Mirrors
/
hf_text-generation-inference
mirror of https://github.com/huggingface/text-generation-inference.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

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

 -->
This commit is contained in:
Nicolas Patry 2024-06-05 14:41:34 +02:00 committed by GitHub
parent 8aece3bd68
commit 0a94fad79f
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 26 additions and 95 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

121
server/text_generation_server/layers/attention/rocm.py
View File

@ -126,40 +126,34 @@ if ENGINE != "triton":
import flash_attn_2_cuda
logger.info("ROCm: using Flash Attention 2 Composable Kernel implementation.")
except ImportError:
try:
import flash_attn_cuda
except ImportError as e:
if major >= 8:
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"
logger.info("ROCm: using Flash Attention 1")
except ImportError as e:
if major >= 8:
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:
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
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:
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,
)
raise RuntimeError(f"Unknown attention engine {ENGINE}")