[Core] Refactor model loading code (#4097)

vllm/model_executor/models/llama.py

@@ -21,7 +21,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Inference-only LLaMA model compatible with HuggingFace weights."""
-from typing import Any, Dict, List, Optional, Tuple
+from typing import Any, Dict, Iterable, List, Optional, Tuple
 
 import torch
 from torch import nn
@@ -42,10 +42,9 @@ from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.sampler import Sampler
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     DEFAULT_VOCAB_PADDING_SIZE, ParallelLMHead, VocabParallelEmbedding)
+from vllm.model_executor.model_loader.weight_utils import (
+    default_weight_loader, kv_cache_scales_loader)
 from vllm.model_executor.sampling_metadata import SamplingMetadata
-from vllm.model_executor.weight_utils import (default_weight_loader,
-                                              hf_model_weights_iterator,
-                                              kv_cache_scales_loader)
 from vllm.sequence import SamplerOutput
 from vllm.utils import is_hip
 
@@ -376,11 +375,7 @@ class LlamaForCausalLM(nn.Module):
         next_tokens = self.sampler(logits, sampling_metadata)
         return next_tokens
 
-    def load_weights(self,
-                     model_name_or_path: str,
-                     cache_dir: Optional[str] = None,
-                     load_format: str = "auto",
-                     revision: Optional[str] = None):
+    def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
         stacked_params_mapping = [
             # (param_name, shard_name, shard_id)
             ("qkv_proj", "q_proj", "q"),
@@ -390,8 +385,7 @@ class LlamaForCausalLM(nn.Module):
             ("gate_up_proj", "up_proj", 1),
         ]
         params_dict = dict(self.named_parameters())
-        for name, loaded_weight in hf_model_weights_iterator(
-                model_name_or_path, cache_dir, load_format, revision):
+        for name, loaded_weight in weights:
             if "rotary_emb.inv_freq" in name:
                 continue
             if ("rotary_emb.cos_cached" in name