Merge branch 'vllm-project:main' into main

.github/mergify.yml

@@ -0,0 +1,58 @@
+pull_request_rules:
+- name: label-documentation
+  description: Automatically apply documentation label
+  conditions:
+    - or:
+      - files~=^[^/]+\.md$
+      - files~=^docs/
+  actions:
+    label:
+      add:
+        - documentation
+
+- name: label-ci-build
+  description: Automatically apply ci/build label
+  conditions:
+    - or:
+      - files~=^\.github/
+      - files~=\.buildkite/
+      - files~=^cmake/
+      - files=CMakeLists.txt
+      - files~=^Dockerfile
+      - files~=^requirements.*\.txt
+      - files=setup.py
+  actions:
+    label:
+      add:
+        - ci/build
+
+- name: label-frontend
+  description: Automatically apply frontend label
+  conditions:
+    - files~=^vllm/entrypoints/
+  actions:
+    label:
+      add:
+        - frontend
+
+- name: ping author on conflicts and add 'needs-rebase' label
+  conditions:
+      - conflict
+      - -closed
+  actions:
+    label:
+      add:
+        - needs-rebase
+    comment:
+      message: |
+       This pull request has merge conflicts that must be resolved before it can be
+       merged. @{{author}} please rebase it. https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork
+
+- name: remove 'needs-rebase' label when conflict is resolved
+  conditions:
+      - -conflict
+      - -closed
+  actions:
+    label:
+      remove:
+        - needs-rebase

.github/workflows/stale.yml

@@ -10,6 +10,7 @@ jobs:
     permissions:
       issues: write
       pull-requests: write
+      actions: write
     runs-on: ubuntu-latest
     steps:
       - uses: actions/stale@28ca1036281a5e5922ead5184a1bbf96e5fc984e # v9.0.0

CMakeLists.txt

@@ -49,7 +49,7 @@ set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx11
 # requirements.txt files and should be kept consistent.  The ROCm torch
 # versions are derived from Dockerfile.rocm
 #
-set(TORCH_SUPPORTED_VERSION_CUDA "2.4.0")
+set(TORCH_SUPPORTED_VERSION_CUDA "2.5.0")
 set(TORCH_SUPPORTED_VERSION_ROCM "2.5.0")
 
 #
@@ -507,7 +507,7 @@ else()
   FetchContent_Declare(
           vllm-flash-attn
           GIT_REPOSITORY https://github.com/vllm-project/flash-attention.git
-          GIT_TAG 013f0c4fc47e6574060879d9734c1df8c5c273bd
+          GIT_TAG 5259c586c403a4e4d8bf69973c159b40cc346fb9
           GIT_PROGRESS TRUE
           # Don't share the vllm-flash-attn build between build types
           BINARY_DIR ${CMAKE_BINARY_DIR}/vllm-flash-attn

CONTRIBUTING.md

@@ -11,12 +11,14 @@ We also believe in the power of community support; thus, answering queries, offe
 
 Finally, one of the most impactful ways to support us is by raising awareness about vLLM. Talk about it in your blog posts and highlight how it's driving your incredible projects. Express your support on social media if you're using vLLM, or simply offer your appreciation by starring our repository!
 
+## License
+
+See [LICENSE](LICENSE).
 
 ## Developing
 
 Depending on the kind of development you'd like to do (e.g. Python, CUDA), you can choose to build vLLM with or without compilation. Check out the [building from source](https://docs.vllm.ai/en/latest/getting_started/installation.html#build-from-source) documentation for details.
 
-
 ## Testing
 
 ```bash
@@ -33,6 +35,14 @@ pytest tests/
 
 ## Contribution Guidelines
 
+### DCO and Signed-off-by
+
+When contributing changes to this project, you must agree to the [DCO](DCO).
+Commits must include a `Signed-off-by:` header which certifies agreement with
+the terms of the [DCO](DCO).
+
+Using `-s` with `git commit` will automatically add this header.
+
 ### Issues
 
 If you encounter a bug or have a feature request, please [search existing issues](https://github.com/vllm-project/vllm/issues?q=is%3Aissue) first to see if it has already been reported. If not, please [file a new issue](https://github.com/vllm-project/vllm/issues/new/choose), providing as much relevant information as possible.

DCO

@@ -0,0 +1,34 @@
+Developer Certificate of Origin
+Version 1.1
+
+Copyright (C) 2004, 2006 The Linux Foundation and its contributors.
+
+Everyone is permitted to copy and distribute verbatim copies of this
+license document, but changing it is not allowed.
+
+
+Developer's Certificate of Origin 1.1
+
+By making a contribution to this project, I certify that:
+
+(a) The contribution was created in whole or in part by me and I
+    have the right to submit it under the open source license
+    indicated in the file; or
+
+(b) The contribution is based upon previous work that, to the best
+    of my knowledge, is covered under an appropriate open source
+    license and I have the right under that license to submit that
+    work with modifications, whether created in whole or in part
+    by me, under the same open source license (unless I am
+    permitted to submit under a different license), as indicated
+    in the file; or
+
+(c) The contribution was provided directly to me by some other
+    person who certified (a), (b) or (c) and I have not modified
+    it.
+
+(d) I understand and agree that this project and the contribution
+    are public and that a record of the contribution (including all
+    personal information I submit with it, including my sign-off) is
+    maintained indefinitely and may be redistributed consistent with
+    this project or the open source license(s) involved.

Dockerfile.rocm

@@ -121,6 +121,8 @@ ARG GIT_REPO_CHECK=0
 RUN --mount=type=bind,source=.git,target=.git \
     if [ "$GIT_REPO_CHECK" != 0 ]; then bash tools/check_repo.sh ; fi
 
+RUN python3 -m pip install --upgrade pip
+
 # Package upgrades for useful functionality or to avoid dependency issues
 RUN --mount=type=cache,target=/root/.cache/pip \
     python3 -m pip install --upgrade numba scipy huggingface-hub[cli] pytest-shard

README.md

@@ -13,9 +13,19 @@ Easy, fast, and cheap LLM serving for everyone
 | <a href="https://docs.vllm.ai"><b>Documentation</b></a> | <a href="https://vllm.ai"><b>Blog</b></a> | <a href="https://arxiv.org/abs/2309.06180"><b>Paper</b></a> | <a href="https://discord.gg/jz7wjKhh6g"><b>Discord</b></a> | <a href="https://x.com/vllm_project"><b>Twitter/X</b></a> | <a href="https://slack.vllm.ai"><b>Developer Slack</b></a> |
 </p>
 
+---
+
+**vLLM x Snowflake Meetup (Wednesday, November 13th, 5:30-8PM PT) at Snowflake HQ, San Mateo**
+
+We are excited to announce the last in-person vLLM meetup of the year!
+Join the vLLM developers and engineers from Snowflake AI Research to chat about the latest LLM inference optimizations and your 2025 vLLM wishlist!
+Register [here](https://lu.ma/h0qvrajz) and be a part of the event!
+
+---
+
 
 *Latest News* 🔥
-- [2024/10] We have just created a developer slack ([slack.vllm.ai](https://slack.vllm.ai)) focusing on coordinating contributions and discussing features. Please feel free to join us there! 
+- [2024/10] We have just created a developer slack ([slack.vllm.ai](https://slack.vllm.ai)) focusing on coordinating contributions and discussing features. Please feel free to join us there!
 - [2024/10] Ray Summit 2024 held a special track for vLLM! Please find the opening talk slides from the vLLM team [here](https://docs.google.com/presentation/d/1B_KQxpHBTRa_mDF-tR6i8rWdOU5QoTZNcEg2MKZxEHM/edit?usp=sharing). Learn more from the [talks](https://raysummit.anyscale.com/flow/anyscale/raysummit2024/landing/page/sessioncatalog?tab.day=20241001&search.sessiontracks=1719251906298001uzJ2) from other vLLM contributors and users!
 - [2024/09] We hosted [the sixth vLLM meetup](https://lu.ma/87q3nvnh) with NVIDIA! Please find the meetup slides [here](https://docs.google.com/presentation/d/1wrLGwytQfaOTd5wCGSPNhoaW3nq0E-9wqyP7ny93xRs/edit?usp=sharing).
 - [2024/07] We hosted [the fifth vLLM meetup](https://lu.ma/lp0gyjqr) with AWS! Please find the meetup slides [here](https://docs.google.com/presentation/d/1RgUD8aCfcHocghoP3zmXzck9vX3RCI9yfUAB2Bbcl4Y/edit?usp=sharing).
@@ -42,7 +52,7 @@ vLLM is fast with:
 - Speculative decoding
 - Chunked prefill
 
-**Performance benchmark**: We include a performance benchmark at the end of [our blog post](https://blog.vllm.ai/2024/09/05/perf-update.html). It compares the performance of vLLM against other LLM serving engines ([TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM), [SGLang](https://github.com/sgl-project/sglang) and [LMDeploy](https://github.com/InternLM/lmdeploy)). The implementation is under [nightly-benchmarks folder](.buildkite/nightly-benchmarks/) and you can [reproduce](https://github.com/vllm-project/vllm/issues/8176) this benchmark using our one-click runnable script. 
+**Performance benchmark**: We include a performance benchmark at the end of [our blog post](https://blog.vllm.ai/2024/09/05/perf-update.html). It compares the performance of vLLM against other LLM serving engines ([TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM), [SGLang](https://github.com/sgl-project/sglang) and [LMDeploy](https://github.com/InternLM/lmdeploy)). The implementation is under [nightly-benchmarks folder](.buildkite/nightly-benchmarks/) and you can [reproduce](https://github.com/vllm-project/vllm/issues/8176) this benchmark using our one-click runnable script.
 
 vLLM is flexible and easy to use with:
 

benchmarks/kernels/benchmark_layernorm.py

@@ -3,8 +3,8 @@
 import torch
 
 from vllm.model_executor.layers.layernorm import RMSNorm
-from vllm.utils import (STR_DTYPE_TO_TORCH_DTYPE, FlexibleArgumentParser,
-                        seed_everything)
+from vllm.platforms import current_platform
+from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE, FlexibleArgumentParser
 
 
 @torch.inference_mode()
@@ -16,7 +16,7 @@ def main(num_tokens: int,
          do_profile: bool = False,
          num_warmup_iters: int = 5,
          num_iters: int = 100) -> None:
-    seed_everything(seed)
+    current_platform.seed_everything(seed)
     torch.set_default_device("cuda")
 
     layer = RMSNorm(hidden_size).to(dtype=dtype)

benchmarks/kernels/benchmark_moe.py

@@ -10,7 +10,8 @@
 from transformers import AutoConfig
 
 from vllm.model_executor.layers.fused_moe.fused_moe import *
-from vllm.utils import FlexibleArgumentParser, seed_everything
+from vllm.platforms import current_platform
+from vllm.utils import FlexibleArgumentParser
 
 
 class BenchmarkConfig(TypedDict):
@@ -88,22 +89,23 @@ def prepare(i: int):
         input_gating.copy_(gating_output[i])
 
     def run():
-        fused_moe(
-            x,
-            w1,
-            w2,
-            input_gating,
-            topk,
-            renormalize=True,
-            inplace=True,
-            override_config=config,
-            use_fp8_w8a8=use_fp8_w8a8,
-            use_int8_w8a16=use_int8_w8a16,
-            w1_scale=w1_scale,
-            w2_scale=w2_scale,
-            a1_scale=a1_scale,
-            a2_scale=a2_scale,
-        )
+        from vllm.model_executor.layers.fused_moe import override_config
+        with override_config(config):
+            fused_moe(
+                x,
+                w1,
+                w2,
+                input_gating,
+                topk,
+                renormalize=True,
+                inplace=True,
+                use_fp8_w8a8=use_fp8_w8a8,
+                use_int8_w8a16=use_int8_w8a16,
+                w1_scale=w1_scale,
+                w2_scale=w2_scale,
+                a1_scale=a1_scale,
+                a2_scale=a2_scale,
+            )
 
     # JIT compilation & warmup
     run()
@@ -166,7 +168,7 @@ class BenchmarkWorker:
 
     def __init__(self, seed: int) -> None:
         torch.set_default_device("cuda")
-        seed_everything(seed)
+        current_platform.seed_everything(seed)
         self.seed = seed
 
     def benchmark(
@@ -180,7 +182,7 @@ def benchmark(
         use_fp8_w8a8: bool,
         use_int8_w8a16: bool,
     ) -> Tuple[Dict[str, int], float]:
-        seed_everything(self.seed)
+        current_platform.seed_everything(self.seed)
         dtype_str = get_config_dtype_str(dtype,
                                          use_int8_w8a16=use_int8_w8a16,
                                          use_fp8_w8a8=use_fp8_w8a8)

benchmarks/kernels/benchmark_paged_attention.py

@@ -5,8 +5,9 @@
 import torch
 
 from vllm import _custom_ops as ops
+from vllm.platforms import current_platform
 from vllm.utils import (STR_DTYPE_TO_TORCH_DTYPE, FlexibleArgumentParser,
-                        create_kv_caches_with_random, seed_everything)
+                        create_kv_caches_with_random)
 
 NUM_BLOCKS = 1024
 PARTITION_SIZE = 512
@@ -28,7 +29,7 @@ def main(
     device: str = "cuda",
     kv_cache_dtype: Optional[str] = None,
 ) -> None:
-    seed_everything(seed)
+    current_platform.seed_everything(seed)
 
     scale = float(1.0 / (head_size**0.5))
     query = torch.empty(num_seqs,

benchmarks/kernels/benchmark_quant.py

@@ -3,8 +3,8 @@
 import torch
 
 from vllm import _custom_ops as ops
-from vllm.utils import (STR_DTYPE_TO_TORCH_DTYPE, FlexibleArgumentParser,
-                        seed_everything)
+from vllm.platforms import current_platform
+from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE, FlexibleArgumentParser
 
 
 @torch.inference_mode()
@@ -17,7 +17,7 @@ def main(num_tokens: int,
          do_profile: bool = False,
          num_warmup_iters: int = 5,
          num_iters: int = 100) -> None:
-    seed_everything(seed)
+    current_platform.seed_everything(seed)
     torch.set_default_device("cuda")
 
     x = torch.randn(num_tokens, hidden_size, dtype=dtype)

benchmarks/kernels/benchmark_rope.py

@@ -6,7 +6,8 @@
 
 from vllm.model_executor.layers.rotary_embedding import (RotaryEmbedding,
                                                          get_rope)
-from vllm.utils import FlexibleArgumentParser, seed_everything
+from vllm.platforms import current_platform
+from vllm.utils import FlexibleArgumentParser
 
 
 def benchmark_rope_kernels_multi_lora(
@@ -22,7 +23,7 @@ def benchmark_rope_kernels_multi_lora(
     max_position: int = 8192,
     base: int = 10000,
 ) -> None:
-    seed_everything(seed)
+    current_platform.seed_everything(seed)
     torch.set_default_device(device)
     if rotary_dim is None:
         rotary_dim = head_size

cmake/utils.cmake

@@ -424,11 +424,7 @@ function (define_gpu_extension_target GPU_MOD_NAME)
   # Don't use `TORCH_LIBRARIES` for CUDA since it pulls in a bunch of
   # dependencies that are not necessary and may not be installed.
   if (GPU_LANGUAGE STREQUAL "CUDA")
-    if ("${CUDA_CUDA_LIB}" STREQUAL "")
-      set(CUDA_CUDA_LIB "${CUDA_CUDA_LIBRARY}")
-    endif()
-    target_link_libraries(${GPU_MOD_NAME} PRIVATE ${CUDA_CUDA_LIB}
-      ${CUDA_LIBRARIES})
+    target_link_libraries(${GPU_MOD_NAME} PRIVATE CUDA::cudart CUDA::cuda_driver)
   else()
     target_link_libraries(${GPU_MOD_NAME} PRIVATE ${TORCH_LIBRARIES})
   endif()

csrc/ops.h

@@ -5,6 +5,30 @@
 
 #include "core/scalar_type.hpp"
 
+#include <vector>
+
+torch::Tensor weak_ref_tensor(torch::Tensor& tensor) {
+  // Ensure tensor is on CUDA
+  if (!tensor.is_cuda()) {
+    throw std::runtime_error("Tensor must be on CUDA device");
+  }
+
+  // Get the raw data pointer
+  void* data_ptr = tensor.data_ptr();
+
+  // Get tensor sizes and strides
+  std::vector<int64_t> sizes = tensor.sizes().vec();
+  std::vector<int64_t> strides = tensor.strides().vec();
+
+  // Get tensor options (dtype, device)
+  auto options = tensor.options();
+
+  // Create a new tensor from the raw data pointer
+  auto new_tensor = torch::from_blob(data_ptr, sizes, strides, options);
+
+  return new_tensor;
+}
+
 void paged_attention_v1(
     torch::Tensor& out, torch::Tensor& query, torch::Tensor& key_cache,
     torch::Tensor& value_cache, int64_t num_kv_heads, double scale,

csrc/torch_bindings.cpp

@@ -18,6 +18,9 @@
 TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
   // vLLM custom ops
 
+  ops.def("weak_ref_tensor(Tensor input) -> Tensor");
+  ops.impl("weak_ref_tensor", torch::kCUDA, &weak_ref_tensor);
+
   // Attention ops
   // Compute the attention between an input query and the cached
   // keys/values using PagedAttention.