feat: add inital build output

.gitattributes

@@ -0,0 +1 @@
+*.so filter=lfs diff=lfs merge=lfs -text

build/torch25-cxx11-cu118-x86_64-linux/flash_mla/__init__.py

@@ -0,0 +1,36 @@
+import torch
+
+from ._ops import ops
+
+
+def get_mla_metadata(seqlens_k: torch.Tensor, s_q: int, h_kv: int):
+    return ops.get_mla_metadata(seqlens_k, s_q, h_kv)
+
+
+def mha_fwd_kvcache_mla(
+    q: torch.Tensor,
+    kcache: torch.Tensor,
+    vcache_: torch.Tensor,
+    head_size_v: int,
+    seqlens_k: torch.Tensor,
+    block_table: torch.Tensor,
+    softmax_scale: float,
+    is_causal_: bool,
+    tile_scheduler_metadata: torch.Tensor,
+    num_splits: torch.Tensor,
+) -> torch.Tensor:
+    # TODO: remove when resolved
+    unknown_param = 0
+    return ops.mha_fwd_kvcache_mla(
+        q,
+        kcache,
+        vcache_,
+        head_size_v,
+        seqlens_k,
+        block_table,
+        softmax_scale,
+        is_causal_,
+        tile_scheduler_metadata,
+        num_splits,
+        unknown_param,
+    )

build/torch25-cxx11-cu118-x86_64-linux/flash_mla/_flash_mla_y6bdeh54o26h6.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c90743918a3a5a7e3f21c3de54c524af2d04d5bf112b9b638ac8b05c5009538
+size 2558728

build/torch25-cxx11-cu118-x86_64-linux/flash_mla/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _flash_mla_y6bdeh54o26h6
+ops = torch.ops._flash_mla_y6bdeh54o26h6
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_flash_mla_y6bdeh54o26h6::{op_name}"

build/torch25-cxx11-cu121-x86_64-linux/flash_mla/__init__.py

@@ -0,0 +1,36 @@
+import torch
+
+from ._ops import ops
+
+
+def get_mla_metadata(seqlens_k: torch.Tensor, s_q: int, h_kv: int):
+    return ops.get_mla_metadata(seqlens_k, s_q, h_kv)
+
+
+def mha_fwd_kvcache_mla(
+    q: torch.Tensor,
+    kcache: torch.Tensor,
+    vcache_: torch.Tensor,
+    head_size_v: int,
+    seqlens_k: torch.Tensor,
+    block_table: torch.Tensor,
+    softmax_scale: float,
+    is_causal_: bool,
+    tile_scheduler_metadata: torch.Tensor,
+    num_splits: torch.Tensor,
+) -> torch.Tensor:
+    # TODO: remove when resolved
+    unknown_param = 0
+    return ops.mha_fwd_kvcache_mla(
+        q,
+        kcache,
+        vcache_,
+        head_size_v,
+        seqlens_k,
+        block_table,
+        softmax_scale,
+        is_causal_,
+        tile_scheduler_metadata,
+        num_splits,
+        unknown_param,
+    )

build/torch25-cxx11-cu121-x86_64-linux/flash_mla/_flash_mla_mytbuokq46mgm.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d20c322601136e544cd1995cee8b21098f6cbc4279b7ebd8e007c59150d5df9
+size 2593856

build/torch25-cxx11-cu121-x86_64-linux/flash_mla/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _flash_mla_mytbuokq46mgm
+ops = torch.ops._flash_mla_mytbuokq46mgm
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_flash_mla_mytbuokq46mgm::{op_name}"

build/torch25-cxx11-cu124-x86_64-linux/flash_mla/__init__.py

@@ -0,0 +1,36 @@
+import torch
+
+from ._ops import ops
+
+
+def get_mla_metadata(seqlens_k: torch.Tensor, s_q: int, h_kv: int):
+    return ops.get_mla_metadata(seqlens_k, s_q, h_kv)
+
+
+def mha_fwd_kvcache_mla(
+    q: torch.Tensor,
+    kcache: torch.Tensor,
+    vcache_: torch.Tensor,
+    head_size_v: int,
+    seqlens_k: torch.Tensor,
+    block_table: torch.Tensor,
+    softmax_scale: float,
+    is_causal_: bool,
+    tile_scheduler_metadata: torch.Tensor,
+    num_splits: torch.Tensor,
+) -> torch.Tensor:
+    # TODO: remove when resolved
+    unknown_param = 0
+    return ops.mha_fwd_kvcache_mla(
+        q,
+        kcache,
+        vcache_,
+        head_size_v,
+        seqlens_k,
+        block_table,
+        softmax_scale,
+        is_causal_,
+        tile_scheduler_metadata,
+        num_splits,
+        unknown_param,
+    )

build/torch25-cxx11-cu124-x86_64-linux/flash_mla/_flash_mla_iohry4qbuggqa.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ccb2bc37dac026a6d0aff1fa3069d07fb3e90d3a0d88bb985508549918e1b454
+size 2574832

build/torch25-cxx11-cu124-x86_64-linux/flash_mla/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _flash_mla_iohry4qbuggqa
+ops = torch.ops._flash_mla_iohry4qbuggqa
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_flash_mla_iohry4qbuggqa::{op_name}"

build/torch25-cxx98-cu118-x86_64-linux/flash_mla/__init__.py

@@ -0,0 +1,36 @@
+import torch
+
+from ._ops import ops
+
+
+def get_mla_metadata(seqlens_k: torch.Tensor, s_q: int, h_kv: int):
+    return ops.get_mla_metadata(seqlens_k, s_q, h_kv)
+
+
+def mha_fwd_kvcache_mla(
+    q: torch.Tensor,
+    kcache: torch.Tensor,
+    vcache_: torch.Tensor,
+    head_size_v: int,
+    seqlens_k: torch.Tensor,
+    block_table: torch.Tensor,
+    softmax_scale: float,
+    is_causal_: bool,
+    tile_scheduler_metadata: torch.Tensor,
+    num_splits: torch.Tensor,
+) -> torch.Tensor:
+    # TODO: remove when resolved
+    unknown_param = 0
+    return ops.mha_fwd_kvcache_mla(
+        q,
+        kcache,
+        vcache_,
+        head_size_v,
+        seqlens_k,
+        block_table,
+        softmax_scale,
+        is_causal_,
+        tile_scheduler_metadata,
+        num_splits,
+        unknown_param,
+    )

build/torch25-cxx98-cu118-x86_64-linux/flash_mla/_flash_mla_ks7izefym4ha2.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:770a95aee0d760d66c11cee67284473199f49ed0a37264c9e98911e520653e5d
+size 2551384

build/torch25-cxx98-cu118-x86_64-linux/flash_mla/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _flash_mla_ks7izefym4ha2
+ops = torch.ops._flash_mla_ks7izefym4ha2
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_flash_mla_ks7izefym4ha2::{op_name}"

build/torch25-cxx98-cu121-x86_64-linux/flash_mla/__init__.py

@@ -0,0 +1,36 @@
+import torch
+
+from ._ops import ops
+
+
+def get_mla_metadata(seqlens_k: torch.Tensor, s_q: int, h_kv: int):
+    return ops.get_mla_metadata(seqlens_k, s_q, h_kv)
+
+
+def mha_fwd_kvcache_mla(
+    q: torch.Tensor,
+    kcache: torch.Tensor,
+    vcache_: torch.Tensor,
+    head_size_v: int,
+    seqlens_k: torch.Tensor,
+    block_table: torch.Tensor,
+    softmax_scale: float,
+    is_causal_: bool,
+    tile_scheduler_metadata: torch.Tensor,
+    num_splits: torch.Tensor,
+) -> torch.Tensor:
+    # TODO: remove when resolved
+    unknown_param = 0
+    return ops.mha_fwd_kvcache_mla(
+        q,
+        kcache,
+        vcache_,
+        head_size_v,
+        seqlens_k,
+        block_table,
+        softmax_scale,
+        is_causal_,
+        tile_scheduler_metadata,
+        num_splits,
+        unknown_param,
+    )

build/torch25-cxx98-cu121-x86_64-linux/flash_mla/_flash_mla_cd6mvrbov7aye.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1dc0f7b58d315d7f713b383732f4866df52131833afeb02fb8c8c0290b867ec9
+size 2590688

build/torch25-cxx98-cu121-x86_64-linux/flash_mla/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _flash_mla_cd6mvrbov7aye
+ops = torch.ops._flash_mla_cd6mvrbov7aye
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_flash_mla_cd6mvrbov7aye::{op_name}"

build/torch25-cxx98-cu124-x86_64-linux/flash_mla/__init__.py

@@ -0,0 +1,36 @@
+import torch
+
+from ._ops import ops
+
+
+def get_mla_metadata(seqlens_k: torch.Tensor, s_q: int, h_kv: int):
+    return ops.get_mla_metadata(seqlens_k, s_q, h_kv)
+
+
+def mha_fwd_kvcache_mla(
+    q: torch.Tensor,
+    kcache: torch.Tensor,
+    vcache_: torch.Tensor,
+    head_size_v: int,
+    seqlens_k: torch.Tensor,
+    block_table: torch.Tensor,
+    softmax_scale: float,
+    is_causal_: bool,
+    tile_scheduler_metadata: torch.Tensor,
+    num_splits: torch.Tensor,
+) -> torch.Tensor:
+    # TODO: remove when resolved
+    unknown_param = 0
+    return ops.mha_fwd_kvcache_mla(
+        q,
+        kcache,
+        vcache_,
+        head_size_v,
+        seqlens_k,
+        block_table,
+        softmax_scale,
+        is_causal_,
+        tile_scheduler_metadata,
+        num_splits,
+        unknown_param,
+    )

build/torch25-cxx98-cu124-x86_64-linux/flash_mla/_flash_mla_muvqsop7ydtdg.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10bbc09993e05062a2e257ea388e3aeaee1c1c54960ac352857bb921973f5dc8
+size 2571656

build/torch25-cxx98-cu124-x86_64-linux/flash_mla/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _flash_mla_muvqsop7ydtdg
+ops = torch.ops._flash_mla_muvqsop7ydtdg
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_flash_mla_muvqsop7ydtdg::{op_name}"

build/torch26-cxx11-cu118-x86_64-linux/flash_mla/__init__.py

@@ -0,0 +1,36 @@
+import torch
+
+from ._ops import ops
+
+
+def get_mla_metadata(seqlens_k: torch.Tensor, s_q: int, h_kv: int):
+    return ops.get_mla_metadata(seqlens_k, s_q, h_kv)
+
+
+def mha_fwd_kvcache_mla(
+    q: torch.Tensor,
+    kcache: torch.Tensor,
+    vcache_: torch.Tensor,
+    head_size_v: int,
+    seqlens_k: torch.Tensor,
+    block_table: torch.Tensor,
+    softmax_scale: float,
+    is_causal_: bool,
+    tile_scheduler_metadata: torch.Tensor,
+    num_splits: torch.Tensor,
+) -> torch.Tensor:
+    # TODO: remove when resolved
+    unknown_param = 0
+    return ops.mha_fwd_kvcache_mla(
+        q,
+        kcache,
+        vcache_,
+        head_size_v,
+        seqlens_k,
+        block_table,
+        softmax_scale,
+        is_causal_,
+        tile_scheduler_metadata,
+        num_splits,
+        unknown_param,
+    )

build/torch26-cxx11-cu118-x86_64-linux/flash_mla/_flash_mla_4gt6haj2eaeve.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7b1010f156fb14c68680155d12b6415ebaf818eb8f8a162b8b1cd0f89f085a4
+size 2563176

build/torch26-cxx11-cu118-x86_64-linux/flash_mla/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _flash_mla_4gt6haj2eaeve
+ops = torch.ops._flash_mla_4gt6haj2eaeve
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_flash_mla_4gt6haj2eaeve::{op_name}"

build/torch26-cxx11-cu124-x86_64-linux/flash_mla/__init__.py

@@ -0,0 +1,36 @@
+import torch
+
+from ._ops import ops
+
+
+def get_mla_metadata(seqlens_k: torch.Tensor, s_q: int, h_kv: int):
+    return ops.get_mla_metadata(seqlens_k, s_q, h_kv)
+
+
+def mha_fwd_kvcache_mla(
+    q: torch.Tensor,
+    kcache: torch.Tensor,
+    vcache_: torch.Tensor,
+    head_size_v: int,
+    seqlens_k: torch.Tensor,
+    block_table: torch.Tensor,
+    softmax_scale: float,
+    is_causal_: bool,
+    tile_scheduler_metadata: torch.Tensor,
+    num_splits: torch.Tensor,
+) -> torch.Tensor:
+    # TODO: remove when resolved
+    unknown_param = 0
+    return ops.mha_fwd_kvcache_mla(
+        q,
+        kcache,
+        vcache_,
+        head_size_v,
+        seqlens_k,
+        block_table,
+        softmax_scale,
+        is_causal_,
+        tile_scheduler_metadata,
+        num_splits,
+        unknown_param,
+    )

build/torch26-cxx11-cu124-x86_64-linux/flash_mla/_flash_mla_fqlv2wdc4vwcs.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13d1bfb8d6fa769168838204b8bcc62bc4c48ed81275e8a75d18f467dab352a0
+size 2575176

build/torch26-cxx11-cu124-x86_64-linux/flash_mla/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _flash_mla_fqlv2wdc4vwcs
+ops = torch.ops._flash_mla_fqlv2wdc4vwcs
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_flash_mla_fqlv2wdc4vwcs::{op_name}"

build/torch26-cxx11-cu126-x86_64-linux/flash_mla/__init__.py

@@ -0,0 +1,36 @@
+import torch
+
+from ._ops import ops
+
+
+def get_mla_metadata(seqlens_k: torch.Tensor, s_q: int, h_kv: int):
+    return ops.get_mla_metadata(seqlens_k, s_q, h_kv)
+
+
+def mha_fwd_kvcache_mla(
+    q: torch.Tensor,
+    kcache: torch.Tensor,
+    vcache_: torch.Tensor,
+    head_size_v: int,
+    seqlens_k: torch.Tensor,
+    block_table: torch.Tensor,
+    softmax_scale: float,
+    is_causal_: bool,
+    tile_scheduler_metadata: torch.Tensor,
+    num_splits: torch.Tensor,
+) -> torch.Tensor:
+    # TODO: remove when resolved
+    unknown_param = 0
+    return ops.mha_fwd_kvcache_mla(
+        q,
+        kcache,
+        vcache_,
+        head_size_v,
+        seqlens_k,
+        block_table,
+        softmax_scale,
+        is_causal_,
+        tile_scheduler_metadata,
+        num_splits,
+        unknown_param,
+    )

build/torch26-cxx11-cu126-x86_64-linux/flash_mla/_flash_mla_b4xwzd3vzzbdy.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c6cfeb75b5f9c8b88624f573fe80afc3f4d9f08828630c64c84e8100afe1101
+size 2555136

build/torch26-cxx11-cu126-x86_64-linux/flash_mla/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _flash_mla_b4xwzd3vzzbdy
+ops = torch.ops._flash_mla_b4xwzd3vzzbdy
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_flash_mla_b4xwzd3vzzbdy::{op_name}"

build/torch26-cxx98-cu118-x86_64-linux/flash_mla/__init__.py

@@ -0,0 +1,36 @@
+import torch
+
+from ._ops import ops
+
+
+def get_mla_metadata(seqlens_k: torch.Tensor, s_q: int, h_kv: int):
+    return ops.get_mla_metadata(seqlens_k, s_q, h_kv)
+
+
+def mha_fwd_kvcache_mla(
+    q: torch.Tensor,
+    kcache: torch.Tensor,
+    vcache_: torch.Tensor,
+    head_size_v: int,
+    seqlens_k: torch.Tensor,
+    block_table: torch.Tensor,
+    softmax_scale: float,
+    is_causal_: bool,
+    tile_scheduler_metadata: torch.Tensor,
+    num_splits: torch.Tensor,
+) -> torch.Tensor:
+    # TODO: remove when resolved
+    unknown_param = 0
+    return ops.mha_fwd_kvcache_mla(
+        q,
+        kcache,
+        vcache_,
+        head_size_v,
+        seqlens_k,
+        block_table,
+        softmax_scale,
+        is_causal_,
+        tile_scheduler_metadata,
+        num_splits,
+        unknown_param,
+    )

build/torch26-cxx98-cu118-x86_64-linux/flash_mla/_flash_mla_iyfleyrgf34la.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25a5dc0c023db2b535161f494470ebbf0cc0c261028906259db65cfa396fd6a5
+size 2551728

build/torch26-cxx98-cu118-x86_64-linux/flash_mla/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _flash_mla_iyfleyrgf34la
+ops = torch.ops._flash_mla_iyfleyrgf34la
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_flash_mla_iyfleyrgf34la::{op_name}"

build/torch26-cxx98-cu124-x86_64-linux/flash_mla/__init__.py

@@ -0,0 +1,36 @@
+import torch
+
+from ._ops import ops
+
+
+def get_mla_metadata(seqlens_k: torch.Tensor, s_q: int, h_kv: int):
+    return ops.get_mla_metadata(seqlens_k, s_q, h_kv)
+
+
+def mha_fwd_kvcache_mla(
+    q: torch.Tensor,
+    kcache: torch.Tensor,
+    vcache_: torch.Tensor,
+    head_size_v: int,
+    seqlens_k: torch.Tensor,
+    block_table: torch.Tensor,
+    softmax_scale: float,
+    is_causal_: bool,
+    tile_scheduler_metadata: torch.Tensor,
+    num_splits: torch.Tensor,
+) -> torch.Tensor:
+    # TODO: remove when resolved
+    unknown_param = 0
+    return ops.mha_fwd_kvcache_mla(
+        q,
+        kcache,
+        vcache_,
+        head_size_v,
+        seqlens_k,
+        block_table,
+        softmax_scale,
+        is_causal_,
+        tile_scheduler_metadata,
+        num_splits,
+        unknown_param,
+    )

build/torch26-cxx98-cu124-x86_64-linux/flash_mla/_flash_mla_fywd4tw2yn6ew.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59ea97594e0550f64aa3b5ad0358e68306f284344acef250e0293ccf50e87273
+size 2572008

build/torch26-cxx98-cu124-x86_64-linux/flash_mla/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _flash_mla_fywd4tw2yn6ew
+ops = torch.ops._flash_mla_fywd4tw2yn6ew
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_flash_mla_fywd4tw2yn6ew::{op_name}"

build/torch26-cxx98-cu126-x86_64-linux/flash_mla/__init__.py

@@ -0,0 +1,36 @@
+import torch
+
+from ._ops import ops
+
+
+def get_mla_metadata(seqlens_k: torch.Tensor, s_q: int, h_kv: int):
+    return ops.get_mla_metadata(seqlens_k, s_q, h_kv)
+
+
+def mha_fwd_kvcache_mla(
+    q: torch.Tensor,
+    kcache: torch.Tensor,
+    vcache_: torch.Tensor,
+    head_size_v: int,
+    seqlens_k: torch.Tensor,
+    block_table: torch.Tensor,
+    softmax_scale: float,
+    is_causal_: bool,
+    tile_scheduler_metadata: torch.Tensor,
+    num_splits: torch.Tensor,
+) -> torch.Tensor:
+    # TODO: remove when resolved
+    unknown_param = 0
+    return ops.mha_fwd_kvcache_mla(
+        q,
+        kcache,
+        vcache_,
+        head_size_v,
+        seqlens_k,
+        block_table,
+        softmax_scale,
+        is_causal_,
+        tile_scheduler_metadata,
+        num_splits,
+        unknown_param,
+    )

build/torch26-cxx98-cu126-x86_64-linux/flash_mla/_flash_mla_duoktlzcb6q5w.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e85fa15e0103badddb1cf6f894e06cf42ce1300431066af499feae6e5df4c66b
+size 2551968

build/torch26-cxx98-cu126-x86_64-linux/flash_mla/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _flash_mla_duoktlzcb6q5w
+ops = torch.ops._flash_mla_duoktlzcb6q5w
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_flash_mla_duoktlzcb6q5w::{op_name}"