[DEPLOY] Complete submission: baseline + all optimizations

Adds ALL files needed for Dockerfile build:
  - qwen3_6_scripts/ (baseline patches + our optimizations)
  - vllm/ (full vllm package)
  - paged_attention_v2_pytorch.py (V2 with single-bmm optimization)
  - Dockerfile + computility-run.yaml

Our optimizations vs baseline:
  1. paged_attn.py: pre-gathered context KV (eliminates 194 gather calls),
     Triton try/fallback, V2 heuristic, threshold 32K→64K
  2. paged_attention_v2_pytorch.py: fills NotImplementedError,
     single-bmm Phase 1 (195 launches → 3)
  3. patch_enable_triton.py: HAS_TRITON=True with safety fallback
  4. patch_triton_tuning.py: BLOCK=64, NUM_WARPS=4 for BI-V100
  5. computility-run.yaml: gpu-memory-utilization 0.9→0.95,
     max-num-batched-tokens 8192→16384

This repo can now be submitted to dev.modelhub.org.cn as-is.

vllm/model_executor/model_loader/utils.py

@@ -0,0 +1,39 @@
+"""Utilities for selecting and loading models."""
+import contextlib
+from typing import Tuple, Type
+
+import torch
+from torch import nn
+
+from vllm.config import ModelConfig
+from vllm.model_executor.models import ModelRegistry
+
+
+@contextlib.contextmanager
+def set_default_torch_dtype(dtype: torch.dtype):
+    """Sets the default torch dtype to the given dtype."""
+    old_dtype = torch.get_default_dtype()
+    torch.set_default_dtype(dtype)
+    yield
+    torch.set_default_dtype(old_dtype)
+
+
+def get_model_architecture(
+        model_config: ModelConfig) -> Tuple[Type[nn.Module], str]:
+    architectures = getattr(model_config.hf_config, "architectures", [])
+    # Special handling for quantized Mixtral.
+    # FIXME(woosuk): This is a temporary hack.
+    mixtral_supported = [
+        "fp8", "compressed-tensors", "gptq_marlin", "awq_marlin"
+    ]
+
+    if (model_config.quantization is not None
+            and model_config.quantization not in mixtral_supported
+            and "MixtralForCausalLM" in architectures):
+        architectures = ["QuantMixtralForCausalLM"]
+
+    return ModelRegistry.resolve_model_cls(architectures)
+
+
+def get_architecture_class_name(model_config: ModelConfig) -> str:
+    return get_model_architecture(model_config)[1]