llms-py 3.0.0b10__py3-none-any.whl → 3.0.2__py3-none-any.whl

llms/extensions/providers/google.py

@@ -79,16 +79,61 @@ def install_google(ctx):
             if "Authorization" in self.headers:
                 del self.headers["Authorization"]
 
-        async def chat(self, chat):
+        def provider_model(self, model):
+            if model.lower().startswith("gemini-"):
+                return model
+            return super().provider_model(model)
+
+        def model_info(self, model):
+            info = super().model_info(model)
+            if info:
+                return info
+            if model.lower().startswith("gemini-"):
+                return {
+                    "id": model,
+                    "name": model,
+                    "cost": {"input": 0, "output": 0},
+                }
+            return None
+
+        async def chat(self, chat, context=None):
             chat["model"] = self.provider_model(chat["model"]) or chat["model"]
+            model_info = (context.get("modelInfo") if context is not None else None) or self.model_info(chat["model"])
 
             chat = await self.process_chat(chat)
             generation_config = {}
+            tools = None
+            supports_tool_calls = model_info.get("tool_call", False)
+
+            if "tools" in chat and supports_tool_calls:
+                function_declarations = []
+                gemini_tools = {}
+
+                for tool in chat["tools"]:
+                    if tool["type"] == "function":
+                        f = tool["function"]
+                        function_declarations.append(
+                            {
+                                "name": f["name"],
+                                "description": f.get("description"),
+                                "parameters": f.get("parameters"),
+                            }
+                        )
+                    elif tool["type"] == "file_search":
+                        gemini_tools["file_search"] = tool["file_search"]
+
+                if function_declarations:
+                    gemini_tools["function_declarations"] = function_declarations
+
+                tools = [gemini_tools] if gemini_tools else None
 
             # Filter out system messages and convert to proper Gemini format
             contents = []
             system_prompt = None
 
+            # Track tool call IDs to names for response mapping
+            tool_id_map = {}
+
             async with aiohttp.ClientSession() as session:
                 for message in chat["messages"]:
                     if message["role"] == "system":
@@ -101,8 +146,55 @@ def install_google(ctx):
                         elif isinstance(content, str):
                             system_prompt = content
                     elif "content" in message:
+                        role = "user"
+                        if "role" in message:
+                            if message["role"] == "user":
+                                role = "user"
+                            elif message["role"] == "assistant":
+                                role = "model"
+                            elif message["role"] == "tool":
+                                role = "function"
+
+                        parts = []
+
+                        # Handle tool calls in assistant messages
+                        if message.get("role") == "assistant" and "tool_calls" in message:
+                            for tool_call in message["tool_calls"]:
+                                tool_id_map[tool_call["id"]] = tool_call["function"]["name"]
+                                parts.append(
+                                    {
+                                        "functionCall": {
+                                            "name": tool_call["function"]["name"],
+                                            "args": json.loads(tool_call["function"]["arguments"]),
+                                        }
+                                    }
+                                )
+
+                        # Handle tool responses from user
+                        if message.get("role") == "tool":
+                            # Gemini expects function response in 'functionResponse' part
+                            # We need to find the name associated with this tool_call_id
+                            tool_call_id = message.get("tool_call_id")
+                            name = tool_id_map.get(tool_call_id)
+                            # If we can't find the name (maybe from previous turn not in history or restart),
+                            # we might have an issue. But let's try to proceed.
+                            # Fallback: if we can't find the name, skip or try to infer?
+                            # Gemini strict validation requires the name.
+                            if name:
+                                # content is the string response
+                                # Some implementations pass the content directly.
+                                # Google docs say: response: { "name": "...", "content": { ... } }
+                                # Actually "response" field in functionResponse is a Struct/Map.
+                                parts.append(
+                                    {
+                                        "functionResponse": {
+                                            "name": name,
+                                            "response": {"name": name, "content": message["content"]},
+                                        }
+                                    }
+                                )
+
                         if isinstance(message["content"], list):
-                            parts = []
                             for item in message["content"]:
                                 if "type" in item:
                                     if item["type"] == "image_url" and "image_url" in item:
@@ -142,23 +234,14 @@ def install_google(ctx):
                                 if "text" in item:
                                     text = item["text"]
                                     parts.append({"text": text})
-                            if len(parts) > 0:
-                                contents.append(
-                                    {
-                                        "role": message["role"]
-                                        if "role" in message and message["role"] == "user"
-                                        else "model",
-                                        "parts": parts,
-                                    }
-                                )
-                        else:
-                            content = message["content"]
+                        elif message["content"]:  # String content
+                            parts.append({"text": message["content"]})
+
+                        if len(parts) > 0:
                             contents.append(
                                 {
-                                    "role": message["role"]
-                                    if "role" in message and message["role"] == "user"
-                                    else "model",
-                                    "parts": [{"text": content}],
+                                    "role": role,
+                                    "parts": parts,
                                 }
                             )
 
@@ -166,6 +249,9 @@ def install_google(ctx):
                     "contents": contents,
                 }
 
+                if tools:
+                    gemini_chat["tools"] = tools
+
                 if self.safety_settings:
                     gemini_chat["safetySettings"] = self.safety_settings
 
@@ -192,18 +278,12 @@ def install_google(ctx):
                 if len(generation_config) > 0:
                     gemini_chat["generationConfig"] = generation_config
 
-                if "tools" in chat:
-                    # gemini_chat["tools"] = chat["tools"]
-                    ctx.log("Error: tools not supported in Gemini")
-                elif self.tools:
-                    # gemini_chat["tools"] = self.tools.copy()
-                    ctx.log("Error: tools not supported in Gemini")
-
                 if "modalities" in chat:
                     generation_config["responseModalities"] = [modality.upper() for modality in chat["modalities"]]
                     if "image" in chat["modalities"] and "image_config" in chat:
                         # delete thinkingConfig
-                        del generation_config["thinkingConfig"]
+                        if "thinkingConfig" in generation_config:
+                            del generation_config["thinkingConfig"]
                         config_map = {
                             "aspect_ratio": "aspectRatio",
                             "image_size": "imageSize",
@@ -212,11 +292,16 @@ def install_google(ctx):
                             config_map[k]: v for k, v in chat["image_config"].items() if k in config_map
                         }
                     if "audio" in chat["modalities"] and self.speech_config:
-                        del generation_config["thinkingConfig"]
+                        if "thinkingConfig" in generation_config:
+                            del generation_config["thinkingConfig"]
                         generation_config["speechConfig"] = self.speech_config.copy()
                         # Currently Google Audio Models only accept AUDIO
                         generation_config["responseModalities"] = ["AUDIO"]
 
+                # Ensure generationConfig is set if we added anything to it
+                if len(generation_config) > 0:
+                    gemini_chat["generationConfig"] = generation_config
+
                 started_at = int(time.time() * 1000)
                 gemini_chat_url = f"https://generativelanguage.googleapis.com/v1beta/models/{chat['model']}:generateContent?key={self.api_key}"
 
@@ -237,6 +322,8 @@ def install_google(ctx):
                             timeout=aiohttp.ClientTimeout(total=120),
                         ) as res:
                             obj = await self.response_json(res)
+                            if context is not None:
+                                context["providerResponse"] = obj
                     except Exception as e:
                         ctx.log(f"Error: {res.status} {res.reason}: {e}")
                         text = await res.text()
@@ -271,7 +358,7 @@ def install_google(ctx):
                     "model": obj.get("modelVersion", chat["model"]),
                 }
                 choices = []
-                for i, candidate in enumerate(obj["candidates"]):
+                for i, candidate in enumerate(obj.get("candidates", [])):
                     role = "assistant"
                     if "content" in candidate and "role" in candidate["content"]:
                         role = "assistant" if candidate["content"]["role"] == "model" else candidate["content"]["role"]
@@ -281,6 +368,8 @@ def install_google(ctx):
                     reasoning = ""
                     images = []
                     audios = []
+                    tool_calls = []
+
                     if "content" in candidate and "parts" in candidate["content"]:
                         text_parts = []
                         reasoning_parts = []
@@ -290,6 +379,16 @@ def install_google(ctx):
                                     reasoning_parts.append(part["text"])
                                 else:
                                     text_parts.append(part["text"])
+                            if "functionCall" in part:
+                                fc = part["functionCall"]
+                                tool_calls.append(
+                                    {
+                                        "id": f"call_{len(tool_calls)}_{int(time.time())}",  # Gemini doesn't return ID, generate one
+                                        "type": "function",
+                                        "function": {"name": fc["name"], "arguments": json.dumps(fc["args"])},
+                                    }
+                                )
+
                             if "inlineData" in part:
                                 inline_data = part["inlineData"]
                                 mime_type = inline_data.get("mimeType", "image/png")
@@ -354,7 +453,7 @@ def install_google(ctx):
                         "finish_reason": candidate.get("finishReason", "stop"),
                         "message": {
                             "role": role,
-                            "content": content,
+                            "content": content if content else None,
                         },
                     }
                     if reasoning:
@@ -363,6 +462,10 @@ def install_google(ctx):
                         choice["message"]["images"] = images
                     if len(audios) > 0:
                         choice["message"]["audios"] = audios
+                    if len(tool_calls) > 0:
+                        choice["message"]["tool_calls"] = tool_calls
+                        # If we have tool calls, content can be null but message should probably exist
+
                     choices.append(choice)
                 response["choices"] = choices
                 if "usageMetadata" in obj:

llms/extensions/providers/nvidia.py

@@ -54,7 +54,7 @@ def install_nvidia(ctx):
                     }
             raise Exception("No artifacts in response")
 
-        async def chat(self, chat, provider=None):
+        async def chat(self, chat, provider=None, context=None):
             headers = self.get_headers(provider, chat)
             if provider is not None:
                 chat["model"] = provider.provider_model(chat["model"]) or chat["model"]
@@ -100,6 +100,6 @@ def install_nvidia(ctx):
                     data=json.dumps(gen_request),
                     timeout=aiohttp.ClientTimeout(total=120),
                 ) as response:
-                    return self.to_response(await self.response_json(response), chat, started_at)
+                    return self.to_response(await self.response_json(response), chat, started_at, context=context)
 
     ctx.add_provider(NvidiaGenAi)

llms/extensions/providers/openai.py

@@ -113,7 +113,7 @@ def install_openai(ctx):
             ctx.log(json.dumps(response, indent=2))
             raise Exception("No 'data' field in response.")
 
-        async def chat(self, chat, provider=None):
+        async def chat(self, chat, provider=None, context=None):
             headers = self.get_headers(provider, chat)
 
             if chat["model"] in self.map_image_models:
@@ -145,7 +145,7 @@ def install_openai(ctx):
                     text = await response.text()
                     ctx.log(text[:1024] + (len(text) > 1024 and "..." or ""))
                     if response.status < 300:
-                        return ctx.log_json(await self.to_response(json.loads(text), chat, started_at))
+                        return ctx.log_json(await self.to_response(json.loads(text), chat, started_at, context=context))
                     else:
                         raise Exception(f"Failed to generate image {response.status}")
 

llms/extensions/providers/openrouter.py

@@ -39,7 +39,7 @@ def install_openrouter(ctx):
 
             return response
 
-        async def chat(self, chat, provider=None):
+        async def chat(self, chat, provider=None, context=None):
             headers = self.get_headers(provider, chat)
             if provider is not None:
                 chat["model"] = provider.provider_model(chat["model"]) or chat["model"]
@@ -67,6 +67,8 @@ def install_openrouter(ctx):
                 ) as response:
                     if metadata:
                         chat["metadata"] = metadata
-                    return ctx.log_json(self.to_response(await self.response_json(response), chat, started_at))
+                    return ctx.log_json(
+                        self.to_response(await self.response_json(response), chat, started_at, context=context)
+                    )
 
     ctx.add_provider(OpenRouterGenerator)

llms/extensions/system_prompts/ui/index.mjs

@@ -142,7 +142,7 @@ const SystemPromptEditor = {
                     System Prompt
                 </label>
                 <div v-if="hasMessages" class="text-sm text-gray-500 dark:text-gray-400">
-                    {{ !threadSystemPrompt ? '' : prompts.find(x => x.value === threadSystemPrompt)?.name || 'Custom' }}
+                    {{ !ext.prefs.systemPrompt ? '' : prompts.find(x => x.value === ext.prefs.systemPrompt)?.name || 'Custom' }}
                 </div>
                 <div v-else class="mb-2 relative" ref="containerRef">
                     <div class="flex items-center gap-2">
@@ -154,16 +154,16 @@ const SystemPromptEditor = {
                             <svg class="size-4 text-gray-500 dark:text-gray-400" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path fill="currentColor" d="M19 6.41L17.59 5L12 10.59L6.41 5L5 6.41L10.59 12L5 17.59L6.41 19L12 13.41L17.59 19L19 17.59L13.41 12z"/></svg>
                         </button>
                         <button type="button" 
-                            @click="showFinder = !showFinder"
+                            @click="ext.setPrefs({ showFinder: !ext.prefs.showFinder })"
                             class="inline-flex items-center gap-x-1.5 rounded-md bg-white dark:bg-gray-900 px-2.5 py-1.5 text-sm font-medium text-gray-700 dark:text-gray-300 shadow-sm border border-gray-300 dark:border-gray-600 hover:bg-gray-50 dark:hover:bg-gray-800">
                             Explore Prompts
                         </button>
                     </div>
-                    <PromptFinder v-model="showFinder" :prompts="prompts" @select="onSelect" />
+                    <PromptFinder v-model="ext.prefs.showFinder" :prompts="prompts" @select="onSelect" />
                 </div>
             </div>
             <div v-if="hasMessages" class="w-full rounded-md border border-gray-300 dark:border-gray-600 bg-white dark:bg-gray-900 text-gray-900 dark:text-gray-100 px-3 py-2 text-sm">
-                {{threadSystemPrompt || 'No System Prompt was used' }}
+                {{$threads.currentThread.value?.systemPrompt || 'No System Prompt was used' }}
             </div>
             <div v-else>
                 <textarea
@@ -186,32 +186,29 @@ const SystemPromptEditor = {
         /**@type {AppContext} */
         const ctx = inject('ctx')
         const containerRef = ref()
-        const showFinder = ref(false)
-        const prefs = ext.getPrefs()
         const hasMessages = computed(() => ctx.threads.currentThread.value?.messages?.length > 0)
-        const threadSystemPrompt = computed(() => ctx.threads.currentThread.value?.systemPrompt || '')
         const selected = computed(() =>
             props.prompts.find(x => x.value === props.modelValue) ?? { name: "Custom", value: props.modelValue })
 
         function onSelect(prompt) {
+            ext.setPrefs({ prompt: prompt }) // {"id","name","value"}
             emit('update:modelValue', prompt.value)
         }
 
         function closeFinder(e) {
-            if (showFinder.value && containerRef.value && !containerRef.value.contains(e.target)) {
-                showFinder.value = false
+            if (ext.prefs.showFinder && containerRef.value && !containerRef.value.contains(e.target)) {
+                ext.setPrefs({ showFinder: false })
             }
         }
 
-        watch(() => props.modelValue, promptValue => {
-            prefs.prompt = selected.value
-            ext.setPrefs(prefs)
+        watch(() => props.modelValue, systemPrompt => {
+            ext.setPrefs({ systemPrompt })
         })
 
         onMounted(() => {
             document.addEventListener('click', closeFinder)
-            if (prefs.prompt) {
-                emit('update:modelValue', prefs.prompt.value)
+            if (ext.prefs.prompt) {
+                emit('update:modelValue', ext.prefs.prompt)
             }
         })
         onUnmounted(() => {
@@ -219,11 +216,10 @@ const SystemPromptEditor = {
         })
 
         return {
-            threadSystemPrompt,
+            ext,
             hasMessages,
             selected,
             containerRef,
-            showFinder,
             onSelect,
         }
     }
@@ -238,9 +234,13 @@ export default {
             PromptFinder,
             SystemPromptEditor,
             SystemPromptsPanel: {
-                template: `<SystemPromptEditor :prompts="$state.prompts" v-model="$state.selectedPrompt" />`,
+                template: `<SystemPromptEditor :prompts="ext.state.prompts" v-model="ext.prefs.prompt" />`,
+                setup() {
+                    return { ext }
+                }
             }
         })
+        ext.setPrefs({ systemPrompt: '' })
 
         ctx.setTopIcons({
             system_prompts: {
@@ -251,12 +251,6 @@ export default {
             }
         })
 
-        ctx.createThreadFilters.push(thread => {
-            const prefs = ext.getPrefs()
-            thread.systemPrompt = prefs?.prompt?.value || ""
-            console.log('createThreadFilters', prefs, thread)
-        })
-
         ctx.chatRequestFilters.push(({ request, thread }) => {
 
             const hasSystemPrompt = request.messages.find(x => x.role === 'system')
@@ -265,11 +259,12 @@ export default {
                 return
             }
 
-            if (thread.systemPrompt) {
+            // Only add the selected system prompt for new requests
+            if (ext.prefs.systemPrompt && request.messages.length <= 1) {
                 // add message to start
                 request.messages.unshift({
                     role: 'system',
-                    content: thread.systemPrompt
+                    content: ext.prefs.systemPrompt
                 })
             }
         })
@@ -280,6 +275,6 @@ export default {
     async load(ctx) {
         const api = await ext.getJson(`/prompts.json`)
         const prompts = api.response || []
-        ctx.setState({ prompts })
+        ext.setState({ prompts })
     }
 }

llms/extensions/system_prompts/ui/prompts.json

@@ -645,7 +645,7 @@
         "value": "I want you to act as an biblical translator. I will speak to you in english and you will translate it and answer in the corrected and improved version of my text, in a biblical dialect. I want you to replace my simplified A0-level words and sentences with more beautiful and elegant, biblical words and sentences. Keep the meaning same. I want you to only reply the correction, the improvements and nothing else, do not write explanations. My first sentence is \"Hello, World!\""
     },
     {
-        "id": "chess-player",
+        "id": "chess-player2",
         "name": "Act as an Chess Player",
         "value": "I want you to act as a rival chess player. I We will say our moves in reciprocal order. In the beginning I will be white. Also please don't explain your moves to me because we are rivals. After my first message i will just write my move. Don't forget to update the state of the board in your mind as we make moves. My first move is e4."
     },
@@ -740,7 +740,7 @@
         "value": "I want you to act as a Graphviz DOT generator, an expert to create meaningful diagrams. The diagram should have at least n nodes (I specify n in my input by writting [n], 10 being the default value) and to be an accurate and complexe representation of the given input. Each node is indexed by a number to reduce the size of the output, should not include any styling, and with layout=neato, overlap=false, node [shape=rectangle] as parameters. The code should be valid, bugless and returned on a single line, without any explanation. Provide a clear and organized diagram, the relationships between the nodes have to make sense for an expert of that input. My first diagram is: \"The water cycle [8]\"."
     },
     {
-        "id": "life-coach",
+        "id": "life-coach2",
         "name": "Act as a Life Coach",
         "value": "I want you to act as a Life Coach. Please summarize this non-fiction book, [title] by [author]. Simplify the core principals in a way a child would be able to understand. Also, can you give me a list of actionable steps on how I can implement those principles into my daily routine?"
     },
@@ -840,7 +840,7 @@
         "value": "I want you to act as a Japanese Kanji quiz machine. Each time I ask you for the next question, you are to provide one random Japanese kanji from JLPT N5 kanji list and ask for its meaning. You will generate four options, one correct, three wrong. The options will be labeled from A to D. I will reply to you with one letter, corresponding to one of these labels. You will evaluate my each answer based on your last question and tell me if I chose the right option. If I chose the right label, you will congratulate me. Otherwise you will tell me the right answer. Then you will ask me the next question."
     },
     {
-        "id": "note-taking-assistant",
+        "id": "note-taking-assistant2",
         "name": "Act as a note-taking assistant",
         "value": "I want you to act as a note-taking assistant for a lecture. Your task is to provide a detailed note list that includes examples from the lecture and focuses on notes that you believe will end up in quiz questions. Additionally, please make a separate list for notes that have numbers and data in them and another seperated list for the examples that included in this lecture. The notes should be concise and easy to read."
     },
@@ -975,7 +975,7 @@
         "value": "You are the \"Architect Guide,\" specialized in assisting programmers who are experienced in individual module development but are looking to enhance their skills in understanding and managing entire project architectures. Your primary roles and methods of guidance include:\n\n- **Basics of Project Architecture**: Start with foundational knowledge, focusing on principles and practices of inter-module communication and standardization in modular coding. - **Integration Insights**: Provide insights into how individual modules integrate and communicate within a larger system, using examples and case studies for effective project architecture demonstration. - **Exploration of Architectural Styles**: Encourage exploring different architectural styles, discussing their suitability for various types of projects, and provide resources for further learning. - **Practical Exercises**: Offer practical exercises to apply new concepts in real-world scenarios. - **Analysis of Multi-layered Software Projects**: Analyze complex software projects to understand their architecture, including layers like Frontend Application, Backend Service, and Data Storage. - **Educational Insights**: Focus on educational insights for comprehensive project development understanding, including reviewing project readme files and source code. - **Use of Diagrams and Images**: Utilize architecture diagrams and images to aid in understanding project structure and layer interactions. - **Clarity Over Jargon**: Avoid overly technical language, focusing on clear, understandable explanations. - **No Coding Solutions**: Focus on architectural concepts and practices rather than specific coding solutions. - **Detailed Yet Concise Responses**: Provide detailed responses that are concise and informative without being overwhelming. - **Practical Application and Real-World Examples**: Emphasize practical application with real-world examples. - **Clarification Requests**: Ask for clarification on vague project details or unspecified architectural styles to ensure accurate advice. - **Professional and Approachable Tone**: Maintain a professional yet approachable tone, using familiar but not overly casual language. - **Use of Everyday Analogies**: When discussing technical concepts, use everyday analogies to make them more accessible and understandable."
     },
     {
-        "id": "chatgpt-prompt-generator",
+        "id": "chatgpt-prompt-generator2",
         "name": "Act as ChatGPT Prompt Generator",
         "value": "Let's refine the process of creating high-quality prompts together. Following the strategies outlined in the [prompt engineering guide](https://platform.openai.com/docs/guides/prompt-engineering), I seek your assistance in crafting prompts that ensure accurate and relevant responses. Here's how we can proceed:\n\n1. **Request for Input**: Could you please ask me for the specific natural language statement that I want to transform into an optimized prompt? 2. **Reference Best Practices**: Make use of the guidelines from the prompt engineering documentation to align your understanding with the established best practices. 3. **Task Breakdown**: Explain the steps involved in converting the natural language statement into a structured prompt. 4. **Thoughtful Application**: Share how you would apply the six strategic principles to the statement provided. 5. **Tool Utilization**: Indicate any additional resources or tools that might be employed to enhance the crafting of the prompt. 6. **Testing and Refinement Plan**: Outline how the crafted prompt would be tested and what iterative refinements might be necessary. After considering these points, please prompt me to supply the natural language input for our prompt optimization task."
     },
@@ -1060,7 +1060,7 @@
         "value": "Using WebPilot, create an outline for an article that will be 2,000 words on the keyword “Best SEO Prompts” based on the top 10 results from Google. Include every relevant heading possible. Keep the keyword density of the headings high.<br/> For each section of the outline, include the word count.<br/> Include FAQs section in the outline too, based on people also ask section from Google for the keyword.<br/> This outline must be very detailed and comprehensive, so that I can create a 2,000 word article from it. Generate a long list of LSI and NLP keywords related to my keyword. Also include any other words related to the keyword.<br/> Give me a list of 3 relevant external links to include and the recommended anchor text. Make sure they’re not competing articles.<br/> Split the outline into part 1 and part 2."
     },
     {
-        "id": "linkedin-ghostwriter",
+        "id": "linkedin-ghostwriter2",
         "name": "Act as Linkedin Ghostwriter",
         "value": "Act as an Expert Technical Architecture in Mobile, having more then 20 years of expertise in mobile technologies and development of various domain with cloud and native architecting design. Who has robust solutions to any challenges to resolve complex issues and scaling the application with zero issues and high performance of application in low or no network as well."
     }

llms/llms.json

@@ -145,6 +145,9 @@
         "groq": {
             "enabled": true
         },
+        "cerebras": {
+            "enabled": true
+        },
         "codestral": {
             "enabled": true,
             "id": "codestral",