Replace retired zephyr default with a fallback chain of supported chat models

HuggingFaceH4/zephyr-7b-beta is no longer carried by any Inference
Provider (router returns model_not_supported), which broke the chat
assistant. chat_update now tries a candidate list in order - an
HF_CHAT_MODEL override first, then Qwen2.5-7B-Instruct,
Mistral-7B-Instruct-v0.3, and Qwen2.5-72B-Instruct - and the run
metadata records whichever model actually served the call
(ACTIVE_MODEL) rather than the configured default.

README.md

@@ -92,9 +92,14 @@ The Step 1 assistant calls the Hugging Face Inference API, so it needs a token:
 - **Locally / as a fallback:** paste a token into the "Use your own
   Hugging Face token" accordion in Step 1; it's only used for that session.
 
-The model defaults to `HuggingFaceH4/zephyr-7b-beta` (ungated). Override
-it with the `HF_CHAT_MODEL` environment variable/secret if you'd rather
-use a different instruct model your token can access.
+The token needs **Inference Providers** permission: use the simple
+"Read" token type, or on a fine-grained token check "Make calls to
+Inference Providers".
+
+The assistant tries `Qwen/Qwen2.5-7B-Instruct` first, then falls back to
+`mistralai/Mistral-7B-Instruct-v0.3` and `Qwen/Qwen2.5-72B-Instruct` if
+a provider doesn't carry it. Set an `HF_CHAT_MODEL` env var/Space
+variable to force a specific model instead.
 
 If the chat call fails or no token is available, the rest of the app
 still works fully via the "Advanced: edit settings manually" accordion

app.py

@@ -242,7 +242,7 @@ def process_files(files, categories, custom_keywords_str, custom_fields_str, onl
         "custom_keywords": "; ".join(custom_keywords),
         "custom_fields": "; ".join(custom_field_labels),
         "only_matches_filter": only_matches,
-        "chat_model": chat_config.DEFAULT_MODEL if chat_history else "",
+        "chat_model": chat_config.ACTIVE_MODEL if chat_history else "",
         "chat_transcript": _format_transcript(chat_history),
     }
     metadata_df = pd.DataFrame([metadata_row], columns=METADATA_COLUMNS)

chat_config.py

@@ -12,7 +12,22 @@ import extractors
 
 CATEGORY_CHOICES = list(extractors.KEYWORD_SETS.keys())
 
-DEFAULT_MODEL = os.environ.get("HF_CHAT_MODEL", "HuggingFaceH4/zephyr-7b-beta")
+# Tried in order until one is accepted by an inference provider. An
+# HF_CHAT_MODEL env var/Space variable always takes first priority.
+CANDIDATE_MODELS = [
+    m for m in [
+        os.environ.get("HF_CHAT_MODEL"),
+        "Qwen/Qwen2.5-7B-Instruct",
+        "mistralai/Mistral-7B-Instruct-v0.3",
+        "Qwen/Qwen2.5-72B-Instruct",
+    ] if m
+]
+
+DEFAULT_MODEL = CANDIDATE_MODELS[0]
+
+# Set to whichever model actually served the most recent successful call,
+# so the run-metadata audit trail records the real model used.
+ACTIVE_MODEL = DEFAULT_MODEL
 
 SYSTEM_PROMPT = f"""You are a configuration assistant for a document OCR/data-extraction tool.
 
@@ -127,17 +142,27 @@ def chat_update(history, user_message, current_settings, hf_token=None):
     messages.extend(history)
     messages.append({"role": "user", "content": user_message})
 
-    try:
-        client = InferenceClient(token=token, provider="auto")
-        response = client.chat_completion(
-            messages=messages, model=DEFAULT_MODEL, max_tokens=400, temperature=0.2
-        )
-        content = response.choices[0].message.content
-    except Exception as exc:
+    global ACTIVE_MODEL
+    client = InferenceClient(token=token, provider="auto")
+    content = None
+    last_exc = None
+    for model in CANDIDATE_MODELS:
+        try:
+            response = client.chat_completion(
+                messages=messages, model=model, max_tokens=400, temperature=0.2
+            )
+            content = response.choices[0].message.content
+            ACTIVE_MODEL = model
+            break
+        except Exception as exc:
+            last_exc = exc
+    if content is None:
         return (
-            f"The chat model call failed ({exc}). Check that HF_TOKEN is valid "
-            f"and that the model '{DEFAULT_MODEL}' is available to your token "
-            "(set the HF_CHAT_MODEL env var to switch models).",
+            f"The chat model call failed for every candidate model "
+            f"({', '.join(CANDIDATE_MODELS)}). Last error: {last_exc}. "
+            "Check that HF_TOKEN is valid and has Inference Providers "
+            "permission, or set the HF_CHAT_MODEL env var to a model your "
+            "account can reach.",
             current_settings,
         )