sglang 0.4.4.post4__py3-none-any.whl → 0.4.5__py3-none-any.whl

sglang/srt/models/mllama4.py

@@ -0,0 +1,154 @@
+# TODO: add Aapted from vllm/mllama4.py
+from collections.abc import Iterable
+from typing import Optional, Set, Tuple
+
+import torch
+from torch import nn
+from transformers import Llama4Config
+
+from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.quantization import QuantizationConfig
+from sglang.srt.model_executor.forward_batch_info import ForwardBatch
+from sglang.srt.model_loader.weight_utils import default_weight_loader
+from sglang.srt.utils import add_prefix
+
+
+class Llama4ForConditionalGeneration(nn.Module):
+    packed_modules_mapping = {
+        "qkv_proj": ["q_proj", "k_proj", "v_proj"],
+    }
+
+    def __init__(
+        self,
+        config: Llama4Config,
+        quant_config: Optional[QuantizationConfig] = None,
+        prefix: str = "",
+    ):
+        super().__init__()
+        self.config = config
+        self.quant_config = quant_config
+
+        # Initialize the language model
+        from sglang.srt.models.llama4 import Llama4ForCausalLM
+
+        self.language_model = Llama4ForCausalLM(
+            config.text_config,
+            quant_config=quant_config,
+            prefix=add_prefix("language_model", prefix),
+        )
+
+        self.logits_processor = LogitsProcessor(config.text_config)
+
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        positions: torch.Tensor,
+        forward_batch: ForwardBatch,
+        **kwargs: object,
+    ) -> torch.Tensor:
+
+        return self.language_model(input_ids, positions, forward_batch)
+
+    def permute_qk_weight_for_rotary(
+        self,
+        name: str,
+        loaded_weight: torch.Tensor,
+    ) -> Tuple[str, torch.Tensor]:
+
+        def permute(w: torch.Tensor, n_heads: int):
+            attn_in = self.language_model.config.head_dim * n_heads
+            attn_out = self.language_model.config.hidden_size
+
+            return (
+                w.view(n_heads, attn_in // n_heads // 2, 2, attn_out)
+                .transpose(1, 2)
+                .reshape(attn_in, attn_out)
+            )
+
+        modules = name.split(".")
+
+        # rotary embeds should be sliced
+        if ("wk" in modules or "k_proj" in modules) and modules[-1] == "weight":
+            loaded_weight = permute(
+                loaded_weight, self.language_model.config.num_key_value_heads
+            )
+        elif ("wq" in modules or "q_proj" in modules) and modules[-1] == "weight":
+            loaded_weight = permute(
+                loaded_weight, self.language_model.config.num_attention_heads
+            )
+
+        return name, loaded_weight
+
+    def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]) -> Set[str]:
+
+        stacked_params_mapping = [
+            # (param_name, shard_name, shard_id)
+            (".self_attn.qkv_proj", ".self_attn.q_proj", "q"),
+            (".self_attn.qkv_proj", ".self_attn.k_proj", "k"),
+            (".self_attn.qkv_proj", ".self_attn.v_proj", "v"),
+            (".shared_expert.gate_up_proj", ".shared_expert.gate_proj", 0),
+            (".shared_expert.gate_up_proj", ".shared_expert.up_proj", 1),
+            (".feed_forward.gate_up_proj", ".feed_forward.gate_proj", 0),
+            (".feed_forward.gate_up_proj", ".feed_forward.up_proj", 1),
+        ]
+
+        params_dict = dict(self.named_parameters())
+
+        num_experts = self.config.text_config.num_local_experts
+
+        for name, loaded_weight in weights:
+
+            if name.startswith("vision_model") or name.startswith(
+                "multi_modal_projector"
+            ):
+                continue
+
+            name, loaded_weight = self.permute_qk_weight_for_rotary(name, loaded_weight)
+
+            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)
+                param = params_dict[name]
+                weight_loader = param.weight_loader
+                weight_loader(param, loaded_weight, shard_id)
+                break
+            else:
+                if ".experts" in name:
+                    if ".gate_up_proj" in name:
+                        name_list = [
+                            name.replace(".experts.gate_up_proj", ".experts.w13_weight")
+                        ] * 2
+                        loaded_weight_list = loaded_weight.chunk(2, dim=-1)
+                        shard_id_list = ["w1", "w3"]
+                    else:
+                        name_list = [
+                            name.replace(".experts.down_proj", ".experts.w2_weight")
+                        ]
+                        shard_id_list = ["w2"]
+                        loaded_weight_list = [loaded_weight]
+                    for name, loaded_weight, shard_id in zip(
+                        name_list, loaded_weight_list, shard_id_list
+                    ):
+                        param = params_dict[name]
+                        weight_loader = param.weight_loader
+                        for expert_id in range(num_experts):
+                            weight_loader(
+                                param,
+                                loaded_weight[expert_id].T,
+                                name,
+                                shard_id=shard_id,
+                                expert_id=expert_id,
+                            )
+                else:
+                    # Skip loading extra bias for GPTQ models.
+                    if name.endswith(".bias") and name not in params_dict:
+                        continue
+                    param = params_dict[name]
+                    weight_loader = getattr(
+                        param, "weight_loader", default_weight_loader
+                    )
+                    weight_loader(param, loaded_weight)
+
+
+EntryClass = Llama4ForConditionalGeneration

sglang/version.py

@@ -1 +1 @@
-__version__ = "0.4.4.post4"
+__version__ = "0.4.5"

{sglang-0.4.4.post4.dist-info → sglang-0.4.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sglang
-Version: 0.4.4.post4
+Version: 0.4.5
 Summary: SGLang is yet another fast serving framework for large language models and vision language models.
 License:                                  Apache License
                                    Version 2.0, January 2004

{sglang-0.4.4.post4.dist-info → sglang-0.4.5.dist-info}/RECORD

@@ -9,9 +9,9 @@ sglang/global_config.py,sha256=xzLdk8W53fneFblNh8iIjGF9C3-7mnzR1-LleD9Btxg,1495
 sglang/launch_server.py,sha256=mDXfwha8LHpWQJekcCosR98QhCQsbmilsBlI5jAIgg0,420
 sglang/llama3_eval.py,sha256=gWSboDchIGybIce88bJlrCG0yiLZ513mw4gcutJlzGM,10017
 sglang/utils.py,sha256=GIcgiRHkZ-gyPxXOdn1qFF41jkg4-YdDxbPc4mzO-qk,16159
-sglang/version.py,sha256=63MgDRjbuREW1iqTD8A1g5qw4uGX6I-V8Zgt5c1nnys,28
+sglang/version.py,sha256=ErkLkI2TDBX1OIqi2GGa20CPeu4ZculEi-XffRbLU6M,22
 sglang/lang/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-sglang/lang/chat_template.py,sha256=xZ-kQpgb4-NY6QOqgRjOODZW5G4EvJnplaqYF6Ng2Ow,18952
+sglang/lang/chat_template.py,sha256=MwNL5dNTe8g_l2ljZubnrazEgT2xEv-9O2D0Ezwxy4I,19658
 sglang/lang/choices.py,sha256=-W1DVw9N9ZliVpvmWrzIXG4cswAah8eMQrHWzkS3D8o,6234
 sglang/lang/compiler.py,sha256=o1C6G3TzhjSlsH-doTPy5oiVehr57dxNTa5oZw5TTAI,7639
 sglang/lang/interpreter.py,sha256=OH1SFCm4rUCPO32MTo8j5V2Z13Jic7_r1GQOP1-aHaw,33234
@@ -27,7 +27,7 @@ sglang/lang/backend/vertexai.py,sha256=O-iBLD-y3vq80UxnrAoJri7bxpgd-_eakZ88Cf8bE
 sglang/srt/_custom_ops.py,sha256=lUBwC5R2UfjFMA1EtC5Kh2IngsqBJM9IuMW46kJWcjE,3647
 sglang/srt/aio_rwlock.py,sha256=6LYtOdeTUY3hkfa1dmYkgsaF2ttrwIF3hUWz2AZ2fqw,2970
 sglang/srt/code_completion_parser.py,sha256=HhEUzdL-FVBsOot9tKDKA1l8Gdx8qsF1RRg-zHNpmLQ,5400
-sglang/srt/conversation.py,sha256=mzNPZX7ndgmm1E8azuK8eytN6bNCEu3WbcPReRFfhd0,27819
+sglang/srt/conversation.py,sha256=WP72AZrZpiqc5RowucT2tW3jVCb1pb4veW_kpwYS4yY,28785
 sglang/srt/custom_op.py,sha256=bIZ__3FiZvkbsN9O_jeLy_49X7ZbYbw0VxoL80uWwaI,3715
 sglang/srt/function_call_parser.py,sha256=buYENeNEP5bhsvD424yGCa9wOqSfVOZSRn6zLiSJp5I,23733
 sglang/srt/hf_transformers_utils.py,sha256=_QYTl9LpU0jmKPlYooHi1etwMvb5v40JIrG_t_Fx06w,9215
@@ -48,7 +48,7 @@ sglang/srt/configs/device_config.py,sha256=kfmpPOECqYxcRoY-ko0QZRhyiBWUGP2CMF51D
 sglang/srt/configs/exaone.py,sha256=Duxd4yQoKy8GWEzZD_kCY_OzmN_67CTJL_Kgn0eXk3g,10731
 sglang/srt/configs/janus_pro.py,sha256=-QtJ4ZGZiAJb0AkOEcuCHzIKLw23nF8nRk3rdCcoUO0,19016
 sglang/srt/configs/load_config.py,sha256=qs-AxuplouBx2tsv9KGBOLZPbwzuVA4vbktbGP_cRp8,3309
-sglang/srt/configs/model_config.py,sha256=-6SEpQw-U_KWONdNYqEe8nXopqDcDtp4B4g-RZQPhHo,20581
+sglang/srt/configs/model_config.py,sha256=ZioUnc5UzsBVEYHE_GgCofYL97MByZm2NfHikS9HwLo,20771
 sglang/srt/configs/utils.py,sha256=3nHUfisMs_Ltuhv8OZTNCJp63YJKJVF43h1QZB1zqx8,670
 sglang/srt/connector/__init__.py,sha256=czLX5JOxuMhH-T9eSJzoc1qv1B4z9chyffDRL5I6wo4,1247
 sglang/srt/connector/base_connector.py,sha256=i6i1TIzsz4NbSEkrdMPq-urb2sN2aLAx8dazga4gB9U,2833
@@ -91,14 +91,14 @@ sglang/srt/layers/linear.py,sha256=HYIGxpRYL6x-jNOkyNtGAw5Ak9Nq8jkntddgTBER_1w,5
 sglang/srt/layers/logits_processor.py,sha256=Vp8ibljVEezTr54xzeOcjiJR7JdYO8ItkO5nLIIMVu0,24206
 sglang/srt/layers/parameter.py,sha256=0OTMtmsNds42e3z3wHTRJiUfxCWFwSL6DHrqgeTgGt8,15151
 sglang/srt/layers/pooler.py,sha256=rj2lygvleBnyLCBZ8I11HGMgpfIDsT0l3PIkshJwdu4,1606
-sglang/srt/layers/radix_attention.py,sha256=UDL0y4Zasay_Rk-_XmIU4kaGbaF26ONvEHX5EQzLrqI,2260
-sglang/srt/layers/rotary_embedding.py,sha256=srAyHlM18hgEQKW7MotYKt1xGkqA_j1PU8g0PO5t3lA,42770
+sglang/srt/layers/radix_attention.py,sha256=4xRq0w9yDfAVdNlBToQpmc7irq-pomJm-GlIfMtpYtk,2328
+sglang/srt/layers/rotary_embedding.py,sha256=1nznPQ6EfVMDSRabKqifEE2xtMFwSri-kEepMaWdZeI,45340
 sglang/srt/layers/sampler.py,sha256=yipSyN5UWGwGS-BC-WzWMmelys4CCDtK_8b1OpaK6sM,11622
 sglang/srt/layers/torchao_utils.py,sha256=Ws24FdRBSkTpyeyA6bQrdDm-W5wfDxKvSIPUSahyMfA,4063
 sglang/srt/layers/vocab_parallel_embedding.py,sha256=QUxd4sELx6p3dHvEKmccPZ-phdd_9EjNdwjH3SJ9zxI,22238
 sglang/srt/layers/attention/base_attn_backend.py,sha256=X_GIbQuU9njtUEGdUP7E_KRhmGxj3UyPHNESlL3QaQ8,3264
 sglang/srt/layers/attention/double_sparsity_backend.py,sha256=2ZRL_gYz14idoVqQzeQ6N77nXer0f_8_TUYw40XUUz0,9161
-sglang/srt/layers/attention/flashattention_backend.py,sha256=IwrJRmBFWLUdYqe7n4Dd2rX4zu9V55qBLjCdNQEmkd0,30306
+sglang/srt/layers/attention/flashattention_backend.py,sha256=ORtcSJUDbV2qfKGkq9ohiy8JJ1SU9R2I5fSMizF4EhI,42572
 sglang/srt/layers/attention/flashinfer_backend.py,sha256=3fxS2NQzCBw7h_gLxBjHcyDkf2quWqBxr_N01lYmfJo,45865
 sglang/srt/layers/attention/flashinfer_mla_backend.py,sha256=pnVhvVEK87iFW8gUb1G7X7c1tqro8R2DSEOFCnlV8Bo,30301
 sglang/srt/layers/attention/flashmla_backend.py,sha256=1RPFNtQOBw6BWxIjrzfJgA9Nx92udLbR-S5KXmqjxS8,10536
@@ -111,7 +111,7 @@ sglang/srt/layers/attention/triton_ops/double_sparsity_attention.py,sha256=BXUY8
 sglang/srt/layers/attention/triton_ops/extend_attention.py,sha256=hbBvnhc2zqu-E3HNROVXyNOZbtDkVRuFus-yTjmE0Sg,13668
 sglang/srt/layers/attention/triton_ops/prefill_attention.py,sha256=Y66gZ37u0GKMPtI8n5MbO6uOxRuGEmKIG0IPbJTOqAM,6213
 sglang/srt/layers/attention/triton_ops/rocm_mla_decode_rope.py,sha256=664WnAJ91EiCUZOcnVDfbTQf4uGJ4ZDZB1CbxpEUFZc,13866
-sglang/srt/layers/moe/fused_moe_native.py,sha256=KF0J5W5biWYWbERUHAduQFnUV5om9VaYv0I9avsVfgU,4330
+sglang/srt/layers/moe/fused_moe_native.py,sha256=bf0po921lY9xnlZivdJly0bGIYFlLqp5v8Mz7tG5bdg,4451
 sglang/srt/layers/moe/router.py,sha256=gvyK7hXlujfCZCmAIFc3oxfgjuAjzlpPe3mp1Blc6Y0,10419
 sglang/srt/layers/moe/topk.py,sha256=iUb-64CaNAUfvBZ1pkgsedcLRQs2sVSIzQ5300WmdXI,10242
 sglang/srt/layers/moe/ep_moe/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -119,8 +119,8 @@ sglang/srt/layers/moe/ep_moe/kernels.py,sha256=ijqRzS-tb0LGnDU5hW-g0JH104ppADrWa
 sglang/srt/layers/moe/ep_moe/layer.py,sha256=1TmWnxv-bW1Qbgru-V-vGnt3ruuTIwHQy0Y5ZA_xzvE,36824
 sglang/srt/layers/moe/ep_moe/token_dispatcher.py,sha256=jnr6KSM8YooftTjZ3gYe0eWpOd1dmkXqk4hKRvLTwCo,19708
 sglang/srt/layers/moe/fused_moe_triton/__init__.py,sha256=h9yMFAL_bagUf-qBED8gSWdCOb7d8IdA-pE-L_nIg8E,842
-sglang/srt/layers/moe/fused_moe_triton/fused_moe.py,sha256=_ufAA7m6alppU4vVxh7QrgSgFOcBSekLQ-UQdHq7TmQ,53988
-sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=CYT3ujh5ifonhqQc1uYSa6maJADUOBJoRHMIWmf-DU4,24100
+sglang/srt/layers/moe/fused_moe_triton/fused_moe.py,sha256=sjAXnjUmLXPpvFFL4VShBce_9xygWY2twAQJ74OJ_ZQ,54500
+sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=43-UL9KEMoaiC0cRSzWFbg2PADtcoxfZqjZ6TOvQ7Vk,24551
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json",sha256=iNGsE2ZeVnQEnN4A8UJ9Jv0d3hbRF2MJ9oBgjup5Szk,2737
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=JJN0hryyLr5Zv3dSS7C8cPFhAwTT6XxUVnBGMZvV6JA,2752
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json",sha256=ouRyZ5PEMPP2njPftCNhs-1g1y6wueWLmhI7G1SjV1k,4131
@@ -132,6 +132,10 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=CYT3ujh5ifonhqQc1uYSa6maJ
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=yf33YmWlVSjjyg0Q4OMAWvc9gjRxvttMrQBUEOfPl4I,4153
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json",sha256=ZWMClYN1moVRUP2f0hYac38di_pUgZggyl9d2D5rnoc,4136
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=C65Q2Mv1LxFQ_qDnv11IZ9nwl7sGZo72nWDflMttu4g,4147
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=128,N=512,device_name=NVIDIA_H100_80GB_HBM3.json",sha256=FsWbV4Q6AzAtgegVuENBDz2ZcSJsqNiwUIVfQbpP7hQ,3244
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=144,N=512,device_name=NVIDIA_H100_80GB_HBM3.json",sha256=pk6VZChh2Y0CsJSzjtUhOnlta1QLTUEWy33aKQU47XY,3244
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=1024,device_name=NVIDIA_H100_80GB_HBM3.json",sha256=Gmk24hc5lVIfQtqSa5wLOcWKedMN8aZUe93DBh6J1AY,3249
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=1024,device_name=NVIDIA_H200.json",sha256=uY_XMPomaXMXxIkTR4ctU_Ybri_jMv2VvCcV-f6O_bw,3255
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=1344,device_name=NVIDIA_A100-SXM4-40GB.json",sha256=pCCKkdUzzuBVtljyk7AEIAbeDf12DUiieXaODZXzm5E,3254
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=1344,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=trX2-c4N6hTTD6zFNi6A2bT3FkhxKjkM2rPl-o1K9ss,3250
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json",sha256=I4d56uD7E1JMXD9RAxq3FebdPquDsnNEkVaIY9Ctm9w,3246
@@ -139,6 +143,7 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=CYT3ujh5ifonhqQc1uYSa6maJ
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=0ejgnIQ_mzJhKjQpnT-I1Vj9-rPfGlTcQ8u0cXgekUw,2746
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json",sha256=lU4fjngJmpzOafi-3_q0vj2pLfZQVVagFnZNoI97etk,4128
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=ilUf76TmbtaL3W0bara6JIBEiV_iPIs29UjLbH_GYtc,4145
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json",sha256=4ye1VakIIzrGUTxGNb_DM5v59djLgok8SeBeU_YI6Go,3252
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=2688,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=G4PKqWxh0MlBhg7QHKj0m--_fP3Ll0gs7VJaeg-NIDM,3254
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json",sha256=bKX9AvcxN6k-i3RUmHSchZZ3rjoYRYb4iBqhCI4L3MY,3257
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json",sha256=OeJUCPVBD1z9CSooZpy3hRyAasjKqFAQaTLcWK6PWno,2741
@@ -152,6 +157,8 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=CYT3ujh5ifonhqQc1uYSa6maJ
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json",sha256=kklgf2qLI5CQYiJJ5e9Gxx2gAfGxcyMDYpdJnIXPV8E,2748
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=800,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json",sha256=8e0tN_DHPwvh_HECVHx9oOF_4WWdaht4s6Nmd_K-aBU,2904
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=160,N=192,device_name=NVIDIA_A800-SXM4-80GB.json",sha256=fRzKfVFIcnxqu6DvGJQNltuFRRGz8F-eaL73bIzBzo8,3255
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=20,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json",sha256=zq1x3KwZ3YuzE42OPPsElXT_VS9sK135CdB1RsotLU4,3252
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=24,N=1024,device_name=NVIDIA_H100_80GB_HBM3.json",sha256=lJbEa7VsBVKL_iRoHniB-2nceMrPctkqtjKszPnX4pk,3248
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json",sha256=RuUDK9XfgXs1eZESWQR9ba4tu-rCRG_UCYwjaJ568sI,3264
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8.json",sha256=wlCi9aoYp7Zc1GThEutvWDbse0kKnNaQgFJsd_L8be0,3259
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json",sha256=tPYxeo_xUOkjQrZMdf9v4IaFrw0RGaZNLGLJPOhjE_g,3260
@@ -243,11 +250,11 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=CYT3ujh5ifonhqQc1uYSa6maJ
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=8192,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json",sha256=-RzUWSIAAsg6iA-8SPMa68hPpBVoUyMJs3dLP7edRu0,4323
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=8192,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json",sha256=sY2nWMPh9lsIkhPCjkHO245wpnfFbrHmzdcZDVFPVww,3265
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=8192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json",sha256=Uz5X80VcNBOaxshwVNUEittHk2zqB4HQCfTJ4TPG5aM,3274
-sglang/srt/layers/quantization/__init__.py,sha256=TxMVIyn2PiP_xrZQugOmzrfZ7DEKEjdNOuFhDNBWBys,12107
+sglang/srt/layers/quantization/__init__.py,sha256=yokDLpqQZ6eIIeaBZggJG-oS4h3TmroXZHLL40YykeM,12159
 sglang/srt/layers/quantization/awq.py,sha256=VImnVCU_QBLFba6S88T0dJ-vLy6SMm3OLIMEdllDfVI,6663
 sglang/srt/layers/quantization/base_config.py,sha256=jWk_egQrVNMYmQgbTI9vkcgzScLFjB5_sywFlAfE5J0,4776
-sglang/srt/layers/quantization/blockwise_int8.py,sha256=BS4nmo18QaC3vyCe9Wwe0Te-7FmFUe-udl7xbhRaU0s,14887
-sglang/srt/layers/quantization/fp8.py,sha256=J9P6SwZ1PIb3dEJoP7X2Uw5VOtU35vU5jugkoYyaZC4,41081
+sglang/srt/layers/quantization/blockwise_int8.py,sha256=yE8ARplbha1sW1Szl-mgsRDzGTRpEZY_zAKkCJIu680,15010
+sglang/srt/layers/quantization/fp8.py,sha256=J5D_KdRYiOQ4NCbjoKfYDHdIgCGMy-tQwHlTiG44pJc,41189
 sglang/srt/layers/quantization/fp8_kernel.py,sha256=JRalHJ-btDpzl3oXu2R_ZoJBu5TzBBmW_wKZDFs-usQ,24384
 sglang/srt/layers/quantization/fp8_utils.py,sha256=CDR2fLrZa_mZ86n5S2dDjYMpVCGa2n7gCXd2BYZjXcM,21391
 sglang/srt/layers/quantization/gptq.py,sha256=e4rMz374-yQQqeAI77WPxfcAaRk38GeN2akEpvnC_Do,15141
@@ -255,10 +262,10 @@ sglang/srt/layers/quantization/int8_kernel.py,sha256=GfRn_imIw8kNgqdtb2lr7Bettjg
 sglang/srt/layers/quantization/int8_utils.py,sha256=YK9CS-lb_n91kNCTKK5o5apYF31V2giDg5G5VKrpcUA,2356
 sglang/srt/layers/quantization/kv_cache.py,sha256=rJi6amyLZsquUMo_V5iLlPMqdsGTLgxh4popN1xUHCQ,4236
 sglang/srt/layers/quantization/modelopt_quant.py,sha256=mne4uKF0R-K0OvWN7X5ZxD4LdXKBc6GvmpZzIW6gkmM,6969
-sglang/srt/layers/quantization/moe_wna16.py,sha256=JlYCiy825MEMlBeQqqWwPuhoH_FbEcgTYlJUzrWj5IE,19248
+sglang/srt/layers/quantization/moe_wna16.py,sha256=3Z8Eq4_ehTN5EEotlYC09FpUNmF8VO8uv7QzUqJa0QI,19371
 sglang/srt/layers/quantization/utils.py,sha256=QqGFwRnFenOm5HfyLoS4D06_LyvNWgOggAiFtZXTpQ4,5637
 sglang/srt/layers/quantization/w8a8_fp8.py,sha256=XcQdgqXA3eKbAf-4_0I81Y5Nvjns3bQTocovnN8141w,6234
-sglang/srt/layers/quantization/w8a8_int8.py,sha256=V5vxn0wmUL1szj38lsJOKeNNEvFHisU7hZZLO4FfoNc,8733
+sglang/srt/layers/quantization/w8a8_int8.py,sha256=oLURfgMpsES8qLf0CIJ-4rfQgBGf452Lo0U6tvq6jH0,8856
 sglang/srt/layers/quantization/compressed_tensors/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py,sha256=ngKVSHfQUNSZzrLMu4Iv_4Fzt2eOoOIZKcO2RNDiwAM,25353
 sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py,sha256=roqRrIJybA9YuN3kqSeoLTJhXfTHOOtJd5MkenpOL8E,25835
@@ -458,6 +465,7 @@ sglang/srt/managers/multimodal_processors/janus_pro.py,sha256=wZs4HZhPov7yvV2VU2
 sglang/srt/managers/multimodal_processors/llava.py,sha256=8mac3vUUpVd12o43k1TyMaLEySZB915ks8Q5epeZmbg,6209
 sglang/srt/managers/multimodal_processors/minicpm.py,sha256=Mq-iH2j90VrGAbSaF3ayYWhTEm9RvWNI6ZhBb6G23dQ,5684
 sglang/srt/managers/multimodal_processors/mlama.py,sha256=MLiGS606LzVtdoXvjWGANx-K_7nE9J_fMVmkXN7Gz8k,1661
+sglang/srt/managers/multimodal_processors/mllama4.py,sha256=K6OKhSZOoaHwrRt0ZVi3gi2vnzMVHWJb5n3fUoStwIs,6188
 sglang/srt/managers/multimodal_processors/qwen_vl.py,sha256=67EmFiAkvZncU-eqiiS0Q4dr3pWcfI-RofYiQnNWvu0,5722
 sglang/srt/mem_cache/base_prefix_cache.py,sha256=NY62Zo0A0tLJ7ObRLOQqQcXCxoJUDZsK8f5U4dNQjKc,973
 sglang/srt/mem_cache/chunk_cache.py,sha256=it5SfL1FwMbrdeOH-I-Eu_i-I9hFB1xL-z_brIUoCkk,1835
@@ -470,7 +478,7 @@ sglang/srt/metrics/collector.py,sha256=aCxHqgsQ6P8ZxsAvq_MoEVsr3KUvIUSOBpGYMgBxm
 sglang/srt/metrics/func_timer.py,sha256=VFyNRrbnKVCwnQsrlLin1lITJfjQpf9m8sGPqL5LIsQ,3438
 sglang/srt/model_executor/cuda_graph_runner.py,sha256=bDLOqlxdwRUyKitG8JyZygnm05N00q-TdNiAayG_T8o,23223
 sglang/srt/model_executor/forward_batch_info.py,sha256=8VI1VxSmyH26lIHnCNeGqYw2XxslbqN_cuSUIEPUtRU,19468
-sglang/srt/model_executor/model_runner.py,sha256=f2e0Ts7eTy5_xCIp7q2If4Mi6Oi3nw45zyBPvNgR_-0,45006
+sglang/srt/model_executor/model_runner.py,sha256=4Xi-1u1tTC34uK_DtYEaj7VtvPjDDgMzRaXeJ5kpsQE,45076
 sglang/srt/model_loader/__init__.py,sha256=zGZkOBz1zx-pkaIy47BasL3fjDlAcxAXUTjInOhXHAE,919
 sglang/srt/model_loader/loader.py,sha256=AUS4SqSFghbQjs29C65lg7_zxR9h1t7N5G0gERjc0Rc,54238
 sglang/srt/model_loader/utils.py,sha256=0NaMR67fESFopaklmsleiL27XH1QUrjZW246MUu1EJ0,1369
@@ -497,7 +505,8 @@ sglang/srt/models/granite.py,sha256=nu_Zl_PYn188gk1uYVZ76y4wwHZV7G0w7uanhqpSFUs,
 sglang/srt/models/grok.py,sha256=pQOXtpHOYVntwt5QQRLffYsnMHmMfPMmGyKMfR0k0Ic,27994
 sglang/srt/models/internlm2.py,sha256=4eh9WVgK4yg13IsnH5qB2xUCWnixj_aLLz7qa_4m2_Q,13017
 sglang/srt/models/internlm2_reward.py,sha256=ndfGmyqYZbVZ7C7rJ-v9oK3wa-EpoBGybS8MlyKZi2E,2522
-sglang/srt/models/llama.py,sha256=FRdD-oCQl1M4hkYwIrWuECQ1tgxBq9rEIYOr-rlgswE,24616
+sglang/srt/models/llama.py,sha256=gcl2YtnM54J_fZQx2Z26LMm7vPbWN7N1CjzlaBEA3zk,24893
+sglang/srt/models/llama4.py,sha256=4WqHX6YPBrlJVA7HoQTMUfdoU_mEhpWSgoFaeKdhdCE,15018
 sglang/srt/models/llama_classification.py,sha256=4QWTFaUZIFKYZvEzs8bx8VkOZNIwdYCLrnwrdAw4QK0,3108
 sglang/srt/models/llama_eagle.py,sha256=OB2lKsjn7BcfCZljklnhk83me8j0PuQmYLou7baNcq4,4866
 sglang/srt/models/llama_eagle3.py,sha256=v3bftBVDIGjnzngQYnu19cy0J_3w7yruHqLP5nsAQDM,6642
@@ -513,6 +522,7 @@ sglang/srt/models/mistral.py,sha256=EYifJUUzN2Z2-iL37eJiNZF_DB0H4pa0mKlgYRIxM70,
 sglang/srt/models/mixtral.py,sha256=6Fse2J-20IMylP-yzpEihIinaH37TmmslATbLcWBRYY,14926
 sglang/srt/models/mixtral_quant.py,sha256=MSa6UKPbgv8Rn8Iv8o1dQhcstAHLNQzE0eepFx_hYSw,15221
 sglang/srt/models/mllama.py,sha256=SsK_cEolaeoXh_HkyXsSF2ueYR3sPv1NvnGH2k6Aqx0,38461
+sglang/srt/models/mllama4.py,sha256=E2mCxJ1zCt6Io4LL4Rtt5uqMj7Jy971234ZcuyJZxSo,5800
 sglang/srt/models/olmo.py,sha256=FJk8A3T3TF5QcTV6rMP8np94QtvxpMWlgCsv_5VwpVE,12632
 sglang/srt/models/olmo2.py,sha256=U0ScFzWazOrb_Q90sfXkpVNAsXT-pgZbNgGh80R40VE,14288
 sglang/srt/models/olmoe.py,sha256=tx5OKWLOr6_pohe2eBcIodCmcuSjtpteHq_tG_QVYCY,15910
@@ -569,8 +579,8 @@ sglang/test/test_programs.py,sha256=VZ3vXtUDBnXz0M7gFdDH8hXg9Wa0j_qI8CVqjEgRN_E,
 sglang/test/test_utils.py,sha256=jUkIDxJ7I8hCPk0XF7F_IWJkOtn6O7eXJG5pI0cduwo,30463
 sglang/test/attention/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/test/attention/test_flashattn_backend.py,sha256=OxS1KsPs19nwZcDtdURj7_liT1cIfEXb6W4FH9KMaaE,10808
-sglang-0.4.4.post4.dist-info/licenses/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
-sglang-0.4.4.post4.dist-info/METADATA,sha256=gHc8ZNyeSnw_PrtGOQ7SAVJwCqonaOP4k0vADEpMqv4,25067
-sglang-0.4.4.post4.dist-info/WHEEL,sha256=CmyFI0kx5cdEMTLiONQRbGQwjIoR1aIYB7eCAQ4KPJ0,91
-sglang-0.4.4.post4.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
-sglang-0.4.4.post4.dist-info/RECORD,,
+sglang-0.4.5.dist-info/licenses/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
+sglang-0.4.5.dist-info/METADATA,sha256=dFvXPJ-aE-juLKgxD5l8wflGgO1cHg2jHjScLX_Ftjw,25061
+sglang-0.4.5.dist-info/WHEEL,sha256=CmyFI0kx5cdEMTLiONQRbGQwjIoR1aIYB7eCAQ4KPJ0,91
+sglang-0.4.5.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
+sglang-0.4.5.dist-info/RECORD,,