[Mypy] Better fixes for the mypy issues in vllm/config (#37902)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

examples/offline_inference/audio_language.py

@@ -9,7 +9,6 @@ on HuggingFace model repository.
 """
 
 import os
-from dataclasses import asdict
 from typing import Any, NamedTuple
 
 from huggingface_hub import snapshot_download
@@ -633,7 +632,7 @@ def main(args):
         req_data.engine_args.limit_mm_per_prompt or {}
     )
 
-    engine_args = asdict(req_data.engine_args) | {"seed": args.seed}
+    engine_args = vars(req_data.engine_args) | {"seed": args.seed}
     if args.tensor_parallel_size is not None:
         engine_args["tensor_parallel_size"] = args.tensor_parallel_size
     llm = LLM(**engine_args)

examples/offline_inference/encoder_decoder_multimodal.py

@@ -8,7 +8,6 @@ the explicit/implicit prompt format on enc-dec LMMs for text generation.
 import os
 import time
 from collections.abc import Sequence
-from dataclasses import asdict
 from typing import NamedTuple
 
 from vllm import LLM, EngineArgs, PromptType, SamplingParams
@@ -91,13 +90,12 @@ def main(args):
     req_data = model_example_map[model]()
 
     # Disable other modalities to save memory
+    engine_args = req_data.engine_args
     default_limits = {"image": 0, "video": 0, "audio": 0}
-    req_data.engine_args.limit_mm_per_prompt = default_limits | dict(
-        req_data.engine_args.limit_mm_per_prompt or {}
-    )
-
-    engine_args = asdict(req_data.engine_args) | {"seed": args.seed}
-    llm = LLM(**engine_args)
+    limit_mm_per_prompt = default_limits | (engine_args.limit_mm_per_prompt or {})
+    engine_args.limit_mm_per_prompt = limit_mm_per_prompt
+    engine_args.seed = args.seed
+    llm = LLM.from_engine_args(engine_args)
 
     prompts = req_data.prompts
 

examples/offline_inference/load_sharded_state.py

@@ -20,8 +20,6 @@ python load_sharded_state.py \
     --max-tokens 50
 """
 
-import dataclasses
-
 from vllm import LLM, EngineArgs, SamplingParams
 from vllm.utils.argparse_utils import FlexibleArgumentParser
 
@@ -64,7 +62,7 @@ def main():
     print(f"Tensor parallel size: {engine_args.tensor_parallel_size}")
 
     # Load the model using engine args
-    llm = LLM(**dataclasses.asdict(engine_args))
+    llm = LLM.from_engine_args(engine_args)
 
     # Prepare sampling parameters
     sampling_params = SamplingParams(

examples/offline_inference/save_sharded_state.py

@@ -21,7 +21,6 @@ llm = LLM(
 )
 """
 
-import dataclasses
 import os
 import shutil
 from pathlib import Path
@@ -60,7 +59,7 @@ def main(args):
     if not Path(model_path).is_dir():
         raise ValueError("model path must be a local directory")
     # Create LLM instance from arguments
-    llm = LLM(**dataclasses.asdict(engine_args))
+    llm = LLM.from_engine_args(engine_args)
     # Prepare output directory
     Path(args.output).mkdir(exist_ok=True)
     # Dump worker states to output directory

examples/offline_inference/vision_language.py

@@ -11,7 +11,6 @@ on HuggingFace model repository.
 import os
 import random
 from contextlib import contextmanager
-from dataclasses import asdict
 from typing import NamedTuple
 
 from huggingface_hub import snapshot_download
@@ -2434,13 +2433,13 @@ def main(args):
         req_data.engine_args.limit_mm_per_prompt or {}
     )
 
-    engine_args = asdict(req_data.engine_args) | {
-        "seed": args.seed,
-        "mm_processor_cache_gb": 0 if args.disable_mm_processor_cache else 4,
-    }
+    engine_args = req_data.engine_args
+    engine_args.seed = args.seed
+    mm_processor_cache_gb = 0 if args.disable_mm_processor_cache else 4
+    engine_args.mm_processor_cache_gb = mm_processor_cache_gb
     if args.tensor_parallel_size is not None:
-        engine_args["tensor_parallel_size"] = args.tensor_parallel_size
-    llm = LLM(**engine_args)
+        engine_args.tensor_parallel_size = args.tensor_parallel_size
+    llm = LLM.from_engine_args(engine_args)
 
     # Don't want to check the flag multiple times, so just hijack `prompts`.
     prompts = (

examples/offline_inference/vision_language_multi_image.py

@@ -8,7 +8,6 @@ using the chat template defined by the model.
 
 import os
 from argparse import Namespace
-from dataclasses import asdict
 from typing import NamedTuple
 
 from huggingface_hub import snapshot_download
@@ -1481,10 +1480,11 @@ def run_generate(
 ):
     req_data = model_example_map[model](question, image_urls)
 
-    engine_args = asdict(req_data.engine_args) | {"seed": seed}
+    engine_args = req_data.engine_args
+    engine_args.seed = seed
     if tensor_parallel_size is not None:
-        engine_args["tensor_parallel_size"] = tensor_parallel_size
-    llm = LLM(**engine_args)
+        engine_args.tensor_parallel_size = tensor_parallel_size
+    llm = LLM.from_engine_args(engine_args)
 
     sampling_params = SamplingParams(
         temperature=0.0, max_tokens=256, stop_token_ids=req_data.stop_token_ids
@@ -1521,10 +1521,11 @@ def run_chat(
         req_data.engine_args.limit_mm_per_prompt or {}
     )
 
-    engine_args = asdict(req_data.engine_args) | {"seed": seed}
+    engine_args = req_data.engine_args
+    engine_args.seed = seed
     if tensor_parallel_size is not None:
-        engine_args["tensor_parallel_size"] = tensor_parallel_size
-    llm = LLM(**engine_args)
+        engine_args.tensor_parallel_size = tensor_parallel_size
+    llm = LLM.from_engine_args(engine_args)
 
     sampling_params = (
         SamplingParams(