Add test_history.py: multi-turn conversation memory test

test_history.py

@@ -0,0 +1,70 @@
+"""
+Test conversation memory: send a follow-up message with history attached
+and verify all three providers handle the new multi-turn format correctly.
+
+Usage:  python test_history.py
+"""
+
+import sys
+import io
+import time
+
+sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding="utf-8", errors="replace")
+
+from config import LLM_OPTIONS
+from rag_chain import stream_llm, retrieve_docs
+
+Q1 = "What is Nietzsche's view on the death of God?"
+Q2 = "How does this relate to the concept of nihilism?"   # follow-up — needs history
+
+# Fake first-turn history the way Gradio passes it back
+MOCK_HISTORY = [
+    {"role": "user", "content": Q1},
+    {
+        "role": "assistant",
+        "content": (
+            "Nietzsche argues that the death of God signals the collapse of "
+            "transcendent moral frameworks, forcing humanity to create its own values. "
+            "As he writes in <em>Thus Spoke Zarathustra</em>, the Übermensch "
+            "must fill the void left by divine authority through acts of self-overcoming."
+            "\n\n---\n**Sources:**\n- `0.30` Nietzsche — *Thus Spoke Zarathustra*"
+        ),
+    },
+]
+
+# Pick one representative model per provider
+PROBE_MODELS = {
+    label: (provider, mid)
+    for label, (provider, mid) in LLM_OPTIONS.items()
+    if label in (
+        "Gemma 4 MoE 26B  [Google]",
+        "Llama 3.1 8B  [Groq]",
+        "Llama 3.3 70B  [OpenRouter]",
+    )
+}
+
+PASS = "\033[92mPASS\033[0m"
+FAIL = "\033[91mFAIL\033[0m"
+
+docs, scores = retrieve_docs(f"{Q1} {Q2}", "All")
+context_str = "\n\n".join(d.page_content for d in docs)
+
+print(f"\nHistory test — follow-up: \"{Q2}\"\n")
+print(f"{'Model':<45} {'Provider':<12} {'Status':<6} {'Time (s)':>8}  Preview")
+print("-" * 110)
+
+for label, (provider, model_id) in PROBE_MODELS.items():
+    sys.stdout.write(f"  {label:<43} {provider:<12} ... ")
+    sys.stdout.flush()
+    t0 = time.perf_counter()
+    try:
+        chunks = list(stream_llm(provider, model_id, context_str, Q2, history=MOCK_HISTORY))
+        elapsed = time.perf_counter() - t0
+        answer = "".join(chunks).replace("\n", " ").strip()
+        preview = answer[:90] + ("…" if len(answer) > 90 else "")
+        print(f"\r  {label:<43} {provider:<12} {PASS}  {elapsed:>8.2f}s  {preview}")
+    except Exception as exc:
+        elapsed = time.perf_counter() - t0
+        print(f"\r  {label:<43} {provider:<12} {FAIL}  {elapsed:>8.2f}s  {str(exc)[:90]}")
+
+print()