Skip to content

【Inference Optimize】support mla cache buffer #4261

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: develop
Choose a base branch
from
+16 −7
Open

【Inference Optimize】support mla cache buffer #4261

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion fastdeploy/engine/sched/resource_manager_v1.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -333,7 +333,7 @@ def schedule(self):
# Prepare decoding task
scheduled_reqs.append(self._prepare_decode_task(request))
num_decoding_req_nums += 1
token_budget -= 1
token_budget -= 1
else: # need to prefill
llm_logger.debug(
f"scheduler prefill task: {request} request.need_prefill_tokens {request.need_prefill_tokens} request.num_computed_tokens {request.num_computed_tokens}"
Expand Down
21 changes: 15 additions & 6 deletions fastdeploy/worker/gpu_model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1167,11 +1167,12 @@ def initialize_kv_cache(self, profile: bool = False) -> None:
fill_value=0,
dtype=cache_type,
)
cache_kvs[f"value_caches_{i}"] = paddle.full(
shape=kv_cache_shape,
fill_value=0,
dtype=cache_type,
)
if "deepseek" not in self.fd_config.model_config.model_type:
cache_kvs[f"value_caches_{i}"] = paddle.full(
shape=kv_cache_shape,
fill_value=0,
dtype=cache_type,
)
if kv_cache_quant_type == "block_wise_fp8":
cache_kvs[f"key_cache_scales_{i}"] = paddle.full(
shape=kv_cache_scale_shape,
Expand Down Expand Up @@ -1898,7 +1899,15 @@ def cal_theortical_kvcache(self):
if self.speculative_method in ["mtp"]
else self.model_config.num_hidden_layers
)
required_memory = byte_of_dtype * 2 * (self.cache_config.block_size * hidden_dim) * num_layers # k + v
if "deepseek" in self.fd_config.model_config.model_type:
required_memory = (
byte_of_dtype
* (self.fd_config.model_config.kv_lora_rank + self.fd_config.model_config.qk_rope_head_dim)
* (self.cache_config.block_size)
* num_layers
) # compress_kv + k_pe
else:
required_memory = byte_of_dtype * 2 * (self.cache_config.block_size * hidden_dim) * num_layers # k + v
return required_memory

def not_need_stop(self) -> bool:
Expand Down
Loading