Update app.py

app.py

@@ -1,8 +1,11 @@
+# app.py
+
 import streamlit as st
 import logging
 import pandas as pd
 import plotly.express as px
-from models.translation_loader import TranslationLoader
+
+from models.model_manager import ModelManager
 from evaluators.evaluator import TranslationEvaluator
 
 # ────────── Logging ──────────
@@ -13,38 +16,39 @@ logging.basicConfig(
 )
 logger = logging.getLogger(__name__)
 
-# ────────── Cached Loader/Evaluator ──────────
+# ────────── Cached Resources ──────────
 @st.cache_resource
 def load_resources():
-    loader = TranslationLoader(
-        model_name="facebook/nllb-200-distilled-600M",
-        quantize=True
-    )
+    """
+    Load and cache the model manager and evaluator on first run.
+    """
+    manager   = ModelManager(quantize=True)
     evaluator = TranslationEvaluator()
-    return loader, evaluator
+    return manager, evaluator
 
 # ────────── Sidebar Model Info ──────────
-def display_model_info(info):
+def display_model_info(info: dict):
     st.sidebar.markdown("### Model Info")
-    st.sidebar.write(f"**Model:** {info['model_name']}")
-    st.sidebar.write(f"**8-bit Quantized:** {info['quantized']}")
-    st.sidebar.write(f"**Device:** {info['device']}")
+    st.sidebar.write(f"**Model:** {info.get('model')}")
+    st.sidebar.write(f"**8-bit Quantized:** {info.get('quantized')}")
+    st.sidebar.write(f"**Device:** {info.get('device')}")
+    st.sidebar.write(f"**Default target:** {info.get('default_tgt')}")
 
 # ────────── Single‐text Processing ──────────
-def process_text(src, ref, loader, evaluator, metrics):
-    # 1) Translate
-    out = loader.translate(src, tgt_lang="tur_Latn")
+def process_text(src: str, ref: str, manager: ModelManager, evaluator: TranslationEvaluator, metrics: list):
+    # 1) Translate (auto-detect source, default target Turkish)
+    out = manager.translate(src)  # returns list of dicts
     hyp = out[0]["translation_text"] if isinstance(out, list) else out["translation_text"]
     # 2) Evaluate
     scores = evaluator.evaluate([src], [ref or ""], [hyp])
     return {
-        "source": src,
-        "reference": ref,
+        "source":     src,
+        "reference":  ref,
         "hypothesis": hyp,
         **{m: scores[m] for m in metrics}
     }
 
-def _show_single_results(res):
+def _show_single_results(res: dict):
     left, right = st.columns(2)
     with left:
         st.markdown("**Source:**")
@@ -56,13 +60,19 @@ def _show_single_results(res):
             st.write(res["reference"])
     with right:
         st.markdown("### Scores")
-        df = pd.DataFrame({k: [v] for k, v in res.items() if k in ["BLEU","BERTScore","BERTurk","COMET"]})
+        df = pd.DataFrame([{k: v for k, v in res.items() if k in ["BLEU","BERTScore","BERTurk","COMET"]}])
         st.table(df)
 
 # ────────── Batch‐CSV Processing ──────────
-def process_file(uploaded, loader, evaluator, metrics, batch_size):
+def process_file(
+    uploaded,
+    manager: ModelManager,
+    evaluator: TranslationEvaluator,
+    metrics: list,
+    batch_size: int
+):
     df = pd.read_csv(uploaded)
-    if not {"src","ref_tr"}.issubset(df.columns):
+    if not {"src", "ref_tr"}.issubset(df.columns):
         raise ValueError("CSV must have `src` and `ref_tr` columns")
     prog = st.progress(0)
     results = []
@@ -72,9 +82,9 @@ def process_file(uploaded, loader, evaluator, metrics, batch_size):
         srcs = batch["src"].tolist()
         refs = batch["ref_tr"].tolist()
         # translate batch
-        outs = loader.translate(srcs, tgt_lang="tur_Latn")
+        outs = manager.translate(srcs)  # list of dicts
         hyps = [o["translation_text"] for o in outs]
-        # evaluate each item individually
+        # evaluate each row
         for s, r, h in zip(srcs, refs, hyps):
             sc = evaluator.evaluate([s], [r], [h])
             entry = {"src": s, "ref_tr": r, "hyp_tr": h}
@@ -83,7 +93,7 @@ def process_file(uploaded, loader, evaluator, metrics, batch_size):
         prog.progress(min(i + batch_size, total) / total)
     return pd.DataFrame(results)
 
-def _show_batch_viz(df, metrics):
+def _show_batch_viz(df: pd.DataFrame, metrics: list):
     for m in metrics:
         st.markdown(f"#### {m} Distribution")
         fig = px.histogram(df, x=m)
@@ -93,7 +103,7 @@ def _show_batch_viz(df, metrics):
 def main():
     st.set_page_config(page_title="🔤 Translation→Turkish Quality", layout="wide")
     st.title("🔤 Translation → TR Quality & COMET")
-    st.markdown("Enter text or upload a CSV to translate into Turkish and evaluate with BLEU, BERTScore, BERTurk & COMET.")
+    st.markdown("Translate any language into Turkish and evaluate with BLEU, BERTScore, BERTurk & COMET.")
 
     # Sidebar
     with st.sidebar:
@@ -101,11 +111,11 @@ def main():
         metrics = st.multiselect(
             "Select metrics",
             ["BLEU", "BERTScore", "BERTurk", "COMET"],
-            default=["BLEU","BERTScore","COMET"]
+            default=["BLEU", "BERTScore", "COMET"]
         )
         batch_size = st.slider("Batch size", 1, 32, 8)
-        loader, evaluator = load_resources()
-        display_model_info(loader.get_info())
+        manager, evaluator = load_resources()
+        display_model_info(manager.get_info())
 
     # Tabs
     tab1, tab2 = st.tabs(["Single Sentence", "Batch CSV"])
@@ -115,22 +125,22 @@ def main():
         ref = st.text_area("Turkish reference (optional):", height=100)
         if st.button("Evaluate"):
             with st.spinner("Translating & evaluating…"):
-                res = process_text(src, ref, loader, evaluator, metrics)
+                res = process_text(src, ref, manager, evaluator, metrics)
             _show_single_results(res)
 
     with tab2:
         uploaded = st.file_uploader("Upload CSV with `src` & `ref_tr` columns", type=["csv"])
         if uploaded:
             with st.spinner("Processing file…"):
-                df_res = process_file(uploaded, loader, evaluator, metrics, batch_size)
+                df_res = process_file(uploaded, manager, evaluator, metrics, batch_size)
             st.markdown("### Batch Results")
             st.dataframe(df_res, use_container_width=True)
             _show_batch_viz(df_res, metrics)
-            st.download_button("Download CSV", df_res.to_csv(index=False), "results.csv")
+            st.download_button("Download results as CSV", df_res.to_csv(index=False), "results.csv")
 
 if __name__ == "__main__":
     try:
         main()
     except Exception as e:
         st.error(f"Unexpected error: {e}")
-        logger.exception("Unhandled exception")
+        logger.exception("Unhandled exception in main()")