[Bug] assert pad >= 0 failed when inter_size is not a multiple of group_size

[Vinkle-hzt]

Checklist

• 1. I have searched related issues but cannot get the expected help.
• 2. The bug has not been fixed in the latest version.
• 3. Please note that if the bug-related issue you submitted lacks corresponding environment info and a minimal reproducible demo, it will be challenging for us to reproduce and resolve the issue, reducing the likelihood of receiving feedback.

Describe the bug

When inter_size=4505, the padding size will be 4480.

# https://github.com/InternLM/lmdeploy/blob/main/lmdeploy/turbomind/deploy/module.py#L47
def pad_out_dims(x: torch.Tensor, dims: int):
    pad = dims - x.size(-1)
    assert pad >= 0 # ERROR: dim=4480, x.size(-1)=4505
    return torch.nn.functional.pad(x, (0, pad), 'constant', 0)

Reproduction

Sorry for not providing the specific model that led to the issue, as I am currently utilizing a non-publicly available pruned model.
Below is the configuration that triggers this bug:

{
  "aligner_config": {
    "cls": "MlpProjector",
    "model_type": "aligner",
    "params": {
      "depth": 2,
      "input_dim": 1024,
      "n_embed": 2048,
      "projector_type": "mlp_gelu"
    }
  },
  "architectures": [
    "MultiModalityCausalLM"
  ],
  "language_config": {
    "hidden_size": 2048,
    "intermediate_size": 4505,
    "max_position_embeddings": 16384,
    "model_type": "llama",
    "num_attention_heads": 16,
    "num_hidden_layers": 24,
    "num_key_value_heads": 16,
    "torch_dtype": "float16",
    "vocab_size": 102400
  },
  "model_type": "multi_modality",
  "torch_dtype": "float16",
  "transformers_version": "4.38.2",
  "vision_config": {
    "cls": "CLIPVisionTower",
    "model_type": "vision",
    "params": {
      "image_size": 384,
      "model_name": "siglip_large_patch16_384",
      "pixel_mean": [
        0.5,
        0.5,
        0.5
      ],
      "pixel_std": [
        0.5,
        0.5,
        0.5
      ],
      "select_feature": "same",
      "select_layer": -1
    }
  }
}

Environment

model: a pruned deepseek-vl-1.3B model with an "intermediate_size" of 4505

lmdeploy version:
lmdeploy==0.6.2.post1

Error traceback

No response