vrr/kvcache-ai-ktransformers
2
0
Fork 0
mirror of https://github.com/kvcache-ai/ktransformers.git synced 2025-09-10 15:29:39 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

fix server warmup

Browse source
This commit is contained in:
Xie Weiyu 2025-02-18 11:39:45 +08:00
parent 9f1da18630
commit f029588b61
2 changed files with 17 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

9
ktransformers/server/backend/interfaces/ktransformers.py
View file

@ -73,13 +73,13 @@ class KTransformersInterface(TransformersInterface):
self._infer_lock = asyncio.Lock()
def decode_one_tokens(self, i):
def decode_one_tokens(self):
device_map = self.model.gguf_loader.tensor_device_map
torch_device = get_device("blk.0.self_attn", device_map)
torch_device = "cuda:0" if torch_device == "cuda" else torch_device
global warm_uped
if self.args.use_cuda_graph and ( (warm_uped == True and int(i) == 1) or (warm_uped == False and int(i) == 2) ):
warm_uped = True
if self.args.use_cuda_graph and warm_uped == True:
if not hasattr(self, "cuda_graph_runner"):
self.cuda_graph_runner = CUDAGraphRunner()
self.cuda_graph_runner.capture(
@ -102,6 +102,9 @@ class KTransformersInterface(TransformersInterface):
logits = logits[0, -1, :]
return self.logits_to_token(logits)
if self.args.use_cuda_graph:
warm_uped = True
if self.use_static_cache:
mask = torch.ones((1, self.seq_length)).to(torch_device)
logits = self.model(

4
ktransformers/server/backend/interfaces/transformers.py
View file

@ -219,7 +219,7 @@ class TransformersInterface(BackendInterfaceBase):
self.ever_generated_ids.add(last)
return last
def decode_one_tokens(self, i):
def decode_one_tokens(self):
if self.use_static_cache:
mask = torch.ones((1, self.seq_length)).to(self.args.device)
logits = self.model(
@ -299,7 +299,7 @@ class TransformersInterface(BackendInterfaceBase):
num_heads=self.model.config.num_attention_heads, head_dim_ckv=self.model.config.kv_lora_rank,
head_dim_kpe=self.model.config.qk_rope_head_dim, page_size=self.cache.page_size,
sm_scale=(self.model.config.qk_rope_head_dim + self.model.config.qk_nope_head_dim) ** (-0.5), q_data_type=torch.bfloat16, kv_data_type=torch.bfloat16)
next_token = self.decode_one_tokens(i)
next_token = self.decode_one_tokens()
self.profiler.inc("decode")
if next_token == self.tokenizer.eos_token_id:
assert self.args.batch_size == 1