faster-AI-1CPU

Dockerfile

@@ -31,6 +31,7 @@ COPY --chown=user . .
 EXPOSE 7860
 
 # Command to run the application using Gunicorn for better stability on HF
-# -w 2: Two workers to keep memory usage balanced
+# -w 1: Single worker to focus all CPU power and save memory
 # -k uvicorn.workers.UvicornWorker: Use Uvicorn for FastAPI
-CMD ["gunicorn", "-w", "2", "-k", "uvicorn.workers.UvicornWorker", "main:app", "--bind", "0.0.0.0:7860", "--timeout", "120"]
+CMD ["gunicorn", "-w", "1", "-k", "uvicorn.workers.UvicornWorker", "main:app", "--bind", "0.0.0.0:7860", "--timeout", "120"]
+

Procfile

@@ -1 +1,2 @@
-web: gunicorn -w 4 -k uvicorn.workers.UvicornWorker main:app --bind 0.0.0.0:$PORT
+web: gunicorn -w 1 -k uvicorn.workers.UvicornWorker main:app --bind 0.0.0.0:$PORT
+

code_commenter_improved.py

@@ -189,6 +189,33 @@ def _invoke_ai(codet5_pipeline, prompt, max_new_tokens=60):
     return ""
 
 
+def _invoke_ai_batch(codet5_pipeline, prompts, max_new_tokens=100, batch_size=8):
+    """Efficiently run multiple prompts through the transformer pipeline in a single batch."""
+    if codet5_pipeline is None or not prompts:
+        return [""] * len(prompts)
+    
+    try:
+        # Pipeline handles list of strings and batching automatically
+        results = codet5_pipeline(
+            prompts, 
+            max_new_tokens=max_new_tokens, 
+            do_sample=False,
+            batch_size=batch_size
+        )
+        
+        outputs = []
+        for res in results:
+            if isinstance(res, list):
+                outputs.append(res[0].get("generated_text", "").strip())
+            else:
+                outputs.append(res.get("generated_text", "").strip())
+        return outputs
+    except Exception as e:
+        logger.warning(f"Batch AI inference failed: {e}")
+        return [""] * len(prompts)
+
+
+
 def _truncate_to_tokens(text: str, max_tokens: int = 400) -> str:
     """Approximate token limit by word count (1 token ~ 0.75 words)."""
     words = text.split()
@@ -231,12 +258,18 @@ def _strip_class_attrs(tag) -> str:
 
 
 def _annotate_html(soup, codet5_pipeline, comment_level):
-    """Three-layer HTML annotator: Structural / Classes / Events."""
+    """Three-layer HTML annotator: Structural / Classes / Events.
+    Uses batch inference for performance.
+    """
     event_attrs = [
         "onclick", "onsubmit", "onchange", "oninput",
         "onload", "onkeydown", "onmouseover",
     ]
 
+    # PASS 1: Identify targets and collect prompts
+    targets = []
+    prompts = []
+    
     for tag in soup.find_all(True):
         has_event = any(tag.get(ea) for ea in event_attrs)
         is_educational_target = tag.name in _TAG_PEDAGOGY
@@ -245,25 +278,54 @@ def _annotate_html(soup, codet5_pipeline, comment_level):
         if not (is_educational_target or has_id_or_class or has_event):
             continue
 
+        should_run_ai = (
+            (tag.name in _SEMANTIC_TAGS or tag.get("id"))
+            and codet5_pipeline is not None
+            and comment_level in ("detailed", "educational")
+        )
+
+        prompt = None
+        if should_run_ai:
+            prompt = (
+                f"Explain this web element for a beginner: WHAT it does, "
+                f"WHY it is placed here in context, and what would "
+                f"BREAK or happen to the visitor if it were removed. "
+                f"IMPORTANT: DO NOT start your answer with the word '{tag.name}'. "
+                f"Context: {_strip_class_attrs(tag)}"
+            )
+            prompts.append(prompt)
+        
+        targets.append({
+            "tag": tag,
+            "has_event": has_event,
+            "prompt_idx": len(prompts) - 1 if prompt else -1
+        })
+
+    # BATCH AI INVOCATION
+    ai_outputs = []
+    if prompts:
+        ai_outputs = _invoke_ai_batch(codet5_pipeline, prompts, max_new_tokens=100)
+
+    # PASS 2: Apply annotations
+    for target in targets:
+        tag = target["tag"]
         role = tag.name.upper()
         structural_parts = []
+        
         pedagogy = _TAG_PEDAGOGY.get(tag.name)
         if pedagogy:
             structural_parts.append(pedagogy)
 
+        # Attribute-aware annotations
         if tag.name == "img":
             src = tag.get("src", "")
             alt = tag.get("alt", "")
             if src:
-                structural_parts.append(
-                    f"source: {src if len(src) <= 60 else src[:57] + '...'}"
-                )
+                structural_parts.append(f"source: {src if len(src) <= 60 else src[:57] + '...'}")
             if alt:
                 structural_parts.append(f'alt text: "{alt}"')
             else:
-                structural_parts.append(
-                    'no alt text (needs alt="description" for accessibility)'
-                )
+                structural_parts.append('no alt text (needs alt="description" for accessibility)')
             if tag.get("width"):
                 structural_parts.append(f"width: {tag.get('width')}px")
         elif tag.name == "a" and tag.get("href"):
@@ -286,25 +348,13 @@ def _annotate_html(soup, codet5_pipeline, comment_level):
             )
             structural_parts.append(f"loads {lib} from CDN")
 
+        # Apply AI summary
         final_summary = ""
-        should_run_model = (
-            (tag.name in _SEMANTIC_TAGS or tag.get("id"))
-            and codet5_pipeline is not None
-            and comment_level in ("detailed", "educational")
-        )
-        if should_run_model:
-            # Enforce the "WHAT, WHY, IMPACT" pedagogy. 
-            # We explicitly tell it NOT to repeat the tag name to avoid the "Header: Header" redundancy.
-            prompt = (
-                f"Explain this web element for a beginner: WHAT it does, "
-                f"WHY it is placed here in context, and what would "
-                f"BREAK or happen to the visitor if it were removed. "
-                f"IMPORTANT: DO NOT start your answer with the word '{tag.name}'. "
-                f"Context: {_strip_class_attrs(tag)}"
-            )
-            raw = _invoke_ai(codet5_pipeline, prompt, max_new_tokens=100)
+        if target["prompt_idx"] != -1 and target["prompt_idx"] < len(ai_outputs):
+            raw = ai_outputs[target["prompt_idx"]]
             final_summary = _sanitize_ai_text(raw, tag.name)
 
+        # Construct final comment
         indicator = ""
         if tag.get("id"):
             indicator = f"#{tag.get('id')}"
@@ -323,6 +373,7 @@ def _annotate_html(soup, codet5_pipeline, comment_level):
         tag.insert_before(soup.new_string("\n"))
         tag.insert_before(Comment(full_structural_comment))
 
+        # Classes annotation
         if tag.get("class"):
             c_val = tag.get("class")
             c_str = " ".join(c_val) if isinstance(c_val, list) else c_val
@@ -333,20 +384,16 @@ def _annotate_html(soup, codet5_pipeline, comment_level):
                     parts = explained.split("\n<!-- WARNING:")
                     tag.insert_before(Comment(f" CLASSES: {parts[0]} "))
                     tag.insert_before(soup.new_string("\n"))
-                    tag.insert_before(
-                        Comment(
-                            f" WARNING:{parts[1].replace('-->', '').strip()} "
-                        )
-                    )
+                    tag.insert_before(Comment(f" WARNING:{parts[1].replace('-->', '').strip()} "))
                 else:
                     tag.insert_before(Comment(f" CLASSES: {explained} "))
 
+        # Events annotation
         for ea in event_attrs:
             if tag.get(ea):
                 tag.insert_before(soup.new_string("\n"))
-                tag.insert_before(
-                    Comment(f" EVENT: triggers {tag.get(ea)} on {ea[2:]} ")
-                )
+                tag.insert_before(Comment(f" EVENT: triggers {tag.get(ea)} on {ea[2:]} "))
+
 
 
 CSS_PROP_COMMENTS = {

github_utils.py

@@ -97,9 +97,13 @@ def get_repository(username, repo_name, token):
         repo = g.get_repo(f"{username}/{repo_name}")
         return repo
     except Exception as e:
-        print(f"Repository not found: {e}")
+        if "404" in str(e):
+            print(f"Info: Repository '{repo_name}' not found on GitHub. A new one will be created.")
+        else:
+            print(f"GitHub Repository lookup error: {e}")
         return None
 
+
 def update_file_in_repo(repo, file_path, content, commit_message="Update file"):
     """
     Update an existing file in the repository.