ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Update utils.py

Browse source
This commit is contained in:
Daniel Han-Chen 2024-01-29 23:07:40 +11:00
commit 7c87d60bc1
1 changed files with 10 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

13
unsloth/kernels/utils.py
View file

@ -191,9 +191,11 @@ def fast_linear_forward(proj, X, temp_lora = None, out = None):
W, W_quant, lora_A, lora_B, lora_S = get_lora_parameters(proj)
bsz = X.shape[0]
if W_quant is None:
out = torch.matmul(X, W.t())
elif X.shape[0] <= 4:
elif bsz <= 4:
# Only batches of 4 are faster with Gemv
out = fast_gemv(X, W, W_quant, out = out)
else:
@ -204,8 +206,13 @@ def fast_linear_forward(proj, X, temp_lora = None, out = None):
# Add in LoRA weights
if lora_A is not None:
dtype = X.dtype
temp_lora = torch.matmul(X, lora_A.to(dtype).t(), out = temp_lora)
out.addmv_(lora_B.to(dtype).t(), temp_lora, alpha = lora_S)
if bsz == 1:
temp_lora = torch.mv(lora_A.to(dtype), X.ravel(), out = temp_lora)
out.addmv_(lora_B.to(dtype).t(), temp_lora, alpha = lora_S)
else:
temp_lora = torch.matmul(X, lora_A.to(dtype).t(), out = temp_lora)
out.addmm_(lora_B.to(dtype).t(), temp_lora, alpha = lora_S)
pass
pass
return out