together 1.5.16__py3-none-any.whl → 1.5.18__py3-none-any.whl

together/cli/api/finetune.py

@@ -139,7 +139,7 @@ def fine_tuning(ctx: click.Context) -> None:
 @click.option(
     "--dpo-beta",
     type=float,
-    default=0.1,
+    default=None,
     help="Beta parameter for DPO training (only used when '--training-method' is 'dpo')",
 )
 @click.option(
@@ -154,7 +154,7 @@ def fine_tuning(ctx: click.Context) -> None:
 @click.option(
     "--rpo-alpha",
     type=float,
-    default=0.0,
+    default=None,
     help=(
         "RPO alpha parameter of DPO training to include NLL in the loss "
         "(only used when '--training-method' is 'dpo')"
@@ -163,7 +163,7 @@ def fine_tuning(ctx: click.Context) -> None:
 @click.option(
     "--simpo-gamma",
     type=float,
-    default=0.0,
+    default=None,
     help="SimPO gamma parameter (only used when '--training-method' is 'dpo')",
 )
 @click.option(
@@ -188,7 +188,7 @@ def fine_tuning(ctx: click.Context) -> None:
 @click.option(
     "--train-on-inputs",
     type=BOOL_WITH_AUTO,
-    default="auto",
+    default=None,
     help="Whether to mask the user messages in conversational data or prompts in instruction data. "
     "`auto` will automatically determine whether to mask the inputs based on the data format.",
 )
@@ -200,6 +200,18 @@ def fine_tuning(ctx: click.Context) -> None:
     "The format: {$JOB_ID/$OUTPUT_MODEL_NAME}:{$STEP}. "
     "The step value is optional, without it the final checkpoint will be used.",
 )
+@click.option(
+    "--hf-api-token",
+    type=str,
+    default=None,
+    help="HF API token to use for uploading a checkpoint to a private repo",
+)
+@click.option(
+    "--hf-output-repo-name",
+    type=str,
+    default=None,
+    help="HF repo to upload the fine-tuned model to",
+)
 def create(
     ctx: click.Context,
     training_file: str,
@@ -229,11 +241,13 @@ def create(
     confirm: bool,
     train_on_inputs: bool | Literal["auto"],
     training_method: str,
-    dpo_beta: float,
+    dpo_beta: float | None,
     dpo_normalize_logratios_by_length: bool,
-    rpo_alpha: float,
-    simpo_gamma: float,
+    rpo_alpha: float | None,
+    simpo_gamma: float | None,
     from_checkpoint: str,
+    hf_api_token: str | None,
+    hf_output_repo_name: str | None,
 ) -> None:
     """Start fine-tuning"""
     client: Together = ctx.obj
@@ -270,6 +284,8 @@ def create(
         rpo_alpha=rpo_alpha,
         simpo_gamma=simpo_gamma,
         from_checkpoint=from_checkpoint,
+        hf_api_token=hf_api_token,
+        hf_output_repo_name=hf_output_repo_name,
     )
 
     if model is None and from_checkpoint is None:
@@ -280,7 +296,7 @@ def create(
         model_name = from_checkpoint.split(":")[0]
 
     model_limits: FinetuneTrainingLimits = client.fine_tuning.get_model_limits(
-        model=model_name
+        model=model_name,
     )
 
     if lora:

together/resources/chat/completions.py

@@ -38,7 +38,7 @@ class ChatCompletions:
         echo: bool | None = None,
         n: int | None = None,
         safety_model: str | None = None,
-        response_format: Dict[str, str | Dict[str, Any]] | None = None,
+        response_format: Dict[str, Any] | None = None,
         tools: List[Dict[str, Any]] | None = None,
         tool_choice: str | Dict[str, str | Dict[str, str]] | None = None,
         **kwargs: Any,

together/resources/finetune.py

@@ -76,6 +76,8 @@ def create_finetune_request(
     rpo_alpha: float | None = None,
     simpo_gamma: float | None = None,
     from_checkpoint: str | None = None,
+    hf_api_token: str | None = None,
+    hf_output_repo_name: str | None = None,
 ) -> FinetuneRequest:
     if model is not None and from_checkpoint is not None:
         raise ValueError(
@@ -183,6 +185,24 @@ def create_finetune_request(
         )
         train_on_inputs = "auto"
 
+    if dpo_beta is not None and training_method != "dpo":
+        raise ValueError("dpo_beta is only supported for DPO training")
+    if dpo_normalize_logratios_by_length and training_method != "dpo":
+        raise ValueError(
+            "dpo_normalize_logratios_by_length=True is only supported for DPO training"
+        )
+    if rpo_alpha is not None:
+        if training_method != "dpo":
+            raise ValueError("rpo_alpha is only supported for DPO training")
+        if not rpo_alpha >= 0.0:
+            raise ValueError(f"rpo_alpha should be non-negative (got {rpo_alpha})")
+
+    if simpo_gamma is not None:
+        if training_method != "dpo":
+            raise ValueError("simpo_gamma is only supported for DPO training")
+        if not simpo_gamma >= 0.0:
+            raise ValueError(f"simpo_gamma should be non-negative (got {simpo_gamma})")
+
     lr_scheduler: FinetuneLRScheduler
     if lr_scheduler_type == "cosine":
         if scheduler_num_cycles <= 0.0:
@@ -244,6 +264,8 @@ def create_finetune_request(
         wandb_name=wandb_name,
         training_method=training_method_cls,
         from_checkpoint=from_checkpoint,
+        hf_api_token=hf_api_token,
+        hf_output_repo_name=hf_output_repo_name,
     )
 
     return finetune_request
@@ -323,6 +345,8 @@ class FineTuning:
         rpo_alpha: float | None = None,
         simpo_gamma: float | None = None,
         from_checkpoint: str | None = None,
+        hf_api_token: str | None = None,
+        hf_output_repo_name: str | None = None,
     ) -> FinetuneResponse:
         """
         Method to initiate a fine-tuning job
@@ -379,6 +403,8 @@ class FineTuning:
             from_checkpoint (str, optional): The checkpoint identifier to continue training from a previous fine-tuning job.
                 The format: {$JOB_ID/$OUTPUT_MODEL_NAME}:{$STEP}.
                 The step value is optional, without it the final checkpoint will be used.
+            hf_api_token (str, optional): API key for the Hugging Face Hub. Defaults to None.
+            hf_output_repo_name (str, optional): HF repo to upload the fine-tuned model to. Defaults to None.
 
         Returns:
             FinetuneResponse: Object containing information about fine-tuning job.
@@ -432,6 +458,8 @@ class FineTuning:
             rpo_alpha=rpo_alpha,
             simpo_gamma=simpo_gamma,
             from_checkpoint=from_checkpoint,
+            hf_api_token=hf_api_token,
+            hf_output_repo_name=hf_output_repo_name,
         )
 
         if verbose:
@@ -744,6 +772,8 @@ class AsyncFineTuning:
         rpo_alpha: float | None = None,
         simpo_gamma: float | None = None,
         from_checkpoint: str | None = None,
+        hf_api_token: str | None = None,
+        hf_output_repo_name: str | None = None,
     ) -> FinetuneResponse:
         """
         Async method to initiate a fine-tuning job
@@ -800,6 +830,8 @@ class AsyncFineTuning:
             from_checkpoint (str, optional): The checkpoint identifier to continue training from a previous fine-tuning job.
                 The format: {$JOB_ID/$OUTPUT_MODEL_NAME}:{$STEP}.
                 The step value is optional, without it the final checkpoint will be used.
+            hf_api_token (str, optional): API key for the Huggging Face Hub. Defaults to None.
+            hf_output_repo_name (str, optional): HF repo to upload the fine-tuned model to. Defaults to None.
 
         Returns:
             FinetuneResponse: Object containing information about fine-tuning job.
@@ -853,6 +885,8 @@ class AsyncFineTuning:
             rpo_alpha=rpo_alpha,
             simpo_gamma=simpo_gamma,
             from_checkpoint=from_checkpoint,
+            hf_api_token=hf_api_token,
+            hf_output_repo_name=hf_output_repo_name,
         )
 
         if verbose:

together/types/chat_completions.py

@@ -28,6 +28,7 @@ class MessageRole(str, Enum):
 class ResponseFormatType(str, Enum):
     JSON_OBJECT = "json_object"
     JSON_SCHEMA = "json_schema"
+    REGEX = "regex"
 
 
 class FunctionCall(BaseModel):
@@ -71,9 +72,15 @@ class ChatCompletionMessage(BaseModel):
 class ResponseFormat(BaseModel):
     type: ResponseFormatType
     schema_: Dict[str, Any] | None = None
+    pattern: str | None = None
 
     def to_dict(self) -> Dict[str, Any]:
-        return {"schema": self.schema_, "type": self.type}
+        result: Dict[str, Any] = {"type": self.type.value}
+        if self.schema_ is not None:
+            result["schema"] = self.schema_
+        if self.pattern is not None:
+            result["pattern"] = self.pattern
+        return result
 
 
 class FunctionTool(BaseModel):

together/types/finetune.py

@@ -212,6 +212,9 @@ class FinetuneRequest(BaseModel):
     )
     # from step
     from_checkpoint: str | None = None
+    # hf related fields
+    hf_api_token: str | None = None
+    hf_output_repo_name: str | None = None
 
 
 class FinetuneResponse(BaseModel):

{together-1.5.16.dist-info → together-1.5.18.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: together
-Version: 1.5.16
+Version: 1.5.18
 Summary: Python client for Together's Cloud Platform!
 License: Apache-2.0
 Author: Together AI

{together-1.5.16.dist-info → together-1.5.18.dist-info}/RECORD

@@ -7,7 +7,7 @@ together/cli/api/chat.py,sha256=2PHRb-9T-lUEKhUJFtc7SxJv3shCVx40gq_8pzfsewM,9234
 together/cli/api/completions.py,sha256=l-Zw5t7hojL3w8xd_mitS2NRB72i5Z0xwkzH0rT5XMc,4263
 together/cli/api/endpoints.py,sha256=f6KafWZvRF6n_ThWdr3y9uhE6wPF37PcD45w_EtgXmY,13289
 together/cli/api/files.py,sha256=QLYEXRkY8J2Gg1SbTCtzGfoTMvosoeACNK83L_oLubs,3397
-together/cli/api/finetune.py,sha256=kyeJHT1hYRVGne10bgaGkSbg5w1hctb1eI2ZyI39cH8,17063
+together/cli/api/finetune.py,sha256=vIAvHQ8K6AxqJn2aqxd2ZPb1ZicLeb509_LpD4A9Thw,17517
 together/cli/api/images.py,sha256=GADSeaNUHUVMtWovmccGuKc28IJ9E_v4vAEwYHJhu5o,2645
 together/cli/api/models.py,sha256=CXw8B1hqNkadogi58GIXhLg_dTJnvTBaE7Kq1_xQ-10,1423
 together/cli/api/utils.py,sha256=IuqYWPnLI38_Bqd7lj8V_SnGdYc59pRmMbQmciS4FsM,1326
@@ -29,13 +29,13 @@ together/resources/audio/__init__.py,sha256=e7xp0Lkp_nMAHXcuFHS7dLXP_YqTPMMZIilW
 together/resources/audio/speech.py,sha256=81ib_gIo-Rxoaipx2Pi9ZsKnOTjeFPwSlBrcUkyX5xk,5211
 together/resources/batch.py,sha256=wSJR30CAFjzZ436vjYdXLCT0ahA5E0ud_qHMS5YvZ1M,3750
 together/resources/chat/__init__.py,sha256=RsTptdP8MeGjcdIjze896-J27cRvCbUoMft0X2BVlQ8,617
-together/resources/chat/completions.py,sha256=jYiNZsWa8RyEacL0VgxWj1egJ857oU4nxIY8uqGHcaU,14459
+together/resources/chat/completions.py,sha256=cBsSFWi9qToQCn4V_3qJ0gwRqORjF6NFDXmHcHfIhOY,14442
 together/resources/code_interpreter.py,sha256=vbN8Mh5MG6HQvqra7p61leIyfebgbgJTM_q2A_Fylhw,2948
 together/resources/completions.py,sha256=5Wa-ZjPCxRcam6CDe7KgGYlTA7yJZMmd5TrRgGCL_ug,11726
 together/resources/embeddings.py,sha256=PTvLb82yjG_-iQOyuhsilp77Fr7gZ0o6WD2KeRnKoxs,2675
 together/resources/endpoints.py,sha256=NNjp-wyzOotzlscGGrANhOHxQBjHTN8f5kTQTH_CLvE,17177
 together/resources/files.py,sha256=y3Ri6UtyAa7fjCJ8_fp26Y2hzzi6Aoo21JKkVgljFl8,5026
-together/resources/finetune.py,sha256=ATQvM_uICoPWBzSehw0d4WO5pevXi3zObpYumTColPs,39548
+together/resources/finetune.py,sha256=IFevk_AwCr9D2f4LssKEFP9OEDbHSuFA5CZMusDHajc,41343
 together/resources/images.py,sha256=LQUjKPaFxWTqOAPnyF1Pp7Rz4NLOYhmoKwshpYiprEM,4923
 together/resources/models.py,sha256=qgmAXv61Cq4oLxytenEZBywA8shldDHYxJ_EAu_4JWQ,3864
 together/resources/rerank.py,sha256=3Ju_aRSyZ1s_3zCSNZnSnEJErUVmt2xa3M8z1nvejMA,3931
@@ -44,7 +44,7 @@ together/types/__init__.py,sha256=_93XstLg1OOWratj_N1bsNN-2aS628uHH3SZj0wszyc,28
 together/types/abstract.py,sha256=1lFQI_3WjsR_t1128AeKW0aTk6EiM6Gh1J3ZuyLLPao,642
 together/types/audio_speech.py,sha256=jlj8BZf3dkIDARF1P11fuenVLj4try8Yx4RN-EAkhOU,2609
 together/types/batch.py,sha256=FP0RuQ3EDy-FV1bh-biPICvyRS7WqLm38GHz5lzKyXM,1112
-together/types/chat_completions.py,sha256=qpBCMXEWtRwW_fmiu6cecm3d4h6mcK8gvr-8JkbAopQ,5104
+together/types/chat_completions.py,sha256=NxJ7tFlWynxoLsRtQHzM7Ka3QxKVjRs6EvtOTYZ79bM,5340
 together/types/code_interpreter.py,sha256=cjF8TKgRkJllHS4i24dWQZBGTRsG557eHSewOiip0Kk,1770
 together/types/common.py,sha256=kxZ-N9xtBsGYZBmbIWnZ0rfT3Pn8PFB7sAbp3iv96pw,1525
 together/types/completions.py,sha256=o3FR5ixsTUj-a3pmOUzbSQg-hESVhpqrC9UD__VCqr4,2971
@@ -52,7 +52,7 @@ together/types/embeddings.py,sha256=J7grkYYn7xhqeKaBO2T-8XQRtHhkzYzymovtGdIUK5A,
 together/types/endpoints.py,sha256=EzNhHOoQ_D9fUdNQtxQPeSWiFzdFLqpNodN0YLmv_h0,4393
 together/types/error.py,sha256=OVlCs3cx_2WhZK4JzHT8SQyRIIqKOP1AZQ4y1PydjAE,370
 together/types/files.py,sha256=i-Ke57p8Svb1MbMZxu-Fo2zxIc6j-mDO2TLGNwPpGu0,1981
-together/types/finetune.py,sha256=6_jXgVVp4OOQXkABh0HKBzGy47H3wYCG2QxtXbdYauw,11079
+together/types/finetune.py,sha256=MRSgM7-s4p_5ZO72UaE78wtEVaTRA7yc53wZuDTty18,11182
 together/types/images.py,sha256=xnC-FZGdZU30WSFTybfGneWxb-kj0ZGufJsgHtB8j0k,980
 together/types/models.py,sha256=nwQIZGHKZpX9I6mK8z56VW70YC6Ry6JGsVa0s99QVxc,1055
 together/types/rerank.py,sha256=qZfuXOn7MZ6ly8hpJ_MZ7OU_Bi1-cgYNSB20Wja8Qkk,1061
@@ -62,8 +62,8 @@ together/utils/api_helpers.py,sha256=2K0O6qeEQ2zVFvi5NBN5m2kjZJaS3-JfKFecQ7SmGaw
 together/utils/files.py,sha256=btWQawwXbNKfPmCtRyObZViG1Xx-IPz45PrAtMXvcy8,16741
 together/utils/tools.py,sha256=H2MTJhEqtBllaDvOyZehIO_IVNK3P17rSDeILtJIVag,2964
 together/version.py,sha256=p03ivHyE0SyWU4jAnRTBi_sOwywVWoZPU4g2gzRgG-Y,126
-together-1.5.16.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-together-1.5.16.dist-info/METADATA,sha256=HsP69DC5NZmuAj6PXR9ja1mQCOWh4ffflnvv3E0Mj3U,15497
-together-1.5.16.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
-together-1.5.16.dist-info/entry_points.txt,sha256=G-b5NKW6lUUf1V1fH8IPTBb7jXnK7lhbX9H1zTEJXPs,50
-together-1.5.16.dist-info/RECORD,,
+together-1.5.18.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+together-1.5.18.dist-info/METADATA,sha256=lr8RNJjGXy_p8hK86PuQolCMt0nKtxGYD4HGrbDe1Ak,15497
+together-1.5.18.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
+together-1.5.18.dist-info/entry_points.txt,sha256=G-b5NKW6lUUf1V1fH8IPTBb7jXnK7lhbX9H1zTEJXPs,50
+together-1.5.18.dist-info/RECORD,,