vllm.model_executor.models.qwen3_dflash ¶

class DFlashQwen3Attention(nn.Module):
    def __init__(
        self,
        hidden_size: int,
        num_heads: int,
        num_kv_heads: int,
        rope_parameters: dict,
        max_position: int = 4096 * 32,
        head_dim: int | None = None,
        rms_norm_eps: float = 1e-06,
        qkv_bias: bool = False,
        cache_config: CacheConfig | None = None,
        quant_config: QuantizationConfig | None = None,
        prefix: str = "",
        attn_type: str = AttentionType.DECODER,
        dual_chunk_attention_config: dict[str, Any] | None = None,
    ) -> None:
        super().__init__()
        self.hidden_size = hidden_size
        tp_size = get_tensor_model_parallel_world_size()
        self.total_num_heads = num_heads
        assert self.total_num_heads % tp_size == 0
        self.num_heads = self.total_num_heads // tp_size
        self.total_num_kv_heads = num_kv_heads
        if self.total_num_kv_heads >= tp_size:
            assert self.total_num_kv_heads % tp_size == 0
        else:
            assert tp_size % self.total_num_kv_heads == 0
        self.num_kv_heads = max(1, self.total_num_kv_heads // tp_size)
        self.head_dim = head_dim or hidden_size // self.total_num_heads
        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.qkv_proj = QKVParallelLinear(
            hidden_size,
            self.head_dim,
            self.total_num_heads,
            self.total_num_kv_heads,
            bias=qkv_bias,
            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,
            max_position=max_position,
            rope_parameters=rope_parameters,
            dual_chunk_attention_config=dual_chunk_attention_config,
        )
        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",
            attn_type=attn_type,
            **{
                "layer_idx": extract_layer_index(prefix),
                "dual_chunk_attention_config": dual_chunk_attention_config,
            }
            if dual_chunk_attention_config
            else {},
        )
        self.q_norm = RMSNorm(self.head_dim, eps=rms_norm_eps)
        self.k_norm = RMSNorm(self.head_dim, eps=rms_norm_eps)

    def forward(
        self,
        positions: torch.Tensor,
        hidden_states: torch.Tensor,
        context_states: torch.Tensor,
    ) -> torch.Tensor:
        # DFlash attends with query tokens against the context+query KV states.
        num_context_tokens = context_states.shape[0]
        concat_states = torch.cat([context_states, hidden_states], dim=0)

        qkv, _ = self.qkv_proj(concat_states)
        q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)

        q_by_head = q.view(*q.shape[:-1], q.shape[-1] // self.head_dim, self.head_dim)
        k_by_head = k.view(*k.shape[:-1], k.shape[-1] // self.head_dim, self.head_dim)
        q = self.q_norm(q_by_head).view(q.shape)
        k = self.k_norm(k_by_head).view(k.shape)

        q, k = self.rotary_emb(positions, q, k)

        q = q[num_context_tokens:]
        attn_output = self.attn(q, k, v)
        output, _ = self.o_proj(attn_output)
        return output

class DFlashQwen3DecoderLayer(nn.Module):
    def __init__(
        self,
        vllm_config: VllmConfig,
        *,
        config: Qwen3Config,
        cache_config: CacheConfig | None = None,
        quant_config: QuantizationConfig | None = None,
        prefix: str = "",
        layer_idx: int = 0,
    ) -> None:
        del layer_idx
        super().__init__()
        self.hidden_size = config.hidden_size
        set_default_rope_theta(config, default_theta=1000000)
        dual_chunk_attention_config = getattr(
            config, "dual_chunk_attention_config", None
        )

        self.self_attn = DFlashQwen3Attention(
            hidden_size=self.hidden_size,
            num_heads=config.num_attention_heads,
            max_position=config.max_position_embeddings,
            num_kv_heads=config.num_key_value_heads,
            rms_norm_eps=config.rms_norm_eps,
            qkv_bias=getattr(config, "attention_bias", False),
            head_dim=getattr(config, "head_dim", None),
            cache_config=cache_config,
            quant_config=quant_config,
            rope_parameters=config.rope_parameters,
            prefix=f"{prefix}.self_attn",
            attn_type=AttentionType.DECODER,
            dual_chunk_attention_config=dual_chunk_attention_config,
        )
        self.mlp = Qwen3MLP(
            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 = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = RMSNorm(
            config.hidden_size, eps=config.rms_norm_eps
        )

    def forward(
        self,
        positions: torch.Tensor,
        hidden_states: torch.Tensor,
        context_states: torch.Tensor,
        residual: torch.Tensor | None,
    ) -> tuple[torch.Tensor, torch.Tensor]:
        if residual is None:
            residual = hidden_states
            hidden_states = self.input_layernorm(hidden_states)
        else:
            hidden_states, residual = self.input_layernorm(hidden_states, residual)

        hidden_states = self.self_attn(
            positions=positions,
            hidden_states=hidden_states,
            context_states=context_states,
        )

        hidden_states, residual = self.post_attention_layernorm(hidden_states, residual)
        hidden_states = self.mlp(hidden_states)
        return hidden_states, residual

class DFlashQwen3ForCausalLM(Qwen3ForCausalLM):
    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
        nn.Module.__init__(self)
        self.config = vllm_config.speculative_config.draft_model_config.hf_config

        if getattr(self.config, "draft_vocab_size", None) is None:
            self.config.draft_vocab_size = getattr(self.config, "vocab_size", None)

        target_layer_num = vllm_config.model_config.get_num_layers(
            vllm_config.parallel_config
        )
        self.config.target_layer_count = target_layer_num

        self.model = DFlashQwen3Model(
            vllm_config=vllm_config,
            prefix="model",
            start_layer_id=target_layer_num,
        )

        logit_scale = getattr(self.config, "logit_scale", 1.0)
        self.lm_head = ParallelLMHead(
            self.config.draft_vocab_size,
            self.config.hidden_size,
            prefix=maybe_prefix(prefix, "lm_head"),
        )
        self.logits_processor = LogitsProcessor(
            self.config.draft_vocab_size, scale=logit_scale
        )

        self.draft_id_to_target_id = nn.Parameter(
            torch.zeros(self.config.draft_vocab_size, dtype=torch.long),
            requires_grad=False,
        )

    def set_aux_hidden_state_layers(self, layers: tuple[int, ...]) -> None:
        dflash_config = getattr(self.config, "dflash_config", None)
        if not isinstance(dflash_config, dict):
            dflash_config = {}
            self.config.dflash_config = dflash_config
        dflash_config["layer_ids"] = list(layers)

    def get_eagle3_aux_hidden_state_layers(self) -> tuple[int, ...]:
        layer_ids = _resolve_dflash_layer_ids(self.config)
        if layer_ids is not None:
            return layer_ids

        target_layer_count = getattr(self.config, "target_layer_count", None)
        if isinstance(target_layer_count, int):
            return _default_aux_layer_ids(target_layer_count)

        return super().get_eagle3_aux_hidden_state_layers()

    def embed_input_ids(
        self,
        input_ids: torch.Tensor,
        multimodal_embeddings: NestedTensors | None = None,
        is_multimodal: torch.Tensor | None = None,
    ) -> torch.Tensor:
        del multimodal_embeddings, is_multimodal
        return self.model.embed_input_ids(input_ids)

    def forward(
        self,
        input_ids: torch.Tensor,
        positions: torch.Tensor,
        hidden_states: torch.Tensor,
        inputs_embeds: torch.Tensor | None = None,
    ) -> tuple[torch.Tensor, torch.Tensor]:
        return self.model(input_ids, positions, hidden_states, inputs_embeds)

    def compute_logits(
        self,
        hidden_states: torch.Tensor,
    ) -> torch.Tensor | None:
        logits = self.logits_processor(self.lm_head, hidden_states)
        if self.draft_id_to_target_id is None:
            assert logits.shape[1] == self.config.vocab_size, (
                "Expected logits to have shape "
                f"(*, {self.config.vocab_size}), but got {logits.shape}"
            )
            return logits

        base = torch.arange(self.config.draft_vocab_size, device=logits.device)
        targets = base + self.draft_id_to_target_id
        logits_new = logits.new_full(
            (
                logits.shape[0],
                self.config.vocab_size,
            ),
            float("-inf"),
        )
        logits_new[:, targets] = logits
        return logits_new

    def combine_hidden_states(
        self,
        hidden_states: torch.Tensor,
    ) -> torch.Tensor:
        if not self.model.use_aux_hidden_state:
            return hidden_states
        return self.model.hidden_norm(self.model.fc(hidden_states))

    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):
        model_weights = {}
        includes_draft_id_mapping = False
        includes_embed_tokens = False

        for name, loaded_weight in weights:
            if "t2d" in name:
                continue
            if "d2t" in name:
                name = name.replace("d2t", "draft_id_to_target_id")
                includes_draft_id_mapping = True
            elif "lm_head" not in name:
                name = "model." + name

            if "embed_tokens" in name:
                includes_embed_tokens = True

            model_weights[name] = loaded_weight
            process_eagle_weight(self, name)

        skip_substrs = []
        if not includes_draft_id_mapping:
            skip_substrs.append("draft_id_to_target_id")
        if not includes_embed_tokens:
            skip_substrs.append("embed_tokens")
        if not self.model.use_aux_hidden_state:
            skip_substrs.append("fc.")

        loader = AutoWeightsLoader(
            self,
            skip_prefixes=None,
            skip_substrs=skip_substrs,
        )
        loader.load_weights(model_weights.items())