modeling_qwen3_guard.py

# coding=utf-8
# Copyright 2025 The Qwen team, Alibaba Group and the HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from typing import Callable, Optional, Union, Tuple, Generator, List, Dict

import torch
from torch import nn
import torch.nn.functional as F
from transformers.activations import ACT2FN
from transformers.cache_utils import Cache, DynamicCache
from transformers.generation import GenerationMixin
from transformers.integrations import use_kernel_forward_from_hub
from transformers.masking_utils import create_causal_mask, create_sliding_window_causal_mask
from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
from transformers.modeling_layers import (
    GenericForQuestionAnswering,
    GenericForSequenceClassification,
    GenericForTokenClassification,
    GradientCheckpointingLayer,
)
from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
from transformers.processing_utils import Unpack
from transformers.utils import TransformersKwargs, auto_docstring, can_return_tuple
from transformers.utils.deprecation import deprecate_kwarg
from transformers.utils.generic import check_model_inputs
from .configuration_qwen3 import Qwen3Config

from dataclasses import dataclass, field

@dataclass
class GuardLogitsOutputWithPast:
    risk_level_logits: torch.FloatTensor = None
    category_logits: torch.FloatTensor = None
    query_risk_level_logits: torch.FloatTensor = None
    query_category_logits: torch.FloatTensor = None
    loss: Optional[torch.FloatTensor] = None
    past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None
    hidden_states: Optional[Tuple[torch.FloatTensor, ...]] = None
    attentions: Optional[Tuple[torch.FloatTensor, ...]] = None


@use_kernel_forward_from_hub("RMSNorm")
class Qwen3RMSNorm(nn.Module):
    def __init__(self, hidden_size, eps: float = 1e-6) -> None:
        """
        Qwen3RMSNorm is equivalent to T5LayerNorm
        """
        super().__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.variance_epsilon = eps

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        input_dtype = hidden_states.dtype
        hidden_states = hidden_states.to(torch.float32)
        variance = hidden_states.pow(2).mean(-1, keepdim=True)
        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
        return self.weight * hidden_states.to(input_dtype)

    def extra_repr(self):
        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"


class Qwen3MLP(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.hidden_size = config.hidden_size
        self.intermediate_size = config.intermediate_size
        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
        self.act_fn = ACT2FN[config.hidden_act]

    def forward(self, x):
        down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
        return down_proj


def rotate_half(x):
    """Rotates half the hidden dims of the input."""
    x1 = x[..., : x.shape[-1] // 2]
    x2 = x[..., x.shape[-1] // 2 :]
    return torch.cat((-x2, x1), dim=-1)


def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
    """Applies Rotary Position Embedding to the query and key tensors.

    Args:
        q (`torch.Tensor`): The query tensor.
        k (`torch.Tensor`): The key tensor.
        cos (`torch.Tensor`): The cosine part of the rotary embedding.
        sin (`torch.Tensor`): The sine part of the rotary embedding.
        position_ids (`torch.Tensor`, *optional*):
            Deprecated and unused.
        unsqueeze_dim (`int`, *optional*, defaults to 1):
            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
    Returns:
        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
    """
    cos = cos.unsqueeze(unsqueeze_dim)
    sin = sin.unsqueeze(unsqueeze_dim)
    q_embed = (q * cos) + (rotate_half(q) * sin)
    k_embed = (k * cos) + (rotate_half(k) * sin)
    return q_embed, k_embed


def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
    """
    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
    """
    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
    if n_rep == 1:
        return hidden_states
    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)


def eager_attention_forward(
    module: nn.Module,
    query: torch.Tensor,
    key: torch.Tensor,
    value: torch.Tensor,
    attention_mask: Optional[torch.Tensor],
    scaling: float,
    dropout: float = 0.0,
    **kwargs: Unpack[TransformersKwargs],
):
    key_states = repeat_kv(key, module.num_key_value_groups)
    value_states = repeat_kv(value, module.num_key_value_groups)

    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
    if attention_mask is not None:
        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
        attn_weights = attn_weights + causal_mask

    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
    attn_output = torch.matmul(attn_weights, value_states)
    attn_output = attn_output.transpose(1, 2).contiguous()

    return attn_output, attn_weights


class Qwen3Attention(nn.Module):
    """Multi-headed attention from 'Attention Is All You Need' paper"""

    def __init__(self, config: Qwen3Config, layer_idx: int):
        super().__init__()
        self.config = config
        self.layer_idx = layer_idx
        self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
        self.scaling = self.head_dim**-0.5
        self.attention_dropout = config.attention_dropout
        self.is_causal = True

        self.q_proj = nn.Linear(
            config.hidden_size, config.num_attention_heads * self.head_dim, bias=config.attention_bias
        )
        self.k_proj = nn.Linear(
            config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
        )
        self.v_proj = nn.Linear(
            config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
        )
        self.o_proj = nn.Linear(
            config.num_attention_heads * self.head_dim, config.hidden_size, bias=config.attention_bias
        )
        self.q_norm = Qwen3RMSNorm(self.head_dim, eps=config.rms_norm_eps)  # unlike olmo, only on the head dim!
        self.k_norm = Qwen3RMSNorm(self.head_dim, eps=config.rms_norm_eps)  # thus post q_norm does not need reshape
        self.sliding_window = config.sliding_window if config.layer_types[layer_idx] == "sliding_attention" else None

    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
    def forward(
        self,
        hidden_states: torch.Tensor,
        position_embeddings: tuple[torch.Tensor, torch.Tensor],
        attention_mask: Optional[torch.Tensor],
        past_key_values: Optional[Cache] = None,
        cache_position: Optional[torch.LongTensor] = None,
        **kwargs: Unpack[FlashAttentionKwargs],
    ) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
        input_shape = hidden_states.shape[:-1]
        hidden_shape = (*input_shape, -1, self.head_dim)

        query_states = self.q_norm(self.q_proj(hidden_states).view(hidden_shape)).transpose(1, 2)
        key_states = self.k_norm(self.k_proj(hidden_states).view(hidden_shape)).transpose(1, 2)
        value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)

        cos, sin = position_embeddings
        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)

        if past_key_values is not None:
            # sin and cos are specific to RoPE models; cache_position needed for the static cache
            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
            key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx, cache_kwargs)

        attention_interface: Callable = eager_attention_forward
        if self.config._attn_implementation != "eager":
            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]

        attn_output, attn_weights = attention_interface(
            self,
            query_states,
            key_states,
            value_states,
            attention_mask,
            dropout=0.0 if not self.training else self.attention_dropout,
            scaling=self.scaling,
            sliding_window=self.sliding_window,  # diff with Llama
            **kwargs,
        )

        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
        attn_output = self.o_proj(attn_output)
        return attn_output, attn_weights


class Qwen3DecoderLayer(GradientCheckpointingLayer):
    def __init__(self, config: Qwen3Config, layer_idx: int):
        super().__init__()
        self.hidden_size = config.hidden_size

        self.self_attn = Qwen3Attention(config=config, layer_idx=layer_idx)

        self.mlp = Qwen3MLP(config)
        self.input_layernorm = Qwen3RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = Qwen3RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.attention_type = config.layer_types[layer_idx]

    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = None,
        use_cache: Optional[bool] = False,
        cache_position: Optional[torch.LongTensor] = None,
        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,  # necessary, but kept here for BC
        **kwargs: Unpack[TransformersKwargs],
    ) -> torch.Tensor:
        residual = hidden_states
        hidden_states = self.input_layernorm(hidden_states)
        # Self Attention
        hidden_states, _ = self.self_attn(
            hidden_states=hidden_states,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            use_cache=use_cache,
            cache_position=cache_position,
            position_embeddings=position_embeddings,
            **kwargs,
        )
        hidden_states = residual + hidden_states

        # Fully Connected
        residual = hidden_states
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states = residual + hidden_states
        return hidden_states


@auto_docstring
class Qwen3PreTrainedModel(PreTrainedModel):
    config: Qwen3Config
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _no_split_modules = ["Qwen3DecoderLayer"]
    _skip_keys_device_placement = ["past_key_values"]
    _supports_flash_attn = True
    _supports_sdpa = True
    _supports_flex_attn = True

    _can_compile_fullgraph = True
    _supports_attention_backend = True
    _can_record_outputs = {
        "hidden_states": Qwen3DecoderLayer,
        "attentions": Qwen3Attention,
    }


class Qwen3RotaryEmbedding(nn.Module):
    inv_freq: torch.Tensor  # fix linting for `register_buffer`

    def __init__(self, config: Qwen3Config, device=None):
        super().__init__()
        # BC: "rope_type" was originally "type"
        if hasattr(config, "rope_scaling") and isinstance(config.rope_scaling, dict):
            self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type"))
        else:
            self.rope_type = "default"
        self.max_seq_len_cached = config.max_position_embeddings
        self.original_max_seq_len = config.max_position_embeddings

        self.config = config
        self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]

        inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device)
        self.register_buffer("inv_freq", inv_freq, persistent=False)
        self.original_inv_freq = self.inv_freq

    @torch.no_grad()
    @dynamic_rope_update  # power user: used with advanced RoPE types (e.g. dynamic rope)
    def forward(self, x, position_ids):
        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to(x.device)
        position_ids_expanded = position_ids[:, None, :].float()

        device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu"
        with torch.autocast(device_type=device_type, enabled=False):  # Force float32
            freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
            emb = torch.cat((freqs, freqs), dim=-1)
            cos = emb.cos() * self.attention_scaling
            sin = emb.sin() * self.attention_scaling

        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)


@auto_docstring
class Qwen3Model(Qwen3PreTrainedModel):
    def __init__(self, config: Qwen3Config):
        super().__init__(config)
        self.padding_idx = config.pad_token_id
        self.vocab_size = config.vocab_size

        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
        self.layers = nn.ModuleList(
            [Qwen3DecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
        )
        self.norm = Qwen3RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.rotary_emb = Qwen3RotaryEmbedding(config=config)
        self.gradient_checkpointing = False
        self.has_sliding_layers = "sliding_attention" in self.config.layer_types

        # Initialize weights and apply final processing
        self.post_init()

    @check_model_inputs
    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        use_cache: Optional[bool] = None,
        cache_position: Optional[torch.LongTensor] = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> BaseModelOutputWithPast:
        if (input_ids is None) ^ (inputs_embeds is not None):
            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")

        if inputs_embeds is None:
            inputs_embeds = self.embed_tokens(input_ids)

        if use_cache and past_key_values is None:
            past_key_values = DynamicCache(config=self.config)

        if cache_position is None:
            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
            cache_position = torch.arange(
                past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device
            )

        if position_ids is None:
            position_ids = cache_position.unsqueeze(0)

        # It may already have been prepared by e.g. `generate`
        if not isinstance(causal_mask_mapping := attention_mask, dict):
            # Prepare mask arguments
            mask_kwargs = {
                "config": self.config,
                "input_embeds": inputs_embeds,
                "attention_mask": attention_mask,
                "cache_position": cache_position,
                "past_key_values": past_key_values,
                "position_ids": position_ids,
            }
            # Create the masks
            causal_mask_mapping = {
                "full_attention": create_causal_mask(**mask_kwargs),
            }
            # The sliding window alternating layers are not always activated depending on the config
            if self.has_sliding_layers:
                causal_mask_mapping["sliding_attention"] = create_sliding_window_causal_mask(**mask_kwargs)

        hidden_states = inputs_embeds

        # create position embeddings to be shared across the decoder layers
        position_embeddings = self.rotary_emb(hidden_states, position_ids)

        for decoder_layer in self.layers[: self.config.num_hidden_layers]:
            hidden_states = decoder_layer(
                hidden_states,
                attention_mask=causal_mask_mapping[decoder_layer.attention_type],
                position_ids=position_ids,
                past_key_values=past_key_values,
                use_cache=use_cache,
                cache_position=cache_position,
                position_embeddings=position_embeddings,
                **kwargs,
            )

        hidden_states = self.norm(hidden_states)
        return BaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            past_key_values=past_key_values if use_cache else None,
        )


@auto_docstring
class Qwen3ForGuardModel(Qwen3PreTrainedModel):

    def __init__(self, config):
        super().__init__(config)
        self.model = Qwen3Model(config)
        self.vocab_size = config.vocab_size

        self.risk_level_category_pre = nn.Linear(config.hidden_size, config.guard_inner_size, bias=False)
        self.risk_level_category_layernorm = Qwen3RMSNorm(config.guard_inner_size, eps=config.rms_norm_eps)
        self.risk_level_head = nn.Linear(config.guard_inner_size, config.num_risk_level, bias=False)
        self.category_head = nn.Linear(config.guard_inner_size, config.num_category, bias=False)

        self.query_risk_level_category_pre = nn.Linear(config.hidden_size, config.guard_inner_size, bias=False)
        self.query_risk_level_category_layernorm = Qwen3RMSNorm(config.guard_inner_size, eps=config.rms_norm_eps)
        self.query_risk_level_head = nn.Linear(config.guard_inner_size, config.num_query_risk_level, bias=False)
        self.query_category_head = nn.Linear(config.guard_inner_size, config.num_query_category, bias=False)

        response_risk_level_map = config.response_risk_level_map
        self.response_risk_level_map = {int(k): v for k, v in response_risk_level_map.items()}
        response_category_map = config.response_category_map
        self.response_category_map = {int(k): v for k, v in response_category_map.items()}

        query_risk_level_map = config.query_risk_level_map
        self.query_risk_level_map = {int(k): v for k, v in query_risk_level_map.items()}
        query_category_map = config.query_category_map
        self.query_category_map = {int(k): v for k, v in query_category_map.items()}

        # Initialize weights and apply final processing
        self.post_init()
 
    @can_return_tuple
    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        cache_position: Optional[torch.LongTensor] = None,
        logits_to_keep: Union[int, torch.Tensor] = 0,
        **kwargs: Unpack[TransformersKwargs],
    ) -> GuardLogitsOutputWithPast:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.

        ```"""
        outputs: BaseModelOutputWithPast = self.model(
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            cache_position=cache_position,
            **kwargs,
        )

        hidden_states = outputs.last_hidden_state
        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
        # modify the mapping here
        risk_level_category_x = self.risk_level_category_pre(hidden_states[:, slice_indices, :])
        risk_level_category_x = self.risk_level_category_layernorm(risk_level_category_x)

        risk_level_logits = self.risk_level_head(risk_level_category_x)
        category_logits = self.category_head(risk_level_category_x)
        
        query_risk_level_category_x = self.query_risk_level_category_pre(hidden_states[:, slice_indices, :])
        query_risk_level_category_x = self.query_risk_level_category_layernorm(query_risk_level_category_x)

        query_risk_level_logits = self.query_risk_level_head(query_risk_level_category_x)
        query_category_logits = self.query_category_head(query_risk_level_category_x)

        loss = None
        return GuardLogitsOutputWithPast(
            loss=loss,
            risk_level_logits=risk_level_logits,
            category_logits=category_logits,
            query_risk_level_logits=query_risk_level_logits,
            query_category_logits=query_category_logits,           
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )


    @torch.no_grad()
    def stream_generate(
        self,
        input_ids: torch.LongTensor
    ) -> Generator[Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor], Optional[torch.LongTensor], None]:

        seq_length = len(input_ids)
        causal_mask = torch.tril(torch.ones((seq_length, seq_length), device=self.device, dtype=torch.bool))
        causal_mask = causal_mask.unsqueeze(0).unsqueeze(0)

        past_key_values = None
        current_input_ids = input_ids

        while True:
            outputs = self.forward(
                input_ids=current_input_ids.unsqueeze(0),
                attention_mask=causal_mask,
                past_key_values=past_key_values
            )
            past_key_values = outputs.past_key_values
            logits_tuple = (
                outputs.risk_level_logits,
                outputs.category_logits,
                outputs.query_risk_level_logits,
                outputs.query_category_logits,
            )
            next_token_id = yield logits_tuple

            if next_token_id is None:
                break
            current_input_ids = torch.cat([current_input_ids, torch.tensor([next_token_id],device=self.device)])   
            cur_len = causal_mask.shape[2]
            new_causal_mask = torch.zeros((1, cur_len+1, cur_len+1), device=causal_mask.device, dtype=torch.bool)
            new_causal_mask[:, :cur_len, :cur_len] = causal_mask.squeeze(0)
            new_causal_mask[:, cur_len, :cur_len+1] = True 
            causal_mask = new_causal_mask.unsqueeze(0)

 
    @torch.no_grad()
    def stream_moderate_from_ids(
        self, 
        token_ids: torch.LongTensor, 
        role: str, 
        stream_state: Optional[Generator] = None
    ) -> Tuple[Dict, Generator]:
        """
        Incrementally processes token_ids to evaluate the risk of the latest tokens.
        Args:
            token_ids (torch.LongTensor): The token IDs to process.
                - For the first call (when `stream_state` is None), this should be the
                full sequence of token IDs for the initial prompt.
                - For subsequent calls, this should ONLY be the new, incremental token id.
                Shape should be (1).
            role (str): The role of the speaker for the provided `token_ids`. 
                        Must be 'user' or 'assistant'.
            stream_state (Generator, optional): The state from the previous call to
                                                this function. Pass `None` to start a
                                                new conversation stream.

        Returns:
            Tuple[Dict, Generator]: A tuple containing:
            - A dictionary with the prediction results for the *last token* processed.
            - The updated stream_state generator to be passed to the next call.
        """
        token_ids = token_ids.to(self.device)

        if stream_state is None:
            stream_state = self.stream_generate(token_ids)
            logits_tuple = next(stream_state)
        else: 
            logits_tuple = stream_state.send(token_ids)
        if role == "user":
            risk_level_logits = logits_tuple[2]  
            category_logits = logits_tuple[3]  
        elif role == "assistant":
            risk_level_logits = logits_tuple[0]  
            category_logits = logits_tuple[1]    
        else:
            raise ValueError("Role must be either 'user' or 'assistant'")
        risk_probs = F.softmax(risk_level_logits.squeeze(1), dim=-1)
        pred_risk_prob, pred_risk_idx = torch.max(risk_probs, dim=-1)
        category_probs = F.softmax(category_logits.squeeze(1), dim=-1)
        pred_cat_prob, pred_cat_idx = torch.max(category_probs, dim=-1)

        if role == "user":
            result = {
                "risk_level": [self.query_risk_level_map[int(i)] for i in pred_risk_idx[0]],
                "risk_prob": [round(float(i),2) for i in pred_risk_prob[0]],
                "category": [self.query_category_map[int(i)] for i in pred_cat_idx[0]],
                "category_prob": [round(float(i),2) for i in pred_cat_prob[0]]
            }
        else:
            result = {
                "risk_level": [self.response_risk_level_map[int(i)] for i in pred_risk_idx[0]],
                "risk_prob": [round(float(i),2) for i in pred_risk_prob[0]],
                "category": [self.response_category_map[int(i)] for i in pred_cat_idx[0]],
                "category_prob": [round(float(i),2) for i in pred_cat_prob[0]]
            }

        return result, stream_state

    @torch.no_grad()
    def close_stream(self, stream_state: Optional[Generator]) -> None:
        if stream_state is not None:
            try:
                stream_state.send(None)
            except StopIteration:
                pass
            finally:
                stream_state.close()

__all__ = [
    "Qwen3PreTrainedModel",
    "Qwen3Model",
    "Qwen3ForGuardModel",
]