syntaxmatrix 2.3.5__py3-none-any.whl → 2.5.5.4__py3-none-any.whl

syntaxmatrix/commentary.py

@@ -160,24 +160,29 @@ def phrase_commentary_vision(context: Dict[str, Any], images_b64: List[str]) ->
     """
     
     _SYSTEM_VISION = (
-    "You are a data analyst. Write a short, precise commentary in UK English that explains what the "
-    "already-rendered visuals mean for the user's question. "
-    "Use information visible in the attached figures and the provided context strings (field names, labels). "
-    "Do not invent numbers. If the figures/context are insufficient, say: 'Insufficient context to comment usefully.'"
+        "You are a data analyst. Write a detailed analysis that explains what the "
+        "already-rendered visuals mean for the user's question. "
+        "You use information visible in the attached figures and the provided context strings (texs, field names, labels). "
+        "You interprete the output without preamble."
     )
 
-    _USER_TMPL_VISION = """\
+    _USER_TMPL_VISION = """
     question:
     {q}
 
     Visible context strings (titles, axes, legends, headers):
     {ctx}
 
-    Write a concise conclusion (~200-260 words) with:
-    - <strong>Headline</strong> (one sentence answering the question).
-    - <strong>Evidence</strong> (6-8 bullets referencing panels/axes/legend groups seen in the figures and explaining the plots/tables  vis-a-vis the query). Explain all the oupupt comprehensively in details.
-    - <strong>Limitations</strong> (1 bullets; avoid quoting numbers unless present in context).
-    - <strong>Next step</strong> (1 bullet).
+    Write a comprehensive conclusion (~250-350 words) as follows:
+    - <b>Headline</b> 
+      2-3 sentence answering the question from an overview of all the output.
+    - <b>Evidence</b> 
+      8-10 bullets referencing the (output-texts/tables/panels/axes/legend groups) seen in the output. 
+      As you reference the visuals, you should interprete them in a way to show how they answer the question.
+    - <b>Limitations</b> 
+      1 bullet; avoid quoting numbers unless present in context.
+    - <b>Next step</b> 
+      1 bullet.
     """
 
     visible = _context_strings(context)
@@ -185,122 +190,139 @@ def phrase_commentary_vision(context: Dict[str, Any], images_b64: List[str]) ->
         q=context.get("question",""),
         ctx=json.dumps(visible, ensure_ascii=False, indent=2)
     )
-
-    prof = _prof.get_profile("image2text") or _prof.get_profile("admin") 
+     
+    prof = _prof.get_profile("vision2text") or _prof.get_profile("admin")
     if not prof:    
         return (
             "<div class='smx-alert smx-alert-warn'>"
                 "No LLM profile is configured for Image2Text. Please, do that in the Admin panel or contact your Administrator."
             "</div>"
         )
-    _client = _prof.get_client(prof)
-    _provider = (prof.get("provider") or "").lower()
-    _model = prof.get("model") or ""
+    
+    prof['client'] = _prof.get_client(prof)
+    _client = prof["client"]
+    _provider = prof["provider"].lower()
+    _model = prof["model"]
 
-    # Google
-    if _provider == "google":
-        try:
-            # Gemini expects a different structure
-            contents = []
-            
-            # Add text part first
-            text_part = {
-                "text": _SYSTEM_VISION + "\n\n" + user
-            }
-            contents.append(text_part)
-            
-            # Add image parts
-            for b64 in images_b64[:4]:
-                if b64:
-                    image_part = {
-                        "inline_data": {
-                            "mime_type": "image/png",
-                            "data": b64
+    try:
+        #1 Google
+        if _provider == "google":
+            try:
+                input_contents = []
+                
+                # Add text part first
+                text_part = {"text": user}
+                input_contents.append(text_part)
+                
+                # Add image parts
+                for b64 in images_b64:
+                    if b64:
+                        image_part = {
+                            "inline_data": {
+                                "mime_type": "image/png",
+                                "data": b64
+                            }
                         }
-                    }
-                    contents.append(image_part)
-            
-            # Correct Gemini API call
-            response = _client.models.generate_content(
-                model=_model,
-                contents=contents
-            )
-            txt = response.text.strip()
-            return txt.strip()
-        except Exception as e:
-            return f"Google Gemini error: {e}"
-           
-    # Openai
-    elif _provider == "openai" and _model in GPT_MODELS_LATEST:
-        # Use the Responses API with multimodal input (text + up to 4 images)
-        try:
-            parts = [{"type": "input_text", "text": user}]
-            for b64 in (images_b64 or [])[:4]:
-                if b64:
-                    parts.append({"type": "input_image", "image_url": f"data:image/png;base64,{b64}"})
-
-            args = set_args(
-                model=_model,
-                instructions=_SYSTEM_VISION,
-                input=[{"role": "user", "content": parts}],
-                previous_id=None,
-                store=False,
-                reasoning_effort="minimal",
-                verbosity="low",
-            )
-            resp = _client.responses.create(**args)
-            txt = _out(resp) or ""
-            if txt.strip():
-                return txt.strip()
-        except Exception:
-            pass  # Fall through to the chat.completions fallback implemented below.
+                        input_contents.append(image_part)
                 
-    # Anthropic
-    elif _provider == "anthropic":
-        try:
-            parts = [{"type":"text","text": user}]
-            for b64 in images_b64[:4]:
-                if b64:
-                    parts.append({"type":"image_url","image_url":{"url": f"data:image/png;base64,{b64}"}})       
-
-            response = _client.messages.create(
-                model=_model,
-                max_tokens=1024,
-                system=_SYSTEM_VISION,
-                messages=[{"role": "user", "content":parts}],
-                stream=False,
-            )
-            return response.content[0].text.strip()   
-        except Exception:
-             pass  # Fall through to the chat.completions fallback implemented below.
-    
-    # OpenAI SDK
-    else: # provider in {"openai","xai","deepseek","moonshotai","alibaba"}:
-        try:
-            parts = [{"type":"text","text": user}]
-            for b64 in images_b64[:4]:
-                if b64:
-                    parts.append({"type":"image_url","image_url":{"url": f"data:image/png;base64,{b64}"}})
-            resp = _client.chat.completions.create(
-                model=_model,
-                temperature=0.3,
-                messages=[
-                    {"role":"system","content":_SYSTEM_VISION},
-                    {"role":"user","content":parts},
-                ],
-                max_tokens=600,
-            )
-            return (resp.choices[0].message.content or "").strip()
-        except Exception:
-             pass  # Fall through to the chat.completions fallback implemented below.
+                response = _client.models.generate_content(
+                    model=_model,
+                    contents=input_contents,
+                    config=types.GenerateContentConfig(
+                        system_instruction=_SYSTEM_VISION,
+                        temperature=0.7,
+                        max_output_tokens=1024,
+                    ),
+                )
+                txt = response.text.strip()
+                return txt.strip()
+            except Exception:
+                pass
+            
+        #2 Openai
+        elif _provider == "openai" and _model in GPT_MODELS_LATEST:
+            try:
+                input_contents = []
+
+                text_part = {"type": "input_text", "text": user}
+                input_contents.append(text_part)
+                for b64 in images_b64:
+                    if b64:
+                        image_part = {
+                            "type": "input_image", 
+                            "image_url": f"data:image/png;base64,{b64}"
+                        }
+                        input_contents.append(image_part)
+
+                args = set_args(
+                    model=_model,
+                    instructions=_SYSTEM_VISION,
+                    input=[{"role": "user", "content": input_contents}],
+                    previous_id=None,
+                    store=False,
+                    reasoning_effort="low",
+                    verbosity="medium",
+                )
+                resp = _client.responses.create(**args)
+                txt = _out(resp) or ""
+                if txt.strip():
+                    return txt.strip()
+            except Exception:
+                pass 
+                    
+        # Anthropic
+        elif _provider == "anthropic":
+            try:
+                input_contents = []
+
+                text_part = {"type":"text","text": user}
+                input_contents.append(text_part)
 
-    # Text-only fallback via Responses API
-    return "Insufficient context to comment usefully."
+                for b64 in images_b64:
+                    if b64:
+                        image_part = {
+                            "type":"image_url",
+                            "image_url":{"url": f"data:image/png;base64,{b64}"}
+                        }
+                        input_contents.append(image_part)       
+
+                response = _client.messages.create(
+                    model=_model,
+                    max_tokens=1024,
+                    system=_SYSTEM_VISION,
+                    messages=[{"role": "user", "content":input_contents}],
+                    stream=False,
+                )
+                return response.content[0].text.strip()   
+            except Exception:
+                pass 
+        
+        # OpenAI SDK
+        else: 
+            try:
+                input_contents = [{"type":"text","text": user}]
+                for b64 in images_b64:
+                    if b64:
+                        input_contents.append({"type":"image_url","image_url":{"url": f"data:image/png;base64,{b64}"}})
+                resp = _client.chat.completions.create(
+                    model=_model,
+                    temperature=1,
+                    messages=[
+                        {"role":"system","content":_SYSTEM_VISION},
+                        {"role":"user","content":input_contents},
+                    ],
+                    max_tokens=1024,
+                )
+                return (resp.choices[0].message.content or "").strip()
+            except Exception:
+                pass
+    except Exception:
+        return "Insufficient context to comment usefully."
 
 def wrap_html(card_text: str) -> str:
     return f"""
         <div class="smx-commentary-card" style="margin-top:1rem;padding:1rem;border:1px solid #e5e7eb;border-radius:0.75rem;background:#fafafa">
-        <div style="font-weight:600;margin-bottom:0.5rem;">Commentary</div>
+        <div style="font-weight:600;margin-bottom:0.5rem;">smxAI Feedback</div>
         <div class="prose" style="white-space:pre-wrap;line-height:1.45">{card_text}</div>
         </div>
     """.strip()