Make together embeddings.create() into OpenAI compatible format and a…

…llow providing a safety_model to Complete.create() (#63) * embeddings take str and list of strings this is to make the method compatible with openai * embeddings take str and list of strings this is to make the method compatible with openai * allows returning object embedding this make the embedding API openai compatible so you can call embed.data[0].embedding * allows returning object embedding this make the embedding API openai compatible so you can call embed.data[0].embedding * Added safety_model to Complete.create this is for the meta safety llama as a placeholder so we can use python in the demo * removed the embeddings api from readme not to be announced yet per heejin * black ruff and mypy * Added TogetherAI() class this allows for both the output = together.Complete.create( form of usage and also the client = TogetherAI() embed = client.embeddings.create( form of usage to keep the python library self consistent but also be OpenAI compatible * Added TogetherAI() class this allows for both the output = together.Complete.create( form of usage and also the client = TogetherAI() embed = client.embeddings.create( form of usage to keep the python library self consistent but also be OpenAI compatible * Added TogetherAI() class this allows for both the output = together.Complete.create( form of usage and also the client = TogetherAI() embed = client.embeddings.create( form of usage to keep the python library self consistent but also be OpenAI compatible * changed TogetherAI to Together class - added safety model to commands and changed Output to EmbeddingsOuput black ruff and mypy * changed TogetherAI to Together class - added safety model to commands and changed Output to EmbeddingsOuput black ruff and mypy --------- Co-authored-by: Carson Lam <[email protected]> Co-authored-by: Carson Lam <[email protected]>
togethercomputer · Dec 6, 2023 · 87a05d2 · 87a05d2
1 parent 8ae1902
commit 87a05d2
Show file tree

Hide file tree

Showing 7 changed files with 76 additions and 22 deletions.
diff --git a/README.md b/README.md
@@ -453,16 +453,6 @@ print(output_text)
 Space Robots are a great way to get your kids interested in science. After all, they are the future!
 ```
 
-## Embeddings API
-
-Embeddings are vector representations of sequences. You can use these vectors for measuring the overall similarity between texts. Embeddings are useful for tasks such as search and retrieval. 
-
-```python
-resp = together.Embeddings.create("embed this sentence into a single vector", model="togethercomputer/bert-base-uncased")
-
-print(resp['data'][0]['embedding']) # [0.06659205, 0.07896972, 0.007910785 ........]
-```
-
 ## Colab Tutorial
 
 Follow along in our Colab (Google Colaboratory) Notebook Tutorial [Example Finetuning Project](https://colab.research.google.com/drive/11DwtftycpDSgp3Z1vnV-Cy68zvkGZL4K?usp=sharing).

diff --git a/pyproject.toml b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "poetry.masonry.api"
 
 [tool.poetry]
 name = "together"
-version = "0.2.8"
+version = "0.2.9"
 authors = [
   "Together AI <[email protected]>"
 ]

diff --git a/src/together/__init__.py b/src/together/__init__.py
@@ -1,6 +1,7 @@
 import os
 import sys
 import urllib.parse
+from typing import Type
 
 from .version import VERSION
 
@@ -41,6 +42,27 @@
 from .models import Models
 
 
+class Together:
+    complete: Type[Complete]
+    completion: Type[Completion]
+    embeddings: Type[Embeddings]
+    files: Type[Files]
+    finetune: Type[Finetune]
+    image: Type[Image]
+    models: Type[Models]
+
+    def __init__(
+        self,
+    ) -> None:
+        self.complete = Complete
+        self.completion = Completion
+        self.embeddings = Embeddings
+        self.files = Files
+        self.finetune = Finetune
+        self.image = Image
+        self.models = Models
+
+
 __all__ = [
     "api_key",
     "api_base",
@@ -63,4 +85,5 @@
     "MISSING_API_KEY_MESSAGE",
     "BACKOFF_FACTOR",
     "min_samples",
+    "Together",
 ]
diff --git a/src/together/commands/complete.py b/src/together/commands/complete.py
@@ -90,6 +90,13 @@ def add_parser(subparsers: argparse._SubParsersAction[argparse.ArgumentParser])
         action="store_true",
         help="temperature for the LM",
     )
+    subparser.add_argument(
+        "--safety-model",
+        "-sm",
+        default=None,
+        type=str,
+        help="The name of the safety model to use for moderation.",
+    )
     subparser.set_defaults(func=_run_complete)
 
 
@@ -142,6 +149,7 @@ def _run_complete(args: argparse.Namespace) -> None:
                 top_k=args.top_k,
                 repetition_penalty=args.repetition_penalty,
                 logprobs=args.logprobs,
+                safety_model=args.safety_model,
             )
         except together.AuthenticationError:
             logger.critical(together.MISSING_API_KEY_MESSAGE)
@@ -159,6 +167,7 @@ def _run_complete(args: argparse.Namespace) -> None:
                 top_p=args.top_p,
                 top_k=args.top_k,
                 repetition_penalty=args.repetition_penalty,
+                safety_model=args.safety_model,
                 raw=args.raw,
             ):
                 if not args.raw:

diff --git a/src/together/commands/embeddings.py b/src/together/commands/embeddings.py
@@ -1,7 +1,6 @@
 from __future__ import annotations
 
 import argparse
-import json
 
 import together
 from together import Embeddings
@@ -42,7 +41,7 @@ def _run_complete(args: argparse.Namespace) -> None:
             model=args.model,
         )
 
-        print(json.dumps(response, indent=4))
+        print([e.embedding for e in response.data])
     except together.AuthenticationError:
         logger.critical(together.MISSING_API_KEY_MESSAGE)
         exit(0)
diff --git a/src/together/complete.py b/src/together/complete.py
@@ -24,6 +24,7 @@ def create(
         logprobs: Optional[int] = None,
         api_key: Optional[str] = None,
         cast: bool = False,
+        safety_model: Optional[str] = None,
     ) -> Union[Dict[str, Any], TogetherResponse]:
         if model == "":
             model = together.default_text_model
@@ -38,6 +39,7 @@ def create(
             "stop": stop,
             "repetition_penalty": repetition_penalty,
             "logprobs": logprobs,
+            "safety_model": safety_model,
         }
 
         # send request
@@ -70,6 +72,7 @@ def create_streaming(
         raw: Optional[bool] = False,
         api_key: Optional[str] = None,
         cast: Optional[bool] = False,
+        safety_model: Optional[str] = None,
     ) -> Union[Iterator[str], Iterator[TogetherResponse]]:
         """
         Prints streaming responses and returns the completed text.
@@ -88,6 +91,7 @@ def create_streaming(
             "stop": stop,
             "repetition_penalty": repetition_penalty,
             "stream_tokens": True,
+            "safety_model": safety_model,
         }
 
         # send request

diff --git a/src/together/embeddings.py b/src/together/embeddings.py
@@ -1,4 +1,5 @@
-from typing import Any, Dict, Optional
+import concurrent.futures
+from typing import Any, Dict, List, Optional, Union
 
 import together
 from together.utils import create_post_request, get_logger
@@ -7,29 +8,57 @@
 logger = get_logger(str(__name__))
 
 
+class DataItem:
+    def __init__(self, embedding: List[float]):
+        self.embedding = embedding
+
+
+class EmbeddingsOutput:
+    def __init__(self, data: List[DataItem]):
+        self.data = data
+
+
 class Embeddings:
     @classmethod
     def create(
-        self,
-        input: str,
+        cls,
+        input: Union[str, List[str]],
         model: Optional[str] = "",
-    ) -> Dict[str, Any]:
+    ) -> EmbeddingsOutput:
         if model == "":
             model = together.default_embedding_model
 
-        parameter_payload = {
-            "input": input,
-            "model": model,
-        }
+        if isinstance(input, str):
+            parameter_payload = {
+                "input": input,
+                "model": model,
+            }
+
+            response = cls._process_input(parameter_payload)
+
+            return EmbeddingsOutput([DataItem(response["data"][0]["embedding"])])
 
+        elif isinstance(input, list):
+            # If input is a list, process each string concurrently
+            with concurrent.futures.ThreadPoolExecutor() as executor:
+                parameter_payloads = [{"input": item, "model": model} for item in input]
+                results = list(executor.map(cls._process_input, parameter_payloads))
+
+            return EmbeddingsOutput(
+                [DataItem(item["data"][0]["embedding"]) for item in results]
+            )
+
+    @classmethod
+    def _process_input(cls, parameter_payload: Dict[str, Any]) -> Dict[str, Any]:
         # send request
         response = create_post_request(
             url=together.api_base_embeddings, json=parameter_payload
         )
 
+        # return the json as a DotDict
         try:
             response_json = dict(response.json())
-
         except Exception as e:
             raise together.JSONError(e, http_status=response.status_code)
+
         return response_json