Sync from v0.13

vllm/model_executor/models/orion.py

@@ -1,55 +1,78 @@
-# coding=utf-8
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
 # Adapted from
 # https://huggingface.co/OrionStarAI/Orion-14B-Base/blob/main/modeling_orion.py
-# Copyright (c) 2024 - 2024 Moore Threads Technology Co., Ltd("Moore Threads"). All rights reserved.
 # Copyright (c) OrionStar Inc.
 # LICENSE: https://huggingface.co/OrionStarAI/Orion-14B-Base/blob/main/LICENSE
 """Inference-only Orion-14B model compatible with HuggingFace weights."""
-from typing import Any, Dict, Iterable, List, Optional, Tuple
+
+from collections.abc import Iterable
+from itertools import islice
+from typing import Any
 
 import torch
 from torch import nn
 from transformers import PretrainedConfig
 
-from vllm.attention import Attention, AttentionMetadata
-from vllm.distributed import get_tensor_model_parallel_world_size
+from vllm.attention.layer import Attention
+from vllm.compilation.decorators import support_torch_compile
+from vllm.config import CacheConfig, VllmConfig
+from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.activation import SiluAndMul
-from vllm.model_executor.layers.linear import (MergedColumnParallelLinear,
-                                               QKVParallelLinear,
-                                               RowParallelLinear)
+from vllm.model_executor.layers.linear import (
+    MergedColumnParallelLinear,
+    QKVParallelLinear,
+    RowParallelLinear,
+)
 from vllm.model_executor.layers.logits_processor import LogitsProcessor
-from vllm.model_executor.layers.quantization.base_config import (
-    QuantizationConfig)
+from vllm.model_executor.layers.quantization import QuantizationConfig
 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 (
-    ParallelLMHead, VocabParallelEmbedding)
+    ParallelLMHead,
+    VocabParallelEmbedding,
+)
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
-from vllm.model_executor.sampling_metadata import SamplingMetadata
-from vllm.sequence import SamplerOutput
+from vllm.sequence import IntermediateTensors
+
+from .interfaces import SupportsPP
+from .utils import (
+    AutoWeightsLoader,
+    is_pp_missing_parameter,
+    make_empty_intermediate_tensors_factory,
+    make_layers,
+    maybe_prefix,
+)
 
 
 class OrionMLP(nn.Module):
-
     def __init__(
         self,
         hidden_size: int,
         intermediate_size: int,
         hidden_act: str,
-        quant_config: Optional[QuantizationConfig] = None,
+        quant_config: QuantizationConfig | None = None,
+        prefix: str = "",
     ) -> None:
         super().__init__()
         self.gate_up_proj = MergedColumnParallelLinear(
-            hidden_size, [intermediate_size] * 2,
+            hidden_size,
+            [intermediate_size] * 2,
             bias=False,
-            quant_config=quant_config)
-        self.down_proj = RowParallelLinear(intermediate_size,
-                                           hidden_size,
-                                           bias=False,
-                                           quant_config=quant_config)
+            quant_config=quant_config,
+            prefix=f"{prefix}.gate_up_proj",
+        )
+        self.down_proj = RowParallelLinear(
+            intermediate_size,
+            hidden_size,
+            bias=False,
+            quant_config=quant_config,
+            prefix=f"{prefix}.down_proj",
+        )
         if hidden_act != "silu":
-            raise ValueError(f"Unsupported activation: {hidden_act}. "
-                             "Only silu is supported for now.")
+            raise ValueError(
+                f"Unsupported activation: {hidden_act}. Only silu is supported for now."
+            )
         self.act_fn = SiluAndMul()
 
     def forward(self, x):
@@ -60,16 +83,16 @@ class OrionMLP(nn.Module):
 
 
 class OrionAttention(nn.Module):
-
     def __init__(
         self,
         hidden_size: int,
         num_heads: int,
         num_kv_heads: int,
-        rope_theta: float = 10000,
-        rope_scaling: Optional[Dict[str, Any]] = None,
+        rope_parameters: dict[str, Any] | None = None,
         max_position_embeddings: int = 8192,
-        quant_config: Optional[QuantizationConfig] = None,
+        cache_config: CacheConfig | None = None,
+        quant_config: QuantizationConfig | None = None,
+        prefix: str = "",
     ) -> None:
         super().__init__()
         self.hidden_size = hidden_size
@@ -91,7 +114,6 @@ class OrionAttention(nn.Module):
         self.q_size = self.num_heads * self.head_dim
         self.kv_size = self.num_kv_heads * self.head_dim
         self.scaling = self.head_dim**-0.5
-        self.rope_theta = rope_theta
         self.max_position_embeddings = max_position_embeddings
 
         self.qkv_proj = QKVParallelLinear(
@@ -101,91 +123,89 @@ class OrionAttention(nn.Module):
             self.total_num_kv_heads,
             bias=False,
             quant_config=quant_config,
+            prefix=f"{prefix}.qkv_proj",
         )
         self.o_proj = RowParallelLinear(
             self.total_num_heads * self.head_dim,
             hidden_size,
             bias=False,
             quant_config=quant_config,
+            prefix=f"{prefix}.o_proj",
         )
 
         self.rotary_emb = get_rope(
             self.head_dim,
-            rotary_dim=self.head_dim,
             max_position=max_position_embeddings,
-            base=rope_theta,
-            rope_scaling=rope_scaling,
+            rope_parameters=rope_parameters,
+        )
+        self.attn = Attention(
+            self.num_heads,
+            self.head_dim,
+            self.scaling,
+            num_kv_heads=self.num_kv_heads,
+            cache_config=cache_config,
+            quant_config=quant_config,
+            prefix=f"{prefix}.attn",
         )
-        self.attn = Attention(self.num_heads,
-                              self.head_dim,
-                              self.scaling,
-                              num_kv_heads=self.num_kv_heads)
 
     def forward(
         self,
         positions: torch.Tensor,
         hidden_states: torch.Tensor,
-        kv_cache: torch.Tensor,
-        attn_metadata: AttentionMetadata,
     ) -> torch.Tensor:
         qkv, _ = self.qkv_proj(hidden_states)
         q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
         q, k = self.rotary_emb(positions, q, k)
-        attn_output = self.attn(q, k, v, kv_cache, attn_metadata)
+        attn_output = self.attn(q, k, v)
         output, _ = self.o_proj(attn_output)
         return output
 
 
 class OrionDecoderLayer(nn.Module):
-
     def __init__(
         self,
         config: PretrainedConfig,
-        quant_config: Optional[QuantizationConfig] = None,
+        cache_config: CacheConfig | None = None,
+        quant_config: QuantizationConfig | None = None,
+        prefix: str = "",
     ) -> None:
         super().__init__()
         self.hidden_size = config.hidden_size
-        rope_theta = getattr(config, "rope_theta", 10000)
-        rope_scaling = getattr(config, "rope_scaling", None)
-        max_position_embeddings = getattr(config, "max_position_embeddings",
-                                          8192)
+        max_position_embeddings = getattr(config, "max_position_embeddings", 8192)
         self.self_attn = OrionAttention(
             hidden_size=self.hidden_size,
             num_heads=config.num_attention_heads,
             num_kv_heads=config.num_key_value_heads,
-            rope_theta=rope_theta,
-            rope_scaling=rope_scaling,
+            rope_parameters=config.rope_parameters,
             max_position_embeddings=max_position_embeddings,
+            cache_config=cache_config,
             quant_config=quant_config,
+            prefix=f"{prefix}.self_attn",
         )
         self.mlp = OrionMLP(
             hidden_size=self.hidden_size,
             intermediate_size=config.intermediate_size,
             hidden_act=config.hidden_act,
             quant_config=quant_config,
+            prefix=f"{prefix}.mlp",
         )
 
-        self.input_layernorm = nn.LayerNorm(config.hidden_size,
-                                            eps=config.rms_norm_eps)
-        self.post_attention_layernorm = nn.LayerNorm(config.hidden_size,
-                                                     eps=config.rms_norm_eps)
+        self.input_layernorm = nn.LayerNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = nn.LayerNorm(
+            config.hidden_size, eps=config.rms_norm_eps
+        )
 
     def forward(
         self,
         positions: torch.Tensor,
         hidden_states: torch.Tensor,
-        kv_cache: torch.Tensor,
-        attn_metadata: AttentionMetadata,
-        residual: Optional[torch.Tensor],
-    ) -> Tuple[torch.Tensor, torch.Tensor]:
+    ) -> tuple[torch.Tensor, torch.Tensor]:
         # Self Attention
         residual = hidden_states
         hidden_states = self.input_layernorm(hidden_states)
         hidden_states = self.self_attn(
             positions=positions,
             hidden_states=hidden_states,
-            kv_cache=kv_cache,
-            attn_metadata=attn_metadata,
         )
 
         hidden_states = residual + hidden_states
@@ -195,93 +215,69 @@ class OrionDecoderLayer(nn.Module):
         hidden_states = self.post_attention_layernorm(hidden_states)
         hidden_states = self.mlp(hidden_states)
         hidden_states = residual + hidden_states
-        return hidden_states, None
+        return hidden_states
 
 
+@support_torch_compile
 class OrionModel(nn.Module):
-
-    def __init__(
-        self,
-        config: PretrainedConfig,
-        quant_config: Optional[QuantizationConfig] = None,
-    ) -> None:
+    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         super().__init__()
+
+        config = vllm_config.model_config.hf_config
+        cache_config = vllm_config.cache_config
+        quant_config = vllm_config.quant_config
+
         self.config = config
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
         self.embed_tokens = VocabParallelEmbedding(
             config.vocab_size,
             config.hidden_size,
         )
-        self.layers = nn.ModuleList([
-            OrionDecoderLayer(config, quant_config)
-            for _ in range(config.num_hidden_layers)
-        ])
+        self.start_layer, self.end_layer, self.layers = make_layers(
+            config.num_hidden_layers,
+            lambda prefix: OrionDecoderLayer(
+                config, cache_config, quant_config, prefix=prefix
+            ),
+            prefix=f"{prefix}.layers",
+        )
         self.norm = nn.LayerNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.make_empty_intermediate_tensors = make_empty_intermediate_tensors_factory(
+            [
+                "hidden_states",
+            ],
+            config.hidden_size,
+        )
+
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.embed_tokens(input_ids)
 
     def forward(
         self,
         input_ids: torch.Tensor,
         positions: torch.Tensor,
-        kv_caches: List[torch.Tensor],
-        attn_metadata: AttentionMetadata,
-    ) -> torch.Tensor:
-        hidden_states = self.embed_tokens(input_ids)
-        residual = None
-        for i in range(len(self.layers)):
-            layer = self.layers[i]
-            hidden_states, residual = layer(
-                positions,
-                hidden_states,
-                kv_caches[i],
-                attn_metadata,
-                residual,
+        intermediate_tensors: IntermediateTensors | None,
+        inputs_embeds: torch.Tensor | None = None,
+    ) -> torch.Tensor | IntermediateTensors:
+        if get_pp_group().is_first_rank:
+            if inputs_embeds is not None:
+                hidden_states = inputs_embeds
+            else:
+                hidden_states = self.embed_input_ids(input_ids)
+        else:
+            assert intermediate_tensors is not None
+            hidden_states = intermediate_tensors["hidden_states"]
+        for layer in islice(self.layers, self.start_layer, self.end_layer):
+            hidden_states = layer(positions, hidden_states)
+        if not get_pp_group().is_last_rank:
+            return IntermediateTensors(
+                {
+                    "hidden_states": hidden_states,
+                }
             )
         hidden_states = self.norm(hidden_states)
         return hidden_states
 
-
-class OrionForCausalLM(nn.Module):
-
-    def __init__(
-        self,
-        config: PretrainedConfig,
-        quant_config: Optional[QuantizationConfig] = None,
-    ) -> None:
-        super().__init__()
-        self.config = config
-        self.quant_config = quant_config
-        self.model = OrionModel(config, quant_config)
-        self.lm_head = ParallelLMHead(config.vocab_size, config.hidden_size)
-        self.logits_processor = LogitsProcessor(config.vocab_size)
-        self.sampler = Sampler()
-
-    def forward(
-        self,
-        input_ids: torch.Tensor,
-        positions: torch.Tensor,
-        kv_caches: List[torch.Tensor],
-        attn_metadata: AttentionMetadata,
-    ) -> torch.Tensor:
-        hidden_states = self.model(input_ids, positions, kv_caches,
-                                   attn_metadata)
-        return hidden_states
-
-    def compute_logits(self, hidden_states: torch.Tensor,
-                       sampling_metadata: SamplingMetadata) -> torch.Tensor:
-        logits = self.logits_processor(self.lm_head.weight, hidden_states,
-                                       sampling_metadata)
-        return logits
-
-    def sample(
-        self,
-        logits: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
-    ) -> Optional[SamplerOutput]:
-        next_tokens = self.sampler(logits, sampling_metadata)
-        return next_tokens
-
-    def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
+    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
         stacked_params_mapping = [
             # (param_name, shard_name, shard_id)
             ("qkv_proj", "q_proj", "q"),
@@ -291,21 +287,17 @@ class OrionForCausalLM(nn.Module):
             ("gate_up_proj", "up_proj", 1),
         ]
         params_dict = dict(self.named_parameters())
+        loaded_params: set[str] = set()
         for name, loaded_weight in weights:
-            if "rotary_emb.inv_freq" in name:
-                continue
-            if ("rotary_emb.cos_cached" in name
-                    or "rotary_emb.sin_cached" in name):
-                # Models trained using ColossalAI may include these tensors in
-                # the checkpoint. Skip them.
-                continue
-            for (param_name, weight_name, shard_id) in stacked_params_mapping:
+            for param_name, weight_name, shard_id in stacked_params_mapping:
                 if weight_name not in name:
                     continue
                 name = name.replace(weight_name, param_name)
                 # Skip loading extra bias for GPTQ models.
                 if name.endswith(".bias") and name not in params_dict:
                     continue
+                if is_pp_missing_parameter(name, self):
+                    continue
                 param = params_dict[name]
                 weight_loader = param.weight_loader
                 weight_loader(param, loaded_weight, shard_id)
@@ -314,7 +306,60 @@ class OrionForCausalLM(nn.Module):
                 # Skip loading extra bias for GPTQ models.
                 if name.endswith(".bias") and name not in params_dict:
                     continue
+                if is_pp_missing_parameter(name, self):
+                    continue
                 param = params_dict[name]
-                weight_loader = getattr(param, "weight_loader",
-                                        default_weight_loader)
+                weight_loader = getattr(param, "weight_loader", default_weight_loader)
                 weight_loader(param, loaded_weight)
+            loaded_params.add(name)
+        return loaded_params
+
+
+class OrionForCausalLM(nn.Module, SupportsPP):
+    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
+        super().__init__()
+        config = vllm_config.model_config.hf_config
+        quant_config = vllm_config.quant_config
+        self.config = config
+        self.quant_config = quant_config
+        self.model = OrionModel(
+            vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model")
+        )
+        self.lm_head = ParallelLMHead(
+            config.vocab_size,
+            config.hidden_size,
+            quant_config=quant_config,
+            prefix=maybe_prefix(prefix, "lm_head"),
+        )
+        if self.config.tie_word_embeddings:
+            self.lm_head.weight = self.model.embed_tokens.weight
+        self.logits_processor = LogitsProcessor(config.vocab_size)
+        self.make_empty_intermediate_tensors = (
+            self.model.make_empty_intermediate_tensors
+        )
+
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
+
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        positions: torch.Tensor,
+        intermediate_tensors: IntermediateTensors | None = None,
+        inputs_embeds: torch.Tensor | None = None,
+    ) -> torch.Tensor | IntermediateTensors:
+        hidden_states = self.model(
+            input_ids, positions, intermediate_tensors, inputs_embeds
+        )
+        return hidden_states
+
+    def compute_logits(
+        self,
+        hidden_states: torch.Tensor,
+    ) -> torch.Tensor | None:
+        logits = self.logits_processor(self.lm_head, hidden_states)
+        return logits
+
+    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
+        loader = AutoWeightsLoader(self)
+        return loader.load_weights(weights)