Suppport qwen model and solve some problems (#75)

README.md

@@ -316,6 +316,7 @@ python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --port
 - Mixtral
 - LLaVA
   - `python3 -m sglang.launch_server --model-path liuhaotian/llava-v1.5-7b --tokenizer-path llava-hf/llava-1.5-7b-hf --port 30000`
+- Qwen
 - AWQ quantization
 
 ## Benchmark And Performance

python/sglang/srt/layers/radix_attention.py

@@ -61,7 +61,6 @@ class RadixAttention(nn.Module):
     def extend_forward_triton(self, q, k, v, input_metadata: InputMetadata):
         o = torch.empty_like(q)
         self.store_kv_cache(k, v, input_metadata)
-
         extend_attention_fwd(
             q.view(-1, self.tp_q_head_num, self.head_dim),
             k.contiguous(),

python/sglang/srt/managers/detokenizer_manager.py

@@ -55,6 +55,7 @@ class DetokenizerManager:
                         first_token = self.tokenizer.convert_ids_to_tokens(
                             int(output_tokens[i][0])
                         )
+                        first_token = first_token.decode("utf-8")
                         if first_token.startswith("▁"):
                             output_strs[i] = " " + output_strs[i]
 

python/sglang/srt/managers/router/model_runner.py

@@ -240,6 +240,7 @@ class ModelRunner:
         from sglang.srt.models.llama2 import LlamaForCausalLM
         from sglang.srt.models.llava import LlavaLlamaForCausalLM
         from sglang.srt.models.mixtral import MixtralForCausalLM
+        from sglang.srt.models.qwen import QWenLMHeadModel
 
         # Select model class
         architectures = getattr(self.model_config.hf_config, "architectures", [])
@@ -258,6 +259,9 @@ class ModelRunner:
             if arch == "MixtralForCausalLM":
                 model_class = MixtralForCausalLM
                 break
+            if arch == "QWenLMHeadModel":
+                model_class = QWenLMHeadModel
+                break
         if model_class is None:
             raise ValueError(f"Unsupported architectures: {architectures}")
 

python/sglang/srt/model_config.py

@@ -20,8 +20,10 @@ class ModelConfig:
         # Unify the config keys for hf_config
         self.context_len = get_context_length(self.hf_config)
         self.head_dim = self.hf_config.hidden_size // self.hf_config.num_attention_heads
-        self.num_key_value_heads = self.hf_config.num_key_value_heads
         self.num_attention_heads = self.hf_config.num_attention_heads
+        self.num_key_value_heads = getattr(self.hf_config, "num_key_value_heads", None)
+        if self.num_key_value_heads is None:
+            self.num_key_value_heads = self.num_attention_heads
         self.hidden_size = self.hf_config.hidden_size
         self.num_hidden_layers = self.hf_config.num_hidden_layers
         self.vocab_size = self.hf_config.vocab_size

python/sglang/srt/models/qwen.py

+from typing import Any, Dict, List, Optional, Tuple
+
+import torch
+from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.radix_attention import RadixAttention
+from sglang.srt.managers.router.model_runner import InputMetadata
+from torch import nn
+from vllm.transformers_utils.configs.qwen import QWenConfig
+from vllm.model_executor.layers.activation import SiluAndMul
+from vllm.model_executor.layers.layernorm import RMSNorm
+from vllm.model_executor.layers.linear import (
+    LinearMethodBase,
+    MergedColumnParallelLinear,
+    QKVParallelLinear,
+    RowParallelLinear,
+)
+from vllm.model_executor.layers.rotary_embedding import get_rope
+from vllm.model_executor.layers.vocab_parallel_embedding import (
+    ParallelLMHead,
+    VocabParallelEmbedding,
+)
+from vllm.model_executor.parallel_utils.parallel_state import (
+    get_tensor_model_parallel_world_size,
+)
+from vllm.model_executor.weight_utils import (
+    default_weight_loader,
+    hf_model_weights_iterator,
+)
+
+class QWenMLP(nn.Module):
+
+    def __init__(
+        self,
+        hidden_size: int,
+        intermediate_size: int,
+        hidden_act: str = "silu",
+    ):
+        super().__init__()
+        self.gate_up_proj = MergedColumnParallelLinear(
+            hidden_size,
+            2 * [intermediate_size],
+            bias=False,
+            gather_output=False,
+        )
+        self.c_proj = RowParallelLinear(
+            intermediate_size,
+            hidden_size,
+            bias=False,
+            input_is_parallel=True,
+        )
+        if hidden_act != "silu":
+            raise ValueError(f"Unsupported activation: {hidden_act}. "
+                             "Only silu is supported for now.")
+        self.act_fn = SiluAndMul()
+
+    def forward(self, x):
+        gate_up, _ = self.gate_up_proj(x)
+        x = self.act_fn(gate_up)
+        x, _ = self.c_proj(x)
+        return x
+
+class QWenAttention(nn.Module):
+
+    def __init__(self,
+                 hidden_size: int,
+                 num_heads: int,
+                 max_position_embeddings: int,
+                 layer_id: int = 0,
+                 rope_theta: float = 10000,
+                 rope_scaling: Optional[Dict[str, Any]] = None):
+        super().__init__()
+        self.hidden_size = hidden_size
+        tensor_model_parallel_world_size = get_tensor_model_parallel_world_size(
+        )
+        self.total_num_heads = num_heads
+        assert self.total_num_heads % tensor_model_parallel_world_size == 0
+        self.num_heads = (self.total_num_heads //
+                          tensor_model_parallel_world_size)
+        self.head_dim = hidden_size // self.total_num_heads
+
+        # pylint: disable=invalid-name
+        self.c_attn = QKVParallelLinear(
+            hidden_size,
+            self.head_dim,
+            self.total_num_heads,
+            bias=True
+        )
+        self.c_proj = RowParallelLinear(
+            self.total_num_heads * self.head_dim,
+            hidden_size,
+            bias=False,
+            input_is_parallel=True,
+        )
+        self.rotary_emb = get_rope(
+            self.head_dim,
+            rotary_dim=self.head_dim,
+            max_position=max_position_embeddings,
+            base=rope_theta,
+            rope_scaling=rope_scaling,
+        )
+        self.scaling = self.head_dim**-0.5
+        self.attn = RadixAttention(
+            self.num_heads,
+            self.head_dim,
+            self.scaling,
+            num_kv_heads=self.num_heads,
+            layer_id=layer_id,
+        )
+
+    def forward(
+        self,
+        positions: torch.Tensor,
+        hidden_states: torch.Tensor,
+        input_metadata: InputMetadata,
+    ) -> torch.Tensor:
+        qkv, _ = self.c_attn(hidden_states)
+        q, k, v = qkv.chunk(chunks=3, dim=-1)
+        q, k = self.rotary_emb(positions, q, k)
+        attn_output = self.attn(q, k, v, input_metadata)
+        output, _ = self.c_proj(attn_output)
+        return output
+
+class QWenBlock(nn.Module):
+
+    def __init__(self, config: QWenConfig,layer_id):
+        super().__init__()
+        self.ln_1 = RMSNorm(config.hidden_size, eps=config.layer_norm_epsilon)
+
+        rope_theta = getattr(config, "rope_theta", 10000)
+        rope_scaling = getattr(config, "rope_scaling", None)
+        self.attn = QWenAttention(config.hidden_size,
+                                  config.num_attention_heads,
+                                  config.max_position_embeddings,
+                                  rope_theta=rope_theta,
+                                  rope_scaling=rope_scaling,
+                                  layer_id=layer_id)
+
+        self.ln_2 = RMSNorm(config.hidden_size, eps=config.layer_norm_epsilon)
+
+        self.mlp = QWenMLP(config.hidden_size, config.intermediate_size // 2)
+
+    def forward(
+        self,
+        positions: torch.Tensor,
+        hidden_states: torch.Tensor,
+        input_metadata: InputMetadata,
+    ) -> torch.Tensor:
+        # Self Attention
+        residual = hidden_states
+        hidden_states = self.ln_1(hidden_states)
+        hidden_states = self.attn(
+            positions=positions,
+            hidden_states=hidden_states,
+            input_metadata=input_metadata,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.ln_2(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+        return hidden_states
+    
+class QWenModel(nn.Module):
+
+    def __init__(self, config:QWenConfig):
+        super().__init__()
+        self.config = config
+        self.vocab_size = config.vocab_size
+
+        vocab_size = ((config.vocab_size + 63) // 64) * 64
+        self.wte = VocabParallelEmbedding(
+            vocab_size,
+            config.hidden_size,
+        )
+        self.h = nn.ModuleList(
+            [QWenBlock(config, i) for i in range(config.num_hidden_layers)])
+        self.ln_f = RMSNorm(config.hidden_size, eps=config.layer_norm_epsilon)
+
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        positions: torch.Tensor,
+        input_metadata: InputMetadata,
+    ) -> torch.Tensor:
+        hidden_states = self.wte(input_ids)
+        for i in range(len(self.h)):
+            layer = self.h[i]
+            hidden_states = layer(
+                positions,
+                hidden_states,
+                input_metadata,
+            )
+        hidden_states = self.ln_f(hidden_states)
+        return hidden_states
+
+class QWenLMHeadModel(nn.Module):
+
+    def __init__(self, config: QWenConfig,linear_method=None):
+        super().__init__()
+        self.config = config
+        self.transformer = QWenModel(config)
+        vocab_size = ((config.vocab_size + 63) // 64) * 64
+        self.lm_head = ParallelLMHead(
+            vocab_size,
+            config.hidden_size
+        )
+        self.logits_processor = LogitsProcessor(config)
+
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        positions: torch.Tensor,
+        input_metadata: InputMetadata
+    ):
+        hidden_states = self.transformer(input_ids, positions,input_metadata)
+        next_tokens = self.logits_processor(
+            input_ids, hidden_states, self.lm_head.weight, input_metadata
+        )
+        return next_tokens
+
+    _column_parallel_weights = []
+    _row_parallel_weights = ["c_proj.weight"]
+
+    def load_weights(
+        self,
+        model_name_or_path: str,
+        cache_dir: Optional[str] = None,
+        load_format: str = "auto",
+        revision: Optional[str] = None,
+    ):
+        stacked_params_mapping = [
+            # (param_name, shard_name, shard_id)
+            ("gate_up_proj", "w2", 0),
+            ("gate_up_proj", "w1", 1),
+        ]
+        params_dict = dict(self.named_parameters())
+        for name, loaded_weight in hf_model_weights_iterator(
+            model_name_or_path, cache_dir, load_format, revision
+        ):
+            if "rotary_emb.inv_freq" in name:
+                continue
+            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)
+                # Skip loading extra bias for GPTQ models.
+                if name.endswith(".bias") and name not in params_dict:
+                    continue
+                param = params_dict[name]
+                weight_loader = param.weight_loader
+                weight_loader(param, loaded_weight, shard_id)
+                break
+            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)

python/sglang/srt/utils.py

@@ -108,9 +108,11 @@ def get_exception_traceback():
 def get_int_token_logit_bias(tokenizer, vocab_size):
     from transformers import LlamaTokenizer, LlamaTokenizerFast
 
+    # a bug when model's vocab size > tokenizer.vocab_size
+    vocab_size = tokenizer.vocab_size
     logit_bias = np.zeros(vocab_size, dtype=np.float32)
     for t_id in range(vocab_size):
-        ss = tokenizer.decode(t_id).strip()
+        ss = tokenizer.decode([t_id]).strip()
         if not (ss.isdigit() or len(ss) == 0 or t_id == tokenizer.eos_token_id):
             logit_bias[t_id] = -1e5
         # else: