Close #28 - Implement model selection

Implement model selection for current sources. The models can be set
with 'model' keys for each source.

.gitignore

@@ -1,6 +1,7 @@
 .coverage
 .directory
 .tox
+.tox-docker
 /env
 __pycache__
 tags

autoload/neural/config.vim

@@ -20,19 +20,21 @@ let s:defaults = {
 \   'source': {
 \       'openai': {
 \           'api_key': '',
+\           'frequency_penalty': 0.1,
+\           'max_tokens': 1024,
+\           'model': 'text-davinci-003',
+\           'presence_penalty': 0.1,
 \           'temperature': 0.2,
 \           'top_p': 1,
-\           'max_tokens': 1024,
-\           'presence_penalty': 0.1,
-\           'frequency_penalty': 0.1,
 \       },
 \       'chatgpt': {
 \           'api_key': '',
+\           'frequency_penalty': 0.1,
+\           'max_tokens': 2048,
+\           'model': 'gpt-3.5-turbo',
+\           'presence_penalty': 0.1,
 \           'temperature': 0.2,
 \           'top_p': 1,
-\           'max_tokens': 2048,
-\           'presence_penalty': 0.1,
-\           'frequency_penalty': 0.1,
 \       },
 \   },
 \}

doc/neural.txt

@@ -215,6 +215,15 @@ g:neural.source.openai.max_tokens           *g:neural.source.openai.max_tokens*
   This translates to roughly `¾` of a word (e.g. `100 tokens ~= 75 words`).
 
 
+g:neural.source.openai.model                     *g:neural.source.openai.model*
+                                             *vim.g.neural.source.openai.model*
+  Type: |String|
+  Default: `'text-davinci-003'`
+
+  The model to use for OpenAI. Please consult OpenAI's documentation for more
+  information on models: https://platform.openai.com/docs/models/overview
+
+
 g:neural.source.openai.presence_penalty
                                       *g:neural.source.openai.presence_penalty*
                                   *vim.g.neural.source.openai.presence_penalty*
@@ -289,6 +298,15 @@ g:neural.source.chatgpt.max_tokens         *g:neural.source.chatgpt.max_tokens*
   See Also: |g:neural.source.openai.max_tokens|
 
 
+g:neural.source.chatgpt.model                   *g:neural.source.chatgpt.model*
+                                            *vim.g.neural.source.chatgpt.model*
+  Type: |String|
+  Default: `'gpt-3.5-turbo'`
+
+  The model to use for ChatGPT. Please consult OpenAI's documentation for more
+  information on models: https://platform.openai.com/docs/models/overview
+
+
 g:neural.source.chatgpt.presence_penalty
                                      *g:neural.source.chatgpt.presence_penalty*
                                  *vim.g.neural.source.chatgpt.presence_penalty*

neural_sources/chatgpt.py

@@ -20,6 +20,7 @@ class Config:
     def __init__(
         self,
         api_key: str,
+        model: str,
         temperature: float,
         top_p: float,
         max_tokens: int,
@@ -27,6 +28,7 @@ class Config:
         frequency_penalty: float,
     ):
         self.api_key = api_key
+        self.model = model
         self.temperature = temperature
         self.top_p = top_p
         self.max_tokens = max_tokens
@@ -43,7 +45,7 @@ def get_chatgpt_completion(
         "Authorization": f"Bearer {config.api_key}"
     }
     data = {
-        "model": "gpt-3.5-turbo",
+        "model": config.model,
         "messages": (
             [{"role": "user", "content": prompt}]
             if isinstance(prompt, str) else
@@ -100,6 +102,11 @@ def load_config(raw_config: Dict[str, Any]) -> Config:
     if not isinstance(api_key, str) or not api_key:  # type: ignore
         raise ValueError("chatgpt.api_key is not defined")
 
+    model = raw_config.get('model')
+
+    if not isinstance(model, str) or not model:
+        raise ValueError("chatgpt.model is not defined")
+
     temperature = raw_config.get('temperature', 0.2)
 
     if not isinstance(temperature, (int, float)):
@@ -127,6 +134,7 @@ def load_config(raw_config: Dict[str, Any]) -> Config:
 
     return Config(
         api_key=api_key,
+        model=model,
         temperature=temperature,
         top_p=top_p,
         max_tokens=max_tokens,

neural_sources/openai.py

@@ -20,6 +20,7 @@ class Config:
     def __init__(
         self,
         api_key: str,
+        model: str,
         temperature: float,
         top_p: float,
         max_tokens: int,
@@ -27,6 +28,7 @@ class Config:
         frequency_penalty: float,
     ):
         self.api_key = api_key
+        self.model = model
         self.temperature = temperature
         self.top_p = top_p
         self.max_tokens = max_tokens
@@ -40,7 +42,7 @@ def get_openai_completion(config: Config, prompt: str) -> None:
         "Authorization": f"Bearer {config.api_key}"
     }
     data = {
-        "model": "text-davinci-003",
+        "model": config.model,
         "prompt": prompt,
         "temperature": config.temperature,
         "max_tokens": config.max_tokens,
@@ -88,6 +90,11 @@ def load_config(raw_config: Dict[str, Any]) -> Config:
     if not isinstance(api_key, str) or not api_key:  # type: ignore
         raise ValueError("openai.api_key is not defined")
 
+    model = raw_config.get('model')
+
+    if not isinstance(model, str) or not model:
+        raise ValueError("openai.model is not defined")
+
     temperature = raw_config.get('temperature', 0.2)
 
     if not isinstance(temperature, (int, float)):
@@ -115,6 +122,7 @@ def load_config(raw_config: Dict[str, Any]) -> Config:
 
     return Config(
         api_key=api_key,
+        model=model,
         temperature=temperature,
         top_p=top_p,
         max_tokens=max_tokens,

test/python/test_chatgpt.py

@@ -14,6 +14,7 @@ from neural_sources import chatgpt
 def get_valid_config() -> Dict[str, Any]:
     return {
         "api_key": ".",
+        "model": "foo",
         "prompt": "say hello",
         "temperature": 1,
         "top_p": 1,
@@ -34,8 +35,10 @@ def test_load_config_errors():
     for modification, expected_error in [
         ({}, "chatgpt.api_key is not defined"),
         ({"api_key": ""}, "chatgpt.api_key is not defined"),
+        ({"api_key": "."}, "chatgpt.model is not defined"),
+        ({"model": ""}, "chatgpt.model is not defined"),
         (
-            {"api_key": ".", "temperature": "x"},
+            {"model": "x", "temperature": "x"},
             "chatgpt.temperature is invalid"
         ),
         (

test/python/test_openai.py

@@ -14,6 +14,7 @@ from neural_sources import openai
 def get_valid_config() -> Dict[str, Any]:
     return {
         "api_key": ".",
+        "model": "foo",
         "prompt": "say hello",
         "temperature": 1,
         "top_p": 1,
@@ -34,8 +35,10 @@ def test_load_config_errors():
     for modification, expected_error in [
         ({}, "openai.api_key is not defined"),
         ({"api_key": ""}, "openai.api_key is not defined"),
+        ({"api_key": "."}, "openai.model is not defined"),
+        ({"model": ""}, "openai.model is not defined"),
         (
-            {"api_key": ".", "temperature": "x"},
+            {"model": "x", "temperature": "x"},
             "openai.temperature is invalid"
         ),
         (

test/script/run-python

@@ -12,7 +12,15 @@ echo '========================================'
 echo 'tox warnings/errors follow:'
 echo
 
-tox_command='tox --workdir .tox-docker --skip-missing-interpreters=false'
+# Use a different tox directory if running outside of docker to avoid issues
+# with the docker mounted directory writing as root, and other differences.
+if [ "$NO_DOCKER" -eq 1 ]; then
+    tox_dir=.tox
+else
+    tox_dir=.tox-docker
+fi
+
+tox_command="tox --workdir $tox_dir --skip-missing-interpreters=false"
 
 set -o pipefail
 

test/vim/test_config.vader

@@ -42,14 +42,30 @@ Execute(The default openai settings should be correct):
   AssertEqual
   \ {
   \   'api_key': '',
+  \   'frequency_penalty': 0.1,
+  \   'max_tokens': 1024,
+  \   'model': 'text-davinci-003',
+  \   'presence_penalty': 0.1,
   \   'temperature': 0.2,
   \   'top_p': 1,
-  \   'max_tokens': 1024,
-  \   'presence_penalty': 0.1,
-  \   'frequency_penalty': 0.1,
   \ },
   \ get(g:neural.source, 'openai')
 
+Execute(The default chatgpt settings should be correct):
+  call neural#config#Load()
+
+  AssertEqual
+  \ {
+  \   'api_key': '',
+  \   'frequency_penalty': 0.1,
+  \   'max_tokens': 2048,
+  \   'model': 'gpt-3.5-turbo',
+  \   'presence_penalty': 0.1,
+  \   'temperature': 0.2,
+  \   'top_p': 1,
+  \ },
+  \ get(g:neural.source, 'chatgpt')
+
 Execute(Settings should be merged correctly):
   for s:i in range(2)
     if s:i == 0