From 8704c091923288d4ff37c37101ce617129df0d98 Mon Sep 17 00:00:00 2001 From: lazymio Date: Mon, 24 Feb 2025 21:01:33 +0800 Subject: [PATCH] Allow temperature and top_p from requests --- .../server/api/openai/legacy/completions.py | 4 ++-- .../server/backend/interfaces/ktransformers.py | 6 +++--- .../server/backend/interfaces/transformers.py | 18 +++++++++++------- .../server/schemas/legacy/completions.py | 2 ++ 4 files changed, 18 insertions(+), 12 deletions(-) diff --git a/ktransformers/server/api/openai/legacy/completions.py b/ktransformers/server/api/openai/legacy/completions.py index be85a29..fe250f4 100644 --- a/ktransformers/server/api/openai/legacy/completions.py +++ b/ktransformers/server/api/openai/legacy/completions.py @@ -20,7 +20,7 @@ async def create_completion(request:Request,create:CompletionCreate): if create.stream: async def inner(): - async for token in interface.inference(create.prompt,id): + async for token in interface.inference(create.prompt,id,create.temperature,create.top_p): d = {'choices':[{'delta':{'content':token}}]} yield f"data:{json.dumps(d)}\n\n" d = {'choices':[{'delta':{'content':''},'finish_reason':''}]} @@ -28,6 +28,6 @@ async def create_completion(request:Request,create:CompletionCreate): return stream_response(request,inner()) else: comp = CompletionObject(id=id,object='text_completion',created=int(time())) - async for token in interface.inference(create.prompt,id): + async for token in interface.inference(create.prompt,id,create.temperature,create.top_p): comp.append_token(token) return comp diff --git a/ktransformers/server/backend/interfaces/ktransformers.py b/ktransformers/server/backend/interfaces/ktransformers.py index 49a3f16..85bfb29 100644 --- a/ktransformers/server/backend/interfaces/ktransformers.py +++ b/ktransformers/server/backend/interfaces/ktransformers.py @@ -14,7 +14,7 @@ from ktransformers.models.custom_cache import StaticCache from ktransformers.util.cuda_graph_runner import CUDAGraphRunner from ktransformers.local_chat import custom_models, default_optimize_rules from ktransformers.util.utils import get_device - +from typing import Optional warm_uped = False @@ -207,7 +207,7 @@ class KTransformersInterface(TransformersInterface): device = self.device_map.get("blk.0.self_attn", {}).get("generate_device", "cuda:0") return torch.tensor([self.seq_length - 1], device=device) - async def inference(self, local_messages, thread_id: str): + async def inference(self, local_messages, thread_id: str, temperature: Optional[float], top_p: Optional[float]): async with self._infer_lock: - async for v in super().inference(local_messages, thread_id): + async for v in super().inference(local_messages, thread_id, temperature, top_p): yield v diff --git a/ktransformers/server/backend/interfaces/transformers.py b/ktransformers/server/backend/interfaces/transformers.py index 8211933..d2e48a4 100644 --- a/ktransformers/server/backend/interfaces/transformers.py +++ b/ktransformers/server/backend/interfaces/transformers.py @@ -202,13 +202,17 @@ class TransformersInterface(BackendInterfaceBase): self.seq_length += 1 return self.streamer.put(new_tokens) - def prepare_logits_wrapper(self, inputs, device): + def prepare_logits_wrapper(self, inputs, device, temperature: Optional[float] = None, top_p: Optional[float] = None): + if temperature is None: + temperature = self.args.temperature + if top_p is None: + top_p = self.args.top_p generation_config, model_kwargs = self.model._prepare_generation_config( None, max_length=self.args.max_new_tokens, do_sample=True, top_k=self.args.top_k, - top_p=self.args.top_p, - temperature=self.args.temperature, + top_p=top_p, + temperature=temperature, repetition_penalty=self.args.repetition_penalty # change this to modify generate config ) self.inputs = inputs @@ -255,7 +259,7 @@ class TransformersInterface(BackendInterfaceBase): return self.logits_to_token(logits) @torch.no_grad - def prefill(self, input_ids: torch.Tensor, is_new: bool): + def prefill(self, input_ids: torch.Tensor, is_new: bool, temperature: Optional[float] = None, top_p: Optional[float] = None): input_ids_length = input_ids.shape[-1] logger.debug(f"input_ids: {input_ids.shape}") @@ -323,7 +327,7 @@ class TransformersInterface(BackendInterfaceBase): else: logits = self.model(inputs_embeds=inputs_embeds, return_dict=False)[0] - self.prepare_logits_wrapper(input_ids, device) + self.prepare_logits_wrapper(input_ids, device, temperature, top_p) next_token = self.logits_to_token(logits[0, -1, :]) yield self.append_new_tokens(next_token) @@ -359,7 +363,7 @@ class TransformersInterface(BackendInterfaceBase): self.last_request_id = thread_id return True - async def inference(self, local_messages, thread_id: str): + async def inference(self, local_messages, thread_id: str, temperature: Optional[float] = None, top_p: Optional[float] = None): self.streamer.reset() self.profiler.create_and_start_timer("tokenize") if isinstance(local_messages, List): @@ -386,7 +390,7 @@ class TransformersInterface(BackendInterfaceBase): print(think, end="",flush=True) yield think - for t in self.prefill(input_ids, self.check_is_new(thread_id)): + for t in self.prefill(input_ids, self.check_is_new(thread_id), temperature, top_p): # output think token after prefill done if t is not None: print(t, end="",flush=True) diff --git a/ktransformers/server/schemas/legacy/completions.py b/ktransformers/server/schemas/legacy/completions.py index 874e556..7be0404 100644 --- a/ktransformers/server/schemas/legacy/completions.py +++ b/ktransformers/server/schemas/legacy/completions.py @@ -9,6 +9,8 @@ class CompletionCreate(BaseModel): model: str prompt: str | List[str] stream: bool = False + temperature: Optional[float] + top_p: Optional[float] def get_tokenizer_messages(self): if isinstance(self.prompt,List):