feat(deps): migrated to the new API version

.gitignore

@@ -2,3 +2,4 @@
 venv/
 llm_together.egg-info/
 __pycache__
+.aider*

llm_together.py

@@ -5,34 +5,26 @@
 
 @llm.hookimpl
 def register_models(register):
+    models = Together().client.models.list()
+    for model in sorted(models, key=lambda m: m.id):
+        register(Together(model))
 
-    together_instance = Together()
-    model_list = together_instance.get_models()
-
-    for model in model_list:
-        if 'isFeaturedModel' in model and model['isFeaturedModel']:
-            register(Together(model)) 
 
 def not_nulls(data) -> dict:
     return {key: value for key, value in data if value is not None}
 
-class Together(llm.Model):
+class Together(llm.KeyModel):
     model_id = "llm-together"
     needs_key = "together"
     key_env_var = "TOGETHER_API_KEY"
-    default_stop = "<human>"
+    default_stop = ["<human>"]
     can_stream = True
 
-    def get_models(self):
-        together.api_key = self.get_key()
-        return together.Models.list()
-
     def __init__(self, model=None):
-        together.api_key = self.get_key()
-
-        if (model is not None):
-            self.model_id = model["name"]
+        self.client = together.Client(api_key=self.get_key())
+        if model:
             self.model = model
+            self.model_id = model.id
 
     class Options(llm.Options):
         temperature: Optional[float] = Field(
@@ -46,7 +38,7 @@ class Options(llm.Options):
             default=None,
         )
         max_tokens: Optional[int] = Field(
-            description="Maximum number of tokens to generate.", default=256
+            description="Maximum number of tokens to generate.", default=8192
         )
         top_p: Optional[float] = Field(
             description=(
@@ -71,7 +63,7 @@ class Options(llm.Options):
             default=None,
         )
 
-    def execute(self, prompt, stream, response, conversation):
+    def execute(self, prompt, stream, response, conversation, key=None):
         kwargs = dict(not_nulls(prompt.options))
 
         user_prompt = "{}\n\n{}".format(prompt.system or "", prompt.prompt)
@@ -82,7 +74,9 @@ def execute(self, prompt, stream, response, conversation):
             if conversation is not None:
                 for message in conversation.responses:
                     if 'prompt_format' in self.model["config"] and self.model["config"]['prompt_format']:
-                        history += self.model["config"]["prompt_format"].format(prompt = message.prompt) + " " + message.text() + "\n"
+                        formatted_prompt = self.model["config"]["prompt_format"].format(prompt=message.prompt)
+                        message_text = message.text()
+                        history += formatted_prompt + " " + message_text + "\n"
                     else:
                         history += "{}\n\n{}".format(message.prompt, message.text())+ "\n"
 
@@ -91,23 +85,28 @@ def execute(self, prompt, stream, response, conversation):
 
 
             if 'stop' in self.model["config"]:
-                stop = self.model["config"]["stop"]
+                config_stop = self.model["config"]["stop"]
+                if isinstance(config_stop, list):
+                    stop = config_stop
+                else:
+                    stop = [config_stop]
 
         if stream:
-            for token in together.Complete.create_streaming(
+            for chunk in self.client.completions.create(
                 prompt =  history + "\n" + user_prompt,
                 model = self.model_id,
+                stream = True,
                 stop = stop,
                 **kwargs,
             ):
-                yield token
-
+                if chunk.choices and len(chunk.choices) > 0:
+                    yield chunk.choices[0].text
         else:
-            output = together.Complete.create(
+            output = self.client.completions.create(
                 prompt =  history + "\n" + user_prompt,
                 model = self.model_id,
                 stop = stop,
                 **kwargs,
             )
-
-            return [output['output']['choices'][0]['text']]
+            if output.choices and len(output.choices) > 0:
+                yield output.choices[0].text

pyproject.toml

@@ -1,16 +1,18 @@
 [project]
 dynamic = ["optional-dependencies"]
 name = "llm-together"
-version = "0.4"
+version = "0.5"
 license = { text="MIT" }
 authors = [
   { name="Kévin Quesada" },
+  { name="Christian Braun" },
 ]
 description = "llm together plugin"
 readme = "README.md"
 requires-python = ">=3.7"
 dependencies = [
-  "llm>=0.5", "together==0.2.9" 
+  "llm>=0.22",
+  "together>=1.4.1"
 ]
 classifiers = [
     "License :: OSI Approved :: Apache Software License"