adaptive-intelligence 3.0.0__tar.gz → 3.0.1__tar.gz

{adaptive_intelligence-3.0.0 → adaptive_intelligence-3.0.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: adaptive-intelligence
-Version: 3.0.0
+Version: 3.0.1
 Summary: Self-improving retrieval orchestration framework with RL-based routing, conditional graph activation, and evaluation-driven learning.
 Author-email: Venkatkumar Rajan <venkatkumarr.vk99@gmail.com>
 License: Apache-2.0
@@ -66,7 +66,7 @@ Dynamic: license-file
 
 Self-improving retrieval orchestration framework. Drop documents, ask questions, the system learns how to retrieve better over time.
 
-[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/VK-Ant/adaptive-intelligence/blob/main/notebooks/adaptive_intelligence_v3_demo.ipynb)
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/VK-Ant/adaptive-intelligence/blob/main/notebooks/adaptive_intelligence_v2_three_way.ipynb)
 
 ## Install
 

{adaptive_intelligence-3.0.0 → adaptive_intelligence-3.0.1}/README.md

@@ -2,7 +2,7 @@
 
 Self-improving retrieval orchestration framework. Drop documents, ask questions, the system learns how to retrieve better over time.
 
-[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/VK-Ant/adaptive-intelligence/blob/main/notebooks/adaptive_intelligence_v3_demo.ipynb)
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/VK-Ant/adaptive-intelligence/blob/main/notebooks/adaptive_intelligence_v2_three_way.ipynb)
 
 ## Install
 

{adaptive_intelligence-3.0.0 → adaptive_intelligence-3.0.1}/adaptive_intelligence/__init__.py

@@ -15,7 +15,7 @@ Usage:
     engine = AdaptiveAI(rl_algorithm="ppo", reranking=True, pretrained_policy=True)
 """
 
-__version__ = "3.0.0"
+__version__ = "3.0.1"
 __author__ = "Venkatkumar Rajan"
 
 from adaptive_intelligence.core.engine import AdaptiveAI

{adaptive_intelligence-3.0.0 → adaptive_intelligence-3.0.1}/adaptive_intelligence/core/engine.py

@@ -74,7 +74,7 @@ class AdaptiveAI:
 
         # Setup
         setup_logging(config.log_level, config.log_file)
-        logger.info(f"Initializing Adaptive Intelligence v3.0.0")
+        logger.info(f"Initializing Adaptive Intelligence v3.0.1")
 
         self._storage_dir = Path(config.storage_dir)
         self._storage_dir.mkdir(parents=True, exist_ok=True)
@@ -224,8 +224,12 @@ class AdaptiveAI:
                 azure_endpoint=self.config.azure_endpoint,
                 deployment_name=self.config.deployment_name,
             )
+            # Quick availability check for Ollama
+            if self.config.llm_backend.value == "ollama" and not self._llm.is_available():
+                logger.info("Ollama not running. Using retrieval-only mode.")
+                self._llm = None
         except Exception as e:
-            logger.warning(f"LLM init failed: {e}. Fallback mode active.")
+            logger.info(f"LLM not available ({e}). Using retrieval-only mode.")
             self._llm = None
 
     # ─── INGESTION ─────────────────────────────────────────
@@ -675,20 +679,29 @@ class AdaptiveAI:
 
     def _fallback_answer(self, query: str, chunks: List[Chunk]) -> str:
         if not chunks:
-            return "I could not find relevant information to answer this query."
+            return "No relevant information found in the ingested documents for this query."
 
-        parts = ["Based on the available documents:\n"]
+        # Build a clean, readable answer from retrieved chunks
+        parts = []
         for i, chunk in enumerate(chunks[:5]):
-            source = chunk.source_document or "Unknown"
-            page = f", Page {chunk.page_number}" if chunk.page_number else ""
-            preview = chunk.content[:300].strip()
-            parts.append(f"[{i+1}] From {source}{page}:\n{preview}\n")
-
-        parts.append(
-            "\nNote: Direct excerpt from sources. "
-            "Connect an LLM provider for synthesized answers."
-        )
-        return "\n".join(parts)
+            source = chunk.source_document or "document"
+            page = f" (page {chunk.page_number})" if chunk.page_number else ""
+            content = chunk.content.strip()
+
+            # Clean up content - take meaningful portion
+            if len(content) > 400:
+                # Cut at sentence boundary
+                cut = content[:400]
+                last_period = cut.rfind(".")
+                if last_period > 100:
+                    content = cut[:last_period + 1]
+                else:
+                    content = cut + "..."
+
+            parts.append(f"From {source}{page}:\n{content}")
+
+        answer = "\n\n".join(parts)
+        return answer
 
     def _extract_citations(self, answer: str, chunks: List[Chunk]) -> List[Citation]:
         citations = []
@@ -817,7 +830,7 @@ class AdaptiveAI:
 
     def status(self) -> Dict[str, Any]:
         return {
-            "version": "3.0.0",
+            "version": "3.0.1",
             "vectorless": self._vectorless,
             "documents_indexed": self.page_index.count() if self._vectorless else (self.vector_index.count() if self.vector_index else 0),
             "total_queries": self._total_queries,

{adaptive_intelligence-3.0.0 → adaptive_intelligence-3.0.1}/adaptive_intelligence/llm/__init__.py

@@ -285,11 +285,20 @@ class HuggingFaceProvider(BaseLLMProvider):
 
         start = time.time()
 
-        full_prompt = ""
-        if system_prompt:
-            full_prompt = f"System: {system_prompt}\n\nUser: {prompt}\n\nAssistant:"
-        else:
-            full_prompt = f"User: {prompt}\n\nAssistant:"
+        # Use chat template if available, else simple format
+        try:
+            messages = []
+            if system_prompt:
+                messages.append({"role": "system", "content": system_prompt})
+            messages.append({"role": "user", "content": prompt})
+            full_prompt = self._tokenizer.apply_chat_template(
+                messages, tokenize=False, add_generation_prompt=True
+            )
+        except Exception:
+            if system_prompt:
+                full_prompt = f"System: {system_prompt}\n\nUser: {prompt}\n\nAssistant:"
+            else:
+                full_prompt = f"User: {prompt}\n\nAssistant:"
 
         inputs = self._tokenizer(full_prompt, return_tensors="pt", truncation=True, max_length=4096)
         inputs = {k: v.to(self._model.device) for k, v in inputs.items()}
@@ -298,18 +307,24 @@ class HuggingFaceProvider(BaseLLMProvider):
         with torch.no_grad():
             outputs = self._model.generate(
                 **inputs,
-                max_new_tokens=max_tokens,
+                max_new_tokens=min(max_tokens, 1024),
                 temperature=max(temperature, 0.01),
                 do_sample=temperature > 0,
-                pad_token_id=self._tokenizer.pad_token_id,
+                pad_token_id=self._tokenizer.pad_token_id or self._tokenizer.eos_token_id,
+                repetition_penalty=1.3,
+                no_repeat_ngram_size=4,
             )
 
         generated = outputs[0][input_len:]
-        text = self._tokenizer.decode(generated, skip_special_tokens=True)
+        text = self._tokenizer.decode(generated, skip_special_tokens=True).strip()
+
+        # Clean garbled output
+        text = self._clean_output(text)
+
         latency = time.time() - start
 
         return LLMResponse(
-            text=text.strip(),
+            text=text,
             model=self.model_name,
             provider="huggingface",
             input_tokens=input_len,
@@ -317,6 +332,53 @@ class HuggingFaceProvider(BaseLLMProvider):
             latency_seconds=latency,
         )
 
+    @staticmethod
+    def _clean_output(text: str) -> str:
+        """Remove garbled/repetitive content from model output."""
+        if not text:
+            return text
+
+        lines = text.split("\n")
+        clean_lines = []
+        seen = set()
+
+        for line in lines:
+            stripped = line.strip()
+            # Skip empty or very short repeated lines
+            if not stripped:
+                if clean_lines and clean_lines[-1] != "":
+                    clean_lines.append("")
+                continue
+
+            # Skip if line is mostly non-alphabetic (garbled)
+            alpha_ratio = sum(c.isalpha() or c.isspace() for c in stripped) / max(len(stripped), 1)
+            if alpha_ratio < 0.5 and len(stripped) > 20:
+                break  # Stop at first garbled line
+
+            # Skip repeated lines
+            key = stripped.lower()[:80]
+            if key in seen and len(key) > 10:
+                continue
+            seen.add(key)
+
+            # Stop if line is just repeated characters
+            if len(set(stripped)) < 3 and len(stripped) > 5:
+                break
+
+            clean_lines.append(line)
+
+        result = "\n".join(clean_lines).strip()
+
+        # If result is too short after cleaning, return what we have
+        if len(result) < 10 and len(text) > 10:
+            # Take first 500 chars of original, cut at last sentence
+            result = text[:500]
+            last_period = result.rfind(".")
+            if last_period > 50:
+                result = result[:last_period + 1]
+
+        return result
+
 
 class LLMManager:
     """Factory and manager for LLM providers."""

{adaptive_intelligence-3.0.0 → adaptive_intelligence-3.0.1}/adaptive_intelligence.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: adaptive-intelligence
-Version: 3.0.0
+Version: 3.0.1
 Summary: Self-improving retrieval orchestration framework with RL-based routing, conditional graph activation, and evaluation-driven learning.
 Author-email: Venkatkumar Rajan <venkatkumarr.vk99@gmail.com>
 License: Apache-2.0
@@ -66,7 +66,7 @@ Dynamic: license-file
 
 Self-improving retrieval orchestration framework. Drop documents, ask questions, the system learns how to retrieve better over time.
 
-[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/VK-Ant/adaptive-intelligence/blob/main/notebooks/adaptive_intelligence_v3_demo.ipynb)
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/VK-Ant/adaptive-intelligence/blob/main/notebooks/adaptive_intelligence_v2_three_way.ipynb)
 
 ## Install
 

{adaptive_intelligence-3.0.0 → adaptive_intelligence-3.0.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "adaptive-intelligence"
-version = "3.0.0"
+version = "3.0.1"
 description = "Self-improving retrieval orchestration framework with RL-based routing, conditional graph activation, and evaluation-driven learning."
 readme = "README.md"
 license = {text = "Apache-2.0"}

{adaptive_intelligence-3.0.0 → adaptive_intelligence-3.0.1}/tests/test_core.py

@@ -118,7 +118,7 @@ class TestAdaptiveAIInit:
             engine = AdaptiveAI(storage_dir=d, log_level="WARNING")
             assert engine.config.llm_backend == LLMBackend.OLLAMA
             status = engine.status()
-            assert status["version"] == "3.0.0"
+            assert status["version"] == "3.0.1"
 
     def test_kwargs_init(self):
         with tempfile.TemporaryDirectory() as d: