Such slow progress...

.env.example

@@ -1,10 +1,19 @@
+
+# Frontend/backend stuff
 API_BASE_URL=http://localhost:8000                  # Temporary btw
 OLLAMA_BASE_URL=http://127.0.0.1:11434              # Ollama's default
 PRECIS_ALLOWED_ORIGINS=http://localhost:5173        # Just front end, might make it more/less strict but I just need something consistent rn
+
+# Model addresses
+DEFAULT_MODEL=phi4-mini:latest
+
+# Update this if you add/remove models
+# Or if you custom-name any of them
+AVAILABLE_MODELS=phi4-mini:latest,qwen3:4b           # Only here so both front and backend have it
+
+# API stuff
 PRECIS_API_KEY=replace-with-a-long-random-secret    # Once the API is actually up, this'll be needed
-PRECIS_DEFAULT_MODEL=phi4-mini:latest
-PRECIS_AVAILABLE_MODELS=phi4-mini:latest,qwen:4b    # Only here so both front and backend have it
 MAX_SUMMARY_TOKENS=120
-TEMPERATURE=0.2                                     # Random choice right now, will probably tweak
+TEMPERATURE=0.2                                     # Randomly selected, will probably tweak later
 PRECIS_MAX_UPLOAD_BYTES=10485760
 PRECIS_MAX_TRANSCRIPT_CHARS=250000

.github/workflows/hf.yml

@@ -12,4 +12,4 @@ jobs:
       - name: Push to hub
         env:
           HF_TOKEN: ${{ secrets.HF_TOKEN }}
-        run: git push --force https://compendious:${HF_TOKEN}@huggingface.co/spaces/compendious/precis main
+        run: git push https://compendious:${HF_TOKEN}@huggingface.co/spaces/compendious/precis main

.gitignore

@@ -3,10 +3,18 @@
 **cache**
 *.ipynb
 *.venv
+temp.py
+
+# Secrets
 .env
 .env.*
 !.env.example
 
+# Leftovers
+*.csv
+*.json*
+/*data*/
+
 # Front end
 node_modules
 package-lock.json

README.md

@@ -27,27 +27,27 @@ All `/summarize/*` endpoints accept an optional `model` field to override the de
 
 ### Prerequisites
 
-- Python 3.11+
-- Node.js 18+ (or [Bun](https://bun.sh))
-- [Ollama](https://ollama.com) installed and running (`ollama serve`)
-- At least one model pulled: `ollama pull phi4-mini:latest`
+- Python 3.11+,
+- Node.js 18+ (or an alternative like [Bun](https://bun.sh)),
+- [Ollama](https://ollama.com) installed and running (`ollama serve` is the command, although it may be on auto-start).
+- At least one model pulled: `ollama pull phi4-mini:latest` (for example)
 
 ### Run the Fine-Tuning
 
-Follow the scripts in `scripts/`, using any model you prefer. This project has been primarily tested with phi4-mini (from Microsoft) and Qwen 3-3b (from Alibaba).
+Follow the scripts in `scripts/`, using any model you prefer. This project has been primarily tested with phi4-mini (from Microsoft) and Qwen 3-4b (from Alibaba) (`ollama pull qwen3:4b` to pull it).
 
-### Backend
+### Start the Backend
 
 ```bash
-cd backend
 # Create a venv or conda environment or whatever else you may want
 pip install -r ../requirements.txt
+cd backend
 uvicorn app:app --reload
 ```
 
 Runs on `http://localhost:8000`. Interactive docs at `/docs`.
 
-### Frontend
+### Run the Frontend
 
 ```bash
 cd frontend

backend/app.py

@@ -42,6 +42,16 @@ def verify_api_key(x_api_key: Optional[str] = Header(default=None, alias="X-API-
         raise HTTPException(status_code=401, detail="Invalid API key.")
 
 
+@app.get("/")
+async def root():
+    return {
+        "service": "Précis API",
+        "docs": "/docs",
+        "health": "/health",
+        "status": "/status",
+    }
+
+
 @app.get("/health")
 async def health():
     return {"status": "healthy", "service": "precis"}
@@ -68,6 +78,18 @@ async def status():
 
 @app.get("/models")
 async def list_models():
+    try:
+        async with httpx.AsyncClient(timeout=5.0) as client:
+            r = await client.get(f"{OLLAMA_BASE_URL}/api/tags")
+            r.raise_for_status()
+            payload = r.json() if r.content else {}
+            installed = [m.get("name") for m in payload.get("models", []) if m.get("name")]
+            if installed:
+                default = DEFAULT_MODEL if DEFAULT_MODEL in installed else installed[0]
+                return {"default": default, "available": installed}
+    except Exception:
+        pass
+
     return {"default": DEFAULT_MODEL, "available": AVAILABLE_MODELS}
 
 

backend/config.py

@@ -23,9 +23,9 @@ def _required_env(name: str) -> str:
 	return value
 
 
-OLLAMA_BASE_URL = _required_env("PRECIS_OLLAMA_BASE_URL")
-DEFAULT_MODEL = _required_env("PRECIS_DEFAULT_MODEL")
-AVAILABLE_MODELS = _csv_env("PRECIS_AVAILABLE_MODELS", [DEFAULT_MODEL])
+OLLAMA_BASE_URL = _required_env("OLLAMA_BASE_URL")
+DEFAULT_MODEL = _required_env("DEFAULT_MODEL")
+AVAILABLE_MODELS = _csv_env("AVAILABLE_MODELS", [DEFAULT_MODEL])
 if DEFAULT_MODEL not in AVAILABLE_MODELS:
 	AVAILABLE_MODELS = [DEFAULT_MODEL, *AVAILABLE_MODELS]
 
@@ -35,7 +35,11 @@ if not ALLOWED_ORIGINS:
 
 API_KEY = _required_env("PRECIS_API_KEY")
 
-MAX_SUMMARY_TOKENS = int(os.getenv("PRECIS_MAX_SUMMARY_TOKENS", "120"))
-TEMPERATURE = float(os.getenv("PRECIS_TEMPERATURE", "0.2"))
+MAX_SUMMARY_TOKENS = int(
+	os.getenv("MAX_SUMMARY_TOKENS", os.getenv("PRECIS_MAX_SUMMARY_TOKENS", "120"))
+)
+TEMPERATURE = float(
+	os.getenv("TEMPERATURE", os.getenv("PRECIS_TEMPERATURE", "0.2"))
+)
 MAX_UPLOAD_BYTES = int(os.getenv("PRECIS_MAX_UPLOAD_BYTES", "10485760"))
 MAX_TRANSCRIPT_CHARS = int(os.getenv("PRECIS_MAX_TRANSCRIPT_CHARS", "120000"))

backend/ollama.py

@@ -1,4 +1,6 @@
 from typing import Optional
+import json
+import os
 
 import httpx
 from fastapi import HTTPException
@@ -25,50 +27,115 @@ def build_prompt(title: Optional[str], text: str) -> str:
         )
     return (
         f"{instructions}\n"
-        "Do not add opinions, commentary, or filler phrases like 'The article discusses'.\n"
-        "Output the summary sentence only — nothing else.\n\n"
+        "Do not add opinions, commentary, or filler phrases like 'The article discusses' or 'This document provides'.\n"
+        "or ANYTHING along those lines whether it be in meaning or phrasing. "
+        "Output the summary sentence only. The sentence should be no longer than 200 characetrs long. Nothing else should be included.\n\n"
         f"Article:\n{text}\n\n"
         "Summary:"
     )
 
 
 def resolve_model(model: Optional[str]) -> str:
-    if not model:
+    requested = model or ""
+
+    # Prefer what Ollama actually has installed.
+    try:
+        with httpx.Client(timeout=5.0) as client:
+            r = client.get(f"{OLLAMA_BASE_URL}/api/tags")
+            r.raise_for_status()
+            payload = r.json() if r.content else {}
+            installed = [m.get("name") for m in payload.get("models", []) if m.get("name")]
+    except Exception:
+        installed = []
+
+    if installed:
+        if not requested:
+            return DEFAULT_MODEL if DEFAULT_MODEL in installed else installed[0]
+        if requested not in installed:
+            raise HTTPException(
+                status_code=400,
+                detail=(
+                    f"Model '{requested}' is not installed in Ollama. "
+                    f"Installed: {installed}. Run `ollama pull {requested}`."
+                ),
+            )
+        return requested
+
+    # Fallback: use configured allowlist when Ollama isn't reachable.
+    if not requested:
         return DEFAULT_MODEL
-    if model not in AVAILABLE_MODELS:
+    if requested not in AVAILABLE_MODELS:
         raise HTTPException(
             status_code=400,
-            detail=f"Unknown model '{model}'. Available: {AVAILABLE_MODELS}",
+            detail=f"Unknown model '{requested}'. Available: {AVAILABLE_MODELS}",
+        )
+    return requested
+
+
+def ensure_ollama_reachable() -> None:
+    try:
+        with httpx.Client(timeout=10.0) as client:
+            response = client.get(f"{OLLAMA_BASE_URL}/api/tags")
+            response.raise_for_status()
+    except httpx.ConnectError:
+        raise HTTPException(
+            status_code=503,
+            detail="Cannot reach Ollama. Make sure `ollama serve` is running.",
+        )
+    except httpx.HTTPError as exc:
+        raise HTTPException(
+            status_code=503,
+            detail=f"Ollama responded with an error: {exc}",
         )
-    return model
 
 
 async def ollama_stream(prompt: str, model: str):
-    """Async generator: yields raw NDJSON lines from Ollama."""
+    """Async generator: yields NDJSON lines from Ollama, filtering out thinking-only chunks."""
+    keep_alive = os.getenv("OLLAMA_KEEP_ALIVE", "30m")
+    # Set num_predict high so thinking tokens don't limit output.
+    num_predict = MAX_SUMMARY_TOKENS * 3
     payload = {
         "model": model,
         "prompt": prompt,
         "stream": True,
+        "keep_alive": keep_alive,
         "options": {
-            "num_predict": MAX_SUMMARY_TOKENS,
+            "num_predict": num_predict,
             "temperature": TEMPERATURE,
-            "stop": ["\n\n", "Article:", "Title:"],
+            "stop": ["Article:", "Title:"],
         },
     }
-    async with httpx.AsyncClient(timeout=120.0) as client:
+    async with httpx.AsyncClient(timeout=300.0) as client:
         try:
             async with client.stream(
                 "POST", f"{OLLAMA_BASE_URL}/api/generate", json=payload,
             ) as resp:
                 resp.raise_for_status()
                 async for line in resp.aiter_lines():
-                    if line:
+                    if not line:
+                        continue
+                    try:
+                        chunk = json.loads(line)
+                        # Skips thinking-only chunks.
+                        if chunk.get("response"):
+                            yield line + "\n"
+                    except json.JSONDecodeError:
                         yield line + "\n"
         except httpx.ConnectError:
-            raise HTTPException(
-                status_code=503,
-                detail="Cannot reach Ollama. Make sure `ollama serve` is running.",
-            )
+            error_line = json.dumps({
+                "error": "Cannot reach Ollama. Make sure `ollama serve` is running.",
+            })
+            yield error_line + "\n"
+        except httpx.TimeoutException:
+            error_line = json.dumps({
+                "error": "Ollama timed out. The model may still be loading — try again in a moment.",
+            })
+            yield error_line + "\n"
+        except httpx.HTTPError as exc:
+            error_line = json.dumps({
+                "error": f"Ollama error: {exc}",
+            })
+            yield error_line + "\n"
 
 
 def stream_summary(
@@ -77,6 +144,7 @@ def stream_summary(
     model: Optional[str] = None,
 ) -> StreamingResponse:
     """Universal funnel: text -> prompt -> Ollama stream -> NDJSON response."""
+    ensure_ollama_reachable()
     resolved = resolve_model(model)
     prompt = build_prompt(title, text)
     return StreamingResponse(

frontend/src/App.jsx

@@ -1,8 +1,8 @@
-import { useState, useRef } from 'react'
+import { useEffect, useState, useRef } from 'react'
 import InlineResult from './components/InlineResult'
 import { useStreaming } from './hooks/useStreaming'
 import logoSvg from './assets/logo.svg'
-import { API_BASE, AVAILABLE_MODELS, DEFAULT_MODEL } from './config'
+import { API_BASE } from './config'
 import './App.css'
 
 function App() {
@@ -10,13 +10,40 @@ function App() {
   const [youtubeUrl, setYoutubeUrl] = useState('')
   const [transcript, setTranscript] = useState('')
   const [selectedFile, setSelectedFile] = useState(null)
-  const [selectedModel, setSelectedModel] = useState(DEFAULT_MODEL)
+  const [models, setModels] = useState([])
+  const [selectedModel, setSelectedModel] = useState('')
   const fileInputRef = useRef(null)
 
   const { loading, response, error, streamingText, submit } = useStreaming()
 
+  useEffect(() => {
+    let cancelled = false
+    ;(async () => {
+      try {
+        const res = await fetch(`${API_BASE}/models`)
+        if (!res.ok) return
+        const data = await res.json()
+        if (cancelled) return
+
+        const available = Array.isArray(data.available) ? data.available : []
+        setModels(available)
+
+        const serverDefault = typeof data.default === 'string' ? data.default : ''
+        setSelectedModel((prev) => prev || serverDefault || available[0] || '')
+      } catch {
+        // Non-fatal: model list stays empty; backend will still pick default if model omitted.
+      }
+    })()
+    return () => { cancelled = true }
+  }, [])
+
   const handleSubmit = () =>
-    submit(activeTab, { youtubeUrl, transcript, selectedFile, selectedModel })
+    submit(activeTab, {
+      youtubeUrl,
+      transcript,
+      selectedFile,
+      selectedModel: selectedModel || undefined,
+    })
 
   const handleFileDrop = (e) => {
     e.preventDefault()
@@ -50,9 +77,9 @@ function App() {
             className="model-select"
             value={selectedModel}
             onChange={(e) => setSelectedModel(e.target.value)}
-            disabled={loading}
+            disabled={loading || models.length === 0}
           >
-            {AVAILABLE_MODELS.map((m) => <option key={m} value={m}>{m}</option>)}
+            {models.map((m) => <option key={m} value={m}>{m}</option>)}
           </select>
           <a href={`${API_BASE}/docs`} target="_blank" rel="noopener noreferrer" className="btn" style={{ textDecoration: 'none' }}>
             API Docs
@@ -102,7 +129,13 @@ function App() {
                     />
                     <p className="form-hint">Paste a YouTube URL. Ctrl+Enter to generate.</p>
                   </div>
-                  {activeTab === 'youtube' && <InlineResult {...resultProps} loadingLabel="Fetching transcript…" />}
+                  {activeTab === 'youtube' && (
+                    <InlineResult
+                      {...resultProps}
+                      loadingLabel="Fetching transcript…"
+                      placeholderText="Fetching transcript…"
+                    />
+                  )}
                 </div>
 
                 {/* Transcript */}
@@ -124,7 +157,13 @@ function App() {
                       {' '}to generate.
                     </p>
                   </div>
-                  {activeTab === 'transcript' && <InlineResult {...resultProps} loadingLabel="Generating…" />}
+                  {activeTab === 'transcript' && (
+                    <InlineResult
+                      {...resultProps}
+                      loadingLabel="Generating…"
+                      placeholderText="Waiting for model…"
+                    />
+                  )}
                 </div>
 
                 {/* File upload */}
@@ -162,7 +201,13 @@ function App() {
                       </div>
                     )}
                   </div>
-                  {activeTab === 'file' && <InlineResult {...resultProps} loadingLabel="Reading file…" />}
+                  {activeTab === 'file' && (
+                    <InlineResult
+                      {...resultProps}
+                      loadingLabel="Reading file…"
+                      placeholderText="Reading file…"
+                    />
+                  )}
                 </div>
 
                 <div className="submit-section">

frontend/src/components/InlineResult.jsx

@@ -1,4 +1,4 @@
-export default function InlineResult({ error, loading, response, streamingText, selectedModel, loadingLabel }) {
+export default function InlineResult({ error, loading, response, streamingText, selectedModel, loadingLabel, placeholderText }) {
     return (
         <>
             {error && (
@@ -17,7 +17,11 @@ export default function InlineResult({ error, loading, response, streamingText,
                         <span className="response-badge" style={{ marginLeft: 'auto' }}>{selectedModel}</span>
                     </div>
                     <p className="inline-result__text">
-                        {streamingText || <span className="streaming-placeholder">Waiting for model…</span>}
+                        {streamingText || (
+                            <span className="streaming-placeholder">
+                                {placeholderText || loadingLabel || 'Waiting for model…'}
+                            </span>
+                        )}
                         <span className="streaming-cursor">▌</span>
                     </p>
                 </div>

frontend/src/config.js

@@ -1,24 +1,21 @@
-const parseCsv = (raw, fallback = []) => {
-  if (!raw || !raw.trim()) return fallback
-  return raw.split(',').map((part) => part.trim()).filter(Boolean)
-}
-
-const requiredEnv = (name) => {
-  const value = import.meta.env[name]
-  if (!value || !String(value).trim()) {
-    throw new Error(`Missing required environment variable: ${name}`)
+const requiredEnv = (names) => {
+  const list = Array.isArray(names) ? names : [names]
+  for (const name of list) {
+    const value = import.meta.env[name]
+    if (value && String(value).trim()) return String(value).trim()
   }
-  return String(value).trim()
+  throw new Error(`Missing required environment variable. Tried: ${list.join(', ')}`)
 }
 
-export const API_BASE = requiredEnv('PRECIS_API_BASE_URL')
-export const API_KEY = requiredEnv('PRECIS_API_KEY')
-
-export const DEFAULT_MODEL = requiredEnv('PRECIS_DEFAULT_MODEL')
-export const AVAILABLE_MODELS = parseCsv(
-  import.meta.env.PRECIS_AVAILABLE_MODELS,
-  [DEFAULT_MODEL],
-)
+export const API_BASE = requiredEnv([
+  'API_BASE_URL',
+  'VITE_API_BASE_URL',
+  'PRECIS_API_BASE_URL',
+])
+export const API_KEY = requiredEnv([
+  'PRECIS_API_KEY',
+  'VITE_API_KEY',
+])
 
 export const authHeaders = (headers = {}) => (
   API_KEY ? { ...headers, 'X-API-Key': API_KEY } : headers

frontend/src/hooks/useStreaming.js

@@ -13,6 +13,7 @@ export function useStreaming() {
         const decoder = new TextDecoder()
         let accumulated = ''
         let buffer = ''
+        let streamError = null
 
         while (true) {
             const { done, value } = await reader.read()
@@ -26,6 +27,10 @@ export function useStreaming() {
                 if (!line.trim()) continue
                 try {
                     const chunk = JSON.parse(line)
+                    if (chunk.error) {
+                        streamError = String(chunk.error)
+                        continue
+                    }
                     if (chunk.response) {
                         accumulated += chunk.response
                         setStreamingText(accumulated)
@@ -34,7 +39,16 @@ export function useStreaming() {
             }
         }
 
-        return accumulated.trim()
+        if (streamError) {
+            throw new Error(streamError)
+        }
+
+        const finalText = accumulated.trim()
+        if (!finalText) {
+            throw new Error('Model returned an empty response. Try again or pick a different model.')
+        }
+
+        return finalText
     }
 
     const streamFrom = async (endpoint, { json, formData } = {}) => {

frontend/vite.config.js

@@ -4,6 +4,6 @@ import react from '@vitejs/plugin-react'
 // https://vite.dev/config/
 export default defineConfig({
   envDir: '..',
-  envPrefix: ['VITE_', 'PRECIS_'],
+  envPrefix: ['VITE_', 'PRECIS_', 'API_', 'DEFAULT_', 'AVAILABLE_'],
   plugins: [react()],
 })

requirements.txt

@@ -1,4 +1,4 @@
-# ML Training vibes
+# Model Trainerz
 torch
 transformers
 accelerate
@@ -10,8 +10,10 @@ sentencepiece
 # API
 fastapi
 uvicorn
-httpx                   # async HTTP client for Ollama calls
-python-multipart        # required by FastAPI for file uploads
-youtube-transcript-api  # YouTube transcript fetching
-python-dotenv           # .env loading for backend/frontend config
+httpx                       # async HTTP client for Ollama calls
+python-multipart            # faster file uploads
 
+python-dotenv
+
+# Prop tools
+youtube-transcript-api