[torch.compile] Adding torch compile annotations to some models (vllm…

…-project#9876)

Signed-off-by: youkaichao <[email protected]>
Co-authored-by: youkaichao <[email protected]>
Signed-off-by: Maxime Fournioux <[email protected]>

docs/source/models/supported_models.rst

@@ -281,7 +281,7 @@ Text Generation
     - ✅︎
   * - :code:`Qwen2ForCausalLM`
     - Qwen2
-    - :code:`Qwen/Qwen2-beta-7B`, :code:`Qwen/Qwen2-beta-7B-Chat`, etc.
+    - :code:`Qwen/Qwen2-7B-Instruct`, :code:`Qwen/Qwen2-7B`, etc.
     - ✅︎
     - ✅︎
   * - :code:`Qwen2MoeForCausalLM`

tests/distributed/test_pipeline_parallel.py

@@ -166,7 +166,7 @@ def iter_params(self, model_name: str):
     "microsoft/Phi-3.5-MoE-instruct": PPTestSettings.fast(trust_remote_code=True),  # noqa: E501
     "adept/persimmon-8b-chat": PPTestSettings.fast(),
     "Qwen/Qwen-7B-Chat": PPTestSettings.fast(trust_remote_code=True),
-    "Qwen/Qwen2-beta-7B-Chat": PPTestSettings.fast(),
+    "Qwen/Qwen2-7B-Instruct": PPTestSettings.fast(),
     "Qwen/Qwen1.5-MoE-A2.7B-Chat": PPTestSettings.fast(),
     "stabilityai/stablelm-3b-4e1t": PPTestSettings.fast(),
     "bigcode/starcoder2-3b": PPTestSettings.fast(),

vllm/model_executor/models/falcon.py

@@ -27,6 +27,7 @@
 from transformers import FalconConfig as HF_FalconConfig
 
 from vllm.attention import Attention, AttentionMetadata
+from vllm.compilation.decorators import support_torch_compile
 from vllm.config import CacheConfig
 from vllm.distributed import (get_pp_group, get_tensor_model_parallel_rank,
                               get_tensor_model_parallel_world_size,
@@ -329,6 +330,7 @@ def forward(
         return output
 
 
+@support_torch_compile
 class FalconModel(nn.Module):
 
     def __init__(

vllm/model_executor/models/phi.py

@@ -42,6 +42,7 @@
 from transformers import PhiConfig
 
 from vllm.attention import Attention, AttentionMetadata
+from vllm.compilation.decorators import support_torch_compile
 from vllm.config import CacheConfig, LoRAConfig
 from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.activation import get_act_fn
@@ -193,6 +194,7 @@ def forward(
         return hidden_states
 
 
+@support_torch_compile
 class PhiModel(nn.Module):
 
     def __init__(self,

vllm/model_executor/models/qwen.py

@@ -20,6 +20,7 @@
 from transformers import PretrainedConfig
 
 from vllm.attention import Attention, AttentionMetadata
+from vllm.compilation.decorators import support_torch_compile
 from vllm.config import CacheConfig, LoRAConfig, MultiModalConfig
 from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
 from vllm.inputs import (INPUT_REGISTRY, DecoderOnlyInputs, InputContext,
@@ -549,6 +550,7 @@ def forward(
         return hidden_states, residual
 
 
+@support_torch_compile
 class QWenModel(nn.Module):
 
     def __init__(

vllm/model_executor/models/qwen2.py

@@ -29,6 +29,7 @@
 from transformers import Qwen2Config
 
 from vllm.attention import Attention, AttentionMetadata
+from vllm.compilation.decorators import support_torch_compile
 from vllm.config import CacheConfig, LoRAConfig
 from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.activation import SiluAndMul
@@ -237,6 +238,7 @@ def forward(
         return hidden_states, residual
 
 
+@support_torch_compile
 class Qwen2Model(nn.Module):
 
     def __init__(

vllm/model_executor/models/qwen2_moe.py

@@ -30,6 +30,7 @@
 from transformers import PretrainedConfig
 
 from vllm.attention import Attention, AttentionMetadata
+from vllm.compilation.decorators import support_torch_compile
 from vllm.config import CacheConfig
 from vllm.distributed import (get_pp_group,
                               get_tensor_model_parallel_world_size,
@@ -312,6 +313,7 @@ def forward(
         return hidden_states, residual
 
 
+@support_torch_compile
 class Qwen2MoeModel(nn.Module):
 
     def __init__(