xinyun/DeepSeek-V3
1
0
Fork 0
mirror of https://git.datalinker.icu/deepseek-ai/DeepSeek-V3.git synced 2025-12-09 04:44:28 +08:00
Code Issues Packages Projects Releases Wiki Activity

add clamp min of 1e-4

Browse Source 
Signed-off-by: youkaichao <youkaichao@gmail.com>
This commit is contained in:
youkaichao 2025-08-26 18:14:52 +08:00
parent 21b2dfe172
commit 484b42ca4e
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
inference/kernel.py
View File

@ -23,7 +23,9 @@ def act_quant_kernel(x_ptr, y_ptr, s_ptr, BLOCK_SIZE: tl.constexpr, scale_fmt: t
pid = tl.program_id(axis=0)
offs = pid * BLOCK_SIZE + tl.arange(0, BLOCK_SIZE)
x = tl.load(x_ptr + offs).to(tl.float32)
s = tl.max(tl.abs(x)) / 448.
amax = tl.max(tl.abs(x))
amax = tl.min(amax, 1e-4)
s = amax / 448.
if scale_fmt == "ue8m0":
exp = tl.math.ceil(tl.math.log2(s))
s = tl.math.exp2(exp)