sglang 0.3.1.post2__py3-none-any.whl → 0.3.2__py3-none-any.whl

sglang/srt/models/grok.py

@@ -28,12 +28,6 @@ from vllm.distributed import (
     get_tensor_model_parallel_rank,
     get_tensor_model_parallel_world_size,
 )
-from vllm.model_executor.layers.linear import (
-    QKVParallelLinear,
-    ReplicatedLinear,
-    RowParallelLinear,
-)
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -44,7 +38,13 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.fused_moe import FusedMoE
 from sglang.srt.layers.layernorm import RMSNorm
+from sglang.srt.layers.linear import (
+    QKVParallelLinear,
+    ReplicatedLinear,
+    RowParallelLinear,
+)
 from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
 

sglang/srt/models/internlm2.py

@@ -23,12 +23,6 @@ from torch import nn
 from transformers import PretrainedConfig
 from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
-from vllm.model_executor.layers.linear import (
-    MergedColumnParallelLinear,
-    QKVParallelLinear,
-    RowParallelLinear,
-)
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -38,7 +32,13 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.layernorm import RMSNorm
+from sglang.srt.layers.linear import (
+    MergedColumnParallelLinear,
+    QKVParallelLinear,
+    RowParallelLinear,
+)
 from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
 

sglang/srt/models/llama.py

@@ -24,12 +24,6 @@ from torch import nn
 from transformers import LlamaConfig
 from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
-from vllm.model_executor.layers.linear import (
-    MergedColumnParallelLinear,
-    QKVParallelLinear,
-    RowParallelLinear,
-)
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -39,7 +33,13 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.layernorm import RMSNorm
+from sglang.srt.layers.linear import (
+    MergedColumnParallelLinear,
+    QKVParallelLinear,
+    RowParallelLinear,
+)
 from sglang.srt.layers.logits_processor import LogitsProcessor, LogitsProcessorOutput
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.layers.torchao_utils import apply_torchao_config_
 from sglang.srt.managers.schedule_batch import global_server_args_dict
@@ -403,6 +403,14 @@ class LlamaForCausalLM(nn.Module):
                 weight_loader = getattr(param, "weight_loader", default_weight_loader)
                 weight_loader(param, loaded_weight)
 
+        if (
+            hasattr(self.config, "tie_word_embeddings")
+            and self.config.tie_word_embeddings
+        ):
+            # Tie output embedding layer to input embedding layer, to solve issues where lm_head.weight is missing
+            param = self.lm_head.weight
+            weight_loader = getattr(param, "weight_loader", default_weight_loader)
+            weight_loader(param, self.model.embed_tokens.weight)
         apply_torchao_config_(self, params_dict, set(["proj.weight"]))
 
 

sglang/srt/models/llama_classification.py

@@ -19,10 +19,10 @@ import torch
 from torch import nn
 from transformers import LlamaConfig
 from vllm.config import CacheConfig
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessorOutput
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
 from sglang.srt.models.llama import LlamaForCausalLM, LlamaModel
 

sglang/srt/models/llava.py

@@ -32,9 +32,9 @@ from transformers import (
 )
 from transformers.models.llava.modeling_llava import LlavaMultiModalProjector
 from vllm.config import CacheConfig
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.mm_utils import (
     get_anyres_image_grid_shape,
     unpad_image,

sglang/srt/models/llavavid.py

@@ -23,9 +23,9 @@ from torch import nn
 from transformers import CLIPVisionModel, LlavaConfig
 from transformers.models.llava.modeling_llava import LlavaMultiModalProjector
 from vllm.config import CacheConfig
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.model_executor.forward_batch_info import ForwardMode, InputMetadata
 from sglang.srt.models.llama import LlamaForCausalLM
 

sglang/srt/models/minicpm.py

@@ -22,12 +22,6 @@ import torch
 from torch import nn
 from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
-from vllm.model_executor.layers.linear import (
-    MergedColumnParallelLinear,
-    QKVParallelLinear,
-    RowParallelLinear,
-)
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -37,7 +31,13 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.layernorm import RMSNorm
+from sglang.srt.layers.linear import (
+    MergedColumnParallelLinear,
+    QKVParallelLinear,
+    RowParallelLinear,
+)
 from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
 

sglang/srt/models/minicpm3.py

@@ -29,7 +29,6 @@ from vllm.model_executor.layers.linear import (
     ReplicatedLinear,
     RowParallelLinear,
 )
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -40,6 +39,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.layernorm import RMSNorm
 from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.managers.schedule_batch import global_server_args_dict
 from sglang.srt.model_executor.forward_batch_info import InputMetadata

sglang/srt/models/mixtral.py

@@ -24,12 +24,6 @@ from transformers import MixtralConfig
 from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.fused_moe import FusedMoE
-from vllm.model_executor.layers.linear import (
-    QKVParallelLinear,
-    ReplicatedLinear,
-    RowParallelLinear,
-)
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     DEFAULT_VOCAB_PADDING_SIZE,
@@ -39,7 +33,13 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.layernorm import RMSNorm
+from sglang.srt.layers.linear import (
+    QKVParallelLinear,
+    ReplicatedLinear,
+    RowParallelLinear,
+)
 from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.layers.torchao_utils import apply_torchao_config_
 from sglang.srt.managers.schedule_batch import global_server_args_dict

sglang/srt/models/mixtral_quant.py

@@ -29,12 +29,6 @@ from vllm.distributed import (
     get_tensor_model_parallel_world_size,
     tensor_model_parallel_all_reduce,
 )
-from vllm.model_executor.layers.linear import (
-    QKVParallelLinear,
-    ReplicatedLinear,
-    RowParallelLinear,
-)
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -43,7 +37,13 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.layernorm import RMSNorm
+from sglang.srt.layers.linear import (
+    QKVParallelLinear,
+    ReplicatedLinear,
+    RowParallelLinear,
+)
 from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
 

sglang/srt/models/olmoe.py

@@ -35,7 +35,6 @@ from vllm.model_executor.layers.linear import (
     ReplicatedLinear,
     RowParallelLinear,
 )
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -47,6 +46,7 @@ from vllm.utils import print_warning_once
 from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.layernorm import RMSNorm
 from sglang.srt.layers.logits_processor import LogitsProcessor, LogitsProcessorOutput
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
 

sglang/srt/models/qwen.py

@@ -22,12 +22,6 @@ from torch import nn
 from transformers import PretrainedConfig
 from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
-from vllm.model_executor.layers.linear import (
-    MergedColumnParallelLinear,
-    QKVParallelLinear,
-    RowParallelLinear,
-)
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -37,7 +31,13 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.layernorm import RMSNorm
+from sglang.srt.layers.linear import (
+    MergedColumnParallelLinear,
+    QKVParallelLinear,
+    RowParallelLinear,
+)
 from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
 

sglang/srt/models/qwen2.py

@@ -22,12 +22,6 @@ import torch
 from torch import nn
 from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
-from vllm.model_executor.layers.linear import (
-    MergedColumnParallelLinear,
-    QKVParallelLinear,
-    RowParallelLinear,
-)
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -37,8 +31,14 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.layernorm import RMSNorm
+from sglang.srt.layers.linear import (
+    MergedColumnParallelLinear,
+    QKVParallelLinear,
+    RowParallelLinear,
+)
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.pooler import Pooler, PoolingType
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
 

sglang/srt/models/qwen2_moe.py

@@ -29,13 +29,6 @@ from vllm.distributed import (
     tensor_model_parallel_all_reduce,
 )
 from vllm.model_executor.layers.fused_moe import FusedMoE
-from vllm.model_executor.layers.linear import (
-    MergedColumnParallelLinear,
-    QKVParallelLinear,
-    ReplicatedLinear,
-    RowParallelLinear,
-)
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -45,7 +38,14 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.layernorm import RMSNorm
+from sglang.srt.layers.linear import (
+    MergedColumnParallelLinear,
+    QKVParallelLinear,
+    ReplicatedLinear,
+    RowParallelLinear,
+)
 from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.layers.torchao_utils import apply_torchao_config_
 from sglang.srt.managers.schedule_batch import global_server_args_dict

sglang/srt/models/stablelm.py

@@ -24,12 +24,6 @@ from torch import nn
 from transformers import PretrainedConfig
 from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
-from vllm.model_executor.layers.linear import (
-    MergedColumnParallelLinear,
-    QKVParallelLinear,
-    RowParallelLinear,
-)
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -38,7 +32,13 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.activation import SiluAndMul
+from sglang.srt.layers.linear import (
+    MergedColumnParallelLinear,
+    QKVParallelLinear,
+    RowParallelLinear,
+)
 from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
 

sglang/srt/models/xverse.py

@@ -31,7 +31,6 @@ from vllm.model_executor.layers.linear import (
     QKVParallelLinear,
     RowParallelLinear,
 )
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -40,6 +39,7 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.model_runner import InputMetadata
 

sglang/srt/models/xverse_moe.py

@@ -34,7 +34,6 @@ from vllm.model_executor.layers.linear import (
     ReplicatedLinear,
     RowParallelLinear,
 )
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -43,6 +42,7 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
 

sglang/srt/models/yivl.py

@@ -21,9 +21,9 @@ import torch
 import torch.nn as nn
 from transformers import CLIPVisionModel, LlavaConfig
 from vllm.config import CacheConfig
-from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.models.llava import LlavaLlamaForCausalLM
 
 

sglang/srt/openai_api/adapter.py

@@ -858,11 +858,18 @@ def v1_chat_generate_request(
                                 openai_compatible_messages.append(
                                     {"role": message.role, "content": content["text"]}
                                 )
+                if openai_compatible_messages[-1]["role"] == "assistant":
+                    assistant_prefix = openai_compatible_messages[-1]["content"]
+                    openai_compatible_messages = openai_compatible_messages[:-1]
+                else:
+                    assistant_prefix = None
                 prompt_ids = tokenizer_manager.tokenizer.apply_chat_template(
                     openai_compatible_messages,
                     tokenize=True,
                     add_generation_prompt=True,
                 )
+                if assistant_prefix:
+                    prompt_ids += tokenizer_manager.tokenizer.encode(assistant_prefix)
                 stop = request.stop
                 image_data = None
                 modalities = []

sglang/srt/utils.py

@@ -26,7 +26,7 @@ import struct
 import time
 from importlib.metadata import PackageNotFoundError, version
 from io import BytesIO
-from typing import List, Optional, Union
+from typing import Any, Dict, List, Optional, Union
 
 import numpy as np
 import psutil
@@ -682,3 +682,23 @@ def replace_submodule(
     target_name = module_name.split(".")[-1]
     setattr(parent, target_name, new_module)
     return new_module
+
+
+def set_weight_attrs(
+    weight: torch.Tensor,
+    weight_attrs: Optional[Dict[str, Any]],
+):
+    """Set attributes on a weight tensor.
+
+    This method is used to set attributes on a weight tensor. This method
+    will not overwrite existing attributes.
+
+    Args:
+        weight: The weight tensor.
+        weight_attrs: A dictionary of attributes to set on the weight tensor.
+    """
+    if weight_attrs is None:
+        return
+    for key, value in weight_attrs.items():
+        assert not hasattr(weight, key), f"Overwriting existing tensor attribute: {key}"
+        setattr(weight, key, value)

sglang/test/runners.py

@@ -21,19 +21,19 @@ from typing import List, Union
 
 import torch
 import torch.nn.functional as F
-from peft import PeftModel
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM
 
+from sglang.srt.hf_transformers_utils import get_tokenizer
 from sglang.srt.server import Runtime
 from sglang.test.test_utils import DEFAULT_PORT_FOR_SRT_TEST_RUNNER
 
 DEFAULT_PROMPTS = [
-    # the output of gemma-2-2b from SRT is unstable on the commented prompt
-    # "The capital of France is",
     "Apple is red. Banana is Yellow. " * 800 + "Apple is",
     "The capital of the United Kingdom is",
     "Today is a sunny day and I like",
     "AI is a field of computer science focused on",
+    # the output of gemma-2-2b from SRT is unstable on the commented prompt
+    # "The capital of France is",
 ]
 
 dirpath = os.path.dirname(__file__)
@@ -93,11 +93,7 @@ class HFRunner:
         self.model_proc.start()
 
     def start_model_process(self, in_queue, out_queue, model_path, torch_dtype):
-        self.tokenizer = AutoTokenizer.from_pretrained(
-            model_path,
-            torch_dtype=torch_dtype,
-        )
-
+        self.tokenizer = get_tokenizer(model_path)
         if self.is_generation:
             self.base_model = AutoModelForCausalLM.from_pretrained(
                 model_path,
@@ -132,6 +128,8 @@ class HFRunner:
                             input_ids = torch.tensor([p], device="cuda")
 
                         if lora_paths is not None and lora_paths[i] is not None:
+                            from peft import PeftModel
+
                             self.model = PeftModel.from_pretrained(
                                 self.base_model,
                                 lora_paths[i],

sglang/test/test_utils.py

@@ -25,11 +25,14 @@ from sglang.utils import get_exception_traceback
 DEFAULT_FP8_MODEL_NAME_FOR_TEST = "neuralmagic/Meta-Llama-3.1-8B-FP8"
 DEFAULT_MODEL_NAME_FOR_TEST = "meta-llama/Meta-Llama-3.1-8B-Instruct"
 DEFAULT_MOE_MODEL_NAME_FOR_TEST = "mistralai/Mixtral-8x7B-Instruct-v0.1"
+DEFAULT_MLA_MODEL_NAME_FOR_TEST = "deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct"
+DEFAULT_MLA_FP8_MODEL_NAME_FOR_TEST = "neuralmagic/DeepSeek-Coder-V2-Lite-Instruct-FP8"
 DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH = 600
 DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_TP1 = "meta-llama/Meta-Llama-3.1-8B-Instruct,mistralai/Mistral-7B-Instruct-v0.3,deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct,google/gemma-2-27b-it"
-DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_TP2 = "meta-llama/Meta-Llama-3.1-70B-Instruct,mistralai/Mixtral-8x7B-Instruct-v0.1,Qwen/Qwen2-57B-A14B-Instruct"
+DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_TP2 = "meta-llama/Meta-Llama-3.1-70B-Instruct,mistralai/Mixtral-8x7B-Instruct-v0.1,Qwen/Qwen2-57B-A14B-Instruct,deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct"
 DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_FP8_TP1 = "neuralmagic/Meta-Llama-3.1-8B-Instruct-FP8,neuralmagic/Mistral-7B-Instruct-v0.3-FP8,neuralmagic/DeepSeek-Coder-V2-Lite-Instruct-FP8,neuralmagic/gemma-2-2b-it-FP8"
-DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_FP8_TP2 = "neuralmagic/Meta-Llama-3.1-70B-Instruct-FP8,neuralmagic/Mixtral-8x7B-Instruct-v0.1-FP8,neuralmagic/Qwen2-72B-Instruct-FP8,neuralmagic/Qwen2-57B-A14B-Instruct-FP8"
+DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_FP8_TP2 = "neuralmagic/Meta-Llama-3.1-70B-Instruct-FP8,neuralmagic/Mixtral-8x7B-Instruct-v0.1-FP8,neuralmagic/Qwen2-72B-Instruct-FP8,neuralmagic/Qwen2-57B-A14B-Instruct-FP8,neuralmagic/DeepSeek-Coder-V2-Lite-Instruct-FP8"
+DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_QUANT_TP1 = "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4,hugging-quants/Meta-Llama-3.1-8B-Instruct-GPTQ-INT4"
 
 
 def is_in_ci():
@@ -585,3 +588,37 @@ def run_bench_latency(model, other_args):
         kill_child_process(process.pid)
 
     return output_throughput
+
+
+def lcs(X, Y):
+    m = len(X)
+    n = len(Y)
+    L = [[0] * (n + 1) for _ in range(m + 1)]
+
+    for i in range(m + 1):
+        for j in range(n + 1):
+            if i == 0 or j == 0:
+                L[i][j] = 0
+            elif X[i - 1] == Y[j - 1]:
+                L[i][j] = L[i - 1][j - 1] + 1
+            else:
+                L[i][j] = max(L[i - 1][j], L[i][j - 1])
+
+    return L[m][n]
+
+
+def calculate_rouge_l(output_strs_list1, output_strs_list2):
+    """calculate the ROUGE-L score"""
+    rouge_l_scores = []
+
+    for s1, s2 in zip(output_strs_list1, output_strs_list2):
+        lcs_len = lcs(s1, s2)
+        precision = lcs_len / len(s1) if len(s1) > 0 else 0
+        recall = lcs_len / len(s2) if len(s2) > 0 else 0
+        if precision + recall > 0:
+            fmeasure = (2 * precision * recall) / (precision + recall)
+        else:
+            fmeasure = 0.0
+        rouge_l_scores.append(fmeasure)
+
+    return rouge_l_scores

sglang/version.py

@@ -1 +1 @@
-__version__ = "0.3.1.post2"
+__version__ = "0.3.2"

{sglang-0.3.1.post2.dist-info → sglang-0.3.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sglang
-Version: 0.3.1.post2
+Version: 0.3.2
 Summary: SGLang is yet another fast serving framework for large language models and vision language models.
 License: Apache License
                                    Version 2.0, January 2004
@@ -318,7 +318,7 @@ pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.4/
 ### Method 2: From source
 ```
 # Use the last release branch
-git clone -b v0.3.1.post2 https://github.com/sgl-project/sglang.git
+git clone -b v0.3.2 https://github.com/sgl-project/sglang.git
 cd sglang
 
 pip install --upgrade pip
@@ -348,9 +348,9 @@ docker run --gpus all \
 <summary>More</summary>
 
 > This method is recommended if you plan to serve it as a service.
-> A better approach is to use the [k8s-sglang-service.yaml](./docker/k8s-sglang-service.yaml).
+> A better approach is to use the [k8s-sglang-service.yaml](docker/k8s-sglang-service.yaml).
 
-1. Copy the [compose.yml](./docker/compose.yaml) to your local machine
+1. Copy the [compose.yml](docker/compose.yaml) to your local machine
 2. Execute the command `docker compose up -d` in your terminal.
 </details>
 
@@ -499,6 +499,7 @@ python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct
 - Llama / Llama 2 / Llama 3 / Llama 3.1
 - Mistral / Mixtral / Mistral NeMo
 - Gemma / Gemma 2
+- OLMoE
 - Qwen / Qwen 2 / Qwen 2 MoE
 - DeepSeek / DeepSeek 2
 - [LLaVA-OneVision](https://llava-vl.github.io/blog/2024-08-05-llava-onevision/)
@@ -520,6 +521,7 @@ python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct
 - BaiChuan2
 - MiniCPM / MiniCPM 3
 - XVERSE / XVERSE MoE
+- SmolLM
 
 
 **Embedding Models**
@@ -528,7 +530,7 @@ python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct
 - gte-Qwen2
   - `python -m sglang.launch_server --model-path Alibaba-NLP/gte-Qwen2-7B-instruct --is-embedding`
 
-Instructions for supporting a new model are [here](https://github.com/sgl-project/sglang/blob/main/docs/en/model_support.md).
+Instructions for supporting a new model are [here](docs/en/model_support.md).
 
 #### Use Models From ModelScope
 <details>
@@ -823,7 +825,7 @@ def chat_example(s):
 Learn more at this [blog](https://lmsys.org/blog/2024-07-25-sglang-llama3/).
 
 ## Roadmap
-[Development Roadmap (2024 Q3)](https://github.com/sgl-project/sglang/issues/634)
+[Development Roadmap (2024 Q4)](https://github.com/sgl-project/sglang/issues/1487)
 
 ## Citation And Acknowledgment
 Please cite our paper, [SGLang: Efficient Execution of Structured Language Model Programs](https://arxiv.org/abs/2312.07104), if you find the project useful.