fix: Construct RequestParameters with all 6 required args (url, task, model, json, data, headers)

src/hf_inference.py

@@ -9,14 +9,13 @@ import os
 import json
 import logging
 import time
-import inspect
 
 logger = logging.getLogger(__name__)
 
 HF_API_TOKEN = os.environ.get("HF_API_TOKEN", "")
+HF_API_BASE = "https://api-inference.huggingface.co/models/"
 
 _client = None
-_RequestParameters = None
 
 
 def _get_client():
@@ -27,14 +26,6 @@ def _get_client():
     return _client
 
 
-def _get_request_params_class():
-    global _RequestParameters
-    if _RequestParameters is None:
-        from huggingface_hub.inference._common import RequestParameters
-        _RequestParameters = RequestParameters
-    return _RequestParameters
-
-
 # Repository IDs
 SUMMARIZATION_REPO = os.environ.get("SUMMARIZATION_REPO_ID", "bayan10/summarization-model")
 SPELLING_REPO = os.environ.get("SPELLING_REPO_ID", "bayan10/AraSpell-Model")
@@ -42,45 +33,32 @@ PUNCTUATION_REPO = os.environ.get("PUNCTUATION_REPO_ID", "bayan10/PuncAra-v1")
 AUTOCOMPLETE_REPO = os.environ.get("AUTOCOMPLETE_REPO_ID", "bayan10/AutoComplete")
 
 
-def _call_model(repo_id, payload, task=None):
+def _call_model(repo_id, payload, task=""):
     """
-    Call HF model using _inner_post with RequestParameters.
+    Call HF model via InferenceClient._inner_post with RequestParameters.
+    
+    RequestParameters(url, task, model, json, data, headers)
     """
+    from huggingface_hub.inference._common import RequestParameters
+
     client = _get_client()
-    RP = _get_request_params_class()
 
     if "options" not in payload:
         payload["options"] = {"wait_for_model": True}
 
-    data_bytes = json.dumps(payload).encode("utf-8")
-    logger.info("Calling HF model: %s (task=%s)", repo_id, task)
+    url = HF_API_BASE + repo_id
+    logger.info("Calling HF model: %s (url=%s, task=%s)", repo_id, url, task)
 
-    # Construct RequestParameters - try with fields we know
-    try:
-        # Inspect what fields RequestParameters accepts
-        sig = inspect.signature(RP)
-        params = sig.parameters
-        logger.info("RequestParameters fields: %s", list(params.keys()))
-
-        # Build kwargs based on what's available
-        rp_kwargs = {}
-        if "model" in params:
-            rp_kwargs["model"] = repo_id
-        if "task" in params and task:
-            rp_kwargs["task"] = task
-        if "data" in params:
-            rp_kwargs["data"] = data_bytes
-        if "json" in params:
-            rp_kwargs["json"] = payload
-
-        rp = RP(**rp_kwargs)
-        response = client._inner_post(rp)
+    rp = RequestParameters(
+        url=url,
+        task=task,
+        model=repo_id,
+        json=payload,
+        data=None,
+        headers={},
+    )
 
-    except Exception as e:
-        logger.warning("RequestParameters construction failed: %s", e)
-        # Last resort: try creating with just positional arg
-        rp = RP(data_bytes)
-        response = client._inner_post(rp)
+    response = client._inner_post(rp)
 
     # Parse response
     if isinstance(response, bytes):
@@ -92,6 +70,11 @@ def _call_model(repo_id, payload, task=None):
 
     logger.info("HF result for %s: type=%s preview=%s",
                 repo_id, type(result).__name__, str(result)[:200])
+
+    # Check for API errors in response
+    if isinstance(result, dict) and "error" in result:
+        raise RuntimeError("HF API error: " + str(result["error"]))
+
     return result
 
 
@@ -106,8 +89,6 @@ def _extract_text(result, fallback=""):
                     or fallback)
         return str(item) if str(item).strip() else fallback
     if isinstance(result, dict):
-        if "error" in result:
-            raise RuntimeError("HF API error: " + str(result["error"]))
         return (result.get("summary_text")
                 or result.get("generated_text")
                 or result.get("translation_text")
@@ -128,12 +109,12 @@ def hf_summarize(text, max_length=128, min_length=30):
 
 
 def hf_correct_spelling(text):
-    result = _call_model(SPELLING_REPO, {"inputs": text})
+    result = _call_model(SPELLING_REPO, {"inputs": text}, task="text2text-generation")
     return _extract_text(result, text)
 
 
 def hf_add_punctuation(text):
-    result = _call_model(PUNCTUATION_REPO, {"inputs": text})
+    result = _call_model(PUNCTUATION_REPO, {"inputs": text}, task="text2text-generation")
     return _extract_text(result, text)
 
 
@@ -166,23 +147,16 @@ def check_hf_api_available():
 
 
 def debug_test_all_models():
-    """Test all HF models with full diagnostics."""
+    """Test all HF models."""
     results = {}
     test_text = "هذا نص تجريبي للاختبار"
     long_text = (test_text + " ") * 5
 
     try:
         import huggingface_hub
-        RP = _get_request_params_class()
-        rp_sig = str(inspect.signature(RP))
-        rp_fields = list(inspect.signature(RP).parameters.keys())
-        results["_info"] = {
-            "hf_hub_version": huggingface_hub.__version__,
-            "RequestParameters_signature": rp_sig[:300],
-            "RequestParameters_fields": rp_fields,
-        }
+        results["_info"] = {"hf_hub_version": huggingface_hub.__version__}
     except Exception as e:
-        results["_info"] = {"error": repr(e)[:300]}
+        results["_info"] = {"error": repr(e)[:200]}
 
     for name, fn, args in [
         ("summarization", hf_summarize, (long_text, 30, 10)),