[SLM] Deepseek Multi-GPU support (#2988)

This PR supports TP function of Deepseek Model

Author: tlopex

python/mlc_llm/model/deepseek/deepseek_model.py

@@ -15,6 +15,7 @@
 from mlc_llm.nn import PagedKVCache, RopeMode
 from mlc_llm.nn.expert import MixtralExperts
 from mlc_llm.support import logging
+from mlc_llm.support import tensor_parallel as tp
 from mlc_llm.support.config import ConfigBase
 from mlc_llm.support.style import bold
 
@@ -48,6 +49,7 @@ class DeepseekConfig(ConfigBase):  # pylint: disable=too-many-instance-attribute
     context_window_size: int = 0
     prefill_chunk_size: int = 0
     tensor_parallel_shards: int = 1
+    head_dim: int = 0
     max_batch_size: int = 1
     num_experts_per_tok: int = 0
     kwargs: Dict[str, Any] = dataclasses.field(default_factory=dict)
@@ -70,6 +72,9 @@ def __post_init__(self):
                     "`context_window_size`, `max_position_embeddings` or `max_sequence_length` is "
                     "provided in `config.json`."
                 )
+        if self.head_dim == 0:
+            self.head_dim = self.hidden_size // self.num_attention_heads
+        assert self.head_dim * self.num_attention_heads == self.hidden_size
         if self.prefill_chunk_size == 0:
             logger.info(
                 "%s defaults to %d",
@@ -85,7 +90,6 @@ def __post_init__(self):
                 min(self.context_window_size, 2048),
             )
             self.prefill_chunk_size = min(self.context_window_size, 2048)
-        assert self.tensor_parallel_shards == 1, "Deepseek currently does not support sharding."
 
 
 # pylint: disable=invalid-name,missing-docstring
@@ -96,17 +100,18 @@ def __init__(self, config: DeepseekConfig):
         super().__init__()  # Make sure to call the parent class constructor
         self.hidden_size = config.hidden_size
         self.rope_theta = config.rope_theta
+        self.tensor_parallel_shards = config.tensor_parallel_shards
         if config.num_attention_heads % config.tensor_parallel_shards != 0:
             raise ValueError(
                 f"Cannot split {config.num_attention_heads} attention heads "
                 f"evenly to {config.tensor_parallel_shards} GPUs."
             )
 
         self.attention_bias = config.attention_bias
-        self.num_heads = config.num_attention_heads
-        self.num_key_value_heads = config.num_key_value_heads
+        self.num_heads = config.num_attention_heads // self.tensor_parallel_shards
+        self.num_key_value_heads = config.num_key_value_heads // self.tensor_parallel_shards
         self.num_key_value_groups = self.num_heads // self.num_key_value_heads
-        self.head_dim = self.hidden_size // self.num_heads
+        self.head_dim = config.head_dim
         self.max_position_embeddings = config.context_window_size
 
         self.wqkv_pack = nn.Linear(
@@ -150,7 +155,7 @@ def __init__(self, config: DeepseekConfig, intermediate_size=None):
             )
         self.intermediate_size = (
             config.intermediate_size if intermediate_size is None else intermediate_size
-        )
+        ) // config.tensor_parallel_shards
 
         self.gate_up_proj = nn.Linear(self.hidden_size, 2 * self.intermediate_size, bias=False)
         self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
@@ -168,16 +173,18 @@ def __init__(self, config: DeepseekConfig):
         self.num_experts_per_tok = config.num_experts_per_tok
         self.gate = nn.Linear(config.hidden_size, config.n_routed_experts, bias=False)
         self.norm_topk_prob = config.norm_topk_prob
-        self.moe_intermediate_size = config.moe_intermediate_size
+        self.moe_intermediate_size = config.moe_intermediate_size // config.tensor_parallel_shards
         self.moe_gate_up_proj = MixtralExperts(
             self.num_local_experts,
             in_features=config.hidden_size,
             out_features=2 * self.moe_intermediate_size,
+            tensor_parallel_shards=config.tensor_parallel_shards,
         )
         self.moe_down_proj = MixtralExperts(
             self.num_local_experts,
             in_features=self.moe_intermediate_size,
             out_features=config.hidden_size,
+            tensor_parallel_shards=config.tensor_parallel_shards,
         )
         self.dtype = "float32"
 
@@ -254,15 +261,64 @@ def __init__(self, config: DeepseekConfig, layer_idx: int):
             config.hidden_size, -1, config.rms_norm_eps, bias=False
         )
 
+        def _set_tp():
+            def _set(layer, hint):
+                layer.attrs["shard_strategy"] = hint
+
+            hd = config.head_dim
+            q = self.self_attn.num_heads * hd
+            k = self.self_attn.num_key_value_heads * hd
+            v = self.self_attn.num_key_value_heads * hd
+
+            if (
+                config.n_routed_experts is not None
+                and layer_idx >= config.first_k_dense_replace
+                and layer_idx % config.moe_layer_freq == 0
+            ):
+                i = self.mlp.moe_intermediate_size
+            else:
+                i = self.mlp.intermediate_size
+            _set(
+                self.self_attn.wqkv_pack.weight,
+                tp.ShardSingleDim("_shard_qkv_weight", dim=0, segs=[q, k, v]),
+            )
+            _set(self.self_attn.o_proj.weight, tp.ShardSingleDim("_shard_o", dim=1))
+
+            if (
+                config.n_routed_experts is not None
+                and layer_idx >= config.first_k_dense_replace
+                and layer_idx % config.moe_layer_freq == 0
+            ):
+                _set(
+                    self.mlp.moe_gate_up_proj.weight,
+                    tp.ShardSingleDim("_shard_mlp_up", segs=[i, i], dim=1),
+                )
+                _set(self.mlp.moe_down_proj.weight, tp.ShardSingleDim("_shard_mlp_down", dim=2))
+
+            else:
+                _set(
+                    self.mlp.gate_up_proj.weight,
+                    tp.ShardSingleDim("_shard_mlp_up", segs=[i, i], dim=0),
+                )
+                _set(self.mlp.down_proj.weight, tp.ShardSingleDim("_shard_mlp_down", dim=1))
+
+        self.tensor_parallel_shards = config.tensor_parallel_shards
+        _set_tp()
+
     def forward(self, hidden_states: Tensor, paged_kv_cache: PagedKVCache, layer_id: int):
         out = self.input_layernorm(hidden_states)
         out = self.self_attn(out, paged_kv_cache, layer_id)
-        hidden_states = hidden_states + out
+        hidden_states = self._apply_residual(hidden_states, residual=out)
         out = self.post_attention_layernorm(hidden_states)
         out = self.mlp(out)  # type: ignore[operator]
-        hidden_states = hidden_states + out
+        hidden_states = self._apply_residual(hidden_states, residual=out)
         return hidden_states
 
+    def _apply_residual(self, out, residual):
+        if self.tensor_parallel_shards > 1:
+            return op.ccl_allreduce(out, "sum") + residual
+        return out + residual
+
 
 class DeepseekModel(nn.Module):
     def __init__(self, config: DeepseekConfig):
@@ -293,7 +349,8 @@ def __init__(self, config: DeepseekConfig):
         self.hidden_size = config.hidden_size
         self.num_attention_heads = config.num_attention_heads
         self.num_key_value_heads = config.num_key_value_heads
-        self.head_dim = config.hidden_size // config.num_attention_heads
+        self.tensor_parallel_shards = config.tensor_parallel_shards
+        self.head_dim = config.head_dim
         self.vocab_size = config.vocab_size
         self.rope_theta = config.rope_theta
         self.dtype = "float32"
@@ -320,6 +377,8 @@ def batch_forward(
         return logits
 
     def embed(self, input_ids: Tensor):
+        if self.tensor_parallel_shards > 1:
+            input_ids = op.ccl_broadcast_from_worker0(input_ids)
         return self.model.embed_tokens(input_ids)
 
     def prefill(self, input_embed: Tensor, paged_kv_cache: PagedKVCache):
@@ -349,6 +408,8 @@ def decode(self, input_embed: Tensor, paged_kv_cache: PagedKVCache):
     def batch_prefill(
         self, input_embeds: Tensor, logit_positions: Tensor, paged_kv_cache: PagedKVCache
     ):
+        if self.tensor_parallel_shards > 1:
+            logit_positions = op.ccl_broadcast_from_worker0(logit_positions)
         logits = self.batch_forward(input_embeds, paged_kv_cache, logit_positions)
         return logits, paged_kv_cache
 
@@ -375,8 +436,8 @@ def create_paged_kv_cache(  # pylint: disable=too-many-arguments
             page_size=page_size,
             support_sliding_window=support_sliding_window,
             num_hidden_layers=self.num_hidden_layers,
-            num_attention_heads=self.num_attention_heads,
-            num_key_value_heads=self.num_key_value_heads,
+            num_attention_heads=self.num_attention_heads // self.tensor_parallel_shards,
+            num_key_value_heads=self.num_key_value_heads // self.tensor_parallel_shards,
             head_dim=self.head_dim,
             rope_mode=RopeMode.NORMAL,
             rope_scale=1,