[None][feat] Address review: remove bundled config, document inline refs

Remove the bundled InternLM3Config from the modeling file. The AD
pipeline loads the config from the HF checkpoint via trust_remote_code=True
(same pattern as DeciLM).

The test file now loads InternLM3Config dynamically from the HF cache.
Inline HF reference classes are kept because the HF modeling_internlm3.py
cannot be imported on the installed transformers version (requires
LossKwargs from transformers >=4.48).

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/modeling_internlm3.py

@@ -19,14 +19,15 @@
 https://huggingface.co/internlm/internlm3-8b-instruct
 
 This implementation differs from the original HuggingFace version in the following ways:
-* Bundled config class to work without trust_remote_code (model not in transformers)
 * Simplified for prefill-only inference (no KV caching)
 * Uses auto_deploy custom ops for export compatibility
 * Removed flash attention variants (uses torch_attention custom op)
 * Removed gradient checkpointing and training code paths
 * Removed attention dropout (inference only)
 * No repeat_kv — AD attention ops handle GQA natively
 
+Config is loaded from the HF checkpoint via trust_remote_code=True (not bundled here).
+
 The InternLM3 model uses GQA with SwiGLU MLP, RMSNorm, and dynamic NTK-scaled RoPE.
 """
 
@@ -35,7 +36,6 @@
 
 import torch
 from torch import nn
-from transformers import AutoConfig, PretrainedConfig
 from transformers.activations import ACT2FN
 from transformers.generation import GenerationMixin
 from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS
@@ -45,79 +45,6 @@
 from tensorrt_llm._torch.auto_deploy.models.hf import AutoModelForCausalLMFactory
 
 
-class InternLM3Config(PretrainedConfig):
-    """Configuration class for InternLM3 model.
-
-    Bundled with the custom model implementation since InternLM3 is not natively
-    registered in transformers (requires trust_remote_code).
-    """
-
-    model_type = "internlm3"
-
-    def __init__(
-        self,
-        vocab_size: int = 128512,
-        hidden_size: int = 4096,
-        intermediate_size: int = 11008,
-        num_hidden_layers: int = 32,
-        num_attention_heads: int = 32,
-        num_key_value_heads: int = 32,
-        hidden_act: str = "silu",
-        max_position_embeddings: int = 32768,
-        initializer_range: float = 0.02,
-        rms_norm_eps: float = 1e-6,
-        tie_word_embeddings: bool = False,
-        rope_theta: float = 10000.0,
-        rope_scaling: Optional[dict] = None,
-        qkv_bias: bool = False,
-        attention_dropout: float = 0.0,
-        bias: bool = False,
-        head_dim: Optional[int] = None,
-        pad_token_id: Optional[int] = None,
-        **kwargs,
-    ):
-        self.vocab_size = vocab_size
-        self.hidden_size = hidden_size
-        self.intermediate_size = intermediate_size
-        self.num_hidden_layers = num_hidden_layers
-        self.num_attention_heads = num_attention_heads
-        self.num_key_value_heads = (
-            num_key_value_heads if num_key_value_heads is not None else num_attention_heads
-        )
-        self.hidden_act = hidden_act
-        self.max_position_embeddings = max_position_embeddings
-        self.initializer_range = initializer_range
-        self.rms_norm_eps = rms_norm_eps
-        self.rope_theta = rope_theta
-        self.rope_scaling = rope_scaling
-        self.qkv_bias = qkv_bias
-        self.attention_dropout = attention_dropout
-        self.bias = bias
-        self.head_dim = (
-            head_dim if head_dim is not None else self.hidden_size // self.num_attention_heads
-        )
-
-        # Normalize rope_scaling type field
-        if self.rope_scaling is not None and "type" in self.rope_scaling:
-            self.rope_scaling["rope_type"] = self.rope_scaling["type"]
-
-        super().__init__(
-            pad_token_id=pad_token_id,
-            tie_word_embeddings=tie_word_embeddings,
-            **kwargs,
-        )
-
-
-# Register config with AutoConfig so it can be loaded from HF hub
-try:
-    AutoConfig.register("internlm3", InternLM3Config, exist_ok=True)
-except TypeError:
-    try:
-        AutoConfig.register("internlm3", InternLM3Config)
-    except ValueError:
-        pass
-
-
 class InternLM3RMSNorm(nn.Module):
     """RMS Normalization using AutoDeploy torch_rmsnorm reference op."""
 
@@ -310,7 +237,6 @@ class InternLM3CausalLMOutput(ModelOutput):
 class InternLM3PreTrainedModel(PreTrainedModel):
     """Base class for InternLM3 models."""
 
-    config_class = InternLM3Config
     base_model_prefix = "model"
     _no_split_modules = ["InternLM3DecoderLayer"]
     supports_gradient_checkpointing = False

tests/unittest/auto_deploy/singlegpu/models/test_internlm3_modeling.py

@@ -20,9 +20,11 @@
 for export compatibility. InternLM3 uses GQA with SwiGLU MLP, RMSNorm,
 and dynamic NTK-scaled RoPE.
 
-Since InternLM3 is not natively in the installed transformers (requires
-trust_remote_code), HF reference classes are defined inline for equivalence
-testing.
+HF reference classes are defined inline for equivalence testing because the
+HF modeling_internlm3.py (from the HF checkpoint) cannot be imported on the
+installed transformers version — it requires ``LossKwargs`` from
+``transformers.utils`` which is only available in transformers >=4.48.
+The config class *can* be loaded via AutoConfig with trust_remote_code.
 """
 
 import math
@@ -33,13 +35,13 @@
 from _model_test_utils import assert_rmse_close
 from torch import nn
 from torch.export import Dim
+from transformers import AutoConfig
 from transformers.activations import ACT2FN
 from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS
 
 from tensorrt_llm._torch.auto_deploy.export import torch_export_to_gm
 from tensorrt_llm._torch.auto_deploy.models.custom.modeling_internlm3 import (
     InternLM3Attention,
-    InternLM3Config,
     InternLM3DecoderLayer,
     InternLM3ForCausalLM,
     InternLM3MLP,
@@ -50,14 +52,26 @@
 
 _BATCH_AND_SEQUENCE_TEST_CASES = ((2, 6), (1, 8))
 
+# Load InternLM3Config from the HF checkpoint cache (trust_remote_code).
+# The config class is not in the installed transformers but is bundled with the HF repo.
+try:
+    _hf_config = AutoConfig.from_pretrained(
+        "internlm/internlm3-8b-instruct", trust_remote_code=True
+    )
+    InternLM3Config = type(_hf_config)
+except Exception:
+    InternLM3Config = None
+
 
 @pytest.fixture(scope="function", autouse=True)
 def set_seed():
     torch.manual_seed(42)
 
 
-def _create_small_config() -> InternLM3Config:
+def _create_small_config():
     """Create a small InternLM3 config for testing."""
+    if InternLM3Config is None:
+        pytest.skip("InternLM3Config not available (internlm/internlm3-8b-instruct not cached)")
     return InternLM3Config(
         vocab_size=1000,
         hidden_size=64,