prompture 0.0.29.dev8__py3-none-any.whl → 0.0.38.dev2__py3-none-any.whl

prompture/drivers/hugging_driver.py

@@ -1,14 +1,14 @@
 import os
+from typing import Any
+
 import requests
+
 from ..driver import Driver
-from typing import Any, Dict
 
 
 class HuggingFaceDriver(Driver):
     # Hugging Face is usage-based (credits/subscription), but we set costs to 0 for now.
-    MODEL_PRICING = {
-        "default": {"prompt": 0.0, "completion": 0.0}
-    }
+    MODEL_PRICING = {"default": {"prompt": 0.0, "completion": 0.0}}
 
     def __init__(self, endpoint: str | None = None, token: str | None = None, model: str = "bert-base-uncased"):
         self.endpoint = endpoint or os.getenv("HF_ENDPOINT")
@@ -22,7 +22,7 @@ class HuggingFaceDriver(Driver):
 
         self.headers = {"Authorization": f"Bearer {self.token}"}
 
-    def generate(self, prompt: str, options: Dict[str, Any]) -> Dict[str, Any]:
+    def generate(self, prompt: str, options: dict[str, Any]) -> dict[str, Any]:
         payload = {
             "inputs": prompt,
             "parameters": options,  # HF allows temperature, max_new_tokens, etc. here
@@ -32,7 +32,7 @@ class HuggingFaceDriver(Driver):
             r.raise_for_status()
             response_data = r.json()
         except Exception as e:
-            raise RuntimeError(f"HuggingFaceDriver request failed: {e}")
+            raise RuntimeError(f"HuggingFaceDriver request failed: {e}") from e
 
         # Different HF models return slightly different response formats
         # Text-generation models usually return [{"generated_text": "..."}]

prompture/drivers/lmstudio_driver.py

@@ -1,59 +1,109 @@
-import os
 import json
-import requests
 import logging
+import os
+from typing import Any, Optional
+
+import requests
+
 from ..driver import Driver
-from typing import Any, Dict
 
 logger = logging.getLogger(__name__)
 
 
 class LMStudioDriver(Driver):
-    # LM Studio is local – costs are always zero.
-    MODEL_PRICING = {
-        "default": {"prompt": 0.0, "completion": 0.0}
-    }
+    supports_json_mode = True
+    supports_json_schema = True
+    supports_vision = True
 
-    def __init__(self, endpoint: str | None = None, model: str = "deepseek/deepseek-r1-0528-qwen3-8b"):
+    # LM Studio is local – costs are always zero.
+    MODEL_PRICING = {"default": {"prompt": 0.0, "completion": 0.0}}
+
+    def __init__(
+        self,
+        endpoint: str | None = None,
+        model: str = "deepseek/deepseek-r1-0528-qwen3-8b",
+        api_key: str | None = None,
+    ):
         # Allow override via env var
-        self.endpoint = endpoint or os.getenv(
-            "LMSTUDIO_ENDPOINT", "http://127.0.0.1:1234/v1/chat/completions"
-        )
+        self.endpoint = endpoint or os.getenv("LMSTUDIO_ENDPOINT", "http://127.0.0.1:1234/v1/chat/completions")
         self.model = model
-        self.options: Dict[str, Any] = {}
+        self.options: dict[str, Any] = {}
+
+        # Derive base_url once for reuse across management endpoints
+        self.base_url = self.endpoint.split("/v1/")[0]
+
+        # API key for LM Studio 0.4.0+ authentication
+        self.api_key = api_key or os.getenv("LMSTUDIO_API_KEY")
+        self._headers = self._build_headers()
 
         # Validate connection to LM Studio server
         self._validate_connection()
 
+    def _build_headers(self) -> dict[str, str]:
+        """Build request headers, including auth if an API key is configured."""
+        headers: dict[str, str] = {"Content-Type": "application/json"}
+        if self.api_key:
+            headers["Authorization"] = f"Bearer {self.api_key}"
+        return headers
+
     def _validate_connection(self):
         """Validate connection to the LM Studio server."""
         try:
-            base_url = self.endpoint.split("/v1/")[0]
-            health_url = f"{base_url}/v1/models"
+            health_url = f"{self.base_url}/v1/models"
 
             logger.debug(f"Validating connection to LM Studio server at: {health_url}")
-            response = requests.get(health_url, timeout=5)
+            response = requests.get(health_url, headers=self._headers, timeout=5)
             response.raise_for_status()
             logger.debug("Connection to LM Studio server validated successfully")
         except requests.exceptions.RequestException as e:
             logger.warning(f"Could not validate connection to LM Studio server: {e}")
 
-    def generate(self, prompt: str, options: Dict[str, Any] = None) -> Dict[str, Any]:
+    supports_messages = True
+
+    def _prepare_messages(self, messages: list[dict[str, Any]]) -> list[dict[str, Any]]:
+        from .vision_helpers import _prepare_openai_vision_messages
+
+        return _prepare_openai_vision_messages(messages)
+
+    def generate(self, prompt: str, options: Optional[dict[str, Any]] = None) -> dict[str, Any]:
+        messages = [{"role": "user", "content": prompt}]
+        return self._do_generate(messages, options)
+
+    def generate_messages(self, messages: list[dict[str, str]], options: dict[str, Any]) -> dict[str, Any]:
+        return self._do_generate(self._prepare_messages(messages), options)
+
+    def _do_generate(self, messages: list[dict[str, str]], options: Optional[dict[str, Any]] = None) -> dict[str, Any]:
         merged_options = self.options.copy()
         if options:
             merged_options.update(options)
 
         payload = {
             "model": merged_options.get("model", self.model),
-            "messages": [{"role": "user", "content": prompt}],
+            "messages": messages,
             "temperature": merged_options.get("temperature", 0.7),
         }
 
+        # Native JSON mode support (LM Studio requires json_schema, not json_object)
+        if merged_options.get("json_mode"):
+            json_schema = merged_options.get("json_schema")
+            if json_schema:
+                payload["response_format"] = {
+                    "type": "json_schema",
+                    "json_schema": {
+                        "name": "extraction",
+                        "schema": json_schema,
+                    },
+                }
+            else:
+                # No schema provided — omit response_format entirely;
+                # LM Studio rejects "json_object" type.
+                pass
+
         try:
             logger.debug(f"Sending request to LM Studio endpoint: {self.endpoint}")
             logger.debug(f"Request payload: {payload}")
 
-            r = requests.post(self.endpoint, json=payload, timeout=120)
+            r = requests.post(self.endpoint, json=payload, headers=self._headers, timeout=120)
             r.raise_for_status()
 
             response_data = r.json()
@@ -70,7 +120,7 @@ class LMStudioDriver(Driver):
             raise
         except Exception as e:
             logger.error(f"Unexpected error in LM Studio request: {e}")
-            raise RuntimeError(f"LM Studio request failed: {e}")
+            raise RuntimeError(f"LM Studio request failed: {e}") from e
 
         # Extract text
         text = response_data["choices"][0]["message"]["content"]
@@ -91,3 +141,31 @@ class LMStudioDriver(Driver):
         }
 
         return {"text": text, "meta": meta}
+
+    # -- Model management (LM Studio 0.4.0+) ----------------------------------
+
+    def list_models(self) -> list[dict[str, Any]]:
+        """List currently loaded models via GET /v1/models (OpenAI-compatible)."""
+        url = f"{self.base_url}/v1/models"
+        r = requests.get(url, headers=self._headers, timeout=10)
+        r.raise_for_status()
+        data = r.json()
+        return data.get("data", [])
+
+    def load_model(self, model: str, context_length: int | None = None) -> dict[str, Any]:
+        """Load a model into LM Studio via POST /api/v1/models/load."""
+        url = f"{self.base_url}/api/v1/models/load"
+        payload: dict[str, Any] = {"model": model}
+        if context_length is not None:
+            payload["context_length"] = context_length
+        r = requests.post(url, json=payload, headers=self._headers, timeout=120)
+        r.raise_for_status()
+        return r.json()
+
+    def unload_model(self, model: str) -> dict[str, Any]:
+        """Unload a model from LM Studio via POST /api/v1/models/unload."""
+        url = f"{self.base_url}/api/v1/models/unload"
+        payload = {"instance_id": model}
+        r = requests.post(url, json=payload, headers=self._headers, timeout=30)
+        r.raise_for_status()
+        return r.json()

prompture/drivers/local_http_driver.py

@@ -1,27 +1,27 @@
 import os
+from typing import Any
+
 import requests
+
 from ..driver import Driver
-from typing import Any, Dict
 
 
 class LocalHTTPDriver(Driver):
     # Default: no cost; extend if your local service has pricing logic
-    MODEL_PRICING = {
-        "default": {"prompt": 0.0, "completion": 0.0}
-    }
+    MODEL_PRICING = {"default": {"prompt": 0.0, "completion": 0.0}}
 
     def __init__(self, endpoint: str | None = None, model: str = "local-model"):
         self.endpoint = endpoint or os.getenv("LOCAL_HTTP_ENDPOINT", "http://localhost:8000/generate")
         self.model = model
 
-    def generate(self, prompt: str, options: Dict[str, Any]) -> Dict[str, Any]:
+    def generate(self, prompt: str, options: dict[str, Any]) -> dict[str, Any]:
         payload = {"prompt": prompt, "options": options}
         try:
             r = requests.post(self.endpoint, json=payload, timeout=options.get("timeout", 30))
             r.raise_for_status()
             response_data = r.json()
         except Exception as e:
-            raise RuntimeError(f"LocalHTTPDriver request failed: {e}")
+            raise RuntimeError(f"LocalHTTPDriver request failed: {e}") from e
 
         # If the local API already provides {"text": "...", "meta": {...}}, just return it
         if "text" in response_data and "meta" in response_data:

prompture/drivers/ollama_driver.py

@@ -1,38 +1,42 @@
-import os
 import json
-import requests
 import logging
+import os
+from collections.abc import Iterator
+from typing import Any, Optional
+
+import requests
+
 from ..driver import Driver
-from typing import Any, Dict
 
 logger = logging.getLogger(__name__)
 
 
 class OllamaDriver(Driver):
+    supports_json_mode = True
+    supports_json_schema = True
+    supports_streaming = True
+    supports_vision = True
+
     # Ollama is free – costs are always zero.
-    MODEL_PRICING = {
-        "default": {"prompt": 0.0, "completion": 0.0}
-    }
+    MODEL_PRICING = {"default": {"prompt": 0.0, "completion": 0.0}}
 
     def __init__(self, endpoint: str | None = None, model: str = "llama3"):
         # Allow override via env var
-        self.endpoint = endpoint or os.getenv(
-            "OLLAMA_ENDPOINT", "http://localhost:11434/api/generate"
-        )
+        self.endpoint = endpoint or os.getenv("OLLAMA_ENDPOINT", "http://localhost:11434/api/generate")
         self.model = model
         self.options = {}  # Initialize empty options dict
-        
+
         # Validate connection to Ollama server
         self._validate_connection()
-        
+
     def _validate_connection(self):
         """Validate connection to the Ollama server."""
         try:
             # Send a simple HEAD request to check if server is accessible
             # Use the base API endpoint without the specific path
-            base_url = self.endpoint.split('/api/')[0]
+            base_url = self.endpoint.split("/api/")[0]
             health_url = f"{base_url}/api/version"
-            
+
             logger.debug(f"Validating connection to Ollama server at: {health_url}")
             response = requests.head(health_url, timeout=5)
             response.raise_for_status()
@@ -42,7 +46,14 @@ class OllamaDriver(Driver):
             # We don't raise an error here to allow for delayed server startup
             # The actual error will be raised when generate() is called
 
-    def generate(self, prompt: str, options: Dict[str, Any] = None) -> Dict[str, Any]:
+    supports_messages = True
+
+    def _prepare_messages(self, messages: list[dict[str, Any]]) -> list[dict[str, Any]]:
+        from .vision_helpers import _prepare_ollama_vision_messages
+
+        return _prepare_ollama_vision_messages(messages)
+
+    def generate(self, prompt: str, options: Optional[dict[str, Any]] = None) -> dict[str, Any]:
         # Merge instance options with call-specific options
         merged_options = self.options.copy()
         if options:
@@ -54,6 +65,11 @@ class OllamaDriver(Driver):
             "stream": False,
         }
 
+        # Native JSON mode / structured output support
+        if merged_options.get("json_mode"):
+            json_schema = merged_options.get("json_schema")
+            payload["format"] = json_schema if json_schema else "json"
+
         # Add any Ollama-specific options from merged_options
         if "temperature" in merged_options:
             payload["temperature"] = merged_options["temperature"]
@@ -65,21 +81,21 @@ class OllamaDriver(Driver):
         try:
             logger.debug(f"Sending request to Ollama endpoint: {self.endpoint}")
             logger.debug(f"Request payload: {payload}")
-            
+
             r = requests.post(self.endpoint, json=payload, timeout=120)
             logger.debug(f"Response status code: {r.status_code}")
-            
+
             r.raise_for_status()
-            
+
             response_text = r.text
             logger.debug(f"Raw response text: {response_text}")
-            
+
             response_data = r.json()
             logger.debug(f"Parsed response data: {response_data}")
-            
+
             if not isinstance(response_data, dict):
                 raise ValueError(f"Expected dict response, got {type(response_data)}")
-                
+
         except requests.exceptions.ConnectionError as e:
             logger.error(f"Connection error to Ollama endpoint: {e}")
             # Preserve original exception
@@ -91,11 +107,11 @@ class OllamaDriver(Driver):
         except json.JSONDecodeError as e:
             logger.error(f"Failed to decode JSON response: {e}")
             # Re-raise JSONDecodeError with more context
-            raise json.JSONDecodeError(f"Invalid JSON response from Ollama: {e.msg}", e.doc, e.pos)
+            raise json.JSONDecodeError(f"Invalid JSON response from Ollama: {e.msg}", e.doc, e.pos) from e
         except Exception as e:
             logger.error(f"Unexpected error in Ollama request: {e}")
             # Only wrap unknown exceptions in RuntimeError
-            raise RuntimeError(f"Ollama request failed: {e}")
+            raise RuntimeError(f"Ollama request failed: {e}") from e
 
         # Extract token counts
         prompt_tokens = response_data.get("prompt_eval_count", 0)
@@ -113,4 +129,133 @@ class OllamaDriver(Driver):
         }
 
         # Ollama returns text in "response"
-        return {"text": response_data.get("response", ""), "meta": meta}
+        return {"text": response_data.get("response", ""), "meta": meta}
+
+    # ------------------------------------------------------------------
+    # Streaming
+    # ------------------------------------------------------------------
+
+    def generate_messages_stream(
+        self,
+        messages: list[dict[str, Any]],
+        options: dict[str, Any],
+    ) -> Iterator[dict[str, Any]]:
+        """Yield response chunks via Ollama streaming API."""
+        merged_options = self.options.copy()
+        if options:
+            merged_options.update(options)
+
+        chat_endpoint = self.endpoint.replace("/api/generate", "/api/chat")
+
+        payload: dict[str, Any] = {
+            "model": merged_options.get("model", self.model),
+            "messages": messages,
+            "stream": True,
+        }
+
+        if merged_options.get("json_mode"):
+            json_schema = merged_options.get("json_schema")
+            payload["format"] = json_schema if json_schema else "json"
+        if "temperature" in merged_options:
+            payload["temperature"] = merged_options["temperature"]
+        if "top_p" in merged_options:
+            payload["top_p"] = merged_options["top_p"]
+        if "top_k" in merged_options:
+            payload["top_k"] = merged_options["top_k"]
+
+        full_text = ""
+        prompt_tokens = 0
+        completion_tokens = 0
+
+        r = requests.post(chat_endpoint, json=payload, timeout=120, stream=True)
+        r.raise_for_status()
+
+        for line in r.iter_lines():
+            if not line:
+                continue
+            chunk = json.loads(line)
+            if chunk.get("done"):
+                prompt_tokens = chunk.get("prompt_eval_count", 0)
+                completion_tokens = chunk.get("eval_count", 0)
+            else:
+                content = chunk.get("message", {}).get("content", "")
+                if content:
+                    full_text += content
+                    yield {"type": "delta", "text": content}
+
+        total_tokens = prompt_tokens + completion_tokens
+        yield {
+            "type": "done",
+            "text": full_text,
+            "meta": {
+                "prompt_tokens": prompt_tokens,
+                "completion_tokens": completion_tokens,
+                "total_tokens": total_tokens,
+                "cost": 0.0,
+                "raw_response": {},
+                "model_name": merged_options.get("model", self.model),
+            },
+        }
+
+    def generate_messages(self, messages: list[dict[str, Any]], options: dict[str, Any]) -> dict[str, Any]:
+        """Use Ollama's /api/chat endpoint for multi-turn conversations."""
+        messages = self._prepare_messages(messages)
+        merged_options = self.options.copy()
+        if options:
+            merged_options.update(options)
+
+        # Derive the chat endpoint from the generate endpoint
+        chat_endpoint = self.endpoint.replace("/api/generate", "/api/chat")
+
+        payload: dict[str, Any] = {
+            "model": merged_options.get("model", self.model),
+            "messages": messages,
+            "stream": False,
+        }
+
+        # Native JSON mode / structured output support
+        if merged_options.get("json_mode"):
+            json_schema = merged_options.get("json_schema")
+            payload["format"] = json_schema if json_schema else "json"
+
+        if "temperature" in merged_options:
+            payload["temperature"] = merged_options["temperature"]
+        if "top_p" in merged_options:
+            payload["top_p"] = merged_options["top_p"]
+        if "top_k" in merged_options:
+            payload["top_k"] = merged_options["top_k"]
+
+        try:
+            logger.debug(f"Sending chat request to Ollama endpoint: {chat_endpoint}")
+            r = requests.post(chat_endpoint, json=payload, timeout=120)
+            r.raise_for_status()
+            response_data = r.json()
+
+            if not isinstance(response_data, dict):
+                raise ValueError(f"Expected dict response, got {type(response_data)}")
+        except requests.exceptions.ConnectionError:
+            raise
+        except requests.exceptions.HTTPError:
+            raise
+        except json.JSONDecodeError as e:
+            raise json.JSONDecodeError(f"Invalid JSON response from Ollama: {e.msg}", e.doc, e.pos) from e
+        except Exception as e:
+            raise RuntimeError(f"Ollama chat request failed: {e}") from e
+
+        prompt_tokens = response_data.get("prompt_eval_count", 0)
+        completion_tokens = response_data.get("eval_count", 0)
+        total_tokens = prompt_tokens + completion_tokens
+
+        meta = {
+            "prompt_tokens": prompt_tokens,
+            "completion_tokens": completion_tokens,
+            "total_tokens": total_tokens,
+            "cost": 0.0,
+            "raw_response": response_data,
+            "model_name": merged_options.get("model", self.model),
+        }
+
+        # Chat endpoint returns response in message.content
+        message = response_data.get("message", {})
+        text = message.get("content", "")
+        return {"text": text, "meta": meta}