Skip to content

Remove transformers dependencies from cache_utils and restructure cache classes #616

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

Draft
quic-mamta wants to merge 6 commits into
base: main
Choose a base branch
from
+681 −690
Draft

Remove transformers dependencies from cache_utils and restructure cache classes #616

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
0ffd2de
remove inheriting transformers cache classes
mamtsing Nov 5, 2025
5334c68
Merge branch 'quic:main' into unify_cache
quic-mamta Nov 7, 2025
a8609f4
Restructure Cache in Layers for HybridCache, HybridChunkedCache and H…
mamtsing Nov 11, 2025
d210a82
Merge branch 'main' into unify_cache
quic-mamta Nov 17, 2025
2ac40fa
Removed redundancies from QEFFHybridCache and QEFFHybridChunkedCache
mamtsing Nov 17, 2025
dac601d
address review comments
mamtsing Nov 19, 2025
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
4 changes: 4 additions & 0 deletions QEfficient/customop/ctx_scatter_gather.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,6 +56,7 @@ def symbolic(g: torch.Graph, data: torch.Value, position_ids: torch.Value, updat
return g.onnxscript_op(CtxScatter, data, position_ids, updates).setTypeAs(data)


# TODO : This function will be deprecated in future.
@onnxscript.script(onnxscript.values.Opset("com.qualcomm.cloud", 1))
def CtxScatter3D(data: onnxscript.FLOAT, position_ids: onnxscript.INT32, updates: onnxscript.FLOAT) -> onnxscript.FLOAT:
# Find dims
Expand All @@ -75,6 +76,7 @@ def CtxScatter3D(data: onnxscript.FLOAT, position_ids: onnxscript.INT32, updates
return ops.ScatterND(data, indices, updates)


# TODO : This function will be deprecated in future.
class CtxScatterFunc3D(torch.autograd.Function):
@staticmethod
def forward(data: torch.Tensor, position_ids: torch.Tensor, updates: torch.Tensor):
Expand All @@ -92,13 +94,15 @@ def symbolic(g: torch.Graph, data: torch.Value, position_ids: torch.Value, updat
return g.onnxscript_op(CtxScatter3D, data, position_ids, updates).setTypeAs(data)


# TODO : This function will be deprecated in future.
@onnxscript.script(onnxscript.values.Opset("com.qualcomm.cloud", 1))
def CtxGather3D(data: onnxscript.FLOAT, ctx_indices: onnxscript.INT32) -> onnxscript.FLOAT:
ctx_indices = ops.Expand(ctx_indices, ops.Slice(ops.Shape(data), starts=[0], ends=[2], axes=[0]))
ctx_indices = ops.Unsqueeze(ctx_indices, [-1])
return ops.GatherND(data, ctx_indices, batch_dims=1)


# TODO : This function will be deprecated in future.
class CtxGatherFunc3D(torch.autograd.Function):
@staticmethod
def forward(data: torch.Tensor, ctx_indices: torch.Tensor):
Expand Down
4 changes: 4 additions & 0 deletions QEfficient/customop/ctx_scatter_gather_cb.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,6 +56,7 @@ def symbolic(
return g.onnxscript_op(CtxScatterCB, data, batch_index, position_ids, updates).setTypeAs(data)


# TODO : This function will be deprecated in future.
@onnxscript.script(onnxscript.values.Opset("com.qualcomm.cloud", 1))
def CtxScatterCB3D(
data: onnxscript.FLOAT, batch_index: onnxscript.INT32, position_ids: onnxscript.INT32, updates: onnxscript.FLOAT
Expand All @@ -76,6 +77,7 @@ def CtxScatterCB3D(
return ops.ScatterND(data, indices, updates)


# TODO : This function will be deprecated in future.
class CtxScatterFuncCB3D(torch.autograd.Function):
@staticmethod
def forward(data: torch.Tensor, batch_index: torch.Tensor, position_ids: torch.Tensor, updates: torch.Tensor):
Expand Down Expand Up @@ -139,6 +141,7 @@ def symbolic(
return g.onnxscript_op(CtxGatherCB, data, batch_index, ctx_indices, comp_ctx_len).setTypeAs(data)


# TODO : This function will be deprecated in future.
@onnxscript.script(onnxscript.values.Opset("com.qualcomm.cloud", 1))
def CtxGatherCB3D(
data: onnxscript.FLOAT, batch_index: onnxscript.INT32, ctx_indices: onnxscript.INT32
Expand All @@ -158,6 +161,7 @@ def CtxGatherCB3D(
return ops.GatherND(data, indices)


# TODO : This function will be deprecated in future.
class CtxGatherFuncCB3D(torch.autograd.Function):
@staticmethod
def forward(data: torch.Tensor, batch_index: torch.Tensor, ctx_indices: torch.Tensor):
Expand Down
Loading
Loading