Skip to content

vllm.model_executor.models.glm4

Inference-only GLM-4-0414 model compatible with HuggingFace weights.

ALL_DECODER_LAYER_TYPES module-attribute 

ALL_DECODER_LAYER_TYPES = {'attention': Glm4DecoderLayer}

Glm4Attention

Bases: Module

Source code in vllm/model_executor/models/glm4.py 
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
class Glm4Attention(nn.Module):

    def __init__(self,
                 config: Glm4Config,
                 hidden_size: int,
                 num_heads: int,
                 num_kv_heads: int,
                 max_position: int = 4096 * 32,
                 head_dim: Optional[int] = None,
                 qkv_bias: bool = False,
                 rope_theta: float = 10000,
                 cache_config: Optional[CacheConfig] = None,
                 quant_config: Optional[QuantizationConfig] = None,
                 rope_scaling: Optional[tuple] = None,
                 prefix: str = "",
                 attn_type: str = AttentionType.DECODER) -> 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:
            # Number of KV heads is greater than TP size, so we partition
            # the KV heads across multiple tensor parallel GPUs.
            assert self.total_num_kv_heads % tp_size == 0
        else:
            # Number of KV heads is less than TP size, so we replicate
            # the KV heads across multiple tensor parallel GPUs.
            assert tp_size % self.total_num_kv_heads == 0
        partial_rotary_factor = getattr(config, "partial_rotary_factor", 0.5)
        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.rotary_dim = self.head_dim
        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.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,
            rotary_dim=self.rotary_dim,
            max_position=max_position,
            base=self.rope_theta,
            rope_scaling=rope_scaling,
            partial_rotary_factor=partial_rotary_factor,
            is_neox_style=False,
        )
        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)

    def forward(
        self,
        positions: torch.Tensor,
        hidden_states: torch.Tensor,
    ) -> 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)
        output, _ = self.o_proj(attn_output)
        return output

attn instance-attribute 

attn = Attention(
    num_heads,
    head_dim,
    scaling,
    num_kv_heads=num_kv_heads,
    cache_config=cache_config,
    quant_config=quant_config,
    prefix=f"{prefix}.attn",
    attn_type=attn_type,
)

head_dim instance-attribute 

head_dim = head_dim or hidden_size // total_num_heads

hidden_size instance-attribute 

hidden_size = hidden_size

kv_size instance-attribute 

kv_size = num_kv_heads * head_dim

num_heads instance-attribute 

num_heads = total_num_heads // tp_size

num_kv_heads instance-attribute 

num_kv_heads = max(1, total_num_kv_heads // tp_size)

o_proj instance-attribute 

o_proj = RowParallelLinear(
    total_num_heads * head_dim,
    hidden_size,
    bias=False,
    quant_config=quant_config,
    prefix=f"{prefix}.o_proj",
)

q_size instance-attribute 

q_size = num_heads * head_dim

qkv_proj instance-attribute 

qkv_proj = QKVParallelLinear(
    hidden_size,
    head_dim,
    total_num_heads,
    total_num_kv_heads,
    bias=qkv_bias,
    quant_config=quant_config,
    prefix=f"{prefix}.qkv_proj",
)

rope_theta instance-attribute 

rope_theta = rope_theta

rotary_dim instance-attribute 

rotary_dim = head_dim

rotary_emb instance-attribute 

rotary_emb = get_rope(
    head_dim,
    rotary_dim=rotary_dim,
    max_position=max_position,
    base=rope_theta,
    rope_scaling=rope_scaling,
    partial_rotary_factor=partial_rotary_factor,
    is_neox_style=False,
)

scaling instance-attribute 

scaling = head_dim ** -0.5

total_num_heads instance-attribute 

total_num_heads = num_heads

total_num_kv_heads instance-attribute 

total_num_kv_heads = num_kv_heads

__init__ 

__init__(
    config: Glm4Config,
    hidden_size: int,
    num_heads: int,
    num_kv_heads: int,
    max_position: int = 4096 * 32,
    head_dim: Optional[int] = None,
    qkv_bias: bool = False,
    rope_theta: float = 10000,
    cache_config: Optional[CacheConfig] = None,
    quant_config: Optional[QuantizationConfig] = None,
    rope_scaling: Optional[tuple] = None,
    prefix: str = "",
    attn_type: str = DECODER,
) -> None
Source code in vllm/model_executor/models/glm4.py 
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
def __init__(self,
             config: Glm4Config,
             hidden_size: int,
             num_heads: int,
             num_kv_heads: int,
             max_position: int = 4096 * 32,
             head_dim: Optional[int] = None,
             qkv_bias: bool = False,
             rope_theta: float = 10000,
             cache_config: Optional[CacheConfig] = None,
             quant_config: Optional[QuantizationConfig] = None,
             rope_scaling: Optional[tuple] = None,
             prefix: str = "",
             attn_type: str = AttentionType.DECODER) -> 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:
        # Number of KV heads is greater than TP size, so we partition
        # the KV heads across multiple tensor parallel GPUs.
        assert self.total_num_kv_heads % tp_size == 0
    else:
        # Number of KV heads is less than TP size, so we replicate
        # the KV heads across multiple tensor parallel GPUs.
        assert tp_size % self.total_num_kv_heads == 0
    partial_rotary_factor = getattr(config, "partial_rotary_factor", 0.5)
    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.rotary_dim = self.head_dim
    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.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,
        rotary_dim=self.rotary_dim,
        max_position=max_position,
        base=self.rope_theta,
        rope_scaling=rope_scaling,
        partial_rotary_factor=partial_rotary_factor,
        is_neox_style=False,
    )
    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)

forward 

forward(positions: Tensor, hidden_states: Tensor) -> Tensor
Source code in vllm/model_executor/models/glm4.py 
125
126
127
128
129
130
131
132
133
134
135
def forward(
    self,
    positions: torch.Tensor,
    hidden_states: torch.Tensor,
) -> 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)
    output, _ = self.o_proj(attn_output)
    return output

Glm4DecoderLayer

Bases: Module

Source code in vllm/model_executor/models/glm4.py 
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
class Glm4DecoderLayer(nn.Module):

    def __init__(
        self,
        config: Glm4Config,
        cache_config: Optional[CacheConfig] = None,
        quant_config: Optional[QuantizationConfig] = None,
        prefix: str = "",
    ) -> None:
        super().__init__()
        self.hidden_size = config.hidden_size
        rope_theta = getattr(config, "rope_theta", 1000000)
        rope_scaling = getattr(config, "rope_scaling", None)

        self.self_attn = Glm4Attention(
            config=config,
            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,
            rope_theta=rope_theta,
            qkv_bias=getattr(config, 'attention_bias', False),
            head_dim=getattr(config, 'head_dim', None),
            cache_config=cache_config,
            quant_config=quant_config,
            rope_scaling=rope_scaling,
            prefix=f"{prefix}.self_attn",
            attn_type=AttentionType.DECODER,
        )
        self.mlp = Glm4MLP(
            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)
        self.post_self_attn_layernorm = RMSNorm(config.hidden_size,
                                                eps=config.rms_norm_eps)
        self.post_mlp_layernorm = RMSNorm(config.hidden_size,
                                          eps=config.rms_norm_eps)

    def forward(
        self,
        positions: torch.Tensor,
        hidden_states: torch.Tensor,
        residual: Optional[torch.Tensor],
    ) -> tuple[torch.Tensor, torch.Tensor]:
        # Self Attention
        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,
        )

        hidden_states = self.post_self_attn_layernorm(hidden_states)

        # Fully Connected
        hidden_states, residual = self.post_attention_layernorm(
            hidden_states, residual)
        hidden_states = self.mlp(hidden_states)
        hidden_states = self.post_mlp_layernorm(hidden_states)

        return hidden_states, residual

hidden_size instance-attribute 

hidden_size = hidden_size

input_layernorm instance-attribute 

input_layernorm = RMSNorm(hidden_size, eps=rms_norm_eps)

mlp instance-attribute 

mlp = LlamaMLP(
    hidden_size=hidden_size,
    intermediate_size=intermediate_size,
    hidden_act=hidden_act,
    quant_config=quant_config,
    prefix=f"{prefix}.mlp",
)

post_attention_layernorm instance-attribute 

post_attention_layernorm = RMSNorm(
    hidden_size, eps=rms_norm_eps
)

post_mlp_layernorm instance-attribute 

post_mlp_layernorm = RMSNorm(hidden_size, eps=rms_norm_eps)

post_self_attn_layernorm instance-attribute 

post_self_attn_layernorm = RMSNorm(
    hidden_size, eps=rms_norm_eps
)

self_attn instance-attribute 

self_attn = Glm4Attention(
    config=config,
    hidden_size=hidden_size,
    num_heads=num_attention_heads,
    max_position=max_position_embeddings,
    num_kv_heads=num_key_value_heads,
    rope_theta=rope_theta,
    qkv_bias=getattr(config, "attention_bias", False),
    head_dim=getattr(config, "head_dim", None),
    cache_config=cache_config,
    quant_config=quant_config,
    rope_scaling=rope_scaling,
    prefix=f"{prefix}.self_attn",
    attn_type=DECODER,
)

__init__ 

__init__(
    config: Glm4Config,
    cache_config: Optional[CacheConfig] = None,
    quant_config: Optional[QuantizationConfig] = None,
    prefix: str = "",
) -> None
Source code in vllm/model_executor/models/glm4.py 
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
def __init__(
    self,
    config: Glm4Config,
    cache_config: Optional[CacheConfig] = None,
    quant_config: Optional[QuantizationConfig] = None,
    prefix: str = "",
) -> None:
    super().__init__()
    self.hidden_size = config.hidden_size
    rope_theta = getattr(config, "rope_theta", 1000000)
    rope_scaling = getattr(config, "rope_scaling", None)

    self.self_attn = Glm4Attention(
        config=config,
        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,
        rope_theta=rope_theta,
        qkv_bias=getattr(config, 'attention_bias', False),
        head_dim=getattr(config, 'head_dim', None),
        cache_config=cache_config,
        quant_config=quant_config,
        rope_scaling=rope_scaling,
        prefix=f"{prefix}.self_attn",
        attn_type=AttentionType.DECODER,
    )
    self.mlp = Glm4MLP(
        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)
    self.post_self_attn_layernorm = RMSNorm(config.hidden_size,
                                            eps=config.rms_norm_eps)
    self.post_mlp_layernorm = RMSNorm(config.hidden_size,
                                      eps=config.rms_norm_eps)

forward 

forward(
    positions: Tensor,
    hidden_states: Tensor,
    residual: Optional[Tensor],
) -> tuple[Tensor, Tensor]
Source code in vllm/model_executor/models/glm4.py 
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
def forward(
    self,
    positions: torch.Tensor,
    hidden_states: torch.Tensor,
    residual: Optional[torch.Tensor],
) -> tuple[torch.Tensor, torch.Tensor]:
    # Self Attention
    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,
    )

    hidden_states = self.post_self_attn_layernorm(hidden_states)

    # Fully Connected
    hidden_states, residual = self.post_attention_layernorm(
        hidden_states, residual)
    hidden_states = self.mlp(hidden_states)
    hidden_states = self.post_mlp_layernorm(hidden_states)

    return hidden_states, residual

Glm4ForCausalLM

Bases: Module, SupportsLoRA, SupportsPP

Source code in vllm/model_executor/models/glm4.py 
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
class Glm4ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
    packed_modules_mapping = {
        "qkv_proj": [
            "q_proj",
            "k_proj",
            "v_proj",
        ],
        "gate_up_proj": [
            "gate_proj",
            "up_proj",
        ],
    }

    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
        super().__init__()
        config = vllm_config.model_config.hf_config
        quant_config = vllm_config.quant_config
        lora_config = vllm_config.lora_config

        self.config = config
        self.lora_config = lora_config

        self.quant_config = quant_config
        self.model = Glm4Model(vllm_config=vllm_config,
                               prefix=maybe_prefix(prefix, "model"))

        if get_pp_group().is_last_rank:
            if config.tie_word_embeddings:
                self.lm_head = self.model.embed_tokens
            else:
                self.lm_head = ParallelLMHead(config.vocab_size,
                                              config.hidden_size,
                                              quant_config=quant_config,
                                              prefix=maybe_prefix(
                                                  prefix, "lm_head"))
        else:
            self.lm_head = PPMissingLayer()

        self.logits_processor = LogitsProcessor(config.vocab_size)

        self.make_empty_intermediate_tensors = (
            self.model.make_empty_intermediate_tensors)

    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
        return self.model.get_input_embeddings(input_ids)

    def forward(
        self,
        input_ids: torch.Tensor,
        positions: torch.Tensor,
        intermediate_tensors: Optional[IntermediateTensors] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
    ) -> Union[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,
        sampling_metadata: SamplingMetadata,
    ) -> Optional[torch.Tensor]:
        logits = self.logits_processor(self.lm_head, hidden_states,
                                       sampling_metadata)
        return logits

    def load_weights(self, weights: Iterable[tuple[str,
                                                   torch.Tensor]]) -> set[str]:
        loader = AutoWeightsLoader(
            self,
            skip_prefixes=(["lm_head."]
                           if self.config.tie_word_embeddings else None),
        )
        return loader.load_weights(weights)

config instance-attribute 

config = config

lm_head instance-attribute 

lm_head = embed_tokens

logits_processor instance-attribute 

logits_processor = LogitsProcessor(vocab_size)

lora_config instance-attribute 

lora_config = lora_config

make_empty_intermediate_tensors instance-attribute 

make_empty_intermediate_tensors = (
    make_empty_intermediate_tensors
)

model instance-attribute 

model = Glm4Model(
    vllm_config=vllm_config,
    prefix=maybe_prefix(prefix, "model"),
)

packed_modules_mapping class-attribute instance-attribute 

packed_modules_mapping = {
    "qkv_proj": ["q_proj", "k_proj", "v_proj"],
    "gate_up_proj": ["gate_proj", "up_proj"],
}

quant_config instance-attribute 

quant_config = quant_config

__init__ 

__init__(*, vllm_config: VllmConfig, prefix: str = '')
Source code in vllm/model_executor/models/glm4.py 
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
    super().__init__()
    config = vllm_config.model_config.hf_config
    quant_config = vllm_config.quant_config
    lora_config = vllm_config.lora_config

    self.config = config
    self.lora_config = lora_config

    self.quant_config = quant_config
    self.model = Glm4Model(vllm_config=vllm_config,
                           prefix=maybe_prefix(prefix, "model"))

    if get_pp_group().is_last_rank:
        if config.tie_word_embeddings:
            self.lm_head = self.model.embed_tokens
        else:
            self.lm_head = ParallelLMHead(config.vocab_size,
                                          config.hidden_size,
                                          quant_config=quant_config,
                                          prefix=maybe_prefix(
                                              prefix, "lm_head"))
    else:
        self.lm_head = PPMissingLayer()

    self.logits_processor = LogitsProcessor(config.vocab_size)

    self.make_empty_intermediate_tensors = (
        self.model.make_empty_intermediate_tensors)

compute_logits 

compute_logits(
    hidden_states: Tensor,
    sampling_metadata: SamplingMetadata,
) -> Optional[Tensor]
Source code in vllm/model_executor/models/glm4.py 
289
290
291
292
293
294
295
296
def compute_logits(
    self,
    hidden_states: torch.Tensor,
    sampling_metadata: SamplingMetadata,
) -> Optional[torch.Tensor]:
    logits = self.logits_processor(self.lm_head, hidden_states,
                                   sampling_metadata)
    return logits

forward 

forward(
    input_ids: Tensor,
    positions: Tensor,
    intermediate_tensors: Optional[
        IntermediateTensors
    ] = None,
    inputs_embeds: Optional[Tensor] = None,
) -> Union[Tensor, IntermediateTensors]
Source code in vllm/model_executor/models/glm4.py 
278
279
280
281
282
283
284
285
286
287
def forward(
    self,
    input_ids: torch.Tensor,
    positions: torch.Tensor,
    intermediate_tensors: Optional[IntermediateTensors] = None,
    inputs_embeds: Optional[torch.Tensor] = None,
) -> Union[torch.Tensor, IntermediateTensors]:
    hidden_states = self.model(input_ids, positions, intermediate_tensors,
                               inputs_embeds)
    return hidden_states

get_input_embeddings 

get_input_embeddings(input_ids: Tensor) -> Tensor
Source code in vllm/model_executor/models/glm4.py 
275
276
def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
    return self.model.get_input_embeddings(input_ids)

load_weights 

load_weights(
    weights: Iterable[tuple[str, Tensor]],
) -> set[str]
Source code in vllm/model_executor/models/glm4.py 
298
299
300
301
302
303
304
305
def load_weights(self, weights: Iterable[tuple[str,
                                               torch.Tensor]]) -> set[str]:
    loader = AutoWeightsLoader(
        self,
        skip_prefixes=(["lm_head."]
                       if self.config.tie_word_embeddings else None),
    )
    return loader.load_weights(weights)

Glm4Model

Bases: LlamaModel

Source code in vllm/model_executor/models/glm4.py 
217
218
219
220
221
222
223
224
225
226
227
228
229
@support_torch_compile(
    dynamic_arg_dims={
        "input_ids": 0,
        "positions": -1,
        "intermediate_tensors": 0,
        "inputs_embeds": 0,
    })
class Glm4Model(LlamaModel):

    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
        super().__init__(vllm_config=vllm_config,
                         prefix=prefix,
                         layer_type=Glm4DecoderLayer)

__init__ 

__init__(*, vllm_config: VllmConfig, prefix: str = '')
Source code in vllm/model_executor/models/glm4.py 
226
227
228
229
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
    super().__init__(vllm_config=vllm_config,
                     prefix=prefix,
                     layer_type=Glm4DecoderLayer)