[Version] Bump version to v0.7.3 (ROCm#453)

* update vllm version to v0.7.3

Signed-off-by: tjtanaa <[email protected]>

* fix linter

Signed-off-by: tjtanaa <[email protected]>

* remove redundant code

Signed-off-by: tjtanaa <[email protected]>

---------

Signed-off-by: tjtanaa <[email protected]>

Author: tjtanaa

benchmarks/kernels/benchmark_mixtral_moe_rocm.py

@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: Apache-2.0
 import argparse
 import json
 import os

benchmarks/test_accuracy.py

@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: Apache-2.0
 import argparse
 import dataclasses
 

setup.py

@@ -499,9 +499,28 @@ def get_gaudi_sw_version():
 
 
 def get_vllm_version() -> str:
-    version = get_version(
+    # Get the version from setuptools_scm but override the base version
+    base_version = os.environ.get('SETUPTOOLS_SCM_PRETEND_VERSION', '0.7.3')
+
+    # Get the full version with git information
+    full_version = get_version(
         write_to="vllm/_version.py",  # TODO: move this to pyproject.toml
     )
+
+    # Extract git information if available
+    if '+' in full_version:
+        # Extract the dev and git parts (e.g., from 0.x.y.devN+gHASH)
+        git_parts = full_version.split('+', 1)[1]
+        # Combine with our base version
+        version = f"{base_version}.{full_version.split('+')[0].split('.')[-1]}+{git_parts}"  # noqa: E501
+    elif '.dev' in full_version:
+        # Handle case where dev number is present but no git hash
+        dev_part = full_version.split('.dev')[1]
+        version = f"{base_version}.dev{dev_part}"
+    else:
+        # No git info available, just use the base version
+        version = base_version
+
     sep = "+" if "+" not in version else "."  # dev versions might contain +
 
     if _no_device():

tests/kernels/test_semi_structured.py

@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: Apache-2.0
 """Tests for sparse cutlass kernels
 
 Run `pytest tests/kernels/test_semi_structured.py`.

vllm/attention/ops/paged_attn_ater.py

@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: Apache-2.0
 from dataclasses import dataclass
 from typing import List, Optional, Tuple
 

vllm/entrypoints/fast_sync_llm.py

@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: Apache-2.0
 import multiprocessing as mp
 from queue import Empty
 from typing import Union
@@ -85,8 +86,8 @@ def run_engine(self):
         try:
             while True:
                 poll_interval -= 1
-                if (self.input_queue.qsize() >=
-                        envs.VLLM_SYNC_SERVER_ACCUM_REQUESTS
+                if (self.input_queue.qsize()
+                        >= envs.VLLM_SYNC_SERVER_ACCUM_REQUESTS
                         or poll_interval <= 0
                         or not self.llm_engine.has_unfinished_requests()):
                     self._poll_requests()

vllm/entrypoints/sync_openai/api_server.py

@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: Apache-2.0
 import asyncio
 import multiprocessing
 import re

vllm/model_executor/model_loader/loader.py

@@ -1063,10 +1063,6 @@ def _load_weights(self, model_config: ModelConfig,
         if hf_to_vllm_mapper := getattr(model, "hf_to_vllm_mapper", None):
             self.weight_mapper = lambda name: hf_to_vllm_mapper._map_name(name)
 
-        # For some models like Molmo, we need to use hf_to_vllm_mapper
-        # to ensure correct loading of weights.
-        if hf_to_vllm_mapper := getattr(model, "hf_to_vllm_mapper", None):
-            self.weight_mapper = lambda name: hf_to_vllm_mapper._map_name(name)
         # Modules whose weights might have fused on disk
         # we need their output_sizes to make shard in flight correctly with TP
         self.maybe_fused_weights_modules: Dict[str, List[int]] = {}