# Adds support for `transformers` as a backend
Following https://github.com/huggingface/transformers/pull/35235, a
bunch of models should already be supported, we are ramping up support
for more models.
Thanks @Isotr0py for the TP support, and @hmellor for his help as well!
This includes:
- `trust_remote_code=True` support: any model on the hub, if it
implements attention the correct way can be natively supported!!
- tensor parallel support
---------
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Signed-off-by: Isotr0py <2037008807@qq.com>
Co-authored-by: Isotr0py <41363108+Isotr0py@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Isotr0py <2037008807@qq.com>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Fix to AWQ quant loading of the new R1 model
The new optimized MoE kernels for a large number of experts `moe_wn16`
uses AWQ quant which requires the attention layers to be in 16bit
The current merge has broken this, and the `get_quant_method` must
return None for it to work correctly again
---------
Signed-off-by: Srikanth Srinivas <srikanth@astrum.ai>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Signed-off-by: Beim <beim2015@outlook.com>
Signed-off-by: rshaw@neuralmagic.com <rshaw@neuralmagic.com>
Signed-off-by: mgoin <michael@neuralmagic.com>
Signed-off-by: npanpaliya <nishidha.panpaliya@partner.ibm.com>
Signed-off-by: Aleksandr Malyshev <maleksan@amd.com>
Signed-off-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>
Signed-off-by: simon-mo <xmo@berkeley.edu>
Signed-off-by: Cody Yu <hao.yu.cody@gmail.com>
Signed-off-by: Chen Zhang <zhangch99@outlook.com>
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
Signed-off-by: Ryan N <ryan.nguyen@centml.ai>
Signed-off-by: Brian Dellabetta <bdellabe@redhat.com>
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Signed-off-by: Rahul Tuli <rahul@neuralmagic.com>
Signed-off-by: Russell Bryant <rbryant@redhat.com>
Signed-off-by: simon-mo <simon.mo@hey.com>
Signed-off-by: Vicente Herrera <vicenteherrera@vicenteherrera.com>
Signed-off-by: Jinzhen Lin <linjinzhen@hotmail.com>
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Signed-off-by: Shawn Du <shawnd200@outlook.com>
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
Signed-off-by: youkaichao <youkaichao@gmail.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Beim <805908499@qq.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
Co-authored-by: mgoin <michael@neuralmagic.com>
Co-authored-by: simon-mo <xmo@berkeley.edu>
Co-authored-by: Nishidha <nishidha.panpaliya@partner.ibm.com>
Co-authored-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>
Co-authored-by: Aleksandr Malyshev <164964928+maleksan85@users.noreply.github.com>
Co-authored-by: Aleksandr Malyshev <maleksan@amd.com>
Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Co-authored-by: simon-mo <simon.mo@hey.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Zhuohan Li <zhuohan123@gmail.com>
Co-authored-by: Tyler Michael Smith <tysmith@redhat.com>
Co-authored-by: Alexander Matveev <59768536+alexm-neuralmagic@users.noreply.github.com>
Co-authored-by: Roger Wang <136131678+ywang96@users.noreply.github.com>
Co-authored-by: Cody Yu <hao.yu.cody@gmail.com>
Co-authored-by: Chen Zhang <zhangch99@outlook.com>
Co-authored-by: Kevin H. Luu <kevin@anyscale.com>
Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com>
Co-authored-by: Ryan Nguyen <96593302+xpbowler@users.noreply.github.com>
Co-authored-by: Brian Dellabetta <brian-dellabetta@users.noreply.github.com>
Co-authored-by: fade_away <1028552010@qq.com>
Co-authored-by: weilong.yu <weilong.yu@shopee.com>
Co-authored-by: Jee Jee Li <pandaleefree@gmail.com>
Co-authored-by: Eldar Kurtic <eldarkurtic314@gmail.com>
Co-authored-by: Rahul Tuli <rahul@neuralmagic.com>
Co-authored-by: Russell Bryant <rbryant@redhat.com>
Co-authored-by: Vicente Herrera <vicenteherrera@vicenteherrera.com>
Co-authored-by: Jinzhen Lin <linjinzhen@hotmail.com>
Co-authored-by: Shawn Du <shawnd200@outlook.com>
Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>
Co-authored-by: youkaichao <youkaichao@gmail.com>
When people use deepseek models, they find that they need to solve cv2
version conflict, see https://zhuanlan.zhihu.com/p/21064432691 .
I added the check, and make all imports of `cv2` lazy.
---------
Signed-off-by: youkaichao <youkaichao@gmail.com>
As more and more people are trying deepseek models with multi-node
inference, https://github.com/vllm-project/vllm/issues/7815 becomes more
frequent. Let's give clear message to users.
Signed-off-by: youkaichao <youkaichao@gmail.com>
- **Add SPDX license headers to python source files**
- **Check for SPDX headers using pre-commit**
commit 9d7ef44c3cfb72ca4c32e1c677d99259d10d4745
Author: Russell Bryant <rbryant@redhat.com>
Date: Fri Jan 31 14:18:24 2025 -0500
Add SPDX license headers to python source files
This commit adds SPDX license headers to python source files as
recommended to
the project by the Linux Foundation. These headers provide a concise way
that is
both human and machine readable for communicating license information
for each
source file. It helps avoid any ambiguity about the license of the code
and can
also be easily used by tools to help manage license compliance.
The Linux Foundation runs license scans against the codebase to help
ensure
we are in compliance with the licenses of the code we use, including
dependencies. Having these headers in place helps that tool do its job.
More information can be found on the SPDX site:
- https://spdx.dev/learn/handling-license-info/
Signed-off-by: Russell Bryant <rbryant@redhat.com>
commit 5a1cf1cb3b80759131c73f6a9dddebccac039dea
Author: Russell Bryant <rbryant@redhat.com>
Date: Fri Jan 31 14:36:32 2025 -0500
Check for SPDX headers using pre-commit
Signed-off-by: Russell Bryant <rbryant@redhat.com>
---------
Signed-off-by: Russell Bryant <rbryant@redhat.com>
As mentioned in RFC https://github.com/vllm-project/vllm/issues/12254,
this PR achieves the task: combine allocate_slots and append_slots.
There should be no functionality change, except that in decode, also
raise exception when num_tokens is zero (like prefill), and change the
unit test case accordingly.
@comaniac @rickyyx @WoosukKwon @youkaichao @heheda12345 @simon-mo
---------
Signed-off-by: Shawn Du <shawnd200@outlook.com>
A small optimization to avoid creating a new `ConstantList` every time `request.kv_block_hashes` is used.
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
I noticed during testing that I was getting a lot of these deprecation
warnings about `local_lora_path`:
```
DeprecationWarning: The 'lora_local_path' attribute is deprecated
and will be removed in a future version.
Please use 'lora_path' instead.
```
The check used for emitting this warning was always True, even when the
parameter was not actually specified. It will always be in
`__struct_fields__`. We should be checking for a non-None value,
instead.
Signed-off-by: Russell Bryant <rbryant@redhat.com>
Signed-off-by: Russell Bryant <rbryant@redhat.com>
From @mgoin in https://github.com/vllm-project/vllm/pull/12638
I cannot push to that branch, therefore a new PR to unblock release.
---------
Signed-off-by: mgoin <michael@neuralmagic.com>
Signed-off-by: simon-mo <simon.mo@hey.com>
Co-authored-by: mgoin <michael@neuralmagic.com>
This PR implements the Deepseek V3 support by performing matrix absorption the fp8 weights
---------
Signed-off-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>
Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Co-authored-by: simon-mo <simon.mo@hey.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Zhuohan Li <zhuohan123@gmail.com>
Co-authored-by: Tyler Michael Smith <tysmith@redhat.com>
Co-authored-by: Alexander Matveev <59768536+alexm-neuralmagic@users.noreply.github.com>
This PR addresses a bug in the Cutlass integration where the
`sparsity_config.ignore` list was not being respected. When only a
subset of modules were configured as Sparse24, the system incorrectly
selected Cutlass for non-sparse modules as well. This update ensures the
correct scheme is selected for non-sparse modules, fixing this behavior.
---
### Changes
- Updated logic to correctly respect `sparsity_config.ignore`.
- Ensured non-sparse modules use the appropriate scheme instead of
defaulting to Cutlass.
---
<details>
<summary>Testing Setup</summary>
The fix has been tested on top of [this
diff](https://github.com/vllm-project/vllm/pull/12097).
#### Steps to Test:
```bash
git checkout -b my-test-branch origin/rahul-bitmask-additions # compressed Cutlass support
git revert --no-edit aa2cd2c # revert Tyler's commit to turn off Cutlass for W16A16
git cherry-pick ca624cddb # this branch
```
#### Additional Patch Required:
```diff
diff --git a/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py b/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py
index a54177c1c..f916dd0c9 100644
--- a/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py
+++ b/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py
@@ -9,7 +9,7 @@ from compressed_tensors.quantization import (QuantizationArgs,
QuantizationStrategy,
QuantizationType)
from pydantic import BaseModel
-
+from vllm.logger import init_logger
from vllm.model_executor.layers.fused_moe import FusedMoE
from vllm.model_executor.layers.linear import (LinearBase, LinearMethodBase,
UnquantizedLinearMethod)
@@ -27,7 +27,7 @@ from vllm.model_executor.layers.quantization.compressed_tensors.utils import (
should_ignore_layer)
from vllm.model_executor.layers.quantization.kv_cache import BaseKVCacheMethod
from vllm.platforms import current_platform
-
+logger = init_logger(__name__)
__all__ = ["CompressedTensorsLinearMethod"]
SPARSITY_CONFIG_NAME: Literal["sparsity_config"] = "sparsity_config"
```
Apply using:
```bash
git apply logging-patch.patch
```
</details>
---
<details>
<summary>Models Tested</summary>
- `nm-testing/TinyLlama-1.1B-Chat-v1.0-gsm8k-partial-24`
- `nm-testing/TinyLlama-1.1B-Chat-v1.0-gsm8k-full-sparse24`
-
`nm-testing/TinyLlama-1.1B-Chat-v1.0-gsm8k-partial-24-entire-fp8-compressed`
-
`nm-testing/TinyLlama-1.1B-Chat-v1.0-gsm8k-partial-24-remaining-fp8-compressed`
</details>
---
<details>
<summary>Example Output</summary>
#### Layers 0-5 (Sparse24)
```
Using scheme: CompressedTensors24 for model.layers.0.self_attn.qkv_proj
Using scheme: CompressedTensors24 for model.layers.0.self_attn.o_proj
Using scheme: CompressedTensors24 for model.layers.0.mlp.gate_up_proj
Using scheme: CompressedTensors24 for model.layers.0.mlp.down_proj
...
```
#### Layers 6+ (Non-Sparse, FP8)
```
Using scheme: CompressedTensorsW8A8Fp8 for model.layers.6.self_attn.qkv_proj
Using scheme: CompressedTensorsW8A8Fp8 for model.layers.6.self_attn.o_proj
Using scheme: CompressedTensorsW8A8Fp8 for model.layers.6.mlp.gate_up_proj
Using scheme: CompressedTensorsW8A8Fp8 for model.layers.6.mlp.down_proj
...
```
</details>
**Note:** Assumed all modules in fused layers such as `QKV_proj` and
`Gate_up_proj` follow the same quantization/pruning scheme.
---
For related tasks using the Asana app for GitHub, refer to [[this
link](https://app.asana.com/0/0/1209227810815160)](https://app.asana.com/0/0/1209227810815160).
Signed-off-by: Rahul Tuli <rahul@neuralmagic.com>
Without this PR
---------------
Quantizing models with llm-compressor and a recipe that explicitly lists
names of layers produces a model that is not loadable by vLLM (i.e.
`vllm serve <model>` fails with `raise ValueError(f"Unable to find
matching target for {module} in the ...`).
Example recipe:
```
recipe = """
quantization_stage:
run_type: oneshot
quantization_modifiers:
GPTQModifier:
ignore: ["lm_head"]
config_groups:
group_0:
weights:
num_bits: 4
type: "int"
symmetric: true
strategy: "group"
group_size: 128
targets: [
"model.layers.0.mlp.down_proj",
"model.layers.2.mlp.down_proj",
"model.layers.3.mlp.down_proj",
"model.layers.4.mlp.down_proj",
"model.layers.5.mlp.down_proj",
"model.layers.6.mlp.down_proj",
"model.layers.7.mlp.down_proj",
"model.layers.8.mlp.down_proj",
"model.layers.9.mlp.down_proj",
"model.layers.10.mlp.down_proj",
"model.layers.11.mlp.down_proj",
"model.layers.12.mlp.down_proj",
"model.layers.13.mlp.down_proj",
"model.layers.14.mlp.down_proj",
"model.layers.15.mlp.down_proj",
"model.layers.16.mlp.down_proj",
"model.layers.17.mlp.down_proj",
"model.layers.19.mlp.down_proj",
"model.layers.21.mlp.down_proj",
"model.layers.22.mlp.down_proj",
.
.
.
]
"""
```
To reproduce the vLLM error:
```bash
vllm serve nm-testing/eldar-test
```
With this PR
------------
Models are loaded correctly without any errors.
**[Guided decoding performance optimization]** Sending the guided
decoding bitmask in xgrammar to the GPU
(`self.token_bitmask.to(scores.device)`) is a blocking operation that
prevents the CPU from pre-launching the sampler kernels. The CPU waits
until decode is complete, then copies the bitmask over. This PR changes
the operation to async via setting `non-blocking=True`.
(Current) The CPU is blocked on a `cudaStreamSynchronize` and only
pre-empts the sampling kernels after bitmask application. Below is the
Nsys profile for one decode phase from Llama 3.1 8B.
With the optimization, this is no longer the case:
---------
Signed-off-by: Ryan N <ryan.nguyen@centml.ai>
SUMMARY:
* previous PR for pulling in block configs also changed defaults
(https://github.com/vllm-project/vllm/pull/11589/files) for FP8
* this broke L4 MoE since there was not enough SHM for the default
configuration
* this reverts the non-block example to the default
Signed-off-by: rshaw@neuralmagic.com <rshaw@neuralmagic.com>
This pr adds extra key to block hash, to generate different hash value
for two blocks with the same token string but different extra_keys in
their parent blocks. For example, it can generate different hash value
for the second block of the following two requests:
```python
request1 = make_request(
request_id=0,
prompt_token_ids=[_ for _ in range(6)],
mm_positions=[{
"offset": 0,
"length": 3
}, {
"offset": 3,
"length": 3
}],
mm_hashes=["hash1", "hash2"],
)
request2 = make_request(
request_id=1,
prompt_token_ids=[_ for _ in range(6)],
mm_positions=[{
"offset": 0,
"length": 3
}, {
"offset": 3,
"length": 3
}],
mm_hashes=["hash3", "hash2"],
)
```
---------
Signed-off-by: Chen Zhang <zhangch99@outlook.com>
Signed-off-by: Ce Gao <cegao@tensorchord.ai>
Co-authored-by: Rafael Vasquez <rafvasq21@gmail.com>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
Co-authored-by: Michael Goin <mgoin@redhat.com>
