edsl 0.1.39.dev3__py3-none-any.whl → 0.1.39.dev4__py3-none-any.whl

edsl/inference_services/PerplexityService.py

@@ -1,163 +1,160 @@
-import aiohttp
-import json
-import requests
-from typing import Any, List, Optional
-from edsl.inference_services.rate_limits_cache import rate_limits
-
-# from edsl.inference_services.InferenceServiceABC import InferenceServiceABC
-from edsl.language_models import LanguageModel
-
-from edsl.inference_services.OpenAIService import OpenAIService
-
-
-class PerplexityService(OpenAIService):
-    """Perplexity service class."""
-
-    _inference_service_ = "perplexity"
-    _env_key_name_ = "PERPLEXITY_API_KEY"
-    _base_url_ = "https://api.perplexity.ai"
-    _models_list_cache: List[str] = []
-    # default perplexity parameters
-    _parameters_ = {
-        "temperature": 0.5,
-        "max_tokens": 1000,
-        "top_p": 1,
-        "logprobs": False,
-        "top_logprobs": 3,
-    }
-
-    @classmethod
-    def available(cls) -> List[str]:
-        return [
-            "llama-3.1-sonar-huge-128k-online",
-            "llama-3.1-sonar-large-128k-online",
-            "llama-3.1-sonar-small-128k-online",
-        ]
-
-    @classmethod
-    def create_model(
-        cls, model_name="llama-3.1-sonar-large-128k-online", model_class_name=None
-    ) -> LanguageModel:
-        if model_class_name is None:
-            model_class_name = cls.to_class_name(model_name)
-
-        class LLM(LanguageModel):
-            """
-            Child class of LanguageModel for interacting with Perplexity models
-            """
-
-            key_sequence = cls.key_sequence
-            usage_sequence = cls.usage_sequence
-            input_token_name = cls.input_token_name
-            output_token_name = cls.output_token_name
-
-            _rpm = cls.get_rpm(cls)
-            _tpm = cls.get_tpm(cls)
-
-            _inference_service_ = cls._inference_service_
-            _model_ = model_name
-
-            _parameters_ = {
-                "temperature": 0.5,
-                "max_tokens": 1000,
-                "top_p": 1,
-                "frequency_penalty": 1,
-                "presence_penalty": 0,
-                # "logprobs": False, # Enable this returns 'Neither or both of logprobs and top_logprobs must be set.
-                # "top_logprobs": 3,
-            }
-
-            def sync_client(self):
-                return cls.sync_client()
-
-            def async_client(self):
-                return cls.async_client()
-
-            @classmethod
-            def available(cls) -> list[str]:
-                return cls.sync_client().models.list()
-
-            def get_headers(self) -> dict[str, Any]:
-                client = self.sync_client()
-                response = client.chat.completions.with_raw_response.create(
-                    messages=[
-                        {
-                            "role": "user",
-                            "content": "Say this is a test",
-                        }
-                    ],
-                    model=self.model,
-                )
-                return dict(response.headers)
-
-            def get_rate_limits(self) -> dict[str, Any]:
-                try:
-                    if "openai" in rate_limits:
-                        headers = rate_limits["openai"]
-
-                    else:
-                        headers = self.get_headers()
-
-                except Exception as e:
-                    return {
-                        "rpm": 10_000,
-                        "tpm": 2_000_000,
-                    }
-                else:
-                    return {
-                        "rpm": int(headers["x-ratelimit-limit-requests"]),
-                        "tpm": int(headers["x-ratelimit-limit-tokens"]),
-                    }
-
-            async def async_execute_model_call(
-                self,
-                user_prompt: str,
-                system_prompt: str = "",
-                files_list: Optional[List["Files"]] = None,
-                invigilator: Optional[
-                    "InvigilatorAI"
-                ] = None,  # TBD - can eventually be used for function-calling
-            ) -> dict[str, Any]:
-                """Calls the OpenAI API and returns the API response."""
-                if files_list:
-                    encoded_image = files_list[0].base64_string
-                    content = [{"type": "text", "text": user_prompt}]
-                    content.append(
-                        {
-                            "type": "image_url",
-                            "image_url": {
-                                "url": f"data:image/jpeg;base64,{encoded_image}"
-                            },
-                        }
-                    )
-                else:
-                    content = user_prompt
-                client = self.async_client()
-
-                messages = [
-                    {"role": "system", "content": system_prompt},
-                    {"role": "user", "content": content},
-                ]
-                if system_prompt == "" and self.omit_system_prompt_if_empty:
-                    messages = messages[1:]
-
-                params = {
-                    "model": self.model,
-                    "messages": messages,
-                    "temperature": self.temperature,
-                    "max_tokens": self.max_tokens,
-                    "top_p": self.top_p,
-                    "frequency_penalty": self.frequency_penalty,
-                    "presence_penalty": self.presence_penalty,
-                    # "logprobs": self.logprobs,
-                    # "top_logprobs": self.top_logprobs if self.logprobs else None,
-                }
-                try:
-                    response = await client.chat.completions.create(**params)
-                except Exception as e:
-                    print(e, flush=True)
-                return response.model_dump()
-
-        LLM.__name__ = "LanguageModel"
-
-        return LLM
+import aiohttp
+import json
+import requests
+from typing import Any, List, Optional
+from edsl.inference_services.rate_limits_cache import rate_limits
+
+# from edsl.inference_services.InferenceServiceABC import InferenceServiceABC
+from edsl.language_models import LanguageModel
+
+from edsl.inference_services.OpenAIService import OpenAIService
+
+
+class PerplexityService(OpenAIService):
+    """Perplexity service class."""
+
+    _inference_service_ = "perplexity"
+    _env_key_name_ = "PERPLEXITY_API_KEY"
+    _base_url_ = "https://api.perplexity.ai"
+    _models_list_cache: List[str] = []
+    # default perplexity parameters
+    _parameters_ = {
+        "temperature": 0.5,
+        "max_tokens": 1000,
+        "top_p": 1,
+        "logprobs": False,
+        "top_logprobs": 3,
+    }
+
+    @classmethod
+    def available(cls) -> List[str]:
+        return [
+            "llama-3.1-sonar-huge-128k-online",
+            "llama-3.1-sonar-large-128k-online",
+            "llama-3.1-sonar-small-128k-online",
+        ]
+
+    @classmethod
+    def create_model(
+        cls, model_name="llama-3.1-sonar-large-128k-online", model_class_name=None
+    ) -> LanguageModel:
+        if model_class_name is None:
+            model_class_name = cls.to_class_name(model_name)
+
+        class LLM(LanguageModel):
+            """
+            Child class of LanguageModel for interacting with Perplexity models
+            """
+
+            key_sequence = cls.key_sequence
+            usage_sequence = cls.usage_sequence
+            input_token_name = cls.input_token_name
+            output_token_name = cls.output_token_name
+
+            _inference_service_ = cls._inference_service_
+            _model_ = model_name
+
+            _parameters_ = {
+                "temperature": 0.5,
+                "max_tokens": 1000,
+                "top_p": 1,
+                "frequency_penalty": 1,
+                "presence_penalty": 0,
+                # "logprobs": False, # Enable this returns 'Neither or both of logprobs and top_logprobs must be set.
+                # "top_logprobs": 3,
+            }
+
+            def sync_client(self):
+                return cls.sync_client()
+
+            def async_client(self):
+                return cls.async_client()
+
+            @classmethod
+            def available(cls) -> list[str]:
+                return cls.sync_client().models.list()
+
+            def get_headers(self) -> dict[str, Any]:
+                client = self.sync_client()
+                response = client.chat.completions.with_raw_response.create(
+                    messages=[
+                        {
+                            "role": "user",
+                            "content": "Say this is a test",
+                        }
+                    ],
+                    model=self.model,
+                )
+                return dict(response.headers)
+
+            def get_rate_limits(self) -> dict[str, Any]:
+                try:
+                    if "openai" in rate_limits:
+                        headers = rate_limits["openai"]
+
+                    else:
+                        headers = self.get_headers()
+
+                except Exception as e:
+                    return {
+                        "rpm": 10_000,
+                        "tpm": 2_000_000,
+                    }
+                else:
+                    return {
+                        "rpm": int(headers["x-ratelimit-limit-requests"]),
+                        "tpm": int(headers["x-ratelimit-limit-tokens"]),
+                    }
+
+            async def async_execute_model_call(
+                self,
+                user_prompt: str,
+                system_prompt: str = "",
+                files_list: Optional[List["Files"]] = None,
+                invigilator: Optional[
+                    "InvigilatorAI"
+                ] = None,  # TBD - can eventually be used for function-calling
+            ) -> dict[str, Any]:
+                """Calls the OpenAI API and returns the API response."""
+                if files_list:
+                    encoded_image = files_list[0].base64_string
+                    content = [{"type": "text", "text": user_prompt}]
+                    content.append(
+                        {
+                            "type": "image_url",
+                            "image_url": {
+                                "url": f"data:image/jpeg;base64,{encoded_image}"
+                            },
+                        }
+                    )
+                else:
+                    content = user_prompt
+                client = self.async_client()
+
+                messages = [
+                    {"role": "system", "content": system_prompt},
+                    {"role": "user", "content": content},
+                ]
+                if system_prompt == "" and self.omit_system_prompt_if_empty:
+                    messages = messages[1:]
+
+                params = {
+                    "model": self.model,
+                    "messages": messages,
+                    "temperature": self.temperature,
+                    "max_tokens": self.max_tokens,
+                    "top_p": self.top_p,
+                    "frequency_penalty": self.frequency_penalty,
+                    "presence_penalty": self.presence_penalty,
+                    # "logprobs": self.logprobs,
+                    # "top_logprobs": self.top_logprobs if self.logprobs else None,
+                }
+                try:
+                    response = await client.chat.completions.create(**params)
+                except Exception as e:
+                    print(e, flush=True)
+                return response.model_dump()
+
+        LLM.__name__ = "LanguageModel"
+
+        return LLM

edsl/inference_services/ServiceAvailability.py

@@ -0,0 +1,135 @@
+from enum import Enum
+from typing import List, Optional, TYPE_CHECKING
+from functools import partial
+import warnings
+
+from edsl.inference_services.data_structures import AvailableModels, ModelNamesList
+
+if TYPE_CHECKING:
+    from edsl.inference_services.InferenceServiceABC import InferenceServiceABC
+
+
+class ModelSource(Enum):
+    LOCAL = "local"
+    COOP = "coop"
+    CACHE = "cache"
+
+
+class ServiceAvailability:
+    """This class is responsible for fetching the available models from different sources."""
+
+    _coop_model_list = None
+
+    def __init__(self, source_order: Optional[List[ModelSource]] = None):
+        """
+        Initialize with custom source order.
+        Default order is LOCAL -> COOP -> CACHE
+        """
+        self.source_order = source_order or [
+            ModelSource.LOCAL,
+            ModelSource.COOP,
+            ModelSource.CACHE,
+        ]
+
+        # Map sources to their fetch functions
+        self._source_fetchers = {
+            ModelSource.LOCAL: self._fetch_from_local_service,
+            ModelSource.COOP: self._fetch_from_coop,
+            ModelSource.CACHE: self._fetch_from_cache,
+        }
+
+    @classmethod
+    def models_from_coop(cls) -> AvailableModels:
+        if not cls._coop_model_list:
+            from edsl.coop.coop import Coop
+
+            c = Coop()
+            coop_model_list = c.fetch_models()
+            cls._coop_model_list = coop_model_list
+        return cls._coop_model_list
+
+    def get_service_available(
+        self, service: "InferenceServiceABC", warn: bool = False
+    ) -> ModelNamesList:
+        """
+        Try to fetch available models from sources in specified order.
+        Returns first successful result.
+        """
+        last_error = None
+
+        for source in self.source_order:
+            try:
+                fetch_func = partial(self._source_fetchers[source], service)
+                result = fetch_func()
+
+                # Cache successful result
+                service._models_list_cache = result
+                return result
+
+            except Exception as e:
+                last_error = e
+                if warn:
+                    self._warn_source_failed(service, source)
+                continue
+
+        # If we get here, all sources failed
+        raise RuntimeError(
+            f"All sources failed to fetch models. Last error: {last_error}"
+        )
+
+    @staticmethod
+    def _fetch_from_local_service(service: "InferenceServiceABC") -> ModelNamesList:
+        """Attempt to fetch models directly from the service."""
+        return service.available()
+
+    @classmethod
+    def _fetch_from_coop(cls, service: "InferenceServiceABC") -> ModelNamesList:
+        """Fetch models from Coop."""
+        models_from_coop = cls.models_from_coop()
+        return models_from_coop.get(service._inference_service_, [])
+
+    @staticmethod
+    def _fetch_from_cache(service: "InferenceServiceABC") -> ModelNamesList:
+        """Fetch models from local cache."""
+        from edsl.inference_services.models_available_cache import models_available
+
+        return models_available.get(service._inference_service_, [])
+
+    def _warn_source_failed(self, service: "InferenceServiceABC", source: ModelSource):
+        """Display appropriate warning message based on failed source."""
+        messages = {
+            ModelSource.LOCAL: f"""Error getting models for {service._inference_service_}. 
+                Check that you have properly stored your Expected Parrot API key and activated remote inference, 
+                or stored your own API keys for the language models that you want to use.
+                See https://docs.expectedparrot.com/en/latest/api_keys.html for instructions on storing API keys.
+                Trying next source.""",
+            ModelSource.COOP: f"Error getting models from Coop for {service._inference_service_}. Trying next source.",
+            ModelSource.CACHE: f"Error getting models from cache for {service._inference_service_}.",
+        }
+        warnings.warn(messages[source], UserWarning)
+
+
+if __name__ == "__main__":
+    # sa = ServiceAvailability()
+    # models_from_coop = sa.models_from_coop()
+    # print(models_from_coop)
+    from edsl.inference_services.OpenAIService import OpenAIService
+
+    openai_models = ServiceAvailability._fetch_from_local_service(OpenAIService())
+    print(openai_models)
+
+# Example usage:
+"""
+# Default order (LOCAL -> COOP -> CACHE)
+availability = ServiceAvailability()
+
+# Custom order (COOP -> LOCAL -> CACHE)
+availability_coop_first = ServiceAvailability([
+    ModelSource.COOP,
+    ModelSource.LOCAL,
+    ModelSource.CACHE
+])
+
+# Get available models using custom order
+models = availability_coop_first.get_service_available(service, warn=True)
+"""

edsl/inference_services/TestService.py

@@ -1,89 +1,90 @@
-from typing import Any, List, Optional
-import os
-import asyncio
-from edsl.inference_services.InferenceServiceABC import InferenceServiceABC
-from edsl.language_models import LanguageModel
-from edsl.inference_services.rate_limits_cache import rate_limits
-from edsl.utilities.utilities import fix_partial_correct_response
-
-from edsl.enums import InferenceServiceType
-import random
-
-
-class TestService(InferenceServiceABC):
-    """OpenAI service class."""
-
-    _inference_service_ = "test"
-    _env_key_name_ = None
-    _base_url_ = None
-
-    _sync_client_ = None
-    _async_client_ = None
-
-    _sync_client_instance = None
-    _async_client_instance = None
-
-    key_sequence = None
-    usage_sequence = None
-    model_exclude_list = []
-    input_token_name = "prompt_tokens"
-    output_token_name = "completion_tokens"
-
-    @classmethod
-    def available(cls) -> list[str]:
-        return ["test"]
-
-    @classmethod
-    def create_model(cls, model_name, model_class_name=None) -> LanguageModel:
-        throw_exception = False
-
-        class TestServiceLanguageModel(LanguageModel):
-            _model_ = "test"
-            _parameters_ = {"temperature": 0.5}
-            _inference_service_ = InferenceServiceType.TEST.value
-            usage_sequence = ["usage"]
-            key_sequence = ["message", 0, "text"]
-            input_token_name = cls.input_token_name
-            output_token_name = cls.output_token_name
-            _rpm = 1000
-            _tpm = 100000
-
-            @property
-            def _canned_response(self):
-                if hasattr(self, "canned_response"):
-                    return self.canned_response
-                else:
-                    return "Hello, world"
-
-            async def async_execute_model_call(
-                self,
-                user_prompt: str,
-                system_prompt: str,
-                # func: Optional[callable] = None,
-                files_list: Optional[List["File"]] = None,
-            ) -> dict[str, Any]:
-                await asyncio.sleep(0.1)
-                # return {"message": """{"answer": "Hello, world"}"""}
-
-                if hasattr(self, "func"):
-                    return {
-                        "message": [
-                            {"text": self.func(user_prompt, system_prompt, files_list)}
-                        ],
-                        "usage": {"prompt_tokens": 1, "completion_tokens": 1},
-                    }
-
-                if hasattr(self, "throw_exception") and self.throw_exception:
-                    if hasattr(self, "exception_probability"):
-                        p = self.exception_probability
-                    else:
-                        p = 1
-
-                    if random.random() < p:
-                        raise Exception("This is a test error")
-                return {
-                    "message": [{"text": f"{self._canned_response}"}],
-                    "usage": {"prompt_tokens": 1, "completion_tokens": 1},
-                }
-
-        return TestServiceLanguageModel
+from typing import Any, List, Optional
+import os
+import asyncio
+from edsl.inference_services.InferenceServiceABC import InferenceServiceABC
+from edsl.language_models.LanguageModel import LanguageModel
+from edsl.inference_services.rate_limits_cache import rate_limits
+from edsl.utilities.utilities import fix_partial_correct_response
+
+from edsl.enums import InferenceServiceType
+import random
+
+
+class TestService(InferenceServiceABC):
+    """OpenAI service class."""
+
+    _inference_service_ = "test"
+    _env_key_name_ = None
+    _base_url_ = None
+
+    _sync_client_ = None
+    _async_client_ = None
+
+    _sync_client_instance = None
+    _async_client_instance = None
+
+    key_sequence = None
+    usage_sequence = None
+    model_exclude_list = []
+    input_token_name = "prompt_tokens"
+    output_token_name = "completion_tokens"
+
+    @classmethod
+    def available(cls) -> list[str]:
+        return ["test"]
+
+    @classmethod
+    def create_model(cls, model_name, model_class_name=None) -> LanguageModel:
+        throw_exception = False
+
+        class TestServiceLanguageModel(LanguageModel):
+            _model_ = "test"
+            _parameters_ = {"temperature": 0.5}
+            _inference_service_ = InferenceServiceType.TEST.value
+            usage_sequence = ["usage"]
+            key_sequence = ["message", 0, "text"]
+            input_token_name = cls.input_token_name
+            output_token_name = cls.output_token_name
+            _rpm = 1000
+            _tpm = 100000
+
+            @property
+            def _canned_response(self):
+                if hasattr(self, "canned_response"):
+
+                    return self.canned_response
+                else:
+                    return "Hello, world"
+
+            async def async_execute_model_call(
+                self,
+                user_prompt: str,
+                system_prompt: str,
+                # func: Optional[callable] = None,
+                files_list: Optional[List["File"]] = None,
+            ) -> dict[str, Any]:
+                await asyncio.sleep(0.1)
+
+                if hasattr(self, "throw_exception") and self.throw_exception:
+                    if hasattr(self, "exception_probability"):
+                        p = self.exception_probability
+                    else:
+                        p = 1
+
+                    if random.random() < p:
+                        raise Exception("This is a test error")
+
+                if hasattr(self, "func"):
+                    return {
+                        "message": [
+                            {"text": self.func(user_prompt, system_prompt, files_list)}
+                        ],
+                        "usage": {"prompt_tokens": 1, "completion_tokens": 1},
+                    }
+
+                return {
+                    "message": [{"text": f"{self._canned_response}"}],
+                    "usage": {"prompt_tokens": 1, "completion_tokens": 1},
+                }
+
+        return TestServiceLanguageModel