together 1.1.3__tar.gz → 1.1.4__tar.gz

{together-1.1.3 → together-1.1.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: together
-Version: 1.1.3
+Version: 1.1.4
 Summary: Python client for Together's Cloud Platform!
 Home-page: https://github.com/togethercomputer/together-python
 License: Apache-2.0

{together-1.1.3 → together-1.1.4}/pyproject.toml

@@ -12,7 +12,7 @@ build-backend = "poetry.masonry.api"
 
 [tool.poetry]
 name = "together"
-version = "1.1.3"
+version = "1.1.4"
 authors = [
     "Together AI <support@together.ai>"
 ]

{together-1.1.3 → together-1.1.4}/src/together/cli/api/chat.py

@@ -28,6 +28,9 @@ class ChatShell(cmd.Cmd):
         top_p: float | None = None,
         top_k: int | None = None,
         repetition_penalty: float | None = None,
+        presence_penalty: float | None = None,
+        frequency_penalty: float | None = None,
+        min_p: float | None = None,
         safety_model: str | None = None,
         system_message: str | None = None,
     ) -> None:
@@ -40,6 +43,9 @@ class ChatShell(cmd.Cmd):
         self.top_p = top_p
         self.top_k = top_k
         self.repetition_penalty = repetition_penalty
+        self.presence_penalty = presence_penalty
+        self.frequency_penalty = frequency_penalty
+        self.min_p = min_p
         self.safety_model = safety_model
         self.system_message = system_message
 
@@ -69,6 +75,9 @@ class ChatShell(cmd.Cmd):
             top_p=self.top_p,
             top_k=self.top_k,
             repetition_penalty=self.repetition_penalty,
+            presence_penalty=self.presence_penalty,
+            frequency_penalty=self.frequency_penalty,
+            min_p=self.min_p,
             safety_model=self.safety_model,
             stream=True,
         ):
@@ -76,13 +85,12 @@ class ChatShell(cmd.Cmd):
             assert isinstance(chunk, ChatCompletionChunk)
             assert chunk.choices
             assert chunk.choices[0].delta
-            assert chunk.choices[0].delta.content
 
             token = chunk.choices[0].delta.content
 
             click.echo(token, nl=False)
 
-            output += token
+            output += token or ""
 
         click.echo("\n")
 
@@ -109,6 +117,10 @@ class ChatShell(cmd.Cmd):
 @click.option("--temperature", type=float, help="Sampling temperature")
 @click.option("--top-p", type=int, help="Top p sampling")
 @click.option("--top-k", type=float, help="Top k sampling")
+@click.option("--repetition-penalty", type=float, help="Repetition penalty")
+@click.option("--presence-penalty", type=float, help="Presence penalty")
+@click.option("--frequency-penalty", type=float, help="Frequency penalty")
+@click.option("--min-p", type=float, help="Minimum p")
 @click.option("--safety-model", type=str, help="Moderation model")
 @click.option("--system-message", type=str, help="System message to use for the chat")
 def interactive(
@@ -120,6 +132,9 @@ def interactive(
     top_p: float | None = None,
     top_k: int | None = None,
     repetition_penalty: float | None = None,
+    presence_penalty: float | None = None,
+    frequency_penalty: float | None = None,
+    min_p: float | None = None,
     safety_model: str | None = None,
     system_message: str | None = None,
 ) -> None:
@@ -135,6 +150,9 @@ def interactive(
         top_p=top_p,
         top_k=top_k,
         repetition_penalty=repetition_penalty,
+        presence_penalty=presence_penalty,
+        frequency_penalty=frequency_penalty,
+        min_p=min_p,
         safety_model=safety_model,
         system_message=system_message,
     ).cmdloop()
@@ -158,6 +176,11 @@ def interactive(
 @click.option("--top-p", type=int, help="Top p sampling")
 @click.option("--top-k", type=float, help="Top k sampling")
 @click.option("--repetition-penalty", type=float, help="Repetition penalty")
+@click.option("--presence-penalty", type=float, help="Presence penalty sampling method")
+@click.option(
+    "--frequency-penalty", type=float, help="Frequency penalty sampling method"
+)
+@click.option("--min-p", type=float, help="Min p sampling")
 @click.option("--no-stream", is_flag=True, help="Disable streaming")
 @click.option("--logprobs", type=int, help="Return logprobs. Only works with --raw.")
 @click.option("--echo", is_flag=True, help="Echo prompt. Only works with --raw.")
@@ -174,6 +197,9 @@ def chat(
     top_p: float | None = None,
     top_k: int | None = None,
     repetition_penalty: float | None = None,
+    presence_penalty: float | None = None,
+    frequency_penalty: float | None = None,
+    min_p: float | None = None,
     no_stream: bool = False,
     logprobs: int | None = None,
     echo: bool | None = None,
@@ -195,6 +221,9 @@ def chat(
         max_tokens=max_tokens,
         stop=stop,
         repetition_penalty=repetition_penalty,
+        presence_penalty=presence_penalty,
+        frequency_penalty=frequency_penalty,
+        min_p=min_p,
         stream=not no_stream,
         logprobs=logprobs,
         echo=echo,

{together-1.1.3 → together-1.1.4}/src/together/cli/api/completions.py

@@ -14,7 +14,6 @@ from together.types.completions import CompletionChoicesChunk, CompletionRespons
 @click.pass_context
 @click.argument("prompt", type=str, required=True)
 @click.option("--model", type=str, required=True, help="Model name")
-@click.option("--no-stream", is_flag=True, help="Disable streaming")
 @click.option("--max-tokens", type=int, help="Max tokens to generate")
 @click.option(
     "--stop", type=str, multiple=True, help="List of strings to stop generation"
@@ -22,6 +21,11 @@ from together.types.completions import CompletionChoicesChunk, CompletionRespons
 @click.option("--temperature", type=float, help="Sampling temperature")
 @click.option("--top-p", type=int, help="Top p sampling")
 @click.option("--top-k", type=float, help="Top k sampling")
+@click.option("--repetition-penalty", type=float, help="Repetition penalty")
+@click.option("--presence-penalty", type=float, help="Presence penalty")
+@click.option("--frequency-penalty", type=float, help="Frequency penalty")
+@click.option("--min-p", type=float, help="Minimum p")
+@click.option("--no-stream", is_flag=True, help="Disable streaming")
 @click.option("--logprobs", type=int, help="Return logprobs. Only works with --raw.")
 @click.option("--echo", is_flag=True, help="Echo prompt. Only works with --raw.")
 @click.option("--n", type=int, help="Number of output generations")
@@ -37,6 +41,9 @@ def completions(
     top_p: float | None = None,
     top_k: int | None = None,
     repetition_penalty: float | None = None,
+    presence_penalty: float | None = None,
+    frequency_penalty: float | None = None,
+    min_p: float | None = None,
     no_stream: bool = False,
     logprobs: int | None = None,
     echo: bool | None = None,
@@ -56,6 +63,9 @@ def completions(
         max_tokens=max_tokens,
         stop=stop,
         repetition_penalty=repetition_penalty,
+        presence_penalty=presence_penalty,
+        frequency_penalty=frequency_penalty,
+        min_p=min_p,
         stream=not no_stream,
         logprobs=logprobs,
         echo=echo,

{together-1.1.3 → together-1.1.4}/src/together/cli/api/finetune.py

@@ -107,10 +107,20 @@ def retrieve(ctx: click.Context, fine_tune_id: str) -> None:
 @fine_tuning.command()
 @click.pass_context
 @click.argument("fine_tune_id", type=str, required=True)
-def cancel(ctx: click.Context, fine_tune_id: str) -> None:
+@click.option(
+    "--quiet", is_flag=True, help="Do not prompt for confirmation before cancelling job"
+)
+def cancel(ctx: click.Context, fine_tune_id: str, quiet: bool = False) -> None:
     """Cancel fine-tuning job"""
     client: Together = ctx.obj
-
+    if not quiet:
+        confirm_response = input(
+            "You will be billed for any completed training steps upon cancellation. "
+            f"Do you want to cancel job {fine_tune_id}? [y/N]"
+        )
+        if "y" not in confirm_response.lower():
+            click.echo({"status": "Cancel not submitted"})
+            return
     response = client.fine_tuning.cancel(fine_tune_id)
 
     click.echo(json.dumps(response.model_dump(), indent=4))

{together-1.1.3 → together-1.1.4}/src/together/resources/chat/completions.py

@@ -28,6 +28,10 @@ class ChatCompletions:
         top_p: float | None = None,
         top_k: int | None = None,
         repetition_penalty: float | None = None,
+        presence_penalty: float | None = None,
+        frequency_penalty: float | None = None,
+        min_p: float | None = None,
+        logit_bias: Dict[str, float] | None = None,
         stream: bool = False,
         logprobs: int | None = None,
         echo: bool | None = None,
@@ -59,6 +63,21 @@ class ChatCompletions:
             repetition_penalty (float, optional): A number that controls the diversity of generated text
                     by reducing the likelihood of repeated sequences. Higher values decrease repetition.
                 Defaults to None.
+            presence_penalty (float, optional): A number that controls the likelihood of tokens based on if they have
+                    appeared in the text. Positive values decrease the likelihood of repeated tokens or phrases.
+                    Must be in the range [-2, 2].
+                Defaults to None.
+            frequency_penalty (float, optional): A number that controls the likelihood of tokens based on the frequency
+                    of their appearance in the text. Positive decrease the likelihood of repeated tokens or phrases.
+                    Must be in the range [-2, 2].
+                Defaults to None.
+            min_p (float, optional): A number that controls the minimum percentage value that a token must reach to
+                be considered during sampling.
+                Must be in the range [0, 1].
+                Defaults to None.
+            logit_bias (Dict[str, float], optional): A dictionary of tokens and their bias values that modify the
+                likelihood of specific tokens being sampled. Bias values must be in the range [-100, 100].
+                Defaults to None.
             stream (bool, optional): Flag indicating whether to stream the generated completions.
                 Defaults to False.
             logprobs (int, optional): Number of top-k logprobs to return
@@ -100,6 +119,10 @@ class ChatCompletions:
             max_tokens=max_tokens,
             stop=stop,
             repetition_penalty=repetition_penalty,
+            presence_penalty=presence_penalty,
+            frequency_penalty=frequency_penalty,
+            min_p=min_p,
+            logit_bias=logit_bias,
             stream=stream,
             logprobs=logprobs,
             echo=echo,
@@ -142,6 +165,10 @@ class AsyncChatCompletions:
         top_p: float | None = None,
         top_k: int | None = None,
         repetition_penalty: float | None = None,
+        presence_penalty: float | None = None,
+        frequency_penalty: float | None = None,
+        min_p: float | None = None,
+        logit_bias: Dict[str, float] | None = None,
         stream: bool = False,
         logprobs: int | None = None,
         echo: bool | None = None,
@@ -173,6 +200,21 @@ class AsyncChatCompletions:
             repetition_penalty (float, optional): A number that controls the diversity of generated text
                     by reducing the likelihood of repeated sequences. Higher values decrease repetition.
                 Defaults to None.
+            presence_penalty (float, optional): A number that controls the likelihood of tokens based on if they have
+                    appeared in the text. Positive values decrease the likelihood of repeated tokens or phrases.
+                    Must be in the range [-2, 2].
+                Defaults to None.
+            frequency_penalty (float, optional): A number that controls the likelihood of tokens based on the frequency
+                    of their appearance in the text. Positive decrease the likelihood of repeated tokens or phrases.
+                    Must be in the range [-2, 2].
+                Defaults to None.
+            min_p (float, optional): A number that controls the minimum percentage value that a token must reach to
+                be considered during sampling.
+                Must be in the range [0, 1].
+                Defaults to None.
+            logit_bias (Dict[str, float], optional): A dictionary of tokens and their bias values that modify the
+                likelihood of specific tokens being sampled. Bias values must be in the range [-100, 100].
+                Defaults to None.
             stream (bool, optional): Flag indicating whether to stream the generated completions.
                 Defaults to False.
             logprobs (int, optional): Number of top-k logprobs to return
@@ -214,6 +256,10 @@ class AsyncChatCompletions:
             max_tokens=max_tokens,
             stop=stop,
             repetition_penalty=repetition_penalty,
+            presence_penalty=presence_penalty,
+            frequency_penalty=frequency_penalty,
+            min_p=min_p,
+            logit_bias=logit_bias,
             stream=stream,
             logprobs=logprobs,
             echo=echo,

{together-1.1.3 → together-1.1.4}/src/together/resources/completions.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from typing import AsyncGenerator, Iterator, List
+from typing import AsyncGenerator, Dict, Iterator, List
 
 from together.abstract import api_requestor
 from together.together_response import TogetherResponse
@@ -28,6 +28,10 @@ class Completions:
         top_p: float | None = None,
         top_k: int | None = None,
         repetition_penalty: float | None = None,
+        presence_penalty: float | None = None,
+        frequency_penalty: float | None = None,
+        min_p: float | None = None,
+        logit_bias: Dict[str, float] | None = None,
         stream: bool = False,
         logprobs: int | None = None,
         echo: bool | None = None,
@@ -55,6 +59,21 @@ class Completions:
             repetition_penalty (float, optional): A number that controls the diversity of generated text
                     by reducing the likelihood of repeated sequences. Higher values decrease repetition.
                 Defaults to None.
+            presence_penalty (float, optional): A number that controls the likelihood of tokens based on if they have
+                    appeared in the text. Positive values decrease the likelihood of repeated tokens or phrases.
+                    Must be in the range [-2, 2].
+                Defaults to None.
+            frequency_penalty (float, optional): A number that controls the likelihood of tokens based on the frequency
+                    of their appearance in the text. Positive decrease the likelihood of repeated tokens or phrases.
+                    Must be in the range [-2, 2].
+                Defaults to None.
+            min_p (float, optional): A number that controls the minimum percentage value that a token must reach to
+                be considered during sampling.
+                Must be in the range [0, 1].
+                Defaults to None.
+            logit_bias (Dict[str, float], optional): A dictionary of tokens and their bias values that modify the
+                likelihood of specific tokens being sampled. Bias values must be in the range [-100, 100].
+                Defaults to None.
             stream (bool, optional): Flag indicating whether to stream the generated completions.
                 Defaults to False.
             logprobs (int, optional): Number of top-k logprobs to return
@@ -85,6 +104,10 @@ class Completions:
             max_tokens=max_tokens,
             stop=stop,
             repetition_penalty=repetition_penalty,
+            presence_penalty=presence_penalty,
+            frequency_penalty=frequency_penalty,
+            min_p=min_p,
+            logit_bias=logit_bias,
             stream=stream,
             logprobs=logprobs,
             echo=echo,
@@ -124,6 +147,10 @@ class AsyncCompletions:
         top_p: float | None = None,
         top_k: int | None = None,
         repetition_penalty: float | None = None,
+        presence_penalty: float | None = None,
+        frequency_penalty: float | None = None,
+        min_p: float | None = None,
+        logit_bias: Dict[str, float] | None = None,
         stream: bool = False,
         logprobs: int | None = None,
         echo: bool | None = None,
@@ -151,6 +178,21 @@ class AsyncCompletions:
             repetition_penalty (float, optional): A number that controls the diversity of generated text
                     by reducing the likelihood of repeated sequences. Higher values decrease repetition.
                 Defaults to None.
+            presence_penalty (float, optional): A number that controls the likelihood of tokens based on if they have
+                    appeared in the text. Positive values decrease the likelihood of repeated tokens or phrases.
+                    Must be in the range [-2, 2].
+                Defaults to None.
+            frequency_penalty (float, optional): A number that controls the likelihood of tokens based on the frequency
+                    of their appearance in the text. Positive decrease the likelihood of repeated tokens or phrases.
+                    Must be in the range [-2, 2].
+                Defaults to None.
+            min_p (float, optional): A number that controls the minimum percentage value that a token must reach to
+                be considered during sampling.
+                Must be in the range [0, 1].
+                Defaults to None.
+            logit_bias (Dict[str, float], optional): A dictionary of tokens and their bias values that modify the
+                likelihood of specific tokens being sampled. Bias values must be in the range [-100, 100].
+                Defaults to None.
             stream (bool, optional): Flag indicating whether to stream the generated completions.
                 Defaults to False.
             logprobs (int, optional): Number of top-k logprobs to return
@@ -181,6 +223,10 @@ class AsyncCompletions:
             max_tokens=max_tokens,
             stop=stop,
             repetition_penalty=repetition_penalty,
+            presence_penalty=presence_penalty,
+            frequency_penalty=frequency_penalty,
+            min_p=min_p,
+            logit_bias=logit_bias,
             stream=stream,
             logprobs=logprobs,
             echo=echo,

{together-1.1.3 → together-1.1.4}/src/together/types/chat_completions.py

@@ -1,9 +1,11 @@
 from __future__ import annotations
 
+import warnings
 from enum import Enum
 from typing import Any, Dict, List
 
-from pydantic import Field
+from pydantic import Field, model_validator
+from typing_extensions import Self
 
 from together.types.abstract import BaseModel
 from together.types.common import (
@@ -87,6 +89,10 @@ class ChatCompletionRequest(BaseModel):
     top_p: float | None = None
     top_k: int | None = None
     repetition_penalty: float | None = None
+    presence_penalty: float | None = None
+    frequency_penalty: float | None = None
+    min_p: float | None = None
+    logit_bias: Dict[str, float] | None = None
     # stream SSE token chunks
     stream: bool = False
     # return logprobs
@@ -103,6 +109,16 @@ class ChatCompletionRequest(BaseModel):
     tools: List[Tools] | None = None
     tool_choice: ToolChoice | ToolChoiceEnum | None = None
 
+    # Raise warning if repetition_penalty is used with presence_penalty or frequency_penalty
+    @model_validator(mode="after")
+    def verify_parameters(self) -> Self:
+        if self.repetition_penalty:
+            if self.presence_penalty or self.frequency_penalty:
+                warnings.warn(
+                    "repetition_penalty is not advisable to be used alongside presence_penalty or frequency_penalty"
+                )
+        return self
+
 
 class ChatCompletionChoicesData(BaseModel):
     index: int | None = None

{together-1.1.3 → together-1.1.4}/src/together/types/completions.py

@@ -1,6 +1,10 @@
 from __future__ import annotations
 
-from typing import List
+import warnings
+from typing import Dict, List
+
+from pydantic import model_validator
+from typing_extensions import Self
 
 from together.types.abstract import BaseModel
 from together.types.common import (
@@ -27,6 +31,10 @@ class CompletionRequest(BaseModel):
     top_p: float | None = None
     top_k: int | None = None
     repetition_penalty: float | None = None
+    presence_penalty: float | None = None
+    frequency_penalty: float | None = None
+    min_p: float | None = None
+    logit_bias: Dict[str, float] | None = None
     # stream SSE token chunks
     stream: bool = False
     # return logprobs
@@ -39,6 +47,16 @@ class CompletionRequest(BaseModel):
     # moderation model
     safety_model: str | None = None
 
+    # Raise warning if repetition_penalty is used with presence_penalty or frequency_penalty
+    @model_validator(mode="after")
+    def verify_parameters(self) -> Self:
+        if self.repetition_penalty:
+            if self.presence_penalty or self.frequency_penalty:
+                warnings.warn(
+                    "repetition_penalty is not advisable to be used alongside presence_penalty or frequency_penalty"
+                )
+        return self
+
 
 class CompletionChoicesData(BaseModel):
     index: int