many styling fixes. cluster label improvements, job ux. update cluste…

…r labels manually on explore. mobile warning page

latentscope/models/chat_models.json

@@ -1,25 +1,7 @@
 [
-    {
-        "provider": "transformers",
-        "name": "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
-        "sanitized_name": "TinyLlama___TinyLlama-1.1B-Chat-v1.0",
-        "id": "transformers-TinyLlama___TinyLlama-1.1B-Chat-v1.0",
-        "params": {
-            "max_tokens": 2048
-        }
-    },
-    {
-      "provider": "transformers",
-      "name": "HuggingFaceH4/zephyr-7b-beta",
-      "sanitized_name": "HuggingFaceH4___zephyr-7b-beta",
-      "id": "transformers-HuggingFaceH4___zephyr-7b-beta",
-      "params": {
-        "max_tokens": 4096
-      }
-    },
     {
         "provider": "openai",
-        "name": "gpt-3.5-turbo",
+        "name": "gpt-3.5-turbo-0125",
         "id": "openai-gpt-3.5-turbo",
         "params": {
             "max_tokens": 4096 
@@ -58,24 +40,22 @@
       }
     },
     {
-        "provider": "cohereai",
-        "name": "embed-english-v3.0",
-        "id": "cohereai-embed-english-v3.0",
+        "provider": "transformers",
+        "name": "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
+        "sanitized_name": "TinyLlama___TinyLlama-1.1B-Chat-v1.0",
+        "id": "transformers-TinyLlama___TinyLlama-1.1B-Chat-v1.0",
         "params": {
-            "input_type": "clustering"
+            "max_tokens": 2048
         }
     },
     {
-        "provider": "togetherai",
-        "name": "togethercomputer/m2-bert-80M-2k-retrieval",
-        "sanitized_name": "togethercomputer___m2-bert-80M-2k-retrieval",
-        "id": "togetherai-togethercomputer___m2-bert-80M-2k-retrieval",
-        "modality": "text",
-        "params": {
-            "truncation": true,
-            "padding": true,
-            "max_tokens": 2048
-        }
+      "provider": "transformers",
+      "name": "HuggingFaceH4/zephyr-7b-beta",
+      "sanitized_name": "HuggingFaceH4___zephyr-7b-beta",
+      "id": "transformers-HuggingFaceH4___zephyr-7b-beta",
+      "params": {
+        "max_tokens": 4096
+      }
     }
 
 ]

latentscope/scripts/embed.py

@@ -112,7 +112,7 @@ def embed(dataset_id, text_column, model_id, prefix, rerun):
 
     for i, batch in enumerate(tqdm(chunked_iterable(sentences, batch_size), total=total_batches)):
         if i < starting_batch:
-            print(f"skipping batch {i}/{total_batches}")
+            print(f"skipping batch {i}/{total_batches}", flush=True)
             continue
         try:
             embeddings = np.array(model.embed(batch))

latentscope/scripts/label_clusters.py

@@ -64,7 +64,7 @@ def labeler(dataset_id, text_column="text", cluster_id="cluster-001", model_id="
     if rerun is not None:
         label_id = rerun
         clusters = pd.read_parquet(os.path.join(cluster_dir, f"{label_id}.parquet"))
-        print(clusters.columns)
+        # print(clusters.columns)
         # find the first row where labeled isnt True
         unlabeled_row = clusters[~clusters['labeled']].first_valid_index()
         print(f"First unlabeled row: {unlabeled_row}")
@@ -124,7 +124,7 @@ def labeler(dataset_id, text_column="text", cluster_id="cluster-001", model_id="
         # print(batch[0])
         if(unlabeled_row > 0):
             if clusters.loc[i, 'labeled']:
-                print("skipping", i, "already labeled", clusters.loc[i, 'label'])
+                print("skipping", i, "already labeled", clusters.loc[i, 'label'], flush=True)
                 time.sleep(0.01)
                 continue
 
@@ -135,7 +135,7 @@ def labeler(dataset_id, text_column="text", cluster_id="cluster-001", model_id="
             ]
             label = model.chat(messages)
             labels.append(label)
-            print("label:\n", label)
+            # print("label:\n", label)
             # do some cleanup of the labels when the model doesn't follow instructions
             clean_label = label.replace("\n", " ")
             clean_label = clean_label.replace('"', '')
@@ -145,7 +145,7 @@ def labeler(dataset_id, text_column="text", cluster_id="cluster-001", model_id="
             clean_label = " ".join(clean_label.split(" ")[0:5])
             clean_labels.append(clean_label)
 
-            print("clean_label:\n", clean_label)
+            # print("clean_label:\n", clean_label)
             clusters.loc[i, 'label'] = clean_label
             clusters.loc[i, 'label_raw'] = label
             clusters.loc[i, 'labeled'] = True

latentscope/server/app.py

@@ -26,7 +26,7 @@ def check_read_only(s):
     return s.lower() in ['true', '1', 't', 'y', 'yes']
 # export LATENT_SCOPE_READ_ONLY=1  
 READ_ONLY = check_read_only(os.getenv("LATENT_SCOPE_READ_ONLY"))
-print("READ ONLY?", READ_ONLY, not READ_ONLY)
+print("READ ONLY?", READ_ONLY)
 
 # in memory cache of dataframes loaded for each dataset
 # used in returning rows for a given index (indexed, get_tags)

latentscope/server/datasets.py

@@ -25,6 +25,7 @@ def get_datasets():
                 jsonData['id'] = dir
                 datasets.append(jsonData)
 
+    datasets.sort(key=lambda x: x.get('length'))
     return jsonify(datasets)
 
 """
@@ -142,8 +143,33 @@ def get_dataset_cluster_labels(dataset, cluster, model):
     file_name = cluster + "-labels-" + model + ".parquet"
     file_path = os.path.join(DATA_DIR, dataset, "clusters", file_name)
     df = pd.read_parquet(file_path)
+    df.reset_index(inplace=True)
     return df.to_json(orient="records")
 
+@datasets_write_bp.route('/<dataset>/clusters/<cluster>/labels/<model>/label/<index>', methods=['GET'])
+def overwrite_dataset_cluster_label(dataset, cluster, model, index):
+    index = int(index)
+    new_label = request.args.get('label')
+    print("write label", index, new_label)
+    if new_label is None:
+        return jsonify({"error": "Missing 'label' in request data"}), 400
+
+    file_name = cluster + "-labels-" + model + ".parquet"
+    file_path = os.path.join(DATA_DIR, dataset, "clusters", file_name)
+    try:
+        df = pd.read_parquet(file_path)
+    except FileNotFoundError:
+        return jsonify({"error": "File not found"}), 404
+
+    if index >= len(df):
+        return jsonify({"error": "Index out of range"}), 400
+
+    df.at[index, 'label'] = new_label
+    df.to_parquet(file_path)
+
+    return jsonify({"success": True, "message": "Label updated successfully"})
+
+
 @datasets_bp.route('/<dataset>/clusters/<cluster>/labels_available', methods=['GET'])
 def get_dataset_cluster_labels_available(dataset, cluster):
     directory_path = os.path.join(DATA_DIR, dataset, "clusters")

latentscope/server/jobs.py

@@ -12,6 +12,8 @@
 jobs_write_bp = Blueprint('jobs_write_bp', __name__)
 DATA_DIR = os.getenv('LATENT_SCOPE_DATA')
 
+TIMEOUT = 60 * 5 # 5 minute timeout TODO: make this a config option
+
 PROCESSES = {}
 
 def run_job(dataset, job_id, command):
@@ -42,8 +44,14 @@ def run_job(dataset, job_id, command):
     process = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True, shell=True)
     PROCESSES[job_id] = process
 
+    last_output_time = time.time()  # Initialize with the current time
+
     while True:
         output = process.stdout.readline()
+        current_time = time.time()  # Update current time on each iteration
+        print(current_time, current_time - last_output_time, TIMEOUT)
+        print("output", output)
+
         if output == '' and process.poll() is not None:
             break
         if output:
@@ -57,6 +65,14 @@ def run_job(dataset, job_id, command):
             job["last_update"] = str(datetime.now())
             with open(progress_file, 'w') as f:
                 json.dump(job, f)
+            last_output_time = current_time
+        elif current_time - last_output_time > TIMEOUT:
+            print(f"Timeout: No output for more than {TIMEOUT} seconds.")
+            print("OUTPUT", output)
+            job["progress"].append(output.strip())
+            job["progress"].append(f"Timeout: No output for more than {TIMEOUT} seconds.")
+            job["status"] = "error"
+            break  # Break the loop
 
     if process.returncode != 0:
         job["status"] = "error"

web/src/App.jsx

@@ -1,4 +1,5 @@
 import { HashRouter as Router, Routes, Route } from 'react-router-dom';
+import Mobile from './pages/Mobile';
 import Home from './components/Home';
 import Explore from './pages/Explore';
 import Setup from './pages/Setup';
@@ -11,18 +12,22 @@ const env = import.meta.env;
 console.log("ENV", env)
 const readonly = import.meta.env.MODE == "read_only"
 
+const isMobileDevice = () => {
+  return /Android|webOS|iPhone|iPad|iPod|BlackBerry|IEMobile|Opera Mini/i.test(navigator.userAgent);
+};
+
 function App() {
   return (
     <Router basename={env.BASE_NAME}>
       <Nav />
       <div className="page">
       <Routes>
-        <Route path="/" element={<Home />} />
+        <Route path="/" element={isMobileDevice() ? <Mobile/> : <Home />} />
+        <Route path="/datasets/:dataset/explore/:scope" element={isMobileDevice() ? <Mobile/> : <Explore />} />
         {readonly ? null : <Route path="/datasets/:dataset/setup" element={<Setup/>} />}
         {readonly ? null : <Route path="/datasets/:dataset/setup/:scope" element={<Setup/>} />}
         {readonly ? null : <Route path="/datasets/:dataset/jobs" element={<Jobs />} />}
         {readonly ? null : <Route path="/datasets/:dataset/jobs/:job" element={<Job />} />}
-        <Route path="/datasets/:dataset/explore/:scope" element={<Explore />} />
       </Routes>
       </div>
     </Router>

web/src/components/AnnotationPlot.jsx

@@ -36,6 +36,7 @@ const AnnotationPlot = ({
       let rw = zScale(size)
       if(!points.length) return
       points.map(point => {
+        if(!point) return;
         if(fill)
           ctx.fillRect(xScale(point[0]) - rw/2, yScale(point[1]) - rw/2, rw, rw)
         if(stroke)

web/src/components/Job/Progress.jsx

@@ -6,11 +6,12 @@ import './Progress.css';
 JobProgress.propTypes = {
   job: PropTypes.object,
   onlyLast: PropTypes.bool,
-  clearJob: PropTypes.func.isRequired,
+  clearJob: PropTypes.func,
   rerunJob: PropTypes.func,
+  killJob: PropTypes.func,
 };
 
-function JobProgress({job, onlyLast, clearJob, rerunJob}) {
+function JobProgress({job, onlyLast, clearJob, rerunJob, killJob}) {
   const preRef = useRef(null);
 
   useEffect(() => {
@@ -19,6 +20,8 @@ function JobProgress({job, onlyLast, clearJob, rerunJob}) {
     }
   }, [job]);
 
+  const secondsSinceLastUpdate = Math.round((+new Date() - +new Date(job?.last_update)) / 1000)
+
   return (
     <>
       { job ? <div className='job-progress'>
@@ -30,13 +33,18 @@ function JobProgress({job, onlyLast, clearJob, rerunJob}) {
         job.progress.join("\n") 
       } 
       </pre>
-      {job.status == "completed" ? <button onClick={clearJob}>👍</button> : null }
+      {clearJob && job.status == "completed" ? <button onClick={clearJob}>👍 Dismiss</button> : null }
+      {killJob && job.status == "running" ? <button onClick={() => {killJob(job)}}>💀 Kill</button> : null}
       {job.status == "error" ? 
         <div className="error-choices">
-          <button onClick={clearJob}>🤬</button> 
-          {rerunJob ? <button onClick={() => rerunJob(job)}>Rerun</button>  : null }
+          {clearJob ? <button onClick={clearJob}>🤬 Dismiss</button> : null }
+          {rerunJob ? <button onClick={() => rerunJob(job)}>🔁 Rerun</button>  : null }
         </div>
       : null }
+      <span className="timer">
+        {secondsSinceLastUpdate} seconds since last update
+
+      </span>
       </div>
       : <></> }
     </>

web/src/components/Job/Run.jsx

@@ -1,7 +1,7 @@
 import { useState, useEffect, useCallback } from 'react';
 const apiUrl = import.meta.env.VITE_API_URL
 
-function jobPolling(dataset, setJob, jobId, intervalms = 200) {
+function jobPolling(dataset, setJob, jobId, intervalms = 500) {
   let intervalId = null
   console.log("start polling", jobId);
   intervalId = setInterval(() => {
@@ -31,7 +31,7 @@ function jobPolling(dataset, setJob, jobId, intervalms = 200) {
         setJob(null)
         // TODO: have some kind of error state persist
       });
-  }, 200);
+  }, intervalms);
   console.log("returning jobPolling cleanup")
   return () => {
     console.log("inside cleanup", intervalId)
@@ -41,7 +41,7 @@ function jobPolling(dataset, setJob, jobId, intervalms = 200) {
   }
 }
 
-function useStartJobPolling(dataset, setJob, url, intervalms = 200) {
+function useStartJobPolling(dataset, setJob, url, intervalms = 500) {
   // const [cleanup, setCleanup] = useState(() => {})
   const startJob = useCallback((params) => {
     fetch(`${url}?dataset=${dataset.id}&${new URLSearchParams(params)}`)

web/src/components/Setup/ClusterLabels.jsx

@@ -18,11 +18,12 @@ ClusterLabels.propTypes = {
   onLabels: PropTypes.func,
   onLabelIds: PropTypes.func,
   onHoverLabel: PropTypes.func,
+  onClickLabel: PropTypes.func,
 };
 
 // This component is responsible for the embeddings state
 // New embeddings update the list
-function ClusterLabels({ dataset, cluster, selectedLabelId, onChange, onLabels, onLabelIds, onHoverLabel}) {
+function ClusterLabels({ dataset, cluster, selectedLabelId, onChange, onLabels, onLabelIds, onHoverLabel, onClickLabel}) {
   const [clusterLabelsJob, setClusterLabelsJob] = useState(null);
   const { startJob: startClusterLabelsJob } = useStartJobPolling(dataset, setClusterLabelsJob, `${apiUrl}/jobs/cluster_label`);
 
@@ -43,6 +44,7 @@ function ClusterLabels({ dataset, cluster, selectedLabelId, onChange, onLabels,
   // the actual labels for the given cluster
   const [clusterLabels, setClusterLabels] = useState([]);
   useEffect(() => {
+    console.log("in cluster labels", dataset, cluster, selectedLabelId)
     if(dataset && cluster && selectedLabelId) {
       fetch(`${apiUrl}/datasets/${dataset.id}/clusters/${cluster.id}/labels/${selectedLabelId}`)
         .then(response => response.json())
@@ -62,11 +64,13 @@ function ClusterLabels({ dataset, cluster, selectedLabelId, onChange, onLabels,
       fetch(`${apiUrl}/datasets/${dataset.id}/clusters/${cluster.id}/labels_available`)
         .then(response => response.json())
         .then(data => {
-          console.log("cluster changed, set label models fetched", cluster.id, data)
-          if(clusterLabelsJob && clusterLabelsJob.status == "completed") {
+          console.log("cluster changed, set label models fetched", cluster.id, data, clusterLabelsJob)
+          if(clusterLabelsJob) {
             let lbl;
-            if(clusterLabelsJob?.job_name == "cluster_label"){
-              lbl = data.find(d => d.id == clusterLabelsJob.run_id)
+            if(clusterLabelsJob?.job_name == "label"){
+              let label_id = clusterLabelsJob.run_id.split("-")[3]
+              lbl = data.find(d => d == label_id)
+              console.log("label_id", label_id, lbl)
             } else if(clusterLabelsJob.job_name == "rm") {
               lbl = data[0]
             }
@@ -123,24 +127,28 @@ function ClusterLabels({ dataset, cluster, selectedLabelId, onChange, onLabels,
           <button type="submit" disabled={!!clusterLabelsJob || !cluster}>Auto Label</button>
         </form>
 
-        <JobProgress job={clusterLabelsJob} clearJob={()=>setClusterLabelsJob(null)} />
+        <JobProgress job={clusterLabelsJob} clearJob={()=>setClusterLabelsJob(null)} killJob={setClusterLabelsJob} />
 
       </div>
       {cluster ? <div className="dataset--setup-cluster-labels-list">
         <label>
-          View Labels:
-          <select 
+          Use Labels: &nbsp;
+          {clusterLabelModels.length > 1 ? <select 
             name="model" 
             value={selectedLabelId}
             onChange={(e) => onChange(e.target.value)}
           >
             {clusterLabelModels.map((model, index) => (
               <option key={index} value={model}>{model}</option>
             ))}
-          </select>
+          </select> : <span>{clusterLabelModels[0]}</span> }
         </label>
         <div className="dataset--setup-labels-list">
-          <DataTable data={clusterLabels.map((d,i) => ({index: i, label: d.label, items: d.indices.length}))} onHover={(index) => onHoverLabel(clusterLabels[index])}/>
+          <DataTable 
+            data={clusterLabels.map((d,i) => ({index: i, label: d.label, items: d.indices.length}))} 
+            onHover={(index) => onHoverLabel(clusterLabels[index])}
+            onClick={(index) => onClickLabel(clusterLabels[index])}
+          />
         </div>
       </div> : null}
     </div>