Skip to content

GitLab

Commit 3f9bbf11 authored by djw's avatar djw
Browse files

support qwen3, dont speak human language

parent f3d842a0
Hide whitespace changes
Inline Side-by-side
Showing with 227 additions and 185 deletions
ktransformers/server/balance_serve/inference/forward_batch.py
View file @ 3f9bbf11
...@@ -281,4 +281,4 @@ class ForwardBatchOutput: ...@@ -281,4 +281,4 @@ class ForwardBatchOutput:
self.generated_tokens_num = [] self.generated_tokens_num = []
self.top_ps = [] self.top_ps = []
self.temperatures = [] self.temperatures = []
pass self.num_batchs = 1
\ No newline at end of file \ No newline at end of file
ktransformers/server/balance_serve/inference/model_runner.py
View file @ 3f9bbf11
...@@ -27,6 +27,8 @@ from ktransformers.server.balance_serve.inference.forward_batch import ForwardBa ...@@ -27,6 +27,8 @@ from ktransformers.server.balance_serve.inference.forward_batch import ForwardBa
from ktransformers.server.config.config import Config from ktransformers.server.config.config import Config
from ktransformers.models.custom_modeling_deepseek_v3 import KDeepseekV3ForCausalLM from ktransformers.models.custom_modeling_deepseek_v3 import KDeepseekV3ForCausalLM
from ktransformers.models.custom_modeling_deepseek_v2 import KDeepseekV2ForCausalLM from ktransformers.models.custom_modeling_deepseek_v2 import KDeepseekV2ForCausalLM
from ktransformers.models.custom_modeling_qwen2_moe import KQwen2MoeForCausalLM
from ktransformers.models.custom_modeling_qwen3_moe import KQwen3MoeForCausalLM
from ktransformers.server.balance_serve.inference.query_manager import QueryManager from ktransformers.server.balance_serve.inference.query_manager import QueryManager
from ktransformers.server.balance_serve.settings import sched_ext from ktransformers.server.balance_serve.settings import sched_ext
...@@ -40,11 +42,11 @@ def deduplicate_and_sort(lst): ...@@ -40,11 +42,11 @@ def deduplicate_and_sort(lst):
class ModelRunner: class ModelRunner:
"""A CudaGraphRunner runs the forward pass of a model with CUDA graph and torch.compile.""" """A CudaGraphRunner runs the forward pass of a model with CUDA graph and torch.compile."""
model: KDeepseekV3ForCausalLM model: KDeepseekV3ForCausalLM | KQwen2MoeForCausalLM | KQwen3MoeForCausalLM
input: ForwardBatchInput | list[ForwardBatchInput] input: ForwardBatchInput | list[ForwardBatchInput]
output: ForwardBatchOutput output: ForwardBatchOutput
def __init__(self, model = None, device = None, use_cuda_graph = False, max_decode_batch_size = 1, max_chunk_size = 4096, num_mini_batches: int = 1, page_size = 256): def __init__(self, model = None, device = None, use_cuda_graph = False, max_decode_batch_size = 1, max_chunk_size = 4096, num_mini_batches: int = 1, page_size = 256, block_num = 8):
self.stream = torch.cuda.Stream(device=device) self.stream = torch.cuda.Stream(device=device)
# 先注释掉 # 先注释掉
...@@ -58,120 +60,92 @@ class ModelRunner: ...@@ -58,120 +60,92 @@ class ModelRunner:
self.use_cuda_graph = use_cuda_graph self.use_cuda_graph = use_cuda_graph
self.model_time = 0 self.model_time = 0
self.page_size = page_size self.page_size = page_size
self.block_num = block_num
# GPU timing for model execution # GPU timing for model execution
self.start_model_event = torch.cuda.Event(enable_timing=True) self.start_model_event = torch.cuda.Event(enable_timing=True)
self.end_model_event = torch.cuda.Event(enable_timing=True) self.end_model_event = torch.cuda.Event(enable_timing=True)
if isinstance(self.cuda_graphs, list):
self.graphs = [torch.cuda.CUDAGraph() for _ in range(len(self.cuda_graphs))] self.graphs = [torch.cuda.CUDAGraph() for _ in range(len(self.cuda_graphs))]
self.page_idx_buf = [torch.zeros([self.cuda_graphs[i]], dtype=torch.int32, device = self.device) for i in range(len(self.cuda_graphs))] self.page_idx_buf = [torch.zeros([self.cuda_graphs[i]], dtype=torch.int32, device = self.device) for i in range(len(self.cuda_graphs))]
self.page_offset_buf = [torch.zeros([self.cuda_graphs[i]], dtype=torch.int32, device = self.device) for i in range(len(self.cuda_graphs))] self.page_offset_buf = [torch.zeros([self.cuda_graphs[i]], dtype=torch.int32, device = self.device) for i in range(len(self.cuda_graphs))]
else:
self.graphs = torch.cuda.CUDAGraph()
self.page_idx_buf = torch.zeros([self.cuda_graphs], dtype=torch.int32, device = self.device)
self.page_offset_buf = torch.zeros([self.cuda_graphs], dtype=torch.int32, device = self.device)
self.num_mini_batches = num_mini_batches self.num_mini_batches = num_mini_batches
self.max_chunk_size = max_chunk_size self.max_chunk_size = max_chunk_size
self.bsz_tensor_buf = torch.empty((1, ),dtype=torch.int32, device=device) self.bsz_tensor_buf = torch.empty((1, ),dtype=torch.int32, device=device)
self.num_tokens_tensor_buf = torch.empty((1, ),dtype=torch.int32, device=device) self.num_tokens_tensor_buf = torch.empty((1, ),dtype=torch.int32, device=device)
def warmup(self):
def capture_graphs(cuda_graph_idx=-1): def model_attn_plan(self, batch, cuda_graph_idx=0):
if cuda_graph_idx != -1: if isinstance(self.model, KDeepseekV3ForCausalLM):
with torch.cuda.graph(self.graphs[cuda_graph_idx], pool=self.graph_memory_pool, stream=self.stream): self.model.flash_infer_attn_plan(batch, self.bsz_tensor_buf, self.num_tokens_tensor_buf,
self.outputs_buf[cuda_graph_idx] = self.model(self.input[cuda_graph_idx], self.features_buf[cuda_graph_idx], self.bsz_tensor_buf, self.num_tokens_tensor_buf, self.page_idx_buf[cuda_graph_idx], self.page_offset_buf[cuda_graph_idx], cuda_graph_idx=cuda_graph_idx) num_heads=self.model.config.num_attention_heads, head_dim_ckv=self.model.config.kv_lora_rank,
self.graph_memory_pool = self.graphs[cuda_graph_idx].pool() head_dim_kpe=self.model.config.qk_rope_head_dim, page_size=self.model.cache.page_size, causal=True,
else: sm_scale=self.model.model.layers[0].self_attn.softmax_scale, q_data_type=torch.bfloat16, kv_data_type=torch.bfloat16)
with torch.cuda.graph(self.graphs, pool=self.graph_memory_pool, stream=self.stream): elif isinstance(self.model, KQwen2MoeForCausalLM) or isinstance(self.model, KQwen3MoeForCausalLM):
self.outputs_buf = self.model(self.input, self.features_buf, self.bsz_tensor_buf, self.num_tokens_tensor_buf, self.page_idx_buf, self.page_offset_buf) self.model.flash_infer_attn_plan(batch, self.bsz_tensor_buf, self.num_tokens_tensor_buf,
self.graph_memory_pool = self.graphs.pool() num_q_heads=self.model.config.num_attention_heads, num_kv_heads=self.model.config.num_key_value_heads,
head_dim=self.model.config.hidden_size // self.model.config.num_attention_heads,
if isinstance(self.cuda_graphs, list): page_size=self.model.cache.page_size, causal=True,
self.input = [] q_data_type=torch.bfloat16, kv_data_type=torch.bfloat16, cuda_graph_idx=cuda_graph_idx)
self.features_buf = [] else:
self.outputs_buf = [] assert False, "model type not supported"
self.bsz_tensor_buf = torch.tensor([0], dtype=torch.int32, device=self.device)
self.num_tokens_tensor_buf = torch.tensor([0], dtype=torch.int32, device=self.device)
for i in range(len(self.cuda_graphs)):
prefill_query_length = (self.cuda_graphs[i] - Config().max_decode_batch_size) // Config().max_prefill_batch_size if self.cuda_graphs[i] > Config().max_decode_batch_size else 0 #@TODO only supprot 2 prefill batch
self.input.append(ForwardBatchInput.gen_max_forward_batch(device=self.device, num_mini_batches = self.num_mini_batches, prefill_query_length=prefill_query_length, prefill_active_length=prefill_query_length, page_size=self.page_size, cuda_lens = self.cuda_graphs[i]))
self.features_buf.append(self.model.batch_embeddings(self.input[i]))
batch_size = self.input[i].minibatch.q_indptr.size(0)-1
num_tokens = self.features_buf[i][0].size(0)
print("capturing cuda graph", batch_size, num_tokens)
self.bsz_tensor_buf[0] = batch_size
self.num_tokens_tensor_buf[0] = num_tokens
self.model.flash_infer_attn_plan(self.input[i], self.bsz_tensor_buf, self.num_tokens_tensor_buf,
num_heads=self.model.config.num_attention_heads, head_dim_ckv=self.model.config.kv_lora_rank,
head_dim_kpe=self.model.config.qk_rope_head_dim, page_size=self.model.cache.page_size, causal=True,
sm_scale=self.model.model.layers[0].self_attn.softmax_scale, q_data_type=torch.bfloat16, kv_data_type=torch.bfloat16)
page_idx, page_offset = self.model.cache.get_page_table(self.input[i].minibatch.position_ids, self.input[i].minibatch.q_indptr, self.input[i].minibatch.kv_indptr, self.input[i].minibatch.kv_indices, self.num_tokens_tensor_buf)
self.page_idx_buf[i][:num_tokens].copy_(page_idx[:num_tokens])
self.page_offset_buf[i][:num_tokens].copy_(page_offset[:num_tokens])
self.page_idx_buf[i][num_tokens:].fill_(self.model.cache.max_cache_len // self.model.cache.page_size -1)
self.outputs_buf.append(None)
torch.cuda.synchronize()
for warm_up_iters in range(11):
with torch.cuda.stream(self.stream):
self.outputs_buf[i] = self.model(self.input[i], self.features_buf[i], self.bsz_tensor_buf, self.num_tokens_tensor_buf, self.page_idx_buf[i], self.page_offset_buf[i])
torch.cuda.synchronize()
capture_graphs(i) def warmup(self):
with torch.cuda.stream(self.stream): def capture_graphs(cuda_graph_idx):
self.graphs[i].replay() with torch.cuda.graph(self.graphs[cuda_graph_idx], pool=self.graph_memory_pool, stream=self.stream):
self.outputs_buf[cuda_graph_idx] = self.model(self.input[cuda_graph_idx], self.features_buf[cuda_graph_idx], self.bsz_tensor_buf, self.num_tokens_tensor_buf, self.page_idx_buf[cuda_graph_idx], self.page_offset_buf[cuda_graph_idx], cuda_graph_idx=cuda_graph_idx)
self.graph_memory_pool = self.graphs[cuda_graph_idx].pool()
self.sync(calc_time=False) self.input = []
print(f"cuda_graph: {i+1}/{len(self.cuda_graphs)}, warmup finished.") self.features_buf = []
else: self.outputs_buf = []
self.input = ForwardBatchInput.gen_max_forward_batch(device=self.device, num_mini_batches = self.num_mini_batches) self.bsz_tensor_buf = torch.tensor([0], dtype=torch.int32, device=self.device)
self.num_tokens_tensor_buf = torch.tensor([0], dtype=torch.int32, device=self.device)
for i in range(len(self.cuda_graphs)):
prefill_query_length = (self.cuda_graphs[i] - Config().max_decode_batch_size) // Config().max_prefill_batch_size if self.cuda_graphs[i] > Config().max_decode_batch_size else 0 #@TODO only supprot 2 prefill batch
self.input.append(ForwardBatchInput.gen_max_forward_batch(device=self.device, num_mini_batches = self.num_mini_batches, prefill_query_length=prefill_query_length, prefill_active_length=prefill_query_length, page_size=self.page_size, cuda_lens=self.cuda_graphs[i]))
self.features_buf = self.model.batch_embeddings(self.input) self.features_buf.append(self.model.batch_embeddings(self.input[i]))
batch_size = self.input.minibatch.q_indptr.size(0)-1 batch_size = self.input[i].minibatch.q_indptr.size(0)-1
num_tokens = self.features_buf[0].size(0) num_tokens = self.features_buf[i][0].size(0)
print("capturing cuda graph", batch_size, num_tokens)
if isinstance(self.model, KQwen2MoeForCausalLM) or isinstance(self.model, KQwen3MoeForCausalLM):
self.bsz_tensor_buf = torch.tensor([batch_size], dtype=torch.int32, device=self.device) self.model.init_wrapper(self.use_cuda_graph, self.device, num_tokens ,batch_size, self.block_num, i) # TODO: 1024 is a magic number(max_batch_tokens)
self.num_tokens_tensor_buf = torch.tensor([num_tokens], dtype=torch.int32, device=self.device)
self.bsz_tensor_buf[0] = batch_size
self.num_tokens_tensor_buf[0] = num_tokens
self.model_attn_plan(self.input[i], i)
page_idx, page_offset = self.model.cache.get_page_table(self.input[i].minibatch.position_ids, self.input[i].minibatch.q_indptr, self.input[i].minibatch.kv_indptr, self.input[i].minibatch.kv_indices, self.num_tokens_tensor_buf)
self.model.flash_infer_attn_plan(self.input, self.bsz_tensor_buf, self.num_tokens_tensor_buf,
num_heads=self.model.config.num_attention_heads, head_dim_ckv=self.model.config.kv_lora_rank,
head_dim_kpe=self.model.config.qk_rope_head_dim, page_size=self.model.cache.page_size, causal=True,
sm_scale=self.model.model.layers[0].self_attn.softmax_scale, q_data_type=torch.bfloat16, kv_data_type=torch.bfloat16)
page_idx, page_offset = self.model.cache.get_page_table(self.input.minibatch.position_ids, self.input.minibatch.q_indptr, self.input.minibatch.kv_indptr, self.input.minibatch.kv_indices, self.num_tokens_tensor_buf)
self.page_idx_buf[:num_tokens].copy_(page_idx[:num_tokens])
self.page_offset_buf[:num_tokens].copy_(page_offset[:num_tokens])
self.page_idx_buf[num_tokens:].fill_(self.model.cache.max_cache_len // self.model.cache.page_size - 1)
self.page_idx_buf[i][:num_tokens].copy_(page_idx[:num_tokens])
self.page_offset_buf[i][:num_tokens].copy_(page_offset[:num_tokens])
self.page_idx_buf[i][num_tokens:].fill_(self.model.cache.max_cache_len // self.model.cache.page_size -1)
self.outputs_buf.append(None)
torch.cuda.synchronize() torch.cuda.synchronize()
for warm_up_iters in range(11): for warm_up_iters in range(11):
with torch.cuda.stream(self.stream): with torch.cuda.stream(self.stream):
self.outputs_buf = self.model(self.input, self.features_buf, self.bsz_tensor_buf, self.num_tokens_tensor_buf, self.page_idx_buf, self.page_offset_buf) self.outputs_buf[i] = self.model(self.input[i], self.features_buf[i], self.bsz_tensor_buf, self.num_tokens_tensor_buf, self.page_idx_buf[i], self.page_offset_buf[i], cuda_graph_idx=i)
torch.cuda.synchronize() torch.cuda.synchronize()
def capture_graphs(): self.outputs_buf[i].num_batchs = batch_size
with torch.cuda.graph(self.graphs, stream=self.stream):
self.outputs_buf = self.model(self.input, self.features_buf, self.bsz_tensor_buf, self.num_tokens_tensor_buf, self.page_idx_buf, self.page_offset_buf)
# self.graph_memory_pool = self.graphs.pool()
capture_graphs() capture_graphs(i)
with torch.cuda.stream(self.stream): with torch.cuda.stream(self.stream):
self.graphs.replay() self.graphs[i].replay()
self.sync(calc_time=False) self.sync(calc_time=False)
print("warmup finished.") print(f"cuda_graph: {i+1}/{len(self.cuda_graphs)}, warmup finished.")
def run(self, batch: sched_ext.BatchQueryTodo = None, query_manager: QueryManager = None): def run(self, batch: sched_ext.BatchQueryTodo = None, query_manager: QueryManager = None):
with torch.cuda.stream(self.stream): with torch.cuda.stream(self.stream):
...@@ -189,107 +163,54 @@ class ModelRunner: ...@@ -189,107 +163,54 @@ class ModelRunner:
if isinstance(self.cuda_graphs, list): # cuda graph idx equal to min idx i in self.cuda_graphs, that self.cuda_graphs[i] > num_tokens
# cuda graph idx equal to min idx i in self.cuda_graphs, that self.cuda_graphs[i] > num_tokens cuda_graph_idx = next((i for i, token in enumerate(self.cuda_graphs) if token >= num_tokens), len(self.cuda_graphs))
cuda_graph_idx = next((i for i, token in enumerate(self.cuda_graphs) if token >= num_tokens), len(self.cuda_graphs)) if not self.use_cuda_graph:
if cuda_graph_idx == len(self.cuda_graphs): cuda_graph_idx = 0
assert False, "num_tokens is too large" # if cuda_graph_idx == len(self.cuda_graphs):
else: # assert False, "num_tokens is too large"
cuda_graph_idx = -1
if self.use_cuda_graph: if self.use_cuda_graph:
if cuda_graph_idx != -1: self.input[cuda_graph_idx].fill(batch, query_manager, self.page_size)
self.input[cuda_graph_idx].fill(batch, query_manager, self.page_size)
else:
self.input.fill(batch, query_manager, self.page_size)
else: else:
self.input = ForwardBatchInput(batch=batch, query_manager=query_manager, device=self.device) self.input = [ForwardBatchInput(batch=batch, query_manager=query_manager, device=self.device)]
if cuda_graph_idx != -1 and self.use_cuda_graph: if self.use_cuda_graph:
self.features = self.model.batch_embeddings(self.input[cuda_graph_idx], device=self.device) self.features = self.model.batch_embeddings(self.input[cuda_graph_idx], device=self.device)
else: else:
self.features = self.model.batch_embeddings(self.input, device=self.device) self.features = self.model.batch_embeddings(self.input[cuda_graph_idx], device=self.device)
self.bsz_tensor_buf.copy_(batch_size) self.bsz_tensor_buf.copy_(batch_size)
self.num_tokens_tensor_buf.copy_(torch.tensor([num_tokens], dtype=torch.int32, device=self.device)) self.num_tokens_tensor_buf.copy_(torch.tensor([num_tokens], dtype=torch.int32, device=self.device))
if self.use_cuda_graph: if self.use_cuda_graph:
if cuda_graph_idx != -1: self.features_buf[cuda_graph_idx][0].copy_(self.features[0], non_blocking=True)
self.features_buf[cuda_graph_idx][0].copy_(self.features[0], non_blocking=True)
else: self.model_attn_plan(self.input[cuda_graph_idx], cuda_graph_idx)
self.features_buf[0].copy_(self.features[0], non_blocking=True) self.start_model_event.record(self.stream)
""" page_idx, page_offset = self.model.cache.get_page_table(self.input[cuda_graph_idx].minibatch.position_ids, self.input[cuda_graph_idx].minibatch.q_indptr, self.input[cuda_graph_idx].minibatch.kv_indptr, self.input[cuda_graph_idx].minibatch.kv_indices, self.num_tokens_tensor_buf)
if num_tokens_0 > 64:
padded_num_tokens_0 = pad_num_tokens(num_tokens_0)
self.features_buf[0][num_tokens_0:padded_num_tokens_0] = 0
"""
#self.input.forward_minibatchs[0].print()
# print([[hash(k[i].float().cpu().numpy().tobytes()) for i in self.input.forward_minibatchs[0].kv_indices] for k in self.model.cache.k_caches])
# print(f"overlap: {overlap}, is_compute_bound: {is_compute_bound}")
# self.model.flash_infer_attn_plan(self.input, self.bsz_tensors, self.num_tokens_tensors)
"""
if self.use_cuda_graph: if self.use_cuda_graph:
print("before replay features_buf", self.features_buf[0]) self.page_idx_buf[cuda_graph_idx][:num_tokens].copy_(page_idx[:num_tokens])
print("features_buf addr", self.features_buf[0].data_ptr()) self.page_offset_buf[cuda_graph_idx][:num_tokens].copy_(page_offset[:num_tokens])
else:
print("before run features", self.features[0]) self.page_idx_buf[cuda_graph_idx][num_tokens:].fill_(self.model.cache.max_cache_len // self.model.cache.page_size -1)
""" self.replay(cuda_graph_idx)
if cuda_graph_idx != -1 and self.use_cuda_graph: self.output = ForwardBatchOutput()
self.model.flash_infer_attn_plan(self.input[cuda_graph_idx], self.bsz_tensor_buf, self.num_tokens_tensor_buf,
num_heads=self.model.config.num_attention_heads, head_dim_ckv=self.model.config.kv_lora_rank, self.output.top_ps.append(self.input[cuda_graph_idx].minibatch.top_ps)
head_dim_kpe=self.model.config.qk_rope_head_dim, page_size=self.model.cache.page_size, causal=True, self.output.temperatures.append(self.input[cuda_graph_idx].minibatch.temperatures)
sm_scale=self.model.model.layers[0].self_attn.softmax_scale, q_data_type=torch.bfloat16, kv_data_type=torch.bfloat16)
self.start_model_event.record(self.stream)
page_idx, page_offset = self.model.cache.get_page_table(self.input[cuda_graph_idx].minibatch.position_ids, self.input[cuda_graph_idx].minibatch.q_indptr, self.input[cuda_graph_idx].minibatch.kv_indptr, self.input[cuda_graph_idx].minibatch.kv_indices, self.num_tokens_tensor_buf) self.output.logits.append(self.outputs_buf[cuda_graph_idx].logits[0][self.input[cuda_graph_idx].minibatch.logits_start].clone())
if self.use_cuda_graph:
self.page_idx_buf[cuda_graph_idx][:num_tokens].copy_(page_idx[:num_tokens])
self.page_offset_buf[cuda_graph_idx][:num_tokens].copy_(page_offset[:num_tokens])
self.page_idx_buf[cuda_graph_idx][num_tokens:].fill_(self.model.cache.max_cache_len // self.model.cache.page_size - 1)
self.replay(cuda_graph_idx)
self.output = ForwardBatchOutput()
self.output.top_ps.append(self.input[cuda_graph_idx].minibatch.top_ps)
self.output.temperatures.append(self.input[cuda_graph_idx].minibatch.temperatures)
self.output.logits.append(self.outputs_buf[cuda_graph_idx].logits[0][self.input[cuda_graph_idx].minibatch.logits_start].clone())
else:
self.output = self.model(self.input[cuda_graph_idx], self.features, self.bsz_tensor_buf, self.num_tokens_tensor_buf, page_idx, page_offset)
self.output.logits[0] = self.output.logits[0][self.input[cuda_graph_idx].minibatch.logits_start]
self.end_model_event.record(self.stream)
else: else:
self.model.flash_infer_attn_plan(self.input, self.bsz_tensor_buf, self.num_tokens_tensor_buf, self.output = self.model(self.input[cuda_graph_idx], self.features, self.bsz_tensor_buf, self.num_tokens_tensor_buf, page_idx, page_offset)
num_heads=self.model.config.num_attention_heads, head_dim_ckv=self.model.config.kv_lora_rank, self.output.logits[0] = self.output.logits[0][self.input[cuda_graph_idx].minibatch.logits_start]
head_dim_kpe=self.model.config.qk_rope_head_dim, page_size=self.model.cache.page_size, causal=True, self.output.top_ps.append(self.input[cuda_graph_idx].minibatch.top_ps)
sm_scale=self.model.model.layers[0].self_attn.softmax_scale, q_data_type=torch.bfloat16, kv_data_type=torch.bfloat16) self.output.temperatures.append(self.input[cuda_graph_idx].minibatch.temperatures)
self.start_model_event.record(self.stream) self.end_model_event.record(self.stream)
page_idx, page_offset = self.model.cache.get_page_table(self.input.minibatch.position_ids, self.input.minibatch.q_indptr, self.input.minibatch.kv_indptr, self.input.minibatch.kv_indices, self.num_tokens_tensor_buf)
if self.use_cuda_graph:
self.page_idx_buf[:num_tokens].copy_(page_idx[:num_tokens])
self.page_offset_buf[:num_tokens].copy_(page_offset[:num_tokens])
self.page_idx_buf[num_tokens:].fill_(self.model.cache.max_cache_len // self.model.cache.page_size - 1)
self.replay(cuda_graph_idx)
self.output = ForwardBatchOutput()
self.output.top_ps.append(self.input.minibatch.top_ps)
self.output.temperatures.append(self.input.minibatch.temperatures)
self.output.logits.append(self.outputs_buf.logits[0][self.input.minibatch.logits_start].clone())
else:
self.output = self.model(self.input, self.features, self.bsz_tensor_buf, self.num_tokens_tensor_buf, page_idx, page_offset)
self.output.logits[0] = self.output.logits[0][self.input.minibatch.logits_start]
self.output.top_ps.append(self.input.minibatch.top_ps)
self.output.temperatures.append(self.input.minibatch.temperatures)
self.end_model_event.record(self.stream)
if not self.use_cuda_graph:
self.output.num_batchs = self.input.batch_size
else:
self.output.num_batchs = self.input[cuda_graph_idx].batch_size
def replay(self, cuda_graph_idx=-1): def replay(self, cuda_graph_idx=-1):
......
ktransformers/server/balance_serve/sched_rpc.py
View file @ 3f9bbf11
...@@ -10,7 +10,7 @@ current_file_path = os.path.abspath(__file__) ...@@ -10,7 +10,7 @@ current_file_path = os.path.abspath(__file__)
# sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "..", "..")) # sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "..", ".."))
import pickle import pickle
import argparse import argparse
from ktransformers.server.balance_serve.settings import sched_ext, create_sched_settings from ktransformers.server.balance_serve.settings import sched_ext, create_sched_settings, create_sched_settings_qwen2moe, create_sched_settings_qwen3moe
...@@ -209,5 +209,10 @@ if __name__ == '__main__': ...@@ -209,5 +209,10 @@ if __name__ == '__main__':
args = parser.parse_args() args = parser.parse_args()
with open(args.config, "rb") as f: with open(args.config, "rb") as f:
main_args = pickle.load(f) main_args = pickle.load(f)
settings = create_sched_settings(main_args) if main_args.architectures == "Qwen2MoeForCausalLM":
settings = create_sched_settings_qwen2moe(main_args)
elif main_args.architectures == "Qwen3MoeForCausalLM":
settings = create_sched_settings_qwen3moe(main_args)
else:
settings = create_sched_settings(main_args)
start_server(settings, main_args) start_server(settings, main_args)
ktransformers/server/balance_serve/settings.py
View file @ 3f9bbf11
...@@ -11,6 +11,8 @@ from time import sleep ...@@ -11,6 +11,8 @@ from time import sleep
import sched_ext import sched_ext
from transformers import AutoConfig from transformers import AutoConfig
from ktransformers.models.configuration_qwen3_moe import Qwen3MoeConfig
def create_sched_settings(args): def create_sched_settings(args):
default_sample_options = sched_ext.SampleOptions() default_sample_options = sched_ext.SampleOptions()
model_name = os.path.basename(os.path.normpath(args.model_dir)) model_name = os.path.basename(os.path.normpath(args.model_dir))
...@@ -64,7 +66,111 @@ def create_sched_settings(args): ...@@ -64,7 +66,111 @@ def create_sched_settings(args):
return settings return settings
def create_sched_settings_qwen2moe(args):
default_sample_options = sched_ext.SampleOptions()
model_name = os.path.basename(os.path.normpath(args.model_dir))
input_model_settings = sched_ext.ModelSettings()
input_model_settings.model_path = args.model_dir
input_model_settings.params_count = int(0)
model_config = AutoConfig.from_pretrained(args.model_dir, trust_remote_code=True)
input_model_settings.layer_count = model_config.num_hidden_layers
input_model_settings.num_k_heads = model_config.num_key_value_heads # model_config["num_key_value_heads"]
input_model_settings.k_head_dim = 128
input_model_settings.bytes_per_params = 2
input_model_settings.bytes_per_kv_cache_element = 2
settings = sched_ext.Settings()
settings.model_name = model_name
settings.quant_type = "BF16"
settings.model_settings = input_model_settings
settings.page_size = args.page_size
settings.gpu_device_count = 1 # tp
settings.gpu_device_id = [i for i in range(settings.gpu_device_count)]
# settings.gpu_memory_size = args.cache_lens*576*2
settings.gpu_memory_size = args.gpu_memory_size
settings.memory_utilization_percentage = args.utilization_percentage
max_batch_size = args.max_batch_size
chunk_size = args.chunk_size
max_decode_batch_size = max_batch_size - 2
settings.max_batch_size = max_batch_size
settings.recommended_chunk_prefill_token_count = (chunk_size - max_decode_batch_size) // 2
settings.sample_options = default_sample_options
settings.sched_metrics_port = args.sched_metrics_port
settings.gpu_only = args.memory_gpu_only
settings.use_self_defined_head_dim = False
settings.self_defined_head_dim = 576
settings.full_kv_cache_on_each_gpu = True
settings.k_cache_on = True
settings.v_cache_on = True
settings.kvc2_root_path = '/mnt/data/persist-kvc'
settings.kvc2_config_path = args.kvc2_config_dir
settings.memory_pool_size_GB = args.cpu_memory_size_GB
settings.evict_count = 40
settings.kvc2_metrics_port = args.kvc2_metrics_port
settings.load_from_disk = False
settings.save_to_disk = True
settings.strategy_name = args.sched_strategy
settings.auto_derive()
return settings
def create_sched_settings_qwen3moe(args):
default_sample_options = sched_ext.SampleOptions()
model_name = os.path.basename(os.path.normpath(args.model_dir))
input_model_settings = sched_ext.ModelSettings()
input_model_settings.model_path = args.model_dir
input_model_settings.params_count = int(0)
model_config = Qwen3MoeConfig.from_pretrained(args.model_dir, trust_remote_code=True)
input_model_settings.layer_count = model_config.num_hidden_layers
input_model_settings.num_k_heads = model_config.num_key_value_heads # model_config["num_key_value_heads"]
input_model_settings.k_head_dim = 128
input_model_settings.bytes_per_params = 2
input_model_settings.bytes_per_kv_cache_element = 2
settings = sched_ext.Settings()
settings.model_name = model_name
settings.quant_type = "BF16"
settings.model_settings = input_model_settings
settings.page_size = args.page_size
settings.gpu_device_count = 1 # tp
settings.gpu_device_id = [i for i in range(settings.gpu_device_count)]
# settings.gpu_memory_size = args.cache_lens*576*2
settings.gpu_memory_size = args.gpu_memory_size
settings.memory_utilization_percentage = args.utilization_percentage
max_batch_size = args.max_batch_size
chunk_size = args.chunk_size
max_decode_batch_size = max_batch_size - 2
settings.max_batch_size = max_batch_size
settings.recommended_chunk_prefill_token_count = (chunk_size - max_decode_batch_size) // 2
settings.sample_options = default_sample_options
settings.sched_metrics_port = args.sched_metrics_port
settings.gpu_only = args.memory_gpu_only
settings.use_self_defined_head_dim = False
settings.self_defined_head_dim = 576
settings.full_kv_cache_on_each_gpu = True
settings.k_cache_on = True
settings.v_cache_on = True
settings.kvc2_root_path = '/mnt/data/persist-kvc'
settings.kvc2_config_path = args.kvc2_config_dir
settings.memory_pool_size_GB = args.cpu_memory_size_GB
settings.evict_count = 40
settings.kvc2_metrics_port = args.kvc2_metrics_port
settings.load_from_disk = False
settings.save_to_disk = True
settings.strategy_name = args.sched_strategy
settings.auto_derive()
return settings
......
ktransformers/server/config/config.py
View file @ 3f9bbf11
...@@ -100,6 +100,7 @@ class Config(metaclass=Singleton): ...@@ -100,6 +100,7 @@ class Config(metaclass=Singleton):
# to make sure it consistent with previous version # to make sure it consistent with previous version
self.model_path: str = self.model_dir self.model_path: str = self.model_dir
self.model_name: str = self.model.get("name", "") self.model_name: str = self.model.get("name", "")
self.architectures: str = self.model.get("name", "")
self.model_device: str = self.model.get("device", "cuda:0") self.model_device: str = self.model.get("device", "cuda:0")
self.gguf_path: Optional[str] = self.model.get("gguf_path", None) self.gguf_path: Optional[str] = self.model.get("gguf_path", None)
self.use_cuda_graph = self.model.get("use_cuda_graph", True) self.use_cuda_graph = self.model.get("use_cuda_graph", True)
......
ktransformers/server/requirements.txt
View file @ 3f9bbf11
torch >= 2.3.0 torch >= 2.3.0
transformers == 4.43.2 transformers == 4.51.3
fastapi >= 0.111.0 fastapi >= 0.111.0
langchain >= 0.2.0 langchain >= 0.2.0
blessed >= 1.20.0 blessed >= 1.20.0
......
ktransformers/util/custom_gguf.py
View file @ 3f9bbf11
...@@ -912,6 +912,9 @@ def translate_name_to_gguf(name): ...@@ -912,6 +912,9 @@ def translate_name_to_gguf(name):
name = name.replace(".self_attn.q_a_proj", ".attn_q_a") name = name.replace(".self_attn.q_a_proj", ".attn_q_a")
name = name.replace(".self_attn.q_a_layernorm", ".attn_q_a_norm") name = name.replace(".self_attn.q_a_layernorm", ".attn_q_a_norm")
name = name.replace(".self_attn.q_b_proj", ".attn_q_b") name = name.replace(".self_attn.q_b_proj", ".attn_q_b")
name = name.replace(".self_attn.q_norm", ".attn_q_norm")
name = name.replace(".self_attn.k_norm", ".attn_k_norm")
name = name.replace(".shared_expert.", ".shared_experts.") name = name.replace(".shared_expert.", ".shared_experts.")
name = name.replace(".shared_expert_", ".shared_experts_") name = name.replace(".shared_expert_", ".shared_experts_")
...@@ -922,17 +925,23 @@ def translate_name_to_gguf(name): ...@@ -922,17 +925,23 @@ def translate_name_to_gguf(name):
name = name.replace(".mlp.shared_experts.gate_proj", ".ffn_gate_shexp") name = name.replace(".mlp.shared_experts.gate_proj", ".ffn_gate_shexp")
name = name.replace(".mlp.shared_experts.up_proj", ".ffn_up_shexp") name = name.replace(".mlp.shared_experts.up_proj", ".ffn_up_shexp")
name = name.replace(".mlp.shared_experts_gate", ".ffn_gate_inp_shexp") name = name.replace(".mlp.shared_experts_gate", ".ffn_gate_inp_shexp")
name = name.replace(".mlp.experts", "") name = name.replace(".mlp.experts", "")
name = name.replace(".mlp.experts.ffn_down_exps", ".ffn_down_exps")
name = name.replace(".mlp.experts.ffn_gate_exps", ".ffn_gate_exps")
name = name.replace(".mlp.experts.ffn_up_exps", ".ffn_up_exps")
name = name.replace(".block_sparse_moe.gate.", ".ffn_gate_inp.") name = name.replace(".block_sparse_moe.gate.", ".ffn_gate_inp.")
name = name.replace(".block_sparse_moe.experts", "") name = name.replace(".block_sparse_moe.experts", "")
name = name.replace(".feed_forward.experts", "")
name = name.replace(".feed_forward.router", ".ffn_gate_inp")
name = name.replace(".feed_forward.shared_experts.down_proj", ".ffn_down_shexp")
name = name.replace(".feed_forward.shared_experts.gate_proj", ".ffn_gate_shexp")
name = name.replace(".feed_forward.shared_experts.up_proj", ".ffn_up_shexp")
return name return name
if __name__ == '__main__': if __name__ == '__main__':
gguf_path = '/mnt/data/model/DeepSeek-Coder-V2-GGUF-WJH' gguf_path = '/mnt/data/model/DeepSeek-Coder-V2-GGUF-WJH'
loader = GGUFLoader(gguf_path) loader = GGUFLoader(gguf_path)
......
pyproject.toml
View file @ 3f9bbf11
...@@ -16,7 +16,7 @@ dynamic = ["version"] ...@@ -16,7 +16,7 @@ dynamic = ["version"]
dependencies = [ dependencies = [
"torch >= 2.3.0", "torch >= 2.3.0",
"transformers == 4.43.2", "transformers == 4.51.3",
"fastapi >= 0.111.0", "fastapi >= 0.111.0",
"uvicorn >= 0.30.1", "uvicorn >= 0.30.1",
"langchain >= 0.2.0", "langchain >= 0.2.0",
......
requirements-local_chat.txt
View file @ 3f9bbf11
fire fire
transformers==4.43.2 transformers==4.51.3
numpy numpy
torch>=2.3.0 torch>=2.3.0
packaging packaging
......
custom_flashinfer @ af4259e8
Compare fd94393f...af4259e8
Subproject commit fd94393fb5b8ba8bae9c0bd6ab1c2a429d81ac76 Subproject commit af4259e8a33f095b419d1fd1733a50b22fc84c49
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment