fix: use groq client directly for explain

pipeline/explainer.py

@@ -1,5 +1,5 @@
 import os
-from huggingface_hub import InferenceClient
+from groq import Groq
 from pathlib import Path
 
 BASE_DIR = Path(__file__).parent.parent
@@ -21,7 +21,7 @@ FALLACY_DEFINITIONS = {
     "intentional": "A deliberate and deceptive use of misleading reasoning"
 }
 
-DEFAULT_MODEL_ID = "HuggingFaceH4/zephyr-7b-beta"
+DEFAULT_MODEL_ID = "llama-3.1-8b-instant"
 
 
 class FallacyExplainer:
@@ -30,16 +30,13 @@ class FallacyExplainer:
 
     def __init__(self, fallacy_classes=None, fallacy_definitions=None):
         self._model_id = os.environ.get("EXPLAIN_MODEL_ID", DEFAULT_MODEL_ID)
-        self._client = InferenceClient(
-            provider="groq",
-            api_key=os.environ["GROQ_API_KEY"],
-        )
+        self._client = Groq(api_key=os.environ["GROQ_API_KEY"])
         self.fallacy_classes = fallacy_classes or self.DEFAULT_FALLACY_CLASSES
         self.fallacy_definitions = fallacy_definitions or FALLACY_DEFINITIONS
         print(f"FallacyExplainer ready (model: {self._model_id} via Groq)")
 
     def _generate_with_prompt(self, prompt_text, max_new_tokens=128):
-        result = self._client.chat_completion(
+        result = self._client.chat.completions.create(
             model=self._model_id,
             messages=[{"role": "user", "content": prompt_text}],
             max_tokens=max_new_tokens,

requirements.txt

@@ -7,3 +7,4 @@ pydantic==2.12.5
 accelerate==1.13.0
 safetensors==0.7.0
 huggingface_hub>=0.25.0
+groq