[None][feat] Add AutoDeploy custom model for OpenELM family

Onboard the OpenELM architecture (apple/OpenELM-270M/1_1B/3B-Instruct)
as a custom AutoDeploy model. This is a heterogeneous transformer with:
- Per-layer varying query/KV head counts (GQA)
- Per-layer varying FFN intermediate sizes
- Fused QKV projection with Q/K normalization
- Shared input/output embeddings (no separate lm_head)
- GLU-style FFN (proj_1 = fused gate+up, proj_2 = down)

Uses canonical AD IR ops: torch_rmsnorm, torch_rope_with_explicit_cos_sin,
torch_attention. Verified numerically equivalent to HF reference (100%
top-1 token match, RMSE < 0.05).

Signed-off-by: Lucas Liebenwein <lliebenwein@nvidia.com>
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

Author: lucaslie

tensorrt_llm/_torch/auto_deploy/models/custom/__init__.py

@@ -1,11 +1,12 @@
 from .modeling_decilm import DeciLMForCausalLM
 from .modeling_deepseek import DeepSeekV3ForCausalLM
 from .modeling_glm4_moe_lite import Glm4MoeLiteForCausalLM
-from .modeling_hunyuan_dense_v1 import HunYuanDenseV1ForCausalLM
 from .modeling_granite_moe_hybrid import GraniteMoeHybridForCausalLM
+from .modeling_hunyuan_dense_v1 import HunYuanDenseV1ForCausalLM
 from .modeling_kimi_k2 import KimiK2ForCausalLM, KimiK25ForConditionalGeneration
 from .modeling_nemotron_flash import NemotronFlashForCausalLM, NemotronFlashPreTrainedTokenizerFast
 from .modeling_nemotron_h import NemotronHForCausalLM
+from .modeling_openelm import OpenELMForCausalLM
 from .modeling_qwen3 import Qwen3ForCausalLM
 from .modeling_qwen3_5_moe import Qwen3_5MoeForCausalLM, Qwen3_5MoeForConditionalGeneration
 
@@ -20,6 +21,7 @@
     "NemotronFlashForCausalLM",
     "NemotronFlashPreTrainedTokenizerFast",
     "NemotronHForCausalLM",
+    "OpenELMForCausalLM",
     "Qwen3ForCausalLM",
     "Qwen3_5MoeForCausalLM",
     "Qwen3_5MoeForConditionalGeneration",