Make using .env work.

Author: johann-petrak

docs/configuration.md

@@ -39,20 +39,33 @@ names must match the provider names known in the [litellm](https://docs.litellm.
 Parameters specified for each of the providers in the `providers`  section apply to every llm in the `llms`  section unless the same 
 parameter is also specified for the llm, in which case that value takes precedence.
 
+IMPORTANT: environment variable based settings, e.g. `api_key_env` will respect any `.env` file in the current directory when 
+the config file is read, updated or an LLM is initialized!
+
 The following parameters are known and supported in the `llms` and/or `providers` sections:
 
 * `llm` (`llms` section only): specifies a specific model using the format `providername/modelid`. 
 * `api_key`: the literal API key to use
-* `api_key_env`:  the environment variable which contains the API key
+* `api_key_env`:  the environment variable which contains the API key, using the value from the current environment or whatever is defined in .env
 * `api_url`: the base URL to use for the model, e.g. for an ollama server. The URL may contain placeholders which will get replaced with
    the model name (`${model}`), or the user and password for basic authentication (`${user}`, `${password}`), e.g.
    `http://${user}:${password}@localhost:11434`
 * `user`, `password`: the user and password to use for basic authentication, this requires `api_url` to also be specified with the 
    corresponding placeholders
+* `user_env`, `password_env`: the environment variable to get the user or password from, this uses the value from the current enviroment or whatever
+   has been set in any `.env`  file in the current directory.
 * `alias` (`llms` section only): an alias name for the model which will have to be used in the API. If no `alias` is specified, the name
    specified for `llm` is used. 
 * `num_retries`: if present, can specify the number of retries to perform if an error occurs before giving up
 * `timeout`: if present, raise timeout error after that many seconds
+* `via_streaming`: the default approach for getting LLM responses is to wait for the complete response to get returned. This can lead to time-outs
+   or other problems with some LLMs. When this is set to true, the response will get retrieved using streaming. However, some information like 
+   cost is not available as part of the `llms_wrapper`  response if streaming is enabled. 
+* `min_delay`: the minimum delay in seconds to ensure between requests sent to the model from code running in the same process and thread. 
+* `cost_per_prompt_token`: set or override the cost per prompt token for the model
+* `cost_per_output_token`: set or override the cost per output token for the model
+* `max_output_tokens`: set or override the maximum output tokens for the model
+* `max_input_tokens`: set or override the maximum input tokens for the model
 
 All other settings are passed as is to the model invocation function. Different providers or APIs may support different parameters, but 
 most will support `temperature`, `max_tokens` and `top_p` 

docs/llms_wrapper/config.html

@@ -98,6 +98,7 @@ <h2 class="section-title" id="header-functions">Functions</h2>
         A dict with the configuration
     &#34;&#34;&#34;
     # read config file as json, yaml or toml, depending on file extension
+    load_dotenv(override=True)
     if filepath.endswith(&#34;.json&#34;):
         with open(filepath, &#39;r&#39;) as f:
             config = json.load(f)
@@ -229,6 +230,7 @@ <h2 id="returns">Returns</h2>
     Returns:
         the updated configuration dict
     &#34;&#34;&#34;
+    load_dotenv(override=True)
     for i, llm in enumerate(config[&#34;llms&#34;]):
         if isinstance(llm, str):
             provider, model = llm.split(&#34;/&#34;)

docs/llms_wrapper/llms.html

@@ -724,6 +724,8 @@ <h3>Methods</h3>
         project name (so far this only works for local phoenix instances). Default URI for a local installation
         is &#34;http://0.0.0.0:6006/v1/traces&#34;
         &#34;&#34;&#34;
+        # before anything, make sure we have loaded any dotenv file to override any env var settings for the api keys
+        load_dotenv(override=True)
         if config is None:
             config = dict(llms=[])
         self.config = deepcopy(config)

llms_wrapper/config.py

@@ -20,6 +20,7 @@
 import hjson
 import tomllib
 import re
+from dotenv import load_dotenv
 
 ## Suppress the annoying litellm warning
 with warnings.catch_warnings():
@@ -69,6 +70,7 @@ def read_config_file(filepath: str, update: bool = True) -> dict:
         A dict with the configuration
     """
     # read config file as json, yaml or toml, depending on file extension
+    load_dotenv(override=True)
     if filepath.endswith(".json"):
         with open(filepath, 'r') as f:
             config = json.load(f)
@@ -148,6 +150,7 @@ def update_llm_config(config: dict):
     Returns:
         the updated configuration dict
     """
+    load_dotenv(override=True)
     for i, llm in enumerate(config["llms"]):
         if isinstance(llm, str):
             provider, model = llm.split("/")

llms_wrapper/llms.py

@@ -12,6 +12,7 @@
 import traceback
 import inspect
 import docstring_parser
+from dotenv import load_dotenv
 from loguru import logger
 import typing
 from typing import Optional, Dict, List, Union, Tuple, Callable, get_args, get_origin
@@ -292,6 +293,8 @@ def __init__(self, config: Dict = None, debug: bool = False, use_phoenix: Option
         project name (so far this only works for local phoenix instances). Default URI for a local installation
         is "http://0.0.0.0:6006/v1/traces"
         """
+        # before anything, make sure we have loaded any dotenv file to override any env var settings for the api keys
+        load_dotenv(override=True)
         if config is None:
             config = dict(llms=[])
         self.config = deepcopy(config)

pyproject.toml

@@ -25,6 +25,7 @@ dependencies = [
     "hjson",
     "loguru",
     "docstring_parser",
+    "python-dotenv",
 ]
 
 [project.optional-dependencies]