apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 08:37:06 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

supports dynamic Cfp8 (#3767)
Some checks failed
CE Compile Job / ce_job_pre_check (push) Has been cancelled
Details
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
Details
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
Details
CE Compile Job / Show Code Archive Output (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
Details
CE Compile Job / CE_UPLOAD (push) Has been cancelled
Details
Deploy GitHub Pages / deploy (push) Has been cancelled
Details

Browse Source 
* supports dynamic Cfp8

* add unittest
This commit is contained in:
lzy
2025-09-08 11:41:29 +08:00
committed by GitHub
parent b5e20e3015
commit af49b81ffd
20 changed files with 1417 additions and 225 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
fastdeploy/worker/gpu_model_runner.py
View File

@@ -1023,6 +1023,8 @@ class GPUModelRunner(ModelRunnerBase):
kv_cache_shape = self.attn_backends[0].get_kv_cache_shape(
max_num_blocks=max_block_num, kv_cache_quant_type=kv_cache_quant_type
)
if kv_cache_quant_type == "block_wise_fp8":
kv_cache_scale_shape = [kv_cache_shape[0], kv_cache_shape[1], kv_cache_shape[2]]
local_rank = self.local_rank % self.parallel_config.tensor_parallel_size
if not profile and (self.cache_config.enable_prefix_caching or self.parallel_config.splitwise_role != "mixed"):
@@ -1050,6 +1052,17 @@ class GPUModelRunner(ModelRunnerBase):
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,
fill_value=0,
dtype=paddle.get_default_dtype(),
)
cache_kvs[f"value_cache_scales_{i}"] = paddle.full(
shape=kv_cache_scale_shape,
fill_value=0,
dtype=paddle.get_default_dtype(),
)
self.share_inputs["caches"] = list(cache_kvs.values())
for value in cache_kvs.values():
del value