.addEventListener("keydown", (event) => {
+  if (event.key === "Escape" && infoModal && !infoModal.classList.contains("hidden")) {
+    closeInfoModal();
+  }
+});
+
 imageInput.addEventListener("change", (event) => {
   const file = event.target.files[0];
   acceptFile(file);

plotextract/static/styles.css

@@ -51,9 +51,18 @@ body {
 
 .hero {
   display: grid;
-  grid-template-columns: repeat(auto-fit, minmax(260px, 1fr));
+  grid-template-columns: minmax(260px, 1fr) auto;
   gap: 24px;
-  align-items: center;
+  align-items: start;
+}
+
+.hero-actions {
+  display: flex;
+  justify-content: flex-end;
+}
+
+.info-button {
+  min-width: 122px;
 }
 
 .brand {
@@ -639,6 +648,65 @@ input[type="file"] {
   border-radius: 9px;
 }
 
+.info-modal-backdrop {
+  position: fixed;
+  inset: 0;
+  z-index: 40;
+  display: grid;
+  place-items: center;
+  padding: 24px;
+  background: rgba(2, 6, 23, 0.78);
+  backdrop-filter: blur(8px);
+}
+
+.info-modal {
+  width: min(680px, 100%);
+  max-height: min(760px, 88vh);
+  overflow: auto;
+  padding: 24px;
+  border: 1px solid rgba(148, 163, 184, 0.28);
+  border-radius: 16px;
+  background: rgba(15, 23, 42, 0.97);
+  box-shadow: var(--shadow);
+}
+
+.info-modal-head {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 16px;
+  margin-bottom: 16px;
+}
+
+.info-modal h2,
+.info-modal h3 {
+  margin: 0;
+}
+
+.info-modal h3 {
+  font-size: 1rem;
+}
+
+.info-modal-body {
+  display: grid;
+  gap: 18px;
+}
+
+.info-modal p,
+.info-modal li {
+  color: var(--muted);
+  line-height: 1.55;
+}
+
+.info-modal p {
+  margin: 8px 0 0;
+}
+
+.info-modal ol {
+  margin: 10px 0 0;
+  padding-left: 22px;
+}
+
 .manual-modal-actions {
   display: flex;
   justify-content: flex-end;
@@ -892,7 +960,24 @@ textarea {
   font-size: 0.9rem;
 }
 
+.footer a {
+  color: var(--accent-2);
+  text-decoration: none;
+}
+
+.footer a:hover {
+  text-decoration: underline;
+}
+
 @media (max-width: 720px) {
+  .hero {
+    grid-template-columns: 1fr;
+  }
+
+  .hero-actions {
+    justify-content: flex-start;
+  }
+
   .actions {
     flex-direction: column;
     align-items: flex-start;

plotextract/templates/index.html

@@ -25,6 +25,9 @@
             Upload or paste a plot image, supply your Anthropic, OpenAI, or Google Gemini API key, and get back CSV data.
           </p>
         </div>
+        <div class="hero-actions">
+          <button class="ghost info-button" id="infoBtn" type="button">Information</button>
+        </div>
       </header>
 
       <section class="panel">
@@ -115,9 +118,52 @@
           issues and are not guaranteed to be correct. Always review and verify results before using them for
           analysis, reporting, or decision-making.
         </p>
+        <p>
+          Reference: Maciej P. Polak and Dane Morgan,
+          <a href="https://arxiv.org/abs/2503.12326" target="_blank" rel="noopener">
+            Leveraging Vision Capabilities of Multimodal LLMs for Automated Data Extraction from Plots
+          </a>,
+          arXiv:2503.12326.
+        </p>
       </footer>
     </main>
 
+    <div class="info-modal-backdrop hidden" id="infoModal" role="dialog" aria-modal="true" aria-labelledby="infoTitle">
+      <div class="info-modal">
+        <div class="info-modal-head">
+          <h2 id="infoTitle">Information</h2>
+          <button class="ghost info-close" id="infoCloseBtn" type="button" aria-label="Close information">Close</button>
+        </div>
+        <div class="info-modal-body">
+          <section>
+            <h3>What this tool does</h3>
+            <p>
+              PlotExtract uses vision-capable language models to read a plot image, infer the plotted data,
+              return CSV data, and generate Python code for a recreated plot.
+            </p>
+          </section>
+          <section>
+            <h3>When to use it</h3>
+            <p>
+              Use it for typical 2D XY plots when you need approximate data from a figure and the original
+              data file is not available. It works best with clear axis labels, readable tick values, and visible
+              markers or curves.
+            </p>
+          </section>
+          <section>
+            <h3>Instructions</h3>
+            <ol>
+              <li>Enter an Anthropic, OpenAI, or Google Gemini API key and choose a vision-capable model.</li>
+              <li>Upload, drag, or paste a PNG/JPG plot image, or upload a ZIP of images.</li>
+              <li>Click Extract Data and review the CSV, generated Python, and replotted output.</li>
+              <li>Use Improve manually to calibrate axes and drag, add, or remove extracted points.</li>
+              <li>Always verify the result before using it for analysis, reporting, or decisions.</li>
+            </ol>
+          </section>
+        </div>
+      </div>
+    </div>
+
     <script src="{{ url_for('static', filename='app.js') }}"></script>
   </body>
 </html>