Fixing rocm. (#2021)

# What does this PR do?

<!--
Congratulations! You've made it this far! You're not quite done yet
though.

Once merged, your PR is going to appear in the release notes with the
title you set, so make sure it's a great title that fully reflects the
extent of your awesome contribution.

Then, please replace this with a description of the change and which
issue is fixed (if applicable). Please also include relevant motivation
and context. List any dependencies (if any) that are required for this
change.

Once you're done, someone will review your PR shortly (see the section
"Who can review?" below to tag some potential reviewers). They may
suggest changes to make the code even better. If no one reviewed your PR
after a week has passed, don't hesitate to post a new comment
@-mentioning the same persons---sometimes notifications get lost.
-->

<!-- Remove if not applicable -->

Fixes # (issue)


## Before submitting
- [ ] This PR fixes a typo or improves the docs (you can dismiss the
other checks if that's the case).
- [ ] Did you read the [contributor
guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests),
      Pull Request section?
- [ ] Was this discussed/approved via a Github issue or the
[forum](https://discuss.huggingface.co/)? Please add a link
      to it if that's the case.
- [ ] Did you make sure to update the documentation with your changes?
Here are the
[documentation
guidelines](https://github.com/huggingface/transformers/tree/main/docs),
and
[here are tips on formatting
docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation).
- [ ] Did you write any new necessary tests?


## Who can review?

Anyone in the community is free to review the PR once the tests have
passed. Feel free to tag
members/contributors who may be interested in your PR.

<!-- Your PR will be replied to more quickly if you can figure out the
right person to tag with @


@OlivierDehaene OR @Narsil

 -->

server/text_generation_server/layers/attention/rocm.py

@@ -126,40 +126,34 @@ if ENGINE != "triton":
         import flash_attn_2_cuda
 
         logger.info("ROCm: using Flash Attention 2 Composable Kernel implementation.")
-    except ImportError:
+    except ImportError as e:
-        try:
+        if major >= 8:
-            import flash_attn_cuda
+            architecture_suffix = f"-{SYSTEM}"
+            raise ImportError(
+                "Flash Attention V2 is not installed.\n"
+                "Use the official Docker image (ghcr.io/huggingface/text-generation-inference:latest) "
+                f"or install flash attention v2 with `cd server && make install install-flash-attention-v2{architecture_suffix}`"
+            )
+        elif is_sm75:
+            raise ImportError(
+                "Flash Attention is not installed.\n"
+                "Use the official Docker image (ghcr.io/huggingface/text-generation-inference:latest) "
+                "or install flash attention with `cd server && make install install-flash-attention`"
+            ) from e
+        else:
 
-            ENGINE = "v1"
+            for idx in range(torch.cuda.device_count()):
-            logger.info("ROCm: using Flash Attention 1")
+                name = torch.cuda.get_device_name(idx)
-        except ImportError as e:
+                if "MI210" not in name and "MI250" not in name:
-            if major >= 8:
+                    raise ImportError(
-                architecture_suffix = f"-{SYSTEM}"
+                        f"AMD GPU {torch.cuda.get_device_name(idx)} does not support flash-attention"
-                raise ImportError(
+                    )
-                    "Flash Attention V2 is not installed.\n"
+            raise ImportError(
-                    "Use the official Docker image (ghcr.io/huggingface/text-generation-inference:latest) "
+                f"AMD GPU with ROCm capability {major} {minor} is not supported"
-                    f"or install flash attention v2 with `cd server && make install install-flash-attention-v2{architecture_suffix}`"
+            ) from e
-                )
-            elif is_sm75:
-                raise ImportError(
-                    "Flash Attention is not installed.\n"
-                    "Use the official Docker image (ghcr.io/huggingface/text-generation-inference:latest) "
-                    "or install flash attention with `cd server && make install install-flash-attention`"
-                ) from e
-            else:
-
-                for idx in range(torch.cuda.device_count()):
-                    name = torch.cuda.get_device_name(idx)
-                    if "MI210" not in name and "MI250" not in name:
-                        raise ImportError(
-                            f"AMD GPU {torch.cuda.get_device_name(idx)} does not support flash-attention"
-                        )
-                raise ImportError(
-                    f"AMD GPU with ROCm capability {major} {minor} is not supported"
-                ) from e
 
 
-SUPPORTS_WINDOWING = ENGINE != "v1"
+SUPPORTS_WINDOWING = False
 if ENGINE == "ck":
 
     def attention(
@@ -186,17 +180,12 @@ if ENGINE == "ck":
             out,
             cu_seqlens,
             cu_seqlens,
-            None,
-            None,
-            None,
             max_s,
             max_s,
             0.0,
             softmax_scale,
             False,
             causal,
-            window_size_left,
-            0,
             False,
             None,
         )
@@ -234,62 +223,4 @@ elif ENGINE == "triton":
         return output
 
 else:
-
+    raise RuntimeError(f"Unknown attention engine {ENGINE}")
-    def attention(
-        q,
-        k,
-        v,
-        out,
-        cu_seqlens,
-        max_s,
-        softmax_scale,
-        window_size_left=-1,
-    ):
-        if window_size_left != -1:
-            raise NotImplementedError(
-                "window_size_left is only available with flash attn v2"
-            )
-
-        # Flash attention v1 requires q, k and v to have the same number of heads
-        if k.shape[1] != q.shape[1]:
-            # MQA expand
-            if k.shape[1] == 1:
-                k = k.expand(-1, q.shape[1], -1)
-            # Grouped attention reshape
-            else:
-                original_shape = k.shape
-                k = (
-                    k.unsqueeze(2)
-                    .expand(-1, -1, q.shape[1] // k.shape[1], -1)
-                    .reshape(original_shape[0], -1, original_shape[2])
-                )
-        if v.shape[1] != q.shape[1]:
-            # MQA expand
-            if v.shape[1] == 1:
-                v = v.expand(-1, q.shape[1], -1)
-            # Grouped attention reshape
-            else:
-                original_shape = v.shape
-                v = (
-                    v.unsqueeze(2)
-                    .expand(-1, -1, q.shape[1] // v.shape[1], -1)
-                    .reshape(original_shape[0], -1, original_shape[2])
-                )
-
-        return flash_attn_cuda.fwd(
-            q,
-            k,
-            v,
-            out,
-            cu_seqlens,
-            cu_seqlens,
-            max_s,
-            max_s,
-            0.0,
-            softmax_scale,
-            False,
-            True,
-            False,
-            0,
-            None,
-        )