[Bugfix] Python package do not have __version__ (#104)

Also fix some format issues in python docstring.

python/flashinfer/__init__.py

@@ -34,3 +34,9 @@
     BatchPrefillWithSharedPrefixPagedKVCacheWrapper,
 )
 from .page import append_paged_kv_cache
+
+try:
+    from ._build_meta import __version__ as __version__
+except ImportError:
+    with open("version.txt") as f:
+        __version__ = f.read().strip()

python/flashinfer/cascade.py

@@ -117,7 +117,7 @@ def merge_state_in_place(
     s_other : torch.Tensor
         The other logsumexp value to be merged, expected to be a float32 tensor,
         shape: ``(seq_len, num_heads)``.
-    
+
     Example
     -------
     >>> import torch
@@ -135,7 +135,7 @@ def merge_state_in_place(
 
 
 def merge_states(v: torch.Tensor, s: torch.Tensor):
-    r"""Merge multiple attention states (v, s). 
+    r"""Merge multiple attention states (v, s).
 
     Parameters
     ----------
@@ -154,7 +154,7 @@ def merge_states(v: torch.Tensor, s: torch.Tensor):
     S : torch.Tensor
         The logsumexp value from the merged KV-segments, shape:
         ``[seq_len, num_heads]``.
-    
+
     Example
     -------
     >>> import torch
@@ -229,7 +229,7 @@ def batch_decode_with_shared_prefix_padded_kv_cache(
     -------
     V : torch.Tensor
         The attention output, shape: ``[batch_size, num_heads, head_dim]``
-    
+
     Example
     -------
     >>> import torch
@@ -312,7 +312,7 @@ class BatchDecodeWithSharedPrefixPagedKVCacheWrapper:
     ... )
     >>> batch_size = 7
     >>> shared_prefix_len = 8192
-    >>> unique_kv_page_indices = torch.arange(max_num_pages).int().to("cuda:0") 
+    >>> unique_kv_page_indices = torch.arange(max_num_pages).int().to("cuda:0")
     >>> unique_kv_page_indptr = torch.tensor(
     ...     [0, 17, 29, 44, 48, 66, 100, 128], dtype=torch.int32, device="cuda:0"
     ... )
@@ -355,7 +355,7 @@ class BatchDecodeWithSharedPrefixPagedKVCacheWrapper:
     ...     # compute batch decode attention, reuse auxiliary data structures for all layers
     ...     o = wrapper.forward(q, k_shared, v_shared, unique_kv_data)
     ...     outputs.append(o)
-    ... 
+    ...
     >>> # clear auxiliary data structures
     >>> wrapper.end_forward()
     >>> outputs[0].shape
@@ -547,7 +547,7 @@ class BatchPrefillWithSharedPrefixPagedKVCacheWrapper:
     >>> qo_indptr = torch.tensor(
     ...     [0, 33, 44, 55, 66, 77, 88, nnz_qo], dtype=torch.int32, device="cuda:0"
     ... )
-    >>> paged_kv_indices = torch.arange(max_num_pages).int().to("cuda:0") 
+    >>> paged_kv_indices = torch.arange(max_num_pages).int().to("cuda:0")
     >>> paged_kv_indptr = torch.tensor(
     ...     [0, 17, 29, 44, 48, 66, 100, 128], dtype=torch.int32, device="cuda:0"
     ... )
@@ -590,7 +590,7 @@ class BatchPrefillWithSharedPrefixPagedKVCacheWrapper:
     ...         q, k_shared, v_shared, kv_data, causal=True
     ...     )
     ...     outputs.append(o)
-    ... 
+    ...
     s[0].shape>>> # clear auxiliary data structures
     >>> prefill_wrapper.end_forward()
     >>> outputs[0].shape

python/flashinfer/decode.py

@@ -179,7 +179,7 @@ def batch_decode_with_padded_kv_cache(
     -------
     torch.Tensor
         The attention output, shape: ``[batch_size, num_qo_heads, head_dim]``.
-    
+
     Examples
     --------
     >>> import torch
@@ -270,7 +270,7 @@ def batch_decode_with_padded_kv_cache_return_lse(
         The attention output, shape: [batch_size, num_qo_heads, head_dim]
     S : torch.Tensor
         The logsumexp of attention scores, Shape: [batch_size, num_qo_heads]
-    
+
     Examples
     --------
     >>> import torch
@@ -342,7 +342,7 @@ class BatchDecodeWithPagedKVCacheWrapper:
     ...     workspace_buffer, "NHD"
     ... )
     >>> batch_size = 7
-    >>> kv_page_indices = torch.arange(max_num_pages).int().to("cuda:0") 
+    >>> kv_page_indices = torch.arange(max_num_pages).int().to("cuda:0")
     >>> kv_page_indptr = torch.tensor(
     ...     [0, 17, 29, 44, 48, 66, 100, 128], dtype=torch.int32, device="cuda:0"
     ... )
@@ -374,7 +374,7 @@ class BatchDecodeWithPagedKVCacheWrapper:
     ...     # compute batch decode attention, reuse auxiliary data structures for all layers
     ...     o = decode_wrapper.forward(q, kv_data)
     ...     outputs.append(o)
-    ... 
+    ...
     >>> # clear auxiliary data structures
     >>> decode_wrapper.end_forward()
     >>> outputs[0].shape
@@ -589,7 +589,7 @@ def forward_return_lse(
             The attention output, shape: ``[batch_size, num_qo_heads, head_dim]``.
         S : torch.Tensor
             The logsumexp of attention scores, Shape: ``[batch_size, num_qo_heads]``.
-        
+
         Notes
         -----
         Please refer to the :ref:`tutorial <recursive-attention>` for a detailed

python/flashinfer/page.py

@@ -67,7 +67,7 @@ def append_paged_kv_cache(
         shape: ``[batch_size]``.
     kv_layout : str
         The layout of the paged kv-cache, either ``NHD`` or ``HND``.
-    
+
     Example
     -------
     >>> import torch
@@ -96,7 +96,7 @@ def append_paged_kv_cache(
     >>> # 25 = (2 - 1) * 16 + 9
     >>> # 22 = (2 - 1) * 16 + 6
     >>> kv_last_page_len = torch.tensor([13, 8, 9, 6], dtype=torch.int32, device="cuda:0")
-    >>> 
+    >>>
     >>> flashinfer.append_paged_kv_cache(
     ...     k_append,
     ...     v_append,
@@ -111,7 +111,7 @@ def append_paged_kv_cache(
     -----
     Please refer to the :ref:`tutorial <recursive-attention>` for a detailed
     explanation of the log-sum-exp function and attention states.
-    
+
     The function assumes that the space for appended k/v have already been allocated,
     which means :attr:`kv_indices`, :attr:`kv_indptr`, :attr:`kv_last_page_len` has
     incorporated appended k/v.

python/flashinfer/prefill.py

@@ -277,7 +277,7 @@ class BatchPrefillWithPagedKVCacheWrapper:
     >>> qo_indptr = torch.tensor(
     ...     [0, 33, 44, 55, 66, 77, 88, nnz_qo], dtype=torch.int32, device="cuda:0"
     ... )
-    >>> paged_kv_indices = torch.arange(max_num_pages).int().to("cuda:0") 
+    >>> paged_kv_indices = torch.arange(max_num_pages).int().to("cuda:0")
     >>> paged_kv_indptr = torch.tensor(
     ...     [0, 17, 29, 44, 48, 66, 100, 128], dtype=torch.int32, device="cuda:0"
     ... )
@@ -308,7 +308,7 @@ class BatchPrefillWithPagedKVCacheWrapper:
     ...         q, kv_data, causal=True
     ...     )
     ...     outputs.append(o)
-    ... 
+    ...
     >>> # clear auxiliary data structures
     >>> prefill_wrapper.end_forward()
     >>> outputs[0].shape
@@ -582,7 +582,7 @@ class BatchPrefillWithRaggedKVCacheWrapper:
     ...         q, k, v, causal=True
     ...     )
     ...     outputs.append(o)
-    ... 
+    ...
     >>> # clear auxiliary data structures
     >>> prefill_wrapper.end_forward()
     >>> outputs[0].shape