Port flat PA from habana_next to habana_main (#169)

FILL IN THE PR DESCRIPTION HERE

FIX #xxxx (*link existing issues this PR will resolve*)

**BEFORE SUBMITTING, PLEASE READ THE CHECKLIST BELOW AND FILL IN THE
DESCRIPTION ABOVE**

---

<details>
<!-- inside this <details> section, markdown rendering does not work, so
we use raw html here. -->
<summary><b> PR Checklist (Click to Expand) </b></summary>

<p>Thank you for your contribution to vLLM! Before submitting the pull
request, please ensure the PR meets the following criteria. This helps
vLLM maintain the code quality and improve the efficiency of the review
process.</p>

<h3>PR Title and Classification</h3>
<p>Only specific types of PRs will be reviewed. The PR title is prefixed
appropriately to indicate the type of change. Please use one of the
following:</p>
<ul>
    <li><code>[Bugfix]</code> for bug fixes.</li>
<li><code>[CI/Build]</code> for build or continuous integration
improvements.</li>
<li><code>[Doc]</code> for documentation fixes and improvements.</li>
<li><code>[Model]</code> for adding a new model or improving an existing
model. Model name should appear in the title.</li>
<li><code>[Frontend]</code> For changes on the vLLM frontend (e.g.,
OpenAI API server, <code>LLM</code> class, etc.) </li>
<li><code>[Kernel]</code> for changes affecting CUDA kernels or other
compute kernels.</li>
<li><code>[Core]</code> for changes in the core vLLM logic (e.g.,
<code>LLMEngine</code>, <code>AsyncLLMEngine</code>,
<code>Scheduler</code>, etc.)</li>
<li><code>[Hardware][Vendor]</code> for hardware-specific changes.
Vendor name should appear in the prefix (e.g.,
<code>[Hardware][AMD]</code>).</li>
<li><code>[Misc]</code> for PRs that do not fit the above categories.
Please use this sparingly.</li>
</ul>
<p><strong>Note:</strong> If the PR spans more than one category, please
include all relevant prefixes.</p>

<h3>Code Quality</h3>

<p>The PR need to meet the following code quality standards:</p>

<ul>
<li>We adhere to <a
href="https://google.github.io/styleguide/pyguide.html">Google Python
style guide</a> and <a
href="https://google.github.io/styleguide/cppguide.html">Google C++
style guide</a>.</li>
<li>Pass all linter checks. Please use <a
href="https://github.com/vllm-project/vllm/blob/main/format.sh"><code>format.sh</code></a>
to format your code.</li>
<li>The code need to be well-documented to ensure future contributors
can easily understand the code.</li>
<li>Include sufficient tests to ensure the project to stay correct and
robust. This includes both unit tests and integration tests.</li>
<li>Please add documentation to <code>docs/source/</code> if the PR
modifies the user-facing behaviors of vLLM. It helps vLLM user
understand and utilize the new features or changes.</li>
</ul>

<h3>Notes for Large Changes</h3>
<p>Please keep the changes as concise as possible. For major
architectural changes (>500 LOC excluding kernel/data/config/test), we
would expect a GitHub issue (RFC) discussing the technical design and
justification. Otherwise, we will tag it with <code>rfc-required</code>
and might not go through the PR.</p>

<h3>What to Expect for the Reviews</h3>

<p>The goal of the vLLM team is to be a <i>transparent reviewing
machine</i>. We would like to make the review process transparent and
efficient and make sure no contributor feel confused or frustrated.
However, the vLLM team is small, so we need to prioritize some PRs over
others. Here is what you can expect from the review process: </p>

<ul>
<li> After the PR is submitted, the PR will be assigned to a reviewer.
Every reviewer will pick up the PRs based on their expertise and
availability.</li>
<li> After the PR is assigned, the reviewer will provide status update
every 2-3 days. If the PR is not reviewed within 7 days, please feel
free to ping the reviewer or the vLLM team.</li>
<li> After the review, the reviewer will put an <code>
action-required</code> label on the PR if there are changes required.
The contributor should address the comments and ping the reviewer to
re-review the PR.</li>
<li> Please respond to all comments within a reasonable time frame. If a
comment isn't clear or you disagree with a suggestion, feel free to ask
for clarification or discuss the suggestion.
 </li>
</ul>

<h3>Thank You</h3>

<p> Finally, thank you for taking the time to read these guidelines and
for your interest in contributing to vLLM. Your contributions make vLLM
a great tool for everyone! </p>


</details>

---------

Co-authored-by: Michal Adamczyk <[email protected]>
Co-authored-by: barak goldberg <[email protected]>
Co-authored-by: Michal Szutenberg <[email protected]>
Co-authored-by: Jan Kaniecki <[email protected]>

README_GAUDI.md

@@ -455,33 +455,33 @@ Environment variables
 -   `VLLM_{phase}_{dim}_BUCKET_{param}` - collection of 12 environment
     variables configuring ranges of bucketing mechanism
     -   `{phase}` is either `PROMPT` or `DECODE`
-    -   `{dim}` is either `BS` or `SEQ`
+    -   `{dim}` is either `BS`, `SEQ` or `BLOCK`
     -   `{param}` is either `MIN`, `STEP` or `MAX`
     -   Default values:
         - Prompt:
            -   batch size min (`VLLM_PROMPT_BS_BUCKET_MIN`): `1`
-           -   batch size step (`VLLM_PROMPT_BS_BUCKET_STEP`): `32`
+           -   batch size step (`VLLM_PROMPT_BS_BUCKET_STEP`): `min(max_num_seqs, 32)`
            -   batch size max (`VLLM_PROMPT_BS_BUCKET_MAX`):
                     `min(max_num_seqs, 64)`
            -   sequence length min (`VLLM_PROMPT_SEQ_BUCKET_MIN`):
                     `block_size`
            -   sequence length step
                     (`VLLM_PROMPT_SEQ_BUCKET_STEP`): `block_size`
            -   sequence length max (`VLLM_PROMPT_SEQ_BUCKET_MAX`):
-                    `1024`
+                    `max_model_len`
 
         - Decode:
-            - batch size min (`VLLM_DECODE_BS_BUCKET_MIN`): `1`
+            - batch size min (`VLLM_DECODE_BS_BUCKET_MIN`): `min(max_num_seqs, 32)`
             -   batch size step (`VLLM_DECODE_BS_BUCKET_STEP`):
-                    `128`
+                    `min(max_num_seqs, 32)`
             -   batch size max (`VLLM_DECODE_BS_BUCKET_MAX`):
                     `max_num_seqs`
-            -   sequence length min (`VLLM_DECODE_SEQ_BUCKET_MIN`):
-                    `block_size`
-            -   sequence length step
-                    (`VLLM_DECODE_SEQ_BUCKET_STEP`): `block_size`
-            -   sequence length max (`VLLM_DECODE_SEQ_BUCKET_MAX`):
-                    `2048`
+            -   block size min (`VLLM_DECODE_BLOCK_BUCKET_MIN`):
+                    `128`
+            -   block size step
+                    (`VLLM_DECODE_BLOCK_BUCKET_STEP`): `128`
+            -   block size max (`VLLM_DECODE_BLOCK_BUCKET_MAX`):
+                    `max(128, (max_num_seqs*max_model_len)/block_size)`
 
 Additionally, there are HPU PyTorch Bridge environment variables
 impacting vLLM execution:

docs/source/getting_started/gaudi-installation.rst

@@ -335,19 +335,19 @@ Environment variables
 
       - Prompt:
          - batch size min (``VLLM_PROMPT_BS_BUCKET_MIN``): ``1``
-         - batch size step (``VLLM_PROMPT_BS_BUCKET_STEP``): ``32``
+         - batch size step (``VLLM_PROMPT_BS_BUCKET_STEP``): ``min(max_num_seqs, 32)``
          - batch size max (``VLLM_PROMPT_BS_BUCKET_MAX``): ``min(max_num_seqs, 64)``
          - sequence length min (``VLLM_PROMPT_SEQ_BUCKET_MIN``): ``block_size``
          - sequence length step (``VLLM_PROMPT_SEQ_BUCKET_STEP``): ``block_size``
-         - sequence length max (``VLLM_PROMPT_SEQ_BUCKET_MAX``): ``1024``
+         - sequence length max (``VLLM_PROMPT_SEQ_BUCKET_MAX``): ``max_model_len``
 
       - Decode:
-         - batch size min (``VLLM_DECODE_BS_BUCKET_MIN``): ``1``
-         - batch size step (``VLLM_DECODE_BS_BUCKET_STEP``): ``128``
+         - batch size min (``VLLM_DECODE_BS_BUCKET_MIN``): ``min(max_num_seqs, 32)``
+         - batch size step (``VLLM_DECODE_BS_BUCKET_STEP``): ``min(max_num_seqs, 32)``
          - batch size max (``VLLM_DECODE_BS_BUCKET_MAX``): ``max_num_seqs``
-         - sequence length min (``VLLM_DECODE_SEQ_BUCKET_MIN``): ``block_size``
-         - sequence length step (``VLLM_DECODE_SEQ_BUCKET_STEP``): ``block_size``
-         - sequence length max (``VLLM_DECODE_SEQ_BUCKET_MAX``): ``2048``
+         - sequence length min (``VLLM_DECODE_SEQ_BUCKET_MIN``): ``128``
+         - sequence length step (``VLLM_DECODE_SEQ_BUCKET_STEP``): ``128``
+         - sequence length max (``VLLM_DECODE_SEQ_BUCKET_MAX``): ``max(128, (max_num_seqs*max_model_len)/block_size)``
 
 
 Additionally, there are HPU PyTorch Bridge environment variables impacting vLLM execution:  

vllm/attention/backends/habana_attn.py

@@ -58,58 +58,14 @@ def copy_blocks(
 
 
 @dataclass
-class HabanaAttentionMetadata(AttentionMetadata, HabanaPagedAttentionMetadata):
-    """Metadata for HabanaAttentionbackend.
-
-    NOTE: Any python object stored here is not updated when it is
-    cuda-graph replayed. If you have values that need to be changed
-    dynamically, it should be stored in tensor. The tensor has to be
-    updated from `CUDAGraphRunner.forward` API.
-    """
+class HabanaAttentionMetadata(HabanaPagedAttentionMetadata, AttentionMetadata):
+    """Metadata for HabanaAttentionbackend."""
     # Currently, input sequences can only contain all prompts
     # or all decoding. True if all sequences are prompts.
     is_prompt: bool
-    # (batch_size,). The sequence length per sequence. Sequence length means
-    # the computed tokens + new tokens None if it is a decoding.
-    seq_lens: Optional[List[int]]
-    # seq_lens stored as a tensor.
+    attn_bias: Optional[torch.Tensor]
     seq_lens_tensor: Optional[torch.Tensor]
 
-    # |---------- N-1 iteration --------|
-    # |---------------- N iteration ---------------------|
-    # |- tokenA -|......................|-- newTokens ---|
-    # |---------- context_len ----------|
-    # |-------------------- seq_len ----------------------|
-    #                                   |-- query_len ---|
-
-    # Maximum query length in the batch.
-    max_query_len: Optional[int]
-    # (batch_size + 1,). The cumulative subquery lengths of the sequences in
-    # the batch, used to index into subquery. E.g., if the subquery length
-    # is [4, 6], it is [0, 4, 10].
-    subquery_start_loc: Optional[torch.Tensor]
-    # FIXME: It is for flash attn.
-    # (batch_size + 1,). The cumulative sequence lengths of the sequences in
-    # the batch, used to index into sequence. E.g., if the sequence length is
-    # [4, 6], it is [0, 4, 10].
-    seq_start_loc: Optional[torch.Tensor]
-    # (batch_size,) A tensor of context lengths (tokens that are computed
-    # so far).
-    context_lens_tensor: Optional[torch.Tensor]
-
-    # Whether or not if cuda graph is enabled.
-    # Cuda-graph is currently enabled for decoding only.
-    # TODO(woosuk): Move `use_cuda_graph` out since it's unrelated to attention.
-    use_cuda_graph: bool
-
-    def __post_init__(self):
-        # Set during the execution of the first attention op.
-        # It is a list because it is needed to set per prompt
-        # when alibi slopes is used. It is because of the limitation
-        # from xformer API.
-        # will not appear in the __repr__ and __init__
-        self.attn_bias: Optional[torch.Tensor] = None
-
 
 class HabanaAttentionImpl(AttentionImpl, torch.nn.Module):
     """
@@ -229,60 +185,48 @@ def forward(
 
         if attn_metadata.is_prompt:
             # Prompt run.
-            if kv_cache is None or attn_metadata.block_tables.numel() == 0:
-                if not self.prefill_usefusedsdpa:
-                    # TODO: move this outside of model
-                    assert attn_metadata.attn_bias is not None, \
+            if not self.prefill_usefusedsdpa:
+                # TODO: move this outside of model
+                assert attn_metadata.attn_bias is not None, \
                         'attn_bias must be set before calling model.forward!'
-                    attn_bias = attn_metadata.attn_bias
-                    if self.alibi_slopes is not None and \
-                        self.position_bias is not None:
-                        attn_bias.add_(self.position_bias[:, :,
-                                                          -attn_bias.size(2):,
-                                                          -attn_bias.size(3):])
-                else:
-                    attn_bias = None
-
-                query_shape = (batch_size, seq_len, self.num_heads,
-                               self.head_size)
-                kv_shape = (batch_size, seq_len_kv, self.num_kv_heads,
-                            self.head_size)
-                out = ops.prompt_attention(
-                    query.view(query_shape),
-                    key.view(kv_shape),
-                    value.view(kv_shape),
-                    attn_bias=attn_bias,
-                    p=0.0,
-                    scale=self.scale,
-                    matmul_qk_op=self.matmul_qk,
-                    softmax_op=self.softmax,
-                    matmul_av_op=self.matmul_av,
-                    valid_seq_lengths=attn_metadata.seq_lens_tensor,
-                )
-                output = out.reshape(batch_size, seq_len, hidden_size)
+                attn_bias = attn_metadata.attn_bias
+                if self.alibi_slopes is not None and \
+                    self.position_bias is not None:
+                    attn_bias.add_(self.position_bias[:, :,
+                                                      -attn_bias.size(2):,
+                                                      -attn_bias.size(3):])
             else:
-                # prefix-enabled attention
-                output = HabanaPagedAttention.forward_prefix(
-                    query,
-                    key,
-                    value,
-                    key_cache,
-                    value_cache,
-                    attn_metadata.block_tables,
-                    attn_metadata.subquery_start_loc,
-                    attn_metadata.seq_lens_tensor,
-                    attn_metadata.context_lens_tensor,
-                    attn_metadata.max_query_len,
-                    self.alibi_slopes,
-                )
+                attn_bias = None
+
+            query_shape = (batch_size, seq_len, self.num_heads, self.head_size)
+            kv_shape = (batch_size, seq_len_kv, self.num_kv_heads,
+                        self.head_size)
+            out = ops.prompt_attention(
+                query.view(query_shape),
+                key.view(kv_shape),
+                value.view(kv_shape),
+                attn_bias=attn_bias,
+                p=0.0,
+                scale=self.scale,
+                matmul_qk_op=self.matmul_qk,
+                softmax_op=self.softmax,
+                matmul_av_op=self.matmul_av,
+            )
+            output = out.reshape(batch_size, seq_len, hidden_size)
         else:
             # Decoding run.
             output = HabanaPagedAttention.forward_decode(
-                query, key_cache, value_cache, attn_metadata.block_tables,
-                attn_metadata.seq_lens_tensor, self.kv_cache_dtype,
-                self.num_kv_heads, self.scale, self.position_bias, k_scale,
-                v_scale, self.matmul_qk, self.softmax, self.matmul_av,
-                self.k_cache, self.v_cache)
+                query=query,
+                key_cache=key_cache,
+                value_cache=value_cache,
+                block_list=attn_metadata.block_list,
+                block_mapping=attn_metadata.block_mapping,
+                block_bias=attn_metadata.attn_bias,
+                scale=self.scale,
+                matmul_qk_op=self.matmul_qk,
+                matmul_av_op=self.matmul_av,
+                keys_fetch_func=self.k_cache.fetch_from_cache,
+                values_fetch_func=self.v_cache.fetch_from_cache)
         # Reshape the output tensor.
         return output.view(batch_size, seq_len, hidden_size)
 

vllm/attention/ops/habana_paged_attn.py

@@ -16,16 +16,9 @@
 @dataclass
 class HabanaPagedAttentionMetadata:
     """Metadata for PagedAttention."""
-    # (batch_size,). The length of sequences (entire tokens seen so far) per
-    # sequence.
-    seq_lens_tensor: Optional[torch.Tensor]
-    # (batch_size, max_blocks_per_seq).
-    # Block addresses per sequence. (Seq id -> list of physical block)
-    # E.g., [0, 1, 2] means tokens are stored in 0th, 1st, and 2nd blocks
-    # in the kv cache. Each block can contain up to block_size tokens.
-    # 2nd dimensions are padded up to max_blocks_per_seq if it is cuda-graph
-    # captured.
-    block_tables: Optional[torch.Tensor]
+    block_list: Optional[torch.Tensor]
+    block_mapping: Optional[torch.Tensor]
+    block_usage: Optional[torch.Tensor]
 
 
 class HabanaPagedAttention:
@@ -63,42 +56,8 @@ def write_to_paged_cache(key: torch.Tensor, value: torch.Tensor,
                                     slot_mapping, kv_cache_dtype, is_prompt)
 
     @staticmethod
-    def forward_decode(
-        query: torch.Tensor,
-        key_cache: torch.Tensor,
-        value_cache: torch.Tensor,
-        block_tables: torch.Tensor,
-        seq_lens: torch.Tensor,
-        kv_cache_dtype: str,
-        num_kv_heads: int,
-        scale: float,
-        alibi_slopes: Optional[torch.Tensor],
-        k_scale: float,
-        v_scale: float,
-        matmul_qk_op,
-        softmax_op,
-        matmul_av_op,
-        k_cache_cls,
-        v_cache_cls,
-    ) -> torch.Tensor:
-        block_size = value_cache.shape[1]
-        return ops.paged_attention_v1(
-            query,
-            key_cache,
-            value_cache,
-            num_kv_heads,
-            scale,
-            block_tables,
-            seq_lens,
-            block_size,
-            alibi_slopes,
-            kv_cache_dtype,
-            matmul_qk_op,
-            softmax_op,
-            matmul_av_op,
-            k_cache_cls,
-            v_cache_cls,
-        )
+    def forward_decode(**kwargs) -> torch.Tensor:
+        return ops.flat_pa(**kwargs)
 
     @staticmethod
     def forward_prefix(