sample_id
stringlengths
21
196
text
stringlengths
105
936k
metadata
dict
category
stringclasses
6 values
vllm-project/vllm:tests/kernels/attention/test_pack_unpack_triton.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import torch from torch.testing import assert_close from vllm.v1.attention.ops.common import pack_seq_triton, unpack_seq_triton def test_pack_seq_basic_fp8(): """Test basic functionality of pack_seq_triton with fp8 and 3D tensors.""" device = "cuda" dtype = torch.float8_e4m3fn # Test cases with 3D tensors (N, H, D) test_cases = [ (6, 8, 4, 2, [3, 3]), # (6, 8, 4) -> (2, 3, 8, 4) (10, 4, 8, 3, [2, 4, 4]), # (10, 4, 8) -> (3, 4, 4, 8) (20, 16, 32, 4, [5, 5, 5, 5]), # (20, 16, 32) -> (4, 5, 16, 32) ] for N, H, D, B, lengths_list in test_cases: # Create input tensor with small values for fp8 x = torch.randn(N, H, D, dtype=torch.float32, device=device) * 0.1 x = x.to(dtype=dtype) lengths = torch.tensor(lengths_list, device=device) # Pack the data packed = pack_seq_triton(x, lengths) # Check output shape and properties expected_shape = (B, max(lengths_list), H, D) assert packed.shape == expected_shape assert packed.dtype == dtype assert packed.device == x.device # Check that valid data is preserved (within fp8 precision) for b in range(B): start_idx = sum(lengths_list[:b]) seq_len = lengths_list[b] expected_data = x[start_idx : start_idx + seq_len].to(torch.float32) actual_data = packed[b, :seq_len].to(torch.float32) assert_close(actual_data, expected_data, rtol=1e-1, atol=1e-2) def test_pack_seq_custom_padding_fp8(): """Test pack_seq_triton with custom padding values for fp8.""" device = "cuda" dtype = torch.float8_e4m3fn N, H, D, B = 20, 8, 16, 2 lengths = torch.tensor([10, 10], device=device) x = torch.randn(N, H, D, dtype=torch.float32, device=device) * 0.1 x = x.to(dtype=dtype) # Test with different padding values for pad_value in [-100.0, -10.0, 0.0, 10.0, 100.0]: result = pack_seq_triton(x, lengths, pad_value=pad_value) # Check valid data for b in range(B): start_idx = b * 10 expected_data = x[start_idx : start_idx + 10].to(torch.float32) actual_data = result[b, :10].to(torch.float32) assert_close(actual_data, expected_data, rtol=1e-1, atol=1e-2) # Check padding (fp8 has limited range, so check for large values) padded_data = result[:, 10:].to(torch.float32) if pad_value < 0: assert torch.all(padded_data < -50) # Large negative values elif pad_value > 0: assert torch.all(padded_data > 50) # Large positive values else: assert torch.allclose(padded_data, torch.zeros_like(padded_data), atol=1e-2) def test_pack_seq_default_negative_inf_padding_fp8(): """Test that pack_seq_triton uses -inf padding by default for fp8.""" device = "cuda" dtype = torch.float8_e4m3fn # B = 2 N, H, D = 20, 8, 16 lengths = torch.tensor([10, 10], device=device) x = torch.randn(N, H, D, dtype=torch.float32, device=device) * 0.1 x = x.to(dtype=dtype) result = pack_seq_triton(x, lengths) # Check that padding is large negative values (fp8 representation of -inf) padded_data = result[:, 10:].to(torch.float32) assert torch.all( padded_data < -100 ) # fp8 -inf is represented as large negative number def test_pack_seq_edge_cases_fp8(): """Test pack_seq_triton with edge cases for fp8.""" device = "cuda" dtype = torch.float8_e4m3fn # Test with single batch element x = torch.randn(10, 8, 16, dtype=torch.float32, device=device) * 0.1 x = x.to(dtype=dtype) lengths = torch.tensor([10], device=device) result = pack_seq_triton(x, lengths) assert result.shape == (1, 10, 8, 16) # Test with very short sequences x = torch.randn(20, 4, 8, dtype=torch.float32, device=device) * 0.1 x = x.to(dtype=dtype) lengths = torch.tensor([1, 1, 1], device=device) result = pack_seq_triton(x, lengths) assert result.shape == (3, 1, 4, 8) # Test with different sequence lengths x = torch.randn(15, 8, 16, dtype=torch.float32, device=device) * 0.1 x = x.to(dtype=dtype) lengths = torch.tensor([5, 7, 3], device=device) result = pack_seq_triton(x, lengths) assert result.shape == (3, 7, 8, 16) def test_pack_seq_different_block_sizes_fp8(): """Test pack_seq_triton with different block sizes for fp8.""" device = "cuda" dtype = torch.float8_e4m3fn N, H, D, B = 100, 16, 32, 4 lengths = torch.tensor([25, 25, 25, 25], device=device) x = torch.randn(N, H, D, dtype=torch.float32, device=device) * 0.1 x = x.to(dtype=dtype) # Test different block sizes for block_t, block_d in [(32, 32), (64, 64), (128, 128)]: result = pack_seq_triton(x, lengths, block_t=block_t, block_d=block_d) assert result.shape == (B, 25, H, D) # Check that valid data is preserved (within fp8 precision) for b in range(B): start_idx = b * 25 expected_data = x[start_idx : start_idx + 25].to(torch.float32) actual_data = result[b, :25].to(torch.float32) assert_close(actual_data, expected_data, rtol=1e-1, atol=1e-2) def test_pack_seq_shape_consistency(): """Test that pack_seq_triton maintains shape consistency.""" device = "cuda" dtype = torch.float8_e4m3fn N, H, D, B = 20, 8, 16, 2 lengths = torch.tensor([10, 10], device=device) x = torch.randn(N, H, D, dtype=torch.float32, device=device) * 0.1 x = x.to(dtype=dtype) result = pack_seq_triton(x, lengths) # Check shape consistency assert result.shape[0] == B # Batch dimension assert result.shape[1] == lengths.max().item() # Max sequence length assert result.shape[2:] == x.shape[1:] # Feature dimensions preserved def test_pack_unpack_roundtrip_fp8(): """Test that pack -> unpack gives us back the original data for fp8.""" device = "cuda" dtype = torch.float8_e4m3fn # Test cases with 3D tensors test_cases = [ (6, 8, 4, 2, [3, 3]), (10, 4, 8, 3, [2, 4, 4]), (20, 16, 32, 4, [5, 5, 5, 5]), (15, 8, 16, 3, [7, 5, 3]), ] for N, H, D, B, lengths_list in test_cases: # Create input tensor with small values for fp8 x = torch.randn(N, H, D, dtype=torch.float32, device=device) * 0.1 x = x.to(dtype=dtype) lengths = torch.tensor(lengths_list, device=device) # Pack the data packed = pack_seq_triton(x, lengths) # Unpack the data unpacked = unpack_seq_triton(packed, lengths) # Check that we get back the original data (within fp8 precision) assert unpacked.shape == x.shape x_f32 = x.to(torch.float32) unpacked_f32 = unpacked.to(torch.float32) assert_close(x_f32, unpacked_f32, rtol=1e-3, atol=1e-3) # Unpack without explicit start locations (computed in kernel) unpacked_with_loc = unpack_seq_triton(packed, lengths) assert_close(x_f32, unpacked_with_loc.to(torch.float32), rtol=1e-3, atol=1e-2) def test_unpack_seq_triton_edge_cases_fp8(): """Test unpack function with edge cases for fp8.""" device = "cuda" dtype = torch.float8_e4m3fn # Test with single batch element x = torch.randn(10, 8, 16, dtype=torch.float32, device=device) * 0.1 x = x.to(dtype=dtype) lengths = torch.tensor([10], device=device) packed = pack_seq_triton(x, lengths) unpacked = unpack_seq_triton(packed, lengths) assert unpacked.shape == x.shape assert_close(x.to(torch.float32), unpacked.to(torch.float32), rtol=1e-1, atol=1e-2) # Test with very short sequences x = torch.randn(20, 4, 8, dtype=torch.float32, device=device) * 0.1 x = x.to(dtype=dtype) lengths = torch.tensor([1, 1, 1], device=device) packed = pack_seq_triton(x, lengths) unpacked = unpack_seq_triton(packed, lengths) # Only compare the first 3 elements that were actually packed assert_close( x[:3].to(torch.float32), unpacked.to(torch.float32), rtol=1e-1, atol=1e-2 ) x = torch.randn(15, 8, 16, dtype=torch.float32, device=device) * 0.1 x = x.to(dtype=dtype) lengths = torch.tensor([5, 7, 3], device=device) packed = pack_seq_triton(x, lengths) unpacked = unpack_seq_triton(packed, lengths) assert unpacked.shape == x.shape assert_close(x.to(torch.float32), unpacked.to(torch.float32), rtol=1e-1, atol=1e-2)
{ "repo_id": "vllm-project/vllm", "file_path": "tests/kernels/attention/test_pack_unpack_triton.py", "license": "Apache License 2.0", "lines": 184, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/v1/attention/test_sparse_mla_backends.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """Unit tests for the sparse MLA backends and utilities.""" import math from types import MethodType, SimpleNamespace import pytest import torch from tests.v1.attention.test_mla_backends import ( BATCH_SPECS, BatchSpec, MockSparseMLAAttentionLayer, create_and_prepopulate_kv_cache, ) from tests.v1.attention.utils import ( create_common_attn_metadata, create_standard_kv_cache_spec, create_vllm_config, ) from vllm import _custom_ops as ops from vllm.config import set_current_vllm_config from vllm.model_executor.layers.linear import ColumnParallelLinear from vllm.platforms import current_platform # TODO: Integrate ROCMAiterMLASparseBackend for ROCm. # The ROCm sparse MLA backend (rocm_aiter_mla_sparse.py) has a compatible # forward_mqa interface but needs validation on ROCm hardware. if not current_platform.is_cuda(): pytest.skip( "Sparse MLA backend tests currently only support CUDA. " "ROCm support requires integrating ROCMAiterMLASparseBackend.", allow_module_level=True, ) from vllm.utils.math_utils import cdiv from vllm.v1.attention.backends.mla.flashinfer_mla_sparse import ( FlashInferMLASparseBackend, ) from vllm.v1.attention.backends.mla.flashmla_sparse import ( FlashMLASparseBackend, triton_convert_req_index_to_global_index, ) from vllm.v1.attention.backends.utils import split_prefill_chunks from vllm.v1.attention.ops import flashmla SPARSE_BACKEND_BATCH_SPECS = { name: BATCH_SPECS[name] for name in [ "mixed_small", "mixed_medium", "small_prefill", "medium_prefill", "single_prefill", ] } SPARSE_BACKEND_BATCH_SPECS["large_q_prefill"] = BatchSpec( seq_lens=[1024] * 2, query_lens=[256] * 2 ) SPARSE_BACKEND_BATCH_SPECS["large_q_pure_prefill"] = BatchSpec( seq_lens=[256] * 2, query_lens=[256] * 2 ) def _float_to_e8m0_truncate(f: float) -> float: """Simulate SM100's float -> e8m0 -> bf16 scale conversion. e8m0 format only stores the exponent (power of 2). cudaRoundZero truncates toward zero, meaning we round down to the nearest power of 2. """ if f <= 0: return 0.0 # e8m0 = floor(log2(f)), then 2^(e8m0) # This is equivalent to truncating to the nearest power of 2 below f exp = math.floor(math.log2(f)) return 2.0**exp def _dequantize_fp8_ds_mla_entry( cache_slice: torch.Tensor, kv_lora_rank: int, rope_dim: int, dtype: torch.dtype, simulate_sm100_e8m0_scales: bool = False, ) -> tuple[torch.Tensor, torch.Tensor]: """Dequantize a single fp8_ds_mla cache entry back to latent + rope. Args: simulate_sm100_e8m0_scales: If True, simulate the SM100 kernel's float -> e8m0 -> bf16 scale conversion path. """ # The first kv_lora_rank bytes store FP8 latent values with one scale per # 128 element tile written as float32 right after the latent payload. scales = cache_slice.view(torch.float32)[kv_lora_rank // 4 : kv_lora_rank // 4 + 4] latent = torch.empty(kv_lora_rank, dtype=torch.float16, device=cache_slice.device) for tile_idx in range(4): tile_start = tile_idx * 128 tile_end = tile_start + 128 scale_val = float(scales[tile_idx].item()) if simulate_sm100_e8m0_scales: # Simulate the lossy float -> e8m0 -> bf16 conversion scale_val = _float_to_e8m0_truncate(scale_val) ops.convert_fp8( latent[tile_start:tile_end], cache_slice[tile_start:tile_end], scale_val, kv_dtype="fp8", ) latent = latent.to(dtype) rope_offset = kv_lora_rank // 2 + 8 rope_vals = cache_slice.view(dtype)[rope_offset : rope_offset + rope_dim] return latent, rope_vals.clone() def _quantize_dequantize_fp8_ds_mla( kv_c: torch.Tensor, k_pe: torch.Tensor, block_size: int, scale: torch.Tensor, simulate_sm100_e8m0_scales: bool = False, ) -> tuple[torch.Tensor, torch.Tensor]: """Round-trip kv_c/k_pe though the fp8_ds_mla cache layout. Args: simulate_sm100_e8m0_scales: If True, simulate the SM100 kernel's float -> e8m0 -> bf16 scale conversion in dequantization. """ if kv_c.numel() == 0: return kv_c.clone(), k_pe.clone() kv_lora_rank = kv_c.shape[-1] rope_dim = k_pe.shape[-1] num_tokens = kv_c.shape[0] num_blocks = max(1, math.ceil(num_tokens / block_size)) entry_size = kv_lora_rank + 4 * 4 + 2 * rope_dim tmp_cache = torch.zeros( num_blocks, block_size, entry_size, dtype=torch.uint8, device=kv_c.device ) slot_mapping = torch.arange(num_tokens, dtype=torch.long, device=kv_c.device) ops.concat_and_cache_mla( kv_c, k_pe, tmp_cache, slot_mapping, kv_cache_dtype="fp8_ds_mla", scale=scale ) dequant_kv_c = torch.empty_like(kv_c) dequant_k_pe = torch.empty_like(k_pe) for token_idx in range(num_tokens): slot = slot_mapping[token_idx].item() block_idx = slot // block_size block_offset = slot % block_size cache_slice = tmp_cache[block_idx, block_offset] latent, rope_vals = _dequantize_fp8_ds_mla_entry( cache_slice, kv_lora_rank, rope_dim, kv_c.dtype, simulate_sm100_e8m0_scales=simulate_sm100_e8m0_scales, ) dequant_kv_c[token_idx] = latent dequant_k_pe[token_idx] = rope_vals return dequant_kv_c, dequant_k_pe @pytest.mark.parametrize( "backend_cls", [FlashMLASparseBackend, FlashInferMLASparseBackend], ids=["FlashMLA", "FlashInfer"], ) @pytest.mark.parametrize("batch_name", list(SPARSE_BACKEND_BATCH_SPECS.keys())) @pytest.mark.parametrize("kv_cache_dtype", ["auto", "fp8", "fp8_ds_mla"]) @pytest.mark.parametrize("tensor_parallel_size", [1, 2, 4]) @pytest.mark.parametrize("block_size", [32, 64]) def test_sparse_backend_decode_correctness( default_vllm_config, dist_init, backend_cls, batch_name, kv_cache_dtype, tensor_parallel_size, block_size, workspace_init, ): if kv_cache_dtype not in backend_cls.supported_kv_cache_dtypes: pytest.skip(f"{backend_cls.get_name()} does not support {kv_cache_dtype}") supported_block_sizes = backend_cls.get_supported_kernel_block_sizes() if block_size not in supported_block_sizes: pytest.skip( f"{backend_cls.get_name()} does not support block_size={block_size}" ) if backend_cls == FlashMLASparseBackend: ok, reason = flashmla.is_flashmla_sparse_supported() if not ok: pytest.skip(reason) elif backend_cls == FlashInferMLASparseBackend: if not current_platform.has_device_capability(100): pytest.skip("FlashInferMLASparseBackend requires SM 10.0 or higher") batch_spec = SPARSE_BACKEND_BATCH_SPECS[batch_name] use_fp8_ds_mla_quantization = kv_cache_dtype == "fp8_ds_mla" device = torch.device("cuda") dtype = torch.bfloat16 # Model hyper-parameters (kept intentionally small for the unit test) total_num_heads = 128 # Compute per-rank heads for simulated TP num_heads = max(1, total_num_heads // tensor_parallel_size) kv_lora_rank = 512 qk_nope_head_dim = 128 qk_rope_head_dim = 64 v_head_dim = 128 head_size = kv_lora_rank + qk_rope_head_dim topk_tokens = 128 max_seqlen = max(batch_spec.seq_lens) total_cache_tokens = sum(batch_spec.seq_lens) # Note: We use TP=1 to avoid multi-GPU requirements in CI. # The test simulates head partitioning via mocked methods below. vllm_config = create_vllm_config( model_name="deepseek-ai/DeepSeek-V2-Lite-Chat", tensor_parallel_size=1, max_model_len=max_seqlen, num_gpu_blocks=max(2048, cdiv(total_cache_tokens, block_size) + 1), block_size=block_size, hf_config_override={ "index_topk": topk_tokens, "attn_module_list_cfg": [{"topk_tokens": topk_tokens}], }, ) model_config = vllm_config.model_config model_config.hf_text_config = SimpleNamespace( q_lora_rank=None, kv_lora_rank=kv_lora_rank, qk_nope_head_dim=qk_nope_head_dim, qk_rope_head_dim=qk_rope_head_dim, v_head_dim=v_head_dim, model_type="deepseek_v2", ) model_config.dtype = dtype model_config.get_num_attention_heads = MethodType( lambda self, parallel_config: num_heads, model_config, ) model_config.get_num_kv_heads = MethodType( lambda self, parallel_config: 1, model_config ) model_config.get_head_size = MethodType(lambda self: head_size, model_config) model_config.get_sliding_window = MethodType(lambda self: None, model_config) kv_cache_spec = create_standard_kv_cache_spec(vllm_config) torch.manual_seed(0) scale = 1.0 / math.sqrt(head_size) # Shared MLA projection weights to keep reference and backend in sync W_UK = torch.rand( kv_lora_rank, num_heads, qk_nope_head_dim, dtype=dtype, device=device ) W_UV = torch.rand(kv_lora_rank, num_heads, v_head_dim, dtype=dtype, device=device) # Build synthetic decode-only workload seq_lens = batch_spec.seq_lens query_lens = batch_spec.query_lens # Pre-compute positions and sparse indices for all tokens. # We need these BEFORE computing the reference to use sparse attention masks. total_query_tokens = sum(query_lens) positions = [] for i in range(batch_spec.batch_size): s_len = seq_lens[i] q_len = query_lens[i] ctx_len = s_len - q_len for q_idx in range(q_len): positions.append(ctx_len + q_idx) # Create sparse indices with UNIQUE per-token offsets to catch bugs where # the kernel uses wrong indices for some tokens (e.g., due to incorrect # tensor shapes like [1, num_tokens, ...] instead of [num_tokens, 1, ...]). # Also include -1 masked indices to verify the kernel handles them correctly. sparse_indices = torch.empty( total_query_tokens, topk_tokens, dtype=torch.int32, device=device ) for tok_idx in range(total_query_tokens): max_valid_idx = positions[tok_idx] offset = tok_idx * 7 # Prime number for varied offsets # Use only half the topk indices as valid, mask the rest with -1 # This tests that the kernel correctly ignores -1 indices num_valid = min(topk_tokens // 2, max_valid_idx + 1) if num_valid > 0: valid_range = torch.arange(num_valid, device=device, dtype=torch.int32) tok_indices = (valid_range + offset) % (max_valid_idx + 1) # Pad with -1 for the remaining positions tok_indices = torch.cat( [ tok_indices, torch.full( (topk_tokens - num_valid,), -1, device=device, dtype=torch.int32 ), ] ) else: tok_indices = torch.full( (topk_tokens,), -1, device=device, dtype=torch.int32 ) tok_indices[0] = 0 # At least one valid index sparse_indices[tok_idx] = tok_indices all_q_vllm, all_kv_c_vllm, all_k_pe_vllm = [], [], [] kv_c_contexts, k_pe_contexts = [], [] reference_outputs = [] kv_cache_scale = torch.tensor(1.0, dtype=torch.float32, device=device) global_token_idx = 0 for i in range(batch_spec.batch_size): s_len = seq_lens[i] q_len = query_lens[i] ctx_len = s_len - q_len q_c = torch.rand( q_len, num_heads, qk_nope_head_dim + qk_rope_head_dim, dtype=dtype, device=device, ) kv_c_full = torch.rand(s_len, kv_lora_rank, dtype=dtype, device=device) k_pe_full = torch.rand(s_len, 1, qk_rope_head_dim, dtype=dtype, device=device) if use_fp8_ds_mla_quantization: is_sm100 = torch.cuda.get_device_capability()[0] >= 10 kv_c_full, k_pe_squeezed = _quantize_dequantize_fp8_ds_mla( kv_c_full, k_pe_full.squeeze(1), block_size=block_size, scale=kv_cache_scale, simulate_sm100_e8m0_scales=is_sm100, ) k_pe_full = k_pe_squeezed.unsqueeze(1) q_nope, q_pe = q_c.split([qk_nope_head_dim, qk_rope_head_dim], dim=-1) ql_nope = torch.einsum("qnh,lnh->qnl", q_nope, W_UK) q_mqa = torch.cat([ql_nope, q_pe], dim=-1) k_mqa = torch.cat([kv_c_full, k_pe_full.squeeze(1)], dim=-1) v_mqa = kv_c_full # Compute sparse SDPA reference per query token using its sparse indices for q_idx in range(q_len): tok_sparse_idx = sparse_indices[global_token_idx] valid_mask = tok_sparse_idx >= 0 valid_indices = tok_sparse_idx[valid_mask].long() q_tok = q_mqa[q_idx : q_idx + 1] # [1, num_heads, head_dim] k_sparse = k_mqa[valid_indices] # [num_valid, head_dim] v_sparse = v_mqa[valid_indices] # [num_valid, kv_lora_rank] k_sparse = k_sparse.unsqueeze(1).expand(-1, num_heads, -1) v_sparse = v_sparse.unsqueeze(1).expand(-1, num_heads, -1) # SDPA: [1, num_heads, 1, head_dim] x [1, num_heads, num_valid, head_dim] q_sdpa_in = q_tok.unsqueeze(0).transpose(1, 2) k_sdpa_in = k_sparse.unsqueeze(0).transpose(1, 2) v_sdpa_in = v_sparse.unsqueeze(0).transpose(1, 2) sdpa_out = torch.nn.functional.scaled_dot_product_attention( q_sdpa_in, k_sdpa_in, v_sdpa_in, scale=scale ) sdpa_out = sdpa_out.transpose(1, 2).squeeze( 0 ) # [1, num_heads, kv_lora_rank] sdpa_out = torch.einsum("qnl,lnv->qnv", sdpa_out, W_UV) reference_outputs.append(sdpa_out.flatten(start_dim=-2)) global_token_idx += 1 all_q_vllm.append(q_c) all_kv_c_vllm.append(kv_c_full[ctx_len:]) all_k_pe_vllm.append(k_pe_full[ctx_len:]) kv_c_contexts.append(kv_c_full[: ctx_len + 1]) k_pe_contexts.append(k_pe_full[: ctx_len + 1]) query_vllm = torch.cat(all_q_vllm, dim=0) kv_c_vllm = torch.cat(all_kv_c_vllm, dim=0) k_pe_vllm = torch.cat(all_k_pe_vllm, dim=0) sdpa_reference = torch.cat(reference_outputs, dim=0) vllm_config.cache_config.cache_dtype = kv_cache_dtype vllm_config.model_config.hf_config.index_topk = topk_tokens common_attn_metadata = create_common_attn_metadata( batch_spec, vllm_config.cache_config.block_size, device, arange_block_indices=True, ) kv_cache = create_and_prepopulate_kv_cache( kv_c_contexts=kv_c_contexts, k_pe_contexts=k_pe_contexts, block_size=vllm_config.cache_config.block_size, head_size=head_size, dtype=dtype, device=device, num_blocks=vllm_config.cache_config.num_gpu_blocks, common_attn_metadata=common_attn_metadata, randomize_blocks=False, kv_cache_dtype=kv_cache_dtype if use_fp8_ds_mla_quantization else "auto", scale=kv_cache_scale, ) builder_cls = backend_cls.get_builder_cls() builder = builder_cls(kv_cache_spec, ["placeholder"], vllm_config, device) metadata = builder.build( common_prefix_len=0, common_attn_metadata=common_attn_metadata ) # Use the pre-computed sparse_indices for the mock indexer mock_indexer = SimpleNamespace(topk_indices_buffer=sparse_indices) kv_b_proj_weight = torch.cat([W_UK, W_UV], dim=-1) kv_b_proj_weight = kv_b_proj_weight.view( kv_lora_rank, num_heads * (qk_nope_head_dim + v_head_dim) ) mock_kv_b_proj = ColumnParallelLinear( input_size=kv_lora_rank, output_size=num_heads * (qk_nope_head_dim + v_head_dim), bias=False, ).to(device=device, dtype=dtype) mock_kv_b_proj.weight = torch.nn.Parameter(kv_b_proj_weight.T.contiguous()) impl_cls = backend_cls.get_impl_cls() with set_current_vllm_config(vllm_config): impl = impl_cls( num_heads=num_heads, head_size=head_size, scale=scale, num_kv_heads=1, alibi_slopes=None, sliding_window=None, kv_cache_dtype=vllm_config.cache_config.cache_dtype, logits_soft_cap=None, attn_type="decoder", kv_sharing_target_layer_name=None, q_lora_rank=None, kv_lora_rank=kv_lora_rank, qk_nope_head_dim=qk_nope_head_dim, qk_rope_head_dim=qk_rope_head_dim, qk_head_dim=qk_nope_head_dim + qk_rope_head_dim, v_head_dim=v_head_dim, kv_b_proj=mock_kv_b_proj, indexer=mock_indexer, ) impl.process_weights_after_loading(dtype) # Create mock sparse MLA layer with weight matrices mock_layer = MockSparseMLAAttentionLayer( impl=impl, num_heads=num_heads, qk_nope_head_dim=qk_nope_head_dim, qk_rope_head_dim=qk_rope_head_dim, v_head_dim=v_head_dim, kv_lora_rank=kv_lora_rank, device=device, W_UK=W_UK, W_UV=W_UV, ) out_buffer = torch.empty( metadata.num_actual_tokens, num_heads * v_head_dim, dtype=dtype, device=device ) with torch.inference_mode(): backend_output = mock_layer.forward_impl( query_vllm, kv_c_vllm, k_pe_vllm, kv_cache, metadata, out_buffer, ) assert backend_output.shape == sdpa_reference.shape assert backend_output.dtype == sdpa_reference.dtype assert torch.isfinite(backend_output).all() # FP8 quantization introduces some error, but should be within reasonable bounds # BF16 (auto) should be very accurate, FP8 allows slightly more tolerance if kv_cache_dtype.startswith("fp8"): torch.testing.assert_close(backend_output, sdpa_reference, rtol=0.05, atol=0.05) else: torch.testing.assert_close(backend_output, sdpa_reference, rtol=0.01, atol=0.01) def _triton_convert_reference_impl( req_ids: torch.Tensor, block_table: torch.Tensor, token_indices: torch.Tensor, block_size: int, num_topk_tokens: int, HAS_PREFILL_WORKSPACE: bool = False, prefill_workspace_request_ids: torch.Tensor | None = None, prefill_workspace_starts: torch.Tensor | None = None, ) -> torch.Tensor: """Reference implementation for triton_convert_req_index_to_global_index.""" num_tokens = req_ids.shape[0] max_blocks_per_req = block_table.shape[1] result = torch.empty( num_tokens, num_topk_tokens, dtype=torch.int32, device=req_ids.device ) for token_id in range(num_tokens): req_id = req_ids[token_id].item() # Determine if this token uses workspace or paged cache use_prefill_workspace = False workspace_start = 0 if HAS_PREFILL_WORKSPACE and prefill_workspace_request_ids is not None: assert prefill_workspace_starts is not None prefill_req_id = prefill_workspace_request_ids[token_id].item() if prefill_req_id >= 0: use_prefill_workspace = True workspace_start = prefill_workspace_starts[prefill_req_id].item() for idx_id in range(num_topk_tokens): token_idx = token_indices[token_id, idx_id].item() if token_idx == -1: result[token_id, idx_id] = -1 elif use_prefill_workspace: # Prefill + using prefill workspace: map to workspace offset result[token_id, idx_id] = workspace_start + token_idx else: # Decode: map to paged cache block_id = token_idx // block_size if block_id >= max_blocks_per_req: result[token_id, idx_id] = -1 else: block_num = block_table[req_id, block_id].item() offset = token_idx % block_size result[token_id, idx_id] = block_num * block_size + offset return result @pytest.mark.parametrize("block_size", [16, 64, 128]) @pytest.mark.parametrize("num_topk_tokens", [128, 256, 512]) @pytest.mark.skipif( torch.cuda.get_device_capability() < (9, 0), reason="FlashMLASparseBackend requires CUDA 9.0 or higher", ) def test_triton_convert_req_index_to_global_index_decode_only( block_size, num_topk_tokens ): device = torch.device("cuda") num_tokens = 8 num_requests = 4 max_blocks_per_req = 10 req_id = torch.randint( 0, num_requests, (num_tokens,), dtype=torch.int32, device=device ) block_table = torch.randint( 0, 100, (num_requests, max_blocks_per_req), dtype=torch.int32, device=device ) token_indices = torch.randint( 0, block_size * max_blocks_per_req, (num_tokens, num_topk_tokens), dtype=torch.int32, device=device, ) # Set some to -1 to test masking token_indices[0, :10] = -1 token_indices[3, 50:60] = -1 # Set some to out of bounds token_indices[2, 100:110] = max_blocks_per_req * block_size token_indices[6, 150:160] = max_blocks_per_req * block_size result = triton_convert_req_index_to_global_index( req_id, block_table, token_indices, BLOCK_SIZE=block_size, NUM_TOPK_TOKENS=num_topk_tokens, ) reference_result = _triton_convert_reference_impl( req_id, block_table, token_indices, block_size, num_topk_tokens, ) torch.testing.assert_close(result, reference_result, rtol=0, atol=0) @pytest.mark.parametrize("block_size", [16]) @pytest.mark.skipif( torch.cuda.get_device_capability() < (9, 0), reason="FlashMLASparseBackend requires CUDA 9.0 or higher", ) def test_triton_convert_req_index_to_global_index_with_prefill_workspace(block_size): device = torch.device("cuda") num_requests = 4 max_blocks_per_req = 8 num_topk_tokens = 128 # First 6 tokens are decode (reqs 0, 1), last 6 are prefill (reqs 2, 3) req_id = torch.tensor( [0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3], dtype=torch.int32, device=device ) prefill_workspace_request_ids = torch.tensor( [-1, -1, -1, -1, -1, -1, 0, 0, 0, 1, 1, 1], dtype=torch.int32, device=device ) # Workspace starts for the 2 prefill reqs: req 2 starts at 0, req 3 starts at 100 prefill_workspace_starts = torch.tensor([0, 100], dtype=torch.int32, device=device) block_table = torch.randint( 0, 50, (num_requests, max_blocks_per_req), dtype=torch.int32, device=device ) token_indices = torch.randint( 0, block_size * max_blocks_per_req, (req_id.shape[0], num_topk_tokens), dtype=torch.int32, device=device, ) # Set some to -1 to test masking token_indices[0, :10] = -1 token_indices[3, 50:60] = -1 # Set some to out of bounds token_indices[2, 100:110] = max_blocks_per_req * block_size token_indices[6, 150:160] = max_blocks_per_req * block_size result = triton_convert_req_index_to_global_index( req_id, block_table, token_indices, BLOCK_SIZE=block_size, NUM_TOPK_TOKENS=num_topk_tokens, HAS_PREFILL_WORKSPACE=True, prefill_workspace_request_ids=prefill_workspace_request_ids, prefill_workspace_starts=prefill_workspace_starts, ) reference_result = _triton_convert_reference_impl( req_id, block_table, token_indices, block_size, num_topk_tokens, HAS_PREFILL_WORKSPACE=True, prefill_workspace_request_ids=prefill_workspace_request_ids, prefill_workspace_starts=prefill_workspace_starts, ) torch.testing.assert_close(result, reference_result, rtol=0, atol=0) @pytest.mark.parametrize( "seq_lens,max_buf,expected", [ # Basic split: totals per chunk ≤ max_buf (torch.tensor([2, 3, 4, 2]), 5, [(0, 2), (2, 3), (3, 4)]), # Exact fits should split between items when adding the next would overflow (torch.tensor([5, 5, 5]), 5, [(0, 1), (1, 2), (2, 3)]), # All requests fit in a single chunk (torch.tensor([1, 1, 1]), 10, [(0, 3)]), # Large buffer (torch.tensor([4, 4, 4]), 100, [(0, 3)]), ], ) def test_split_prefill_chunks(seq_lens, max_buf, expected): out = split_prefill_chunks(seq_lens, max_buf) assert out == expected def test_triton_convert_returns_valid_counts(): """Test that return_valid_counts correctly counts non-negative indices.""" device = torch.device("cuda") num_tokens = 8 num_requests = 2 max_blocks_per_req = 10 block_size = 64 num_topk_tokens = 128 req_id = torch.tensor([0, 0, 0, 0, 1, 1, 1, 1], dtype=torch.int32, device=device) block_table = torch.arange( num_requests * max_blocks_per_req, dtype=torch.int32, device=device ).view(num_requests, max_blocks_per_req) # Create token indices with varying numbers of valid entries # Token 0: 64 valid, 64 invalid (-1) # Token 1: 32 valid, 96 invalid # Token 2: 128 valid (all) # Token 3: 1 valid, 127 invalid # etc. token_indices = torch.full( (num_tokens, num_topk_tokens), -1, dtype=torch.int32, device=device ) expected_valid = [] for i in range(num_tokens): num_valid = [64, 32, 128, 1, 64, 32, 128, 1][i] token_indices[i, :num_valid] = torch.arange( num_valid, dtype=torch.int32, device=device ) % (block_size * max_blocks_per_req) expected_valid.append(num_valid) expected_valid_tensor = torch.tensor( expected_valid, dtype=torch.int32, device=device ) # Test with return_valid_counts=True result, valid_counts = triton_convert_req_index_to_global_index( req_id, block_table, token_indices, BLOCK_SIZE=block_size, NUM_TOPK_TOKENS=num_topk_tokens, return_valid_counts=True, ) torch.testing.assert_close(valid_counts, expected_valid_tensor, rtol=0, atol=0) # Test that return_valid_counts=False returns only the indices result_only = triton_convert_req_index_to_global_index( req_id, block_table, token_indices, BLOCK_SIZE=block_size, NUM_TOPK_TOKENS=num_topk_tokens, return_valid_counts=False, ) assert isinstance(result_only, torch.Tensor) torch.testing.assert_close(result_only, result, rtol=0, atol=0)
{ "repo_id": "vllm-project/vllm", "file_path": "tests/v1/attention/test_sparse_mla_backends.py", "license": "Apache License 2.0", "lines": 652, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/v1/attention/backends/mla/flashmla_sparse.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from dataclasses import dataclass from typing import TYPE_CHECKING, ClassVar import numpy as np import torch from vllm import _custom_ops as ops from vllm.config import VllmConfig, get_current_vllm_config from vllm.config.cache import CacheDType from vllm.logger import init_logger from vllm.model_executor.layers.attention.mla_attention import ( get_mla_dims, ) from vllm.platforms import current_platform from vllm.platforms.interface import DeviceCapability from vllm.utils.platform_utils import num_compute_units from vllm.v1.attention.backend import ( AttentionBackend, AttentionCGSupport, AttentionLayer, AttentionMetadata, AttentionMetadataBuilder, CommonAttentionMetadata, MultipleOf, SparseMLAAttentionImpl, ) from vllm.v1.attention.backends.mla.sparse_utils import ( triton_convert_req_index_to_global_index, ) from vllm.v1.attention.backends.utils import ( reshape_attn_output_for_spec_decode, reshape_query_for_spec_decode, split_decodes_and_prefills, split_prefill_chunks, ) from vllm.v1.attention.ops.flashmla import ( FlashMLASchedMeta, flash_mla_sparse_fwd, flash_mla_with_kvcache, get_mla_metadata, ) from vllm.v1.kv_cache_interface import AttentionSpec from vllm.v1.worker.workspace import current_workspace_manager if TYPE_CHECKING: from vllm.model_executor.models.deepseek_v2 import Indexer logger = init_logger(__name__) # For FP8 sparse attention we have two impelementations: # 1. Mixed batch mode: use the FP8 decode kernel for both prefill and decode this is # done by treating all tokens as single batch. # 2. Separate prefill and decode mode: use the BF16 prefill kernel for prefill # (upconverting the FP8 cache to BF16 then calling the prefill kernel) and using # the FP8 decode kernel for decode. # Currently we use #1 when the number of heads per rank is low (i.e. TP) since the BF16 # prefill kernel requires padding the numer of heads to 128 while the decode does not # so when the per ranke head count is below MIN_HEADS_FOR_BF16_PREFILL we use the mixed # batch mode (#2). MIN_HEADS_FOR_BF16_PREFILL = 32 """ NOTE: FlashMLA Sparse uses an fp8 cache with the following format In the "FP8 with scale" format, each token's KV cache is 656 Bytes, structured as: - **First 512 bytes:** The "quantized NoPE" part, containing 512 `float8_e4m3` values. - **Next 16 bytes:** Scale factors, containing 4 `float32` values. The first `float32` is the scale for the first 128 `float8_e4m3` values, the second for the next 128, and so on. - **Last 128 bytes:** The "RoPE" part, containing 64 `bfloat16` values. This part is not quantized for accuracy. """ class FlashMLASparseBackend(AttentionBackend): accept_output_buffer: bool = True supported_dtypes: ClassVar[list[torch.dtype]] = [torch.bfloat16] supported_kv_cache_dtypes: ClassVar[list[CacheDType]] = [ "auto", "bfloat16", "fp8_ds_mla", ] @staticmethod def get_supported_kernel_block_sizes() -> list[int | MultipleOf]: return [64] @staticmethod def get_name() -> str: return "FLASHMLA_SPARSE" @staticmethod def get_builder_cls() -> type["FlashMLASparseMetadataBuilder"]: return FlashMLASparseMetadataBuilder @staticmethod def get_impl_cls() -> type["FlashMLASparseImpl"]: return FlashMLASparseImpl @classmethod def get_supported_head_sizes(cls) -> list[int]: return [576] @classmethod def is_mla(cls) -> bool: return True @classmethod def is_sparse(cls) -> bool: return True @classmethod def supports_compute_capability(cls, capability: DeviceCapability) -> bool: return capability.major in [9, 10] @staticmethod def get_kv_cache_shape( num_blocks: int, block_size: int, num_kv_heads: int, # assumed to be 1 for MLA head_size: int, cache_dtype_str: str = "auto", ) -> tuple[int, ...]: if cache_dtype_str == "fp8_ds_mla": # custom storage fromat is 656 bytes # see FlashMLA readme.md for details return (num_blocks, block_size, 656) else: return (num_blocks, block_size, head_size) @dataclass class FlashMLASparseMetadata(AttentionMetadata): num_reqs: int max_query_len: int max_seq_len: int num_actual_tokens: int # Number of tokens excluding padding. query_start_loc: torch.Tensor slot_mapping: torch.Tensor block_table: torch.Tensor req_id_per_token: torch.Tensor block_size: int = 64 topk_tokens: int = 2048 @dataclass class FP8KernelMetadata: scheduler_metadata: FlashMLASchedMeta dummy_block_table: torch.Tensor cache_lens: torch.Tensor @dataclass class FP8SeparatePrefillDecode: @dataclass class Decode: kernel_metadata: "FlashMLASparseMetadata.FP8KernelMetadata" decode_query_len: int # needed for reshape in spec decode @dataclass class Prefill: # Sequence lengths (context + query) for prefill requests # Shape: [num_prefill_reqs] seq_lens: torch.Tensor # Request ID for each token: -1 for decode tokens, request index # (0, 1, 2, ...) for prefill tokens. # Shape: [num_actual_tokens] request_ids: torch.Tensor # Workspace start offsets for all prefill requests # Shape: [num_prefill_reqs], adjusted in-place per chunk to be # 0-indexed within each chunk. Used to map prefill tokens to workspace # offsets in convert_logical_index_to_physical_index workspace_starts: torch.Tensor @dataclass class Chunk: """Metadata for a chunk of prefill requests. Prefill requests may be chunked to fit within the fixed workspace size. """ seq_lens: torch.Tensor tokens_slice: slice block_table: torch.Tensor req_start_idx: int workspace_starts: torch.Tensor chunk_tot_seqlen: int chunks: list[Chunk] num_prefills: int = 0 num_decodes: int = 0 num_prefill_tokens: int = 0 num_decode_tokens: int = 0 decode: Decode | None = None prefill: Prefill | None = None fp8_extra_metadata: FP8SeparatePrefillDecode | FP8KernelMetadata | None = None fp8_use_mixed_batch: bool = False def get_prefill_workspace_size(max_model_len: int): # NOTE(Lucas): 5 is a magic number for controlling the prefill buffer size. # May be tuned later. # Memory usage: 5 * max_model_len * 576 * 2 bytes # Example: DeepSeek-V3.2 with max_model_len=163840 -> # 5 * 163840 * 576 * 2 = ~900 MB # This fits nicely below the typical MoE workspace size of >2GB so this is "free" return max_model_len * 5 class FlashMLASparseMetadataBuilder(AttentionMetadataBuilder[FlashMLASparseMetadata]): _cudagraph_support: ClassVar[AttentionCGSupport] = AttentionCGSupport.UNIFORM_BATCH def __init__( self, kv_cache_spec: AttentionSpec, layer_names: list[str], vllm_config: VllmConfig, device: torch.device, ) -> None: self.vllm_config = vllm_config self.layer_names = layer_names cache_config = vllm_config.cache_config self.kv_cache_spec = kv_cache_spec self.model_config = vllm_config.model_config parallel_config = vllm_config.parallel_config self.device = device # Treat requests with query length <= 1 as decodes to match the # DeepGEMM indexer constraint (fp8_paged_mqa_logits only supports next_n <= 2) self._init_reorder_batch_threshold(1, supports_spec_as_decode=True) sm_count = num_compute_units(device.index) self.num_heads = self.model_config.get_num_attention_heads(parallel_config) self.mla_dims = get_mla_dims(self.model_config) # FP8 decode kernel only supports h_q = 64 or 128, so we need to pad self.fp8_decode_padded_heads = ( FlashMLASparseImpl._compute_fp8_decode_padded_heads(self.num_heads) ) self.topk_tokens = vllm_config.model_config.hf_config.index_topk self.use_fp8_kv_cache = cache_config.cache_dtype == "fp8_ds_mla" max_num_seqs = vllm_config.scheduler_config.max_num_seqs # Shape: [max_num_seqs], all elements = topk_tokens (constant for full-CG) self.topk_tokens_tensor = torch.full( (max_num_seqs,), self.topk_tokens, device=device, dtype=torch.int32 ) # Shape: [max_num_seqs], all elements = max_model_len self.max_model_len_tensor = torch.full( (max_num_seqs,), self.model_config.max_model_len, device=device, dtype=torch.int32, ) # this is ignored by `flash_mla_with_kvcache` if indices not None self.dummy_block_table = torch.empty( (max_num_seqs, 1), dtype=torch.int32, device=self.device ) # Equation taken from FlashMLA/csrc/api/sparse_decode.h # For sparse FP8 decode, the formula depends on architecture: # - SM90 (Hopper): num_sm_parts = num_sms / s_q / (h_q/64) # - SM100 (Blackwell head64/head64x2): num_sm_parts = num_sms / s_q # - SM100 (Blackwell head128): num_sm_parts = num_sms / s_q / 2 # For max buffer size, use s_q = 1 (the case that produces largest output) # Use padded head count since that's what will be passed to the kernel h_q = self.fp8_decode_padded_heads if current_platform.is_device_capability_family(100): # SM100 head64 or head64x2 uses full SM count max_num_sm_parts = sm_count else: # SM90 uses h_q/64 divisor max_num_sm_parts = sm_count // max(1, h_q // 64) self.tile_scheduler_metadata_buffer = torch.empty( # TileSchedulerMetaDataSize = 8 # see: FlashMLA/csrc/params.h (max_num_sm_parts, 8), dtype=torch.int32, device=device, ) # Sized for per-request batching (num_decodes + 1) self.num_splits_buffer = torch.empty( (max_num_seqs + 1,), dtype=torch.int32, device=device, ) self.req_id_per_token_buffer = torch.empty( (vllm_config.scheduler_config.max_num_batched_tokens,), dtype=torch.int32, device=device, ) def _build_fp8_mixed_decode_prefill( self, common_attn_metadata: CommonAttentionMetadata, ) -> "FlashMLASparseMetadata.FP8KernelMetadata": """Build FP8 metadata treating all tokens as one mixed batch. This matches main branch's approach and avoids the BF16 prefill kernel which has head padding overhead when num_heads is small (high TP case). """ num_tokens = common_attn_metadata.num_actual_tokens # Use padded head count since that's what the kernel will see padded_heads = self.fp8_decode_padded_heads # Build metadata for all tokens as a single batch scheduler_metadata, _ = get_mla_metadata( cache_seqlens=self.topk_tokens_tensor[:1], # Single batch num_q_tokens_per_head_k=num_tokens * padded_heads, topk=self.topk_tokens, num_heads_q=padded_heads, num_heads_k=1, is_fp8_kvcache=True, ) fp8_metadata = FlashMLASparseMetadata.FP8KernelMetadata( scheduler_metadata=scheduler_metadata, cache_lens=self.max_model_len_tensor[:1], dummy_block_table=self.dummy_block_table[:1], ) return fp8_metadata def _build_fp8_separate_prefill_decode( self, common_attn_metadata: CommonAttentionMetadata, ) -> "FlashMLASparseMetadata.FP8SeparatePrefillDecode": num_tokens = common_attn_metadata.num_actual_tokens (num_decodes, num_prefills, num_decode_tokens, num_prefill_tokens) = ( split_decodes_and_prefills( common_attn_metadata, decode_threshold=self.reorder_batch_threshold or 1, require_uniform=True, ) ) FP8Meta = FlashMLASparseMetadata.FP8SeparatePrefillDecode fp8_metadata = FP8Meta( num_decodes=num_decodes, num_prefills=num_prefills, num_decode_tokens=num_decode_tokens, num_prefill_tokens=num_prefill_tokens, ) # Extract prefill sequence lengths (context + query, not just query) # Decode requests come first in the batch, prefill requests follow prefill_seq_lens = None prefill_request_id = None prefill_workspace_starts = None prefill_chunks = None # For pure decode batches, prefill_request_id will be None # For mixed batches, it will have -1 for decode and request_id for prefill if num_prefills > 0: seq_lens_cpu = common_attn_metadata.seq_lens.cpu() seq_lens = common_attn_metadata.seq_lens query_start_loc_cpu = common_attn_metadata.query_start_loc_cpu prefill_seq_lens_cpu = seq_lens_cpu[num_decodes:] prefill_seq_lens = seq_lens[num_decodes:] # Build prefill_request_id: -1 for decode, request index for # prefill. This enables a single # convert_logical_index_to_physical_index call for all tokens prefill_request_id = torch.full( (num_tokens,), -1, dtype=torch.int32, device=self.device ) # Map prefill tokens to their request IDs (0, 1, 2, ...) for req_idx in range(num_prefills): # Get query token range for this prefill request global_req_idx = num_decodes + req_idx req_query_start = query_start_loc_cpu[global_req_idx] req_query_end = query_start_loc_cpu[global_req_idx + 1] prefill_request_id[req_query_start:req_query_end] = req_idx # will be adjusted by chunk loop prefill_workspace_starts_cpu = torch.zeros( num_prefills, dtype=torch.int32, pin_memory=True ) prefill_workspace_starts_cpu[1:] = torch.cumsum( prefill_seq_lens_cpu[:-1], dim=0 ) # populated by non-blocking copy after prefill_workspace_starts_cpu is # updated by each chunk prefill_workspace_starts = torch.empty( num_prefills, dtype=torch.int32, device=self.device ) # Chunk prefill requests to fit within workspace size max_prefill_buffer_size = get_prefill_workspace_size( self.vllm_config.model_config.max_model_len ) chunk_bounds = split_prefill_chunks( prefill_seq_lens_cpu, max_prefill_buffer_size ) prefill_chunks = [] for chunk_start, chunk_end in chunk_bounds: # Adjust workspace_starts in-place per chunk to be # 0-indexed within each chunk # Example: seq_lens=[10,15,20,5], chunks=[[0,2],[2,4]] # Initial: workspace_starts=[0,10,25,45] # After: workspace_starts=[0,10,0,20] # (chunk 0 starts at 0, chunk 1 starts at 0) offset = prefill_workspace_starts_cpu[chunk_start].item() prefill_workspace_starts_cpu[chunk_start:chunk_end] -= offset chunk_seq_lens = prefill_seq_lens[chunk_start:chunk_end] chunk_tot_seqlen = prefill_seq_lens_cpu[chunk_start:chunk_end].sum() token_start = query_start_loc_cpu[num_decodes + chunk_start].item() token_end = query_start_loc_cpu[num_decodes + chunk_end].item() tokens_slice = slice(token_start, token_end) # Create chunk view of gpu tensor chunk_workspace_starts = prefill_workspace_starts[chunk_start:chunk_end] chunk_block_table = common_attn_metadata.block_table_tensor[ num_decodes + chunk_start : num_decodes + chunk_end ] prefill_chunks.append( FP8Meta.Prefill.Chunk( seq_lens=chunk_seq_lens, tokens_slice=tokens_slice, block_table=chunk_block_table, req_start_idx=chunk_start, workspace_starts=chunk_workspace_starts, chunk_tot_seqlen=chunk_tot_seqlen, ) ) prefill_workspace_starts.copy_( prefill_workspace_starts_cpu, non_blocking=True ) fp8_metadata.prefill = FP8Meta.Prefill( seq_lens=prefill_seq_lens, request_ids=prefill_request_id, workspace_starts=prefill_workspace_starts, chunks=prefill_chunks, ) if num_decodes > 0: # Compute decode_query_len for spec decode (uniform due to require_uniform) query_start_loc_cpu = common_attn_metadata.query_start_loc_cpu decode_query_len = (query_start_loc_cpu[1] - query_start_loc_cpu[0]).item() # Use padded head count since that's what the kernel will see padded_heads = self.fp8_decode_padded_heads scheduler_metadata, _ = get_mla_metadata( cache_seqlens=self.topk_tokens_tensor[:num_decodes], num_q_tokens_per_head_k=decode_query_len * padded_heads, topk=self.topk_tokens, num_heads_q=padded_heads, num_heads_k=1, is_fp8_kvcache=True, ) kernel_meta = FlashMLASparseMetadata.FP8KernelMetadata( scheduler_metadata=scheduler_metadata, dummy_block_table=self.dummy_block_table[:num_decodes], cache_lens=self.max_model_len_tensor[:num_decodes], ) fp8_metadata.decode = FP8Meta.Decode( kernel_metadata=kernel_meta, decode_query_len=decode_query_len, ) return fp8_metadata def build( self, common_prefix_len: int, common_attn_metadata: CommonAttentionMetadata, fast_build: bool = False, ) -> FlashMLASparseMetadata: cm = common_attn_metadata num_tokens = cm.num_actual_tokens starts = np.asarray(cm.query_start_loc_cpu, dtype=np.int32) seg_lengths = np.diff(starts) req_id_per_token = np.repeat( np.arange(seg_lengths.shape[0], dtype=np.int32), seg_lengths ) # Zero-fill for cudagraphs self.req_id_per_token_buffer.fill_(0) self.req_id_per_token_buffer[: req_id_per_token.shape[0]].copy_( torch.from_numpy(req_id_per_token), non_blocking=True ) req_id_per_token = self.req_id_per_token_buffer[:num_tokens] fp8_extra_metadata: ( FlashMLASparseMetadata.FP8SeparatePrefillDecode | FlashMLASparseMetadata.FP8KernelMetadata | None ) = None fp8_use_mixed_batch = self.num_heads < MIN_HEADS_FOR_BF16_PREFILL if self.use_fp8_kv_cache: if fp8_use_mixed_batch: fp8_extra_metadata = self._build_fp8_mixed_decode_prefill(cm) else: fp8_extra_metadata = self._build_fp8_separate_prefill_decode(cm) metadata = FlashMLASparseMetadata( num_reqs=cm.num_reqs, max_query_len=cm.max_query_len, max_seq_len=cm.max_seq_len, num_actual_tokens=cm.num_actual_tokens, query_start_loc=cm.query_start_loc, slot_mapping=cm.slot_mapping, block_table=cm.block_table_tensor, req_id_per_token=req_id_per_token, block_size=self.kv_cache_spec.block_size, topk_tokens=self.topk_tokens, fp8_extra_metadata=fp8_extra_metadata, fp8_use_mixed_batch=fp8_use_mixed_batch, ) return metadata class FlashMLASparseImpl(SparseMLAAttentionImpl[FlashMLASparseMetadata]): @staticmethod def _compute_fp8_decode_padded_heads(num_heads: int) -> int: # FP8 decode kernel only supports h_q = 64 or 128 # Compute padded head count for decode return 64 if num_heads <= 64 else 128 def __init__( self, num_heads: int, head_size: int, scale: float, num_kv_heads: int, alibi_slopes: list[float] | None, sliding_window: int | None, kv_cache_dtype: str, logits_soft_cap: float | None, attn_type: str, kv_sharing_target_layer_name: str | None, # MLA Specific Arguments topk_indice_buffer: torch.Tensor | None = None, indexer: "Indexer | None" = None, **mla_args, ) -> None: self.num_heads = num_heads self.head_size = head_size self.scale = float(scale) self.num_kv_heads = num_kv_heads self.kv_cache_dtype = kv_cache_dtype self.kv_lora_rank: int = mla_args["kv_lora_rank"] self.softmax_scale = scale assert indexer is not None self.topk_indices_buffer: torch.Tensor | None = indexer.topk_indices_buffer # Prefill BF16 kernel requires 64 on Hopper, 128 on Blackwell self.prefill_padding = ( 128 if current_platform.is_device_capability_family(100) else 64 ) self.fp8_decode_padded_heads = self._compute_fp8_decode_padded_heads(num_heads) if kv_cache_dtype == "fp8_ds_mla": # Reserve workspace during initialization vllm_config = get_current_vllm_config() assert vllm_config is not None and vllm_config.model_config is not None prefill_workspace_size = get_prefill_workspace_size( vllm_config.model_config.max_model_len ) self.prefill_workspace_shape = (prefill_workspace_size, head_size) (self.prefill_bf16_workspace,) = ( current_workspace_manager().get_simultaneous( (self.prefill_workspace_shape, torch.bfloat16) ) ) def _forward_bf16_kv( self, q: torch.Tensor, kv_c_and_k_pe_cache: torch.Tensor, topk_indices: torch.Tensor, attn_metadata: FlashMLASparseMetadata, ) -> torch.Tensor: # Convert per-request indices to global slots (decode) or workspace # offsets (prefill). topk_indices = triton_convert_req_index_to_global_index( attn_metadata.req_id_per_token, attn_metadata.block_table, topk_indices, BLOCK_SIZE=attn_metadata.block_size, NUM_TOPK_TOKENS=topk_indices.shape[1], ) return self._bf16_flash_mla_kernel(q, kv_c_and_k_pe_cache, topk_indices) def _forward_fp8_kv_separate_prefill_decode( self, q: torch.Tensor, kv_c_and_k_pe_cache: torch.Tensor, topk_indices: torch.Tensor, attn_metadata: FlashMLASparseMetadata, ) -> torch.Tensor: fp8_metadata = attn_metadata.fp8_extra_metadata assert isinstance(fp8_metadata, FlashMLASparseMetadata.FP8SeparatePrefillDecode) num_decodes = fp8_metadata.num_decodes prefill_request_ids = None prefill_workspace_starts = None has_prefill_workspace = False if fp8_metadata.prefill is not None: prefill_request_ids = fp8_metadata.prefill.request_ids prefill_workspace_starts = fp8_metadata.prefill.workspace_starts has_prefill_workspace = True # Convert per-request indices to global slots (decode) or workspace # offsets (prefill). # For FP8 cache: prefill uses workspace mapping (upconverted to BF16) # For BF16 cache: always use global cache slots (no workspace) # prefill_workspace_starts has been adjusted in-place per chunk so # prefill indices automatically come out chunk-local topk_indices = triton_convert_req_index_to_global_index( attn_metadata.req_id_per_token, attn_metadata.block_table, topk_indices, BLOCK_SIZE=attn_metadata.block_size, NUM_TOPK_TOKENS=topk_indices.shape[1], HAS_PREFILL_WORKSPACE=has_prefill_workspace, prefill_workspace_request_ids=prefill_request_ids, prefill_workspace_starts=prefill_workspace_starts, ) fp8_metadata = attn_metadata.fp8_extra_metadata assert isinstance(fp8_metadata, FlashMLASparseMetadata.FP8SeparatePrefillDecode) def _fp8_decode(q: torch.Tensor, topk_indices: torch.Tensor) -> torch.Tensor: # Reshape q: (num_decode_tokens, num_heads, head_dim) # -> (num_decodes, seq_len, num_heads, head_dim) q = reshape_query_for_spec_decode(q, num_decodes) seq_len = q.shape[1] # Reshape topk_indices: (num_decode_tokens, topk) # -> (num_decodes, seq_len, topk) topk_indices = topk_indices.view(num_decodes, seq_len, -1) assert fp8_metadata.decode is not None attn_out, _ = self._fp8_flash_mla_kernel( q=q, kv_c_and_k_pe_cache=kv_c_and_k_pe_cache, topk_indices=topk_indices, kernel_metadata=fp8_metadata.decode.kernel_metadata, ) # Reshape output: (num_decodes, seq_len, num_heads, head_dim_v) # -> (num_decode_tokens, num_heads, head_dim_v) return reshape_attn_output_for_spec_decode(attn_out) num_decode_tokens = fp8_metadata.num_decode_tokens num_prefill_tokens = fp8_metadata.num_prefill_tokens # Pure decode: direct call without allocation if num_decode_tokens > 0 and num_prefill_tokens == 0: assert fp8_metadata.decode is not None attn_out = _fp8_decode(q, topk_indices) else: # Mixed or pure prefill: allocate output tensor attn_out = q.new_empty( (attn_metadata.num_actual_tokens, self.num_heads, self.kv_lora_rank), dtype=q.dtype, device=q.device, ) if num_decode_tokens > 0: attn_out[:num_decode_tokens] = _fp8_decode( q[:num_decode_tokens], topk_indices[:num_decode_tokens] ) assert fp8_metadata.prefill is not None for chunk in fp8_metadata.prefill.chunks: chunk_workspace = self.prefill_bf16_workspace[: chunk.chunk_tot_seqlen] ops.cp_gather_and_upconvert_fp8_kv_cache( kv_c_and_k_pe_cache, chunk_workspace, chunk.block_table, chunk.seq_lens, chunk.workspace_starts, len(chunk.block_table), ) chunk_q = q[chunk.tokens_slice] chunk_topk_indices_workspace = topk_indices[chunk.tokens_slice] attn_out[chunk.tokens_slice] = self._bf16_flash_mla_kernel( chunk_q, chunk_workspace, chunk_topk_indices_workspace, ) return attn_out def _forward_fp8_kv_mixed_batch( self, q: torch.Tensor, kv_c_and_k_pe_cache: torch.Tensor, topk_indices: torch.Tensor, attn_metadata: FlashMLASparseMetadata, ) -> torch.Tensor: """Mixed batch FP8 forward path that treats all tokens as one batch. This is equivalent to main branch's approach and avoids the BF16 prefill kernel which has head padding overhead when num_heads is small. Used when use_mixed_batch is True. """ # Convert per-request indices to global slots (decode) or workspace # offsets (prefill). topk_indices = triton_convert_req_index_to_global_index( attn_metadata.req_id_per_token, attn_metadata.block_table, topk_indices, BLOCK_SIZE=attn_metadata.block_size, NUM_TOPK_TOKENS=topk_indices.shape[1], ) assert attn_metadata.fp8_extra_metadata is not None assert isinstance( attn_metadata.fp8_extra_metadata, FlashMLASparseMetadata.FP8KernelMetadata ) fp8_metadata = attn_metadata.fp8_extra_metadata _attn_out, _ = self._fp8_flash_mla_kernel( q=q.unsqueeze(0), # unsqueeze to add batch_dim: (T, H, D) -> (1, T, H, D) kv_c_and_k_pe_cache=kv_c_and_k_pe_cache, topk_indices=topk_indices.unsqueeze(0), # (T, topk) -> (1, T, topk) kernel_metadata=fp8_metadata, ) # Output is (1, T, H, D_v), squeeze back to (T, H, D_v) return _attn_out.squeeze(0) def _fp8_flash_mla_kernel( self, q: torch.Tensor, kv_c_and_k_pe_cache: torch.Tensor, topk_indices: torch.Tensor, kernel_metadata: FlashMLASparseMetadata.FP8KernelMetadata, ) -> tuple[torch.Tensor, torch.Tensor]: # q shape: (batch, seq_len, num_heads, head_dim) actual_num_heads = q.size(2) padded_num_heads = self.fp8_decode_padded_heads # Pad query if needed (kernel only supports h_q = 64 or 128) if actual_num_heads < padded_num_heads: logger.warning_once( f"Padding num_heads from {actual_num_heads} to " f"{padded_num_heads} for FP8 sparse decode kernel" ) q_padded = q.new_zeros((q.size(0), q.size(1), padded_num_heads, q.size(3))) q_padded[:, :, :actual_num_heads, :] = q q = q_padded out, lse = flash_mla_with_kvcache( q=q, k_cache=kv_c_and_k_pe_cache.view(torch.uint8).unsqueeze(-2), block_table=kernel_metadata.dummy_block_table, head_dim_v=512, cache_seqlens=kernel_metadata.cache_lens, tile_scheduler_metadata=kernel_metadata.scheduler_metadata, is_fp8_kvcache=True, indices=topk_indices, softmax_scale=self.softmax_scale, ) # Slice output back to actual head count if we padded if actual_num_heads < padded_num_heads: out = out[:, :, :actual_num_heads, :] return out, lse def _bf16_flash_mla_kernel( self, q: torch.Tensor, kv_c_and_k_pe_cache: torch.Tensor, topk_indices: torch.Tensor, ) -> torch.Tensor: num_tokens = q.shape[0] kv_c_and_k_pe_cache = kv_c_and_k_pe_cache.view( -1, 1, kv_c_and_k_pe_cache.shape[-1] ) # NOTE(Chen): kernel requires num_local_head to be a multiple of # 64 on hopper and 128 on blackwell if self.num_heads % self.prefill_padding != 0: assert self.prefill_padding % self.num_heads == 0 logger.warning_once( f"Padding num_heads from {self.num_heads} to " f"{self.prefill_padding} for BF16 sparse prefill kernel" ) q_padded = q.new_empty((q.shape[0], self.prefill_padding, q.shape[2])) q_padded[:, : self.num_heads, :] = q q = q_padded topk_indices = topk_indices.view(num_tokens, 1, -1) output = flash_mla_sparse_fwd( q, kv_c_and_k_pe_cache, topk_indices, self.softmax_scale )[0] output = output[:, : self.num_heads, :] return output def forward_mqa( self, q: torch.Tensor | tuple[torch.Tensor, torch.Tensor], kv_c_and_k_pe_cache: torch.Tensor, attn_metadata: FlashMLASparseMetadata, layer: AttentionLayer, ) -> tuple[torch.Tensor, torch.Tensor | None]: # NOTE(lucas): for the sparse FlashMLA kernels the kernels want to use # MQA 576/512 approach for both prefill and decode # Concatenate q if it's a tuple (ql_nope, q_pe) if isinstance(q, tuple): q = torch.cat(q, dim=-1) num_actual_toks = q.shape[0] # Get topk indices assert self.topk_indices_buffer is not None topk_indices = self.topk_indices_buffer[:num_actual_toks] use_fp8_cache = self.kv_cache_dtype == "fp8_ds_mla" if not use_fp8_cache: attn_out = self._forward_bf16_kv( q, kv_c_and_k_pe_cache, topk_indices, attn_metadata ) elif attn_metadata.fp8_use_mixed_batch: attn_out = self._forward_fp8_kv_mixed_batch( q, kv_c_and_k_pe_cache, topk_indices, attn_metadata ) else: attn_out = self._forward_fp8_kv_separate_prefill_decode( q, kv_c_and_k_pe_cache, topk_indices, attn_metadata ) return attn_out, None
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/attention/backends/mla/flashmla_sparse.py", "license": "Apache License 2.0", "lines": 736, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/v1/attention/backends/mla/indexer.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from dataclasses import dataclass from typing import ClassVar import torch from vllm.config import VllmConfig from vllm.logger import init_logger from vllm.platforms import current_platform from vllm.utils.deep_gemm import ( get_paged_mqa_logits_metadata, is_deep_gemm_supported, ) from vllm.utils.platform_utils import num_compute_units from vllm.v1.attention.backend import ( AttentionBackend, AttentionCGSupport, AttentionMetadataBuilder, CommonAttentionMetadata, MultipleOf, ) from vllm.v1.attention.backends.utils import ( split_decodes_and_prefills, split_prefill_chunks, ) logger = init_logger(__name__) class DeepseekV32IndexerBackend(AttentionBackend): @staticmethod def get_name() -> str: return "DEEPSEEK_V32_INDEXER" @staticmethod def get_supported_kernel_block_sizes() -> list[int | MultipleOf]: return [1 if current_platform.is_rocm() else 64] @classmethod def get_supported_head_sizes(cls) -> list[int]: return [32, 64, 128] @staticmethod def get_builder_cls() -> type["DeepseekV32IndexerMetadataBuilder"]: return DeepseekV32IndexerMetadataBuilder @staticmethod def get_kv_cache_shape( num_blocks: int, block_size: int, num_kv_heads: int, head_size: int, cache_dtype_str: str = "auto", ) -> tuple[int, ...]: assert num_kv_heads == 1 return (num_blocks, block_size, head_size) @staticmethod def get_kv_cache_stride_order( include_num_layers_dimension: bool = False, ) -> tuple[int, ...]: if include_num_layers_dimension: return (0, 1, 2, 3) return (0, 1, 2) @dataclass class DeepseekV32IndexerPrefillChunkMetadata: block_table: torch.Tensor cu_seqlen_ks: torch.Tensor cu_seqlen_ke: torch.Tensor cu_seq_lens: torch.Tensor token_to_seq: torch.Tensor total_seq_lens: int token_start: int token_end: int num_reqs: int @dataclass class DeepseekV32IndexerPrefillMetadata: chunks: list[DeepseekV32IndexerPrefillChunkMetadata] @dataclass class DeepSeekV32IndexerDecodeMetadata: block_table: torch.Tensor seq_lens: torch.Tensor decode_lens: torch.Tensor requires_padding: bool schedule_metadata: torch.Tensor use_large_context_topk: bool offsets: torch.Tensor | None # Precomputed offsets for speculative decoding @dataclass class DeepseekV32IndexerMetadata: # FIXME (zyongye) # hacky way to access the data now, need to be in chunked meta seq_lens: torch.Tensor num_reqs: int max_query_len: int max_seq_len: int num_actual_tokens: int # Number of tokens excluding padding. query_start_loc: torch.Tensor slot_mapping: torch.Tensor # The dimension of the attention heads head_dim: int # New for MLA (compared to FlashAttention) # For handling prefill decode split num_decodes: int num_decode_tokens: int num_prefills: int num_prefill_tokens: int decode: DeepSeekV32IndexerDecodeMetadata | None = None prefill: DeepseekV32IndexerPrefillMetadata | None = None # TODO (zyongye) optimize this, this is now vibe coded def kv_spans_from_batches( start_seq_loc: torch.Tensor, seq_len_per_batch: torch.Tensor, device: torch.device ) -> tuple[torch.Tensor, torch.Tensor]: """ Args: start_seq_loc: 1D long tensor [B+1], cumulative counts of selected tokens per batch. Example: [0, 2, 4, 7] -> batch sizes (selected) [2, 2, 3], N=7 tokens total. seq_len_per_batch: 1D long tensor [B], full sequence length (KV length) of each batch. Example: [5, 9, 4]. Returns: start_tensor: 1D long tensor [N], start offset in the concatenated KV cache for each token's batch. end_location: 1D long tensor [N], **exclusive** end = start + token's local position. (So the attended KV slice is kv[start:end].) Assumes each batch contributes its full `seq_len_per_batch[i]` keys to the KV cache, andthe selected tokens within a batch are the **last** `counts[i]` positions of that sequence. """ q = start_seq_loc.to(dtype=torch.long) L = seq_len_per_batch.to(dtype=torch.long) assert q.dim() == 1 and L.dim() == 1 assert q.numel() == L.numel() + 1, "start_seq_loc must have length B+1" # Selected tokens per batch and totals counts = q[1:] - q[:-1] # [B] N = int(q[-1].item()) # total selected tokens B = L.numel() if N == 0: return ( torch.empty(0, dtype=torch.long, device=device), torch.empty(0, dtype=torch.long, device=device), ) # KV start offsets per batch in the concatenated KV cache kv_starts_per_batch = torch.cumsum(L, dim=0) - L # [B] # For each selected token, which batch does it belong to? batch_id = torch.repeat_interleave(torch.arange(B), counts) # [N] # Map batch KV start to each token start_tensor = kv_starts_per_batch[batch_id] # [N] # End-align local positions inside each batch: # local_pos = L[b] - counts[b] + (1..counts[b]) for each batch b L_expand = torch.repeat_interleave(L, counts) # [N] m_expand = torch.repeat_interleave(counts, counts) # [N] # position within the selected block: 1..counts[b] pos_within = ( torch.arange(N, dtype=torch.long) - torch.repeat_interleave(q[:-1], counts) + 1 ) local_pos = L_expand - m_expand + pos_within # [N], 1-based end_location = start_tensor + local_pos # exclusive end return start_tensor.int().to(device), end_location.int().to(device) def get_max_prefill_buffer_size(vllm_config: VllmConfig): max_model_len = vllm_config.model_config.max_model_len # NOTE(Chen): 40 is a magic number for controlling the prefill buffer size. # Each entry is 128 fp8 bytes and 4 scale bytes for a total of 132 bytes. # The flashmla_sparse backend uses a workspace size of 5 * max_model_len. # The memory usage of the workspace there is 576 * 2 bytes; so we size this as # (576 * 2 // 132) * 5 = 40 to maximize this workspace size while still fitting # within the flashmla_sparse workspace. # For DeepSeek-V3.2, the max_model_len is 163840. # 40 * 163840 * 132 = 865075200 bytes = 825 MB return max_model_len * 40 class DeepseekV32IndexerMetadataBuilder(AttentionMetadataBuilder): _cudagraph_support: ClassVar[AttentionCGSupport] = AttentionCGSupport.UNIFORM_BATCH reorder_batch_threshold: int = 1 def __init__(self, *args, **kwargs): super().__init__(*args, **kwargs) scheduler_config = self.vllm_config.scheduler_config # NOTE(Chen):an estimated max size of flattened_kv. Need to double check. self.max_prefill_buffer_size = get_max_prefill_buffer_size(self.vllm_config) self.num_speculative_tokens = ( self.vllm_config.speculative_config.num_speculative_tokens if self.vllm_config.speculative_config else 0 ) if self.num_speculative_tokens > 1: raise ValueError( "Sparse MLA only supports " "num_speculative_tokens <= 1 because the DeepGEMM " "fp8_paged_mqa_logits kernel does not support next_n > 2. " f"Got num_speculative_tokens={self.num_speculative_tokens}." ) self.reorder_batch_threshold += self.num_speculative_tokens sm_count = num_compute_units(self.device.index) self.num_sms = sm_count self.decode_lens_buffer = torch.empty( (scheduler_config.max_num_seqs,), dtype=torch.int32, device=self.device ) # See: DeepGMM/csrc/apis/attention.hpp self.scheduler_metadata_buffer = torch.empty( (self.num_sms + 1, 2), dtype=torch.int32, device=self.device ) def build_one_prefill_chunk( self, reqs_start, reqs_end, query_start_loc_cpu, seq_lens_cpu, block_table ): prefill_query_start_loc = ( query_start_loc_cpu[reqs_start : reqs_end + 1] - query_start_loc_cpu[reqs_start] ) cu_seqlen_ks, cu_seqlen_ke = kv_spans_from_batches( prefill_query_start_loc, seq_lens_cpu[reqs_start:reqs_end], self.device ) token_start = query_start_loc_cpu[reqs_start].item() token_end = query_start_loc_cpu[reqs_end].item() total_seq_lens = seq_lens_cpu[reqs_start:reqs_end].sum() seq_idx = torch.arange(0, reqs_end - reqs_start, dtype=torch.int32) token_to_seq = torch.repeat_interleave( seq_idx, seq_lens_cpu[reqs_start:reqs_end] ).to(self.device) assert total_seq_lens <= self.max_prefill_buffer_size cu_seq_lens = ( torch.cat( [ torch.zeros(1, dtype=torch.int32), seq_lens_cpu[reqs_start:reqs_end].cumsum(dim=0), ] ) .to(torch.int32) .to(self.device) ) return DeepseekV32IndexerPrefillChunkMetadata( cu_seqlen_ks=cu_seqlen_ks, cu_seqlen_ke=cu_seqlen_ke, cu_seq_lens=cu_seq_lens, token_to_seq=token_to_seq, total_seq_lens=total_seq_lens, block_table=block_table[reqs_start:reqs_end], token_start=token_start, token_end=token_end, num_reqs=reqs_end - reqs_start, ) def build( self, common_prefix_len: int, common_attn_metadata: CommonAttentionMetadata, fast_build: bool = False, ) -> DeepseekV32IndexerMetadata: num_reqs = common_attn_metadata.num_reqs num_tokens = common_attn_metadata.num_actual_tokens query_start_loc_cpu = common_attn_metadata.query_start_loc_cpu num_decodes, num_prefills, num_decode_tokens, num_prefill_tokens = ( split_decodes_and_prefills( common_attn_metadata, decode_threshold=self.reorder_batch_threshold ) ) assert num_decodes + num_prefills == num_reqs assert num_decode_tokens + num_prefill_tokens == num_tokens prefill_metadata = None if num_prefills > 0: chunk_seq_ids = split_prefill_chunks( common_attn_metadata.seq_lens_cpu[num_decodes:], self.max_prefill_buffer_size, request_offset=num_decodes, ) chunks = [ self.build_one_prefill_chunk( reqs_start, reqs_end, query_start_loc_cpu, common_attn_metadata.seq_lens_cpu, common_attn_metadata.block_table_tensor, ) for reqs_start, reqs_end in chunk_seq_ids ] prefill_metadata = DeepseekV32IndexerPrefillMetadata( chunks=chunks, ) decode_metadata = None if num_decodes > 0: torch.diff( common_attn_metadata.query_start_loc[: num_decodes + 1], out=self.decode_lens_buffer[:num_decodes], ) decode_lens = self.decode_lens_buffer[:num_decodes] decode_lens_cpu = torch.diff( common_attn_metadata.query_start_loc_cpu[: num_decodes + 1] ) # Use CPU to avoid GPU sync; breaking async scheduling requires_padding = (decode_lens_cpu.max() > decode_lens_cpu.min()).item() # Decide which top-k kernel to use based on batch size and sequence length batch_size = num_decodes _is_large_context = common_attn_metadata.max_seq_len > 8192 # Decision logic based on micro-benchmark results: # - large_context_topk wins for batch <= 128 and seq_len > 8K # - top_k_per_row_decode wins for batch > 128 or seq_len <= 8K use_large_context_topk = batch_size <= 128 and _is_large_context next_n = 1 + self.num_speculative_tokens if next_n > 1: offsets = torch.arange(next_n, device=self.device, dtype=torch.int32) else: offsets = None seq_lens = common_attn_metadata.seq_lens[:num_decodes] # DeepGEMM is required for the paged MQA logits on CUDA devices if current_platform.is_cuda() and is_deep_gemm_supported(): self.scheduler_metadata_buffer[:] = get_paged_mqa_logits_metadata( seq_lens, self.kv_cache_spec.block_size, self.num_sms ) block_table = common_attn_metadata.block_table_tensor[:num_decodes, ...] # Padded CUDA graph requests have block_table entries of -1. # Clamp to 0 to prevent OOB access in the DeepGEMM kernel. # This is safe because padded requests have seq_lens=0, so the # kernel produces no meaningful output for those rows. block_table.clamp_(min=0) decode_metadata = DeepSeekV32IndexerDecodeMetadata( block_table=block_table, seq_lens=common_attn_metadata.seq_lens[:num_decodes], decode_lens=decode_lens, requires_padding=requires_padding, schedule_metadata=self.scheduler_metadata_buffer, use_large_context_topk=use_large_context_topk, offsets=offsets, ) attn_metadata = DeepseekV32IndexerMetadata( seq_lens=common_attn_metadata.seq_lens, num_reqs=common_attn_metadata.num_reqs, max_query_len=common_attn_metadata.max_query_len, max_seq_len=common_attn_metadata.max_seq_len, num_actual_tokens=common_attn_metadata.num_actual_tokens, query_start_loc=common_attn_metadata.query_start_loc, slot_mapping=common_attn_metadata.slot_mapping, head_dim=128, num_decodes=num_decodes, num_decode_tokens=num_decode_tokens, num_prefills=num_prefills, num_prefill_tokens=num_prefill_tokens, prefill=prefill_metadata, decode=decode_metadata, ) # if get_tensor_model_parallel_rank() == 0: # logger.info(f"attn_metadata: {attn_metadata}") return attn_metadata
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/attention/backends/mla/indexer.py", "license": "Apache License 2.0", "lines": 332, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/config/vllm.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import copy import getpass import json import os import tempfile import threading import time from contextlib import contextmanager from dataclasses import is_dataclass from datetime import datetime from enum import IntEnum from functools import lru_cache from pathlib import Path from typing import TYPE_CHECKING, Any, Literal, TypeVar, get_args import torch from pydantic import ConfigDict, Field, model_validator import vllm.envs as envs from vllm.logger import enable_trace_function_call, init_logger from vllm.transformers_utils.runai_utils import is_runai_obj_uri from vllm.utils import random_uuid from vllm.utils.hashing import safe_hash from .attention import AttentionConfig from .cache import CacheConfig from .compilation import CompilationConfig, CompilationMode, CUDAGraphMode from .device import DeviceConfig from .ec_transfer import ECTransferConfig from .kernel import KernelConfig from .kv_events import KVEventsConfig from .kv_transfer import KVTransferConfig from .load import LoadConfig from .lora import LoRAConfig from .model import ModelConfig from .observability import ObservabilityConfig from .offload import OffloadConfig from .parallel import ParallelConfig from .profiler import ProfilerConfig from .scheduler import SchedulerConfig from .speculative import EagleModelTypes, SpeculativeConfig from .structured_outputs import StructuredOutputsConfig from .utils import SupportsHash, config, replace from .weight_transfer import WeightTransferConfig if TYPE_CHECKING: from transformers import PretrainedConfig from vllm.model_executor.layers.quantization.base_config import QuantizationConfig from vllm.v1.kv_cache_interface import KVCacheConfig else: PretrainedConfig = Any QuantizationConfig = Any KVCacheConfig = Any logger = init_logger(__name__) class OptimizationLevel(IntEnum): """Optimization level enum.""" O0 = 0 """O0 : No optimization. no compilation, no cudagraphs, no other optimization, just starting up immediately""" O1 = 1 """O1: Quick optimizations. Dynamo+Inductor compilation and Piecewise cudagraphs""" O2 = 2 """O2: Full optimizations. -O1 as well as Full and Piecewise cudagraphs.""" O3 = 3 """O3: Currently the same as -O2s.""" PerformanceMode = Literal["balanced", "interactivity", "throughput"] IS_QUANTIZED = False IS_DENSE = False # The optimizations that depend on these properties currently set to False # in all cases. # if model_config is not None: # IS_QUANTIZED = lambda c: c.model_config.is_quantized() # IS_DENSE = lambda c: not c.model_config.is_model_moe() # See https://github.com/vllm-project/vllm/issues/25689. def enable_norm_fusion(cfg: "VllmConfig") -> bool: """Enable if either RMS norm or quant FP8 custom op is active; otherwise Inductor handles fusion.""" return cfg.compilation_config.is_custom_op_enabled( "rms_norm" ) or cfg.compilation_config.is_custom_op_enabled("quant_fp8") def enable_act_fusion(cfg: "VllmConfig") -> bool: """ Enable if either SiLU+Mul or quant FP8 custom op is active; otherwise Inductor handles fusion. Also enable for FP4 models as FP4 quant is always custom so Inductor cannot fuse it. """ return ( cfg.compilation_config.is_custom_op_enabled("silu_and_mul") or cfg.compilation_config.is_custom_op_enabled("quant_fp8") or (cfg.model_config is not None and cfg.model_config.is_nvfp4_quantized()) ) def enable_allreduce_rms_fusion(cfg: "VllmConfig") -> bool: """Enable if TP > 1 and Hopper/Blackwell and flashinfer installed.""" from vllm.platforms import current_platform from vllm.utils.flashinfer import has_flashinfer return ( cfg.parallel_config.tensor_parallel_size > 1 and current_platform.is_cuda() and has_flashinfer() and ( current_platform.is_device_capability(100) or current_platform.is_device_capability(90) ) # tp-dp combination broken: # https://github.com/vllm-project/vllm/issues/34458 and cfg.parallel_config.data_parallel_size == 1 # tp-pp combination broken: # https://github.com/vllm-project/vllm/issues/35426 and cfg.parallel_config.pipeline_parallel_size == 1 ) def enable_rope_kvcache_fusion(cfg: "VllmConfig") -> bool: """Enable if rotary embedding custom op is active and use_inductor_graph_partition is enabled. """ from vllm._aiter_ops import rocm_aiter_ops return ( rocm_aiter_ops.is_enabled() and cfg.compilation_config.is_custom_op_enabled("rotary_embedding") and cfg.compilation_config.use_inductor_graph_partition ) def enable_norm_pad_fusion(cfg: "VllmConfig") -> bool: """Enable if using AITER RMSNorm and AITER Triton GEMMs and hidden size is 2880 i.e. gpt-oss; otherwise Inductor handles fusion.""" from vllm._aiter_ops import rocm_aiter_ops return ( rocm_aiter_ops.is_rmsnorm_enabled() and not rocm_aiter_ops.is_triton_gemm_enabled() and cfg.model_config is not None and cfg.model_config.get_hidden_size() == 2880 ) OPTIMIZATION_LEVEL_00 = { "compilation_config": { "pass_config": { "fuse_norm_quant": False, "fuse_act_quant": False, "fuse_allreduce_rms": False, "fuse_attn_quant": False, "enable_sp": False, "fuse_gemm_comms": False, "fuse_act_padding": False, "fuse_rope_kvcache": False, }, "cudagraph_mode": CUDAGraphMode.NONE, "use_inductor_graph_partition": False, }, "kernel_config": { "enable_flashinfer_autotune": False, }, } OPTIMIZATION_LEVEL_01 = { "compilation_config": { "pass_config": { "fuse_norm_quant": enable_norm_fusion, "fuse_act_quant": enable_act_fusion, "fuse_allreduce_rms": False, "fuse_attn_quant": False, "enable_sp": False, "fuse_gemm_comms": False, "fuse_act_padding": enable_norm_pad_fusion, "fuse_rope_kvcache": enable_rope_kvcache_fusion, }, "cudagraph_mode": CUDAGraphMode.PIECEWISE, "use_inductor_graph_partition": False, }, "kernel_config": { "enable_flashinfer_autotune": True, }, } OPTIMIZATION_LEVEL_02 = { "compilation_config": { "pass_config": { "fuse_norm_quant": enable_norm_fusion, "fuse_act_quant": enable_act_fusion, "fuse_allreduce_rms": enable_allreduce_rms_fusion, "fuse_attn_quant": IS_QUANTIZED, "enable_sp": IS_DENSE, "fuse_gemm_comms": IS_DENSE, "fuse_act_padding": enable_norm_pad_fusion, "fuse_rope_kvcache": enable_rope_kvcache_fusion, }, "cudagraph_mode": CUDAGraphMode.FULL_AND_PIECEWISE, "use_inductor_graph_partition": False, }, "kernel_config": { "enable_flashinfer_autotune": True, }, } OPTIMIZATION_LEVEL_03 = { "compilation_config": { "pass_config": { "fuse_norm_quant": enable_norm_fusion, "fuse_act_quant": enable_act_fusion, "fuse_allreduce_rms": enable_allreduce_rms_fusion, "fuse_attn_quant": IS_QUANTIZED, "enable_sp": IS_DENSE, "fuse_gemm_comms": IS_DENSE, "fuse_act_padding": enable_norm_pad_fusion, "fuse_rope_kvcache": enable_rope_kvcache_fusion, }, "cudagraph_mode": CUDAGraphMode.FULL_AND_PIECEWISE, "use_inductor_graph_partition": False, }, "kernel_config": { "enable_flashinfer_autotune": True, }, } OPTIMIZATION_LEVEL_TO_CONFIG = { OptimizationLevel.O0: OPTIMIZATION_LEVEL_00, OptimizationLevel.O1: OPTIMIZATION_LEVEL_01, OptimizationLevel.O2: OPTIMIZATION_LEVEL_02, OptimizationLevel.O3: OPTIMIZATION_LEVEL_03, } @config(config=ConfigDict(arbitrary_types_allowed=True)) class VllmConfig: """Dataclass which contains all vllm-related configuration. This simplifies passing around the distinct configurations in the codebase. """ # TODO: use default_factory once default constructing ModelConfig doesn't # try to download a model model_config: ModelConfig = Field(default=None) """Model configuration.""" cache_config: CacheConfig = Field(default_factory=CacheConfig) """Cache configuration.""" parallel_config: ParallelConfig = Field(default_factory=ParallelConfig) """Parallel configuration.""" scheduler_config: SchedulerConfig = Field( default_factory=SchedulerConfig.default_factory, ) """Scheduler configuration.""" device_config: DeviceConfig = Field(default_factory=DeviceConfig) """Device configuration.""" load_config: LoadConfig = Field(default_factory=LoadConfig) """Load configuration.""" offload_config: OffloadConfig = Field(default_factory=OffloadConfig) """Model weight offloading configuration.""" attention_config: AttentionConfig = Field(default_factory=AttentionConfig) """Attention configuration.""" kernel_config: KernelConfig = Field(default_factory=KernelConfig) """Kernel configuration.""" lora_config: LoRAConfig | None = None """LoRA configuration.""" speculative_config: SpeculativeConfig | None = None """Speculative decoding configuration.""" structured_outputs_config: StructuredOutputsConfig = Field( default_factory=StructuredOutputsConfig ) """Structured outputs configuration.""" observability_config: ObservabilityConfig = Field( default_factory=ObservabilityConfig ) """Observability configuration.""" quant_config: QuantizationConfig | None = None """Quantization configuration.""" compilation_config: CompilationConfig = Field(default_factory=CompilationConfig) """`torch.compile` and cudagraph capture configuration for the model. As a shorthand, one can append compilation arguments via -cc.parameter=argument such as `-cc.mode=3` (same as `-cc='{"mode":3}'`). You can specify the full compilation config like so: `{"mode": 3, "cudagraph_capture_sizes": [1, 2, 4, 8]}` """ profiler_config: ProfilerConfig = Field(default_factory=ProfilerConfig) """Profiling configuration.""" kv_transfer_config: KVTransferConfig | None = None """The configurations for distributed KV cache transfer.""" kv_events_config: KVEventsConfig | None = None """The configurations for event publishing.""" ec_transfer_config: ECTransferConfig | None = None """The configurations for distributed EC cache transfer.""" # some opaque config, only used to provide additional information # for the hash computation, mainly used for testing, debugging or out of # tree config registration. additional_config: dict | SupportsHash = Field(default_factory=dict) """Additional config for specified platform. Different platforms may support different configs. Make sure the configs are valid for the platform you are using. Contents must be hashable.""" instance_id: str = "" """The ID of the vLLM instance.""" optimization_level: OptimizationLevel = OptimizationLevel.O2 """The optimization level. These levels trade startup time cost for performance, with -O0 having the best startup time and -O3 having the best performance. -O2 is used by default. See OptimizationLevel for full description.""" performance_mode: PerformanceMode = "balanced" """Performance mode for runtime behavior, 'balanced' is the default. 'interactivity' favors low end-to-end per-request latency at small batch sizes (fine-grained CUDA graphs, latency-oriented kernels). 'throughput' favors aggregate tokens/sec at high concurrency (larger CUDA graphs, more aggressive batching, throughput-oriented kernels).""" weight_transfer_config: WeightTransferConfig | None = None """The configurations for weight transfer during RL training.""" def compute_hash(self) -> str: """ WARNING: Whenever a new field is added to this config, ensure that it is included in the factors list if it affects the computation graph. Provide a hash that uniquely identifies all the configs that affect the structure of the computation graph from input ids/embeddings to the final hidden states, excluding anything before input ids/embeddings and after the final hidden states. """ factors: list[Any] = [] # summarize vllm config vllm_factors: list[Any] = [] from vllm import __version__ vllm_factors.append(__version__) if self.model_config: vllm_factors.append(self.model_config.compute_hash()) if ( self.compilation_config and getattr(self.compilation_config, "compile_mm_encoder", False) and self.model_config.multimodal_config ): vllm_factors.append(self.model_config.multimodal_config.compute_hash()) else: vllm_factors.append("None") if self.cache_config: vllm_factors.append(self.cache_config.compute_hash()) else: vllm_factors.append("None") if self.parallel_config: vllm_factors.append(self.parallel_config.compute_hash()) else: vllm_factors.append("None") if self.scheduler_config: vllm_factors.append(self.scheduler_config.compute_hash()) else: vllm_factors.append("None") if self.device_config: vllm_factors.append(self.device_config.compute_hash()) else: vllm_factors.append("None") if self.load_config: vllm_factors.append(self.load_config.compute_hash()) else: vllm_factors.append("None") if self.offload_config: vllm_factors.append(self.offload_config.compute_hash()) else: vllm_factors.append("None") if self.attention_config: vllm_factors.append(self.attention_config.compute_hash()) else: vllm_factors.append("None") if self.lora_config: vllm_factors.append(self.lora_config.compute_hash()) else: vllm_factors.append("None") if self.speculative_config: vllm_factors.append(self.speculative_config.compute_hash()) else: vllm_factors.append("None") if self.structured_outputs_config: vllm_factors.append(self.structured_outputs_config.compute_hash()) if self.profiler_config: vllm_factors.append(self.profiler_config.compute_hash()) else: vllm_factors.append("None") vllm_factors.append(self.observability_config.compute_hash()) if self.quant_config: pass # should be captured by model_config.quantization if self.compilation_config: vllm_factors.append(self.compilation_config.compute_hash()) else: vllm_factors.append("None") if self.kv_transfer_config: vllm_factors.append(self.kv_transfer_config.compute_hash()) else: vllm_factors.append("None") if self.ec_transfer_config: vllm_factors.append(self.ec_transfer_config.compute_hash()) else: vllm_factors.append("None") if self.additional_config: if isinstance(additional_config := self.additional_config, dict): additional_config_hash = safe_hash( json.dumps(additional_config, sort_keys=True).encode(), usedforsecurity=False, ).hexdigest() else: additional_config_hash = additional_config.compute_hash() vllm_factors.append(additional_config_hash) else: vllm_factors.append("None") factors.append(vllm_factors) hash_str = safe_hash(str(factors).encode(), usedforsecurity=False).hexdigest()[ :10 ] return hash_str @property def num_speculative_tokens(self) -> int: if ( self.speculative_config is not None and self.speculative_config.num_speculative_tokens is not None ): return self.speculative_config.num_speculative_tokens return 0 @property def needs_dp_coordinator(self) -> bool: """ Determine if the DPCoordinator process is needed. The DPCoordinator is needed in two cases: 1. For MoE models with DP > 1: to handle wave coordination (even in external LB mode, since wave coordination runs in the coordinator) 2. For non-MoE models in internal/hybrid LB mode: to collect and publish queue stats for load balancing across DP ranks Returns: True if DPCoordinator process is needed, False otherwise. """ # For non-MoE models, only need coordinator in internal/hybrid LB mode # (for stats collection). return self.parallel_config.data_parallel_size > 1 and ( self.model_config is None or self.model_config.is_moe or not self.parallel_config.data_parallel_external_lb ) def enable_trace_function_call_for_thread(self) -> None: """ Set up function tracing for the current thread, if enabled via the `VLLM_TRACE_FUNCTION` environment variable. """ if envs.VLLM_TRACE_FUNCTION: tmp_dir = tempfile.gettempdir() # add username to tmp_dir to avoid permission issues tmp_dir = os.path.join(tmp_dir, getpass.getuser()) filename = ( f"VLLM_TRACE_FUNCTION_for_process_{os.getpid()}" f"_thread_{threading.get_ident()}_at_{datetime.now()}.log" ).replace(" ", "_") log_path = os.path.join( tmp_dir, "vllm", f"vllm-instance-{self.instance_id}", filename, ) os.makedirs(os.path.dirname(log_path), exist_ok=True) enable_trace_function_call(log_path) @staticmethod def _get_quantization_config( model_config: ModelConfig, load_config: LoadConfig ) -> QuantizationConfig | None: """Get the quantization config.""" from vllm.platforms import current_platform if model_config.quantization is not None: from vllm.model_executor.model_loader.weight_utils import get_quant_config quant_config = get_quant_config(model_config, load_config) capability_tuple = current_platform.get_device_capability() if capability_tuple is not None: capability = capability_tuple.to_int() if capability < quant_config.get_min_capability(): raise ValueError( f"The quantization method {model_config.quantization} " "is not supported for the current GPU. Minimum " f"capability: {quant_config.get_min_capability()}. " f"Current capability: {capability}." ) supported_dtypes = quant_config.get_supported_act_dtypes() if model_config.dtype not in supported_dtypes: raise ValueError( f"{model_config.dtype} is not supported for quantization " f"method {model_config.quantization}. Supported dtypes: " f"{supported_dtypes}" ) quant_config.maybe_update_config(model_config.model) return quant_config return None @staticmethod def get_quantization_config( model_config: ModelConfig, load_config: LoadConfig ) -> QuantizationConfig | None: import copy # For some reason, the _ version of this modifies the model_config # object, so using deepcopy to avoid this problem. return VllmConfig._get_quantization_config( copy.deepcopy(model_config), load_config ) def with_hf_config( self, hf_config: PretrainedConfig, architectures: list[str] | None = None, ) -> "VllmConfig": if architectures is not None: hf_config = copy.deepcopy(hf_config) hf_config.architectures = architectures model_config = copy.deepcopy(self.model_config) if ( model_config.is_multimodal_model and hasattr(model_config.hf_config, "tie_word_embeddings") and not hasattr(hf_config.get_text_config(), "tie_word_embeddings") ): # In Transformers v5, tie_word_embeddings belongs to the config of the class # that can see both layers to be tied. For example: # # SomeVLModel: # self.language_model = SomeLanguageModel() # self.vision_model = SomeVisionModel() # # SomeVLModelForMultimodalLM: # self.model = SomeVLModel() # self.lm_head = nn.Linear() # # Therefore, tie_word_embeddings is defined in SomeVLModelForMultimodalLM's # config and is not present in SomeVLModel's config. In vLLM, the lm_head # belongs to the language_model, so we must ensure that tie_word_embeddings # is set in the language_model's config. tie_word_embeddings = model_config.hf_config.tie_word_embeddings hf_config.get_text_config().tie_word_embeddings = tie_word_embeddings model_config.hf_config = hf_config model_config.model_arch_config = model_config.get_model_arch_config() return replace(self, model_config=model_config) def _set_config_default(self, config_obj: Any, key: str, value: Any) -> None: """Set config attribute to default if not already set by user. Args: config_obj: Configuration object to update. key: Attribute name. value: Default value (static or callable). """ if getattr(config_obj, key) is None: # Some config values are known before initialization and are # hard coded. # Other values depend on the user given configuration, so they are # implemented with lambda functions and decided at run time. setattr(config_obj, key, value(self) if callable(value) else value) def _apply_optimization_level_defaults(self, defaults: dict[str, Any]) -> None: """Apply optimization level defaults using self as root. Recursively applies values from defaults into nested config objects. Only fields present in defaults are overwritten. If the user configuration does not specify a value for a default field and if the default field is still None after all user selections are applied, then default values will be applied to the field. User speciied fields will not be overridden by the default. Args: defaults: Dictionary of default values to apply. """ def apply_recursive(config_obj: Any, config_defaults: dict[str, Any]) -> None: """Recursively apply defaults to config_obj, using self as root.""" for key, value in config_defaults.items(): if not hasattr(config_obj, key): continue current = getattr(config_obj, key) if isinstance(value, dict) and is_dataclass(current): apply_recursive(current, value) else: self._set_config_default(config_obj, key, value) apply_recursive(self, defaults) def _post_init_kv_transfer_config(self) -> None: """Update KVTransferConfig based on top-level configs in VllmConfig. Right now, this function reads the offloading settings from CacheConfig and configures the KVTransferConfig accordingly. """ # KV offloading is only activated when kv_offloading_size is set. if (kv_offloading_size := self.cache_config.kv_offloading_size) is None: return kv_offloading_backend = self.cache_config.kv_offloading_backend # If no KVTransferConfig is provided, create a default one. if self.kv_transfer_config is None: self.kv_transfer_config = KVTransferConfig() num_kv_ranks = ( self.parallel_config.tensor_parallel_size * self.parallel_config.pipeline_parallel_size ) if kv_offloading_backend == "native": self.kv_transfer_config.kv_connector = "OffloadingConnector" self.kv_transfer_config.kv_connector_extra_config.update( {"cpu_bytes_to_use": kv_offloading_size * (1 << 30)} ) elif kv_offloading_backend == "lmcache": self.kv_transfer_config.kv_connector = "LMCacheConnectorV1" kv_gb_per_rank = kv_offloading_size / num_kv_ranks self.kv_transfer_config.kv_connector_extra_config = { "lmcache.local_cpu": True, "lmcache.max_local_cpu_size": kv_gb_per_rank, } # This is the same for all backends self.kv_transfer_config.kv_role = "kv_both" def __post_init__(self): """Verify configs are valid & consistent with each other.""" # To give each torch profile run a unique instance name. self.instance_id = f"{time.time_ns()}" if self.performance_mode != "balanced": logger.info_once( "Performance mode set to '%s'.", self.performance_mode, scope="local" ) self.try_verify_and_update_config() if self.model_config is not None: self.model_config.verify_with_parallel_config(self.parallel_config) self.model_config.verify_dual_chunk_attention_config(self.load_config) self.parallel_config.is_moe_model = self.model_config.is_moe self.cache_config.verify_with_parallel_config(self.parallel_config) if self.lora_config is not None: self.lora_config.verify_with_model_config(self.model_config) if self.quant_config is None and self.model_config is not None: self.quant_config = VllmConfig._get_quantization_config( self.model_config, self.load_config ) executor_backend = self.parallel_config.distributed_executor_backend executor_supports_async_sched = executor_backend in ( "mp", "uni", "external_launcher", ) if self.scheduler_config.async_scheduling: # Async scheduling explicitly enabled, hard fail any incompatibilities. # Currently, async scheduling only support eagle speculative # decoding. if self.speculative_config is not None: if ( self.speculative_config.method not in get_args(EagleModelTypes) and self.speculative_config.method != "draft_model" ): raise ValueError( "Currently, async scheduling is only supported " "with EAGLE/MTP/Draft Model kind of speculative decoding." ) if self.speculative_config.disable_padded_drafter_batch: raise ValueError( "Async scheduling is not compatible with " "disable_padded_drafter_batch=True." ) if not executor_supports_async_sched: raise ValueError( "Currently, async scheduling only supports `mp`, `uni`, or " "`external_launcher` distributed executor backend, but you chose " f"`{executor_backend}`." ) elif self.scheduler_config.async_scheduling is None: # Enable async scheduling unless there is an incompatible option. if ( self.speculative_config is not None and self.speculative_config.method not in get_args(EagleModelTypes) ): logger.warning_once( "Async scheduling not supported with %s-based " "speculative decoding and will be disabled.", self.speculative_config.method, scope="local", ) self.scheduler_config.async_scheduling = False elif ( self.speculative_config is not None and self.speculative_config.disable_padded_drafter_batch ): logger.warning_once( "Async scheduling is not compatible with " "disable_padded_drafter_batch=True and will be disabled.", scope="local", ) self.scheduler_config.async_scheduling = False elif not executor_supports_async_sched: logger.warning_once( "Async scheduling will be disabled because it is not supported " "with the `%s` distributed executor backend (only `mp`, `uni`, and " "`external_launcher` are supported).", executor_backend, scope="local", ) self.scheduler_config.async_scheduling = False else: self.scheduler_config.async_scheduling = True logger.info_once( "Asynchronous scheduling is %s.", "enabled" if self.scheduler_config.async_scheduling else "disabled", ) if self.parallel_config.disable_nccl_for_dp_synchronization is None: if self.scheduler_config.async_scheduling: if self.parallel_config.data_parallel_size > 1 and ( self.model_config is None or self.model_config.is_moe ): logger.info_once( "Disabling NCCL for DP synchronization " "when using async scheduling.", scope="local", ) self.parallel_config.disable_nccl_for_dp_synchronization = True else: self.parallel_config.disable_nccl_for_dp_synchronization = False from vllm.platforms import current_platform if ( self.model_config is not None and self.scheduler_config.enable_chunked_prefill and self.model_config.dtype == torch.float32 and current_platform.get_device_capability() == (7, 5) ): logger.warning_once( "Turing devices tensor cores do not support float32 matmul. " "To workaround this limitation, vLLM will set 'ieee' input " "precision for chunked prefill triton kernels." ) if self.model_config is not None and self.model_config.enforce_eager: logger.warning( "Enforce eager set, disabling torch.compile and CUDAGraphs. " "This is equivalent to setting -cc.mode=none -cc.cudagraph_mode=none" ) self.compilation_config.mode = CompilationMode.NONE self.compilation_config.cudagraph_mode = CUDAGraphMode.NONE if self.compilation_config.backend == "eager" or ( self.compilation_config.mode is not None and self.compilation_config.mode != CompilationMode.VLLM_COMPILE ): logger.warning( "Inductor compilation was disabled by user settings, " "optimizations settings that are only active during " "inductor compilation will be ignored." ) def has_blocked_weights(): if self.quant_config is not None: if hasattr(self.quant_config, "weight_block_size"): return self.quant_config.weight_block_size is not None elif hasattr(self.quant_config, "has_blocked_weights"): return self.quant_config.has_blocked_weights() return False # Enable quant_fp8 CUDA ops (TODO disable in follow up) # On H100 the CUDA kernel is faster than # native implementation # https://github.com/vllm-project/vllm/issues/25094 if has_blocked_weights(): custom_ops = self.compilation_config.custom_ops if "-quant_fp8" not in custom_ops: custom_ops.append("+quant_fp8") current_platform.apply_config_platform_defaults(self) if self.compilation_config.mode is None: if self.optimization_level > OptimizationLevel.O0: self.compilation_config.mode = CompilationMode.VLLM_COMPILE else: self.compilation_config.mode = CompilationMode.NONE if all(s not in self.compilation_config.custom_ops for s in ("all", "none")): if ( self.compilation_config.backend == "inductor" and self.compilation_config.mode != CompilationMode.NONE ): self.compilation_config.custom_ops.append("none") else: self.compilation_config.custom_ops.append("all") default_config = OPTIMIZATION_LEVEL_TO_CONFIG[self.optimization_level] self._apply_optimization_level_defaults(default_config) if self.kernel_config.enable_flashinfer_autotune is None: raise ValueError( "KernelConfig.enable_flashinfer_autotune must be set after applying " "optimization level defaults." ) if ( self.compilation_config.cudagraph_mode.requires_piecewise_compilation() and self.compilation_config.mode != CompilationMode.VLLM_COMPILE ): logger.info( "Cudagraph mode %s is not compatible with compilation mode %s." "Overriding to NONE.", self.compilation_config.cudagraph_mode, self.compilation_config.mode, ) self.compilation_config.cudagraph_mode = CUDAGraphMode.NONE # async tp is built on top of sequence parallelism # and requires it to be enabled. if self.compilation_config.pass_config.fuse_gemm_comms: self.compilation_config.pass_config.enable_sp = True if self.compilation_config.pass_config.enable_sp: if self.parallel_config.tensor_parallel_size == 1: logger.warning("Sequence Parallelism requires TP>1, disabling") self.compilation_config.pass_config.enable_sp = False self.compilation_config.pass_config.fuse_gemm_comms = False else: # Compute SP threshold early; disable if None (model too # small for SP to be beneficial). pass_config = self.compilation_config.pass_config if pass_config.sp_min_token_num is None: from vllm.compilation.passes.fusion.sequence_parallelism import ( get_sequence_parallelism_threshold, ) tp_size = self.parallel_config.tensor_parallel_size hidden_size = self.model_config.get_hidden_size() element_size = self.model_config.dtype.itemsize pass_config.sp_min_token_num = get_sequence_parallelism_threshold( hidden_size, tp_size, element_size ) if pass_config.sp_min_token_num is None: logger.warning( "Model hidden_size too small for the SP " "threshold heuristic, disabling. To force SP, " "set pass_config.sp_min_token_num manually." ) self.compilation_config.pass_config.enable_sp = False self.compilation_config.pass_config.fuse_gemm_comms = False from vllm.utils.torch_utils import HAS_OPAQUE_TYPE if HAS_OPAQUE_TYPE: # On torch >= 2.11 the hoisted OpaqueObject approach supersedes # fast_moe_cold_start, so force it off. self.compilation_config.fast_moe_cold_start = False elif self.compilation_config.fast_moe_cold_start is None: # resolve default behavior: try to be as safe as possible # this config is unsafe if any spec decoding draft model has a MOE. # We'll conservatively turn it off if we see spec decoding. self.compilation_config.fast_moe_cold_start = ( self.speculative_config is None ) self._set_max_num_scheduled_tokens() if current_platform.support_static_graph_mode(): # if cudagraph_mode has full cudagraphs, we need to check support if model_config := self.model_config: if ( self.compilation_config.cudagraph_mode.has_full_cudagraphs() and model_config.pooler_config is not None ): logger.warning_once( "Pooling models do not support full cudagraphs. " "Overriding cudagraph_mode to PIECEWISE." ) self.compilation_config.cudagraph_mode = CUDAGraphMode.PIECEWISE elif ( model_config.is_encoder_decoder and self.compilation_config.cudagraph_mode not in (CUDAGraphMode.NONE, CUDAGraphMode.FULL_DECODE_ONLY) ): logger.info_once( "Encoder-decoder models do not support %s. " "Overriding cudagraph_mode to FULL_DECODE_ONLY.", self.compilation_config.cudagraph_mode.name, ) self.compilation_config.cudagraph_mode = ( CUDAGraphMode.FULL_DECODE_ONLY ) # disable cudagraph when enforce eager execution if self.model_config is not None and self.model_config.enforce_eager: logger.info("Cudagraph is disabled under eager mode") self.compilation_config.cudagraph_mode = CUDAGraphMode.NONE # override related settings when enforce eager self.compilation_config.max_cudagraph_capture_size = 0 self.compilation_config.cudagraph_capture_sizes = [] else: self.compilation_config.cudagraph_num_of_warmups = 1 self._set_cudagraph_sizes() else: self.compilation_config.cudagraph_mode = CUDAGraphMode.NONE if self.cache_config.kv_sharing_fast_prefill: if ( self.speculative_config is not None and self.speculative_config.use_eagle() ): raise ValueError( "Fast prefill optimization for KV sharing is not " "compatible with EAGLE as EAGLE requires correct logits " "for all tokens while fast prefill gives incorrect logits " "for prompt tokens." ) logger.warning_once( "--kv-sharing-fast-prefill requires changes on model side for " "correctness and to realize prefill savings." ) # TODO: Move after https://github.com/vllm-project/vllm/pull/26847 lands self._set_compile_ranges() if ( self.model_config and self.model_config.architecture == "WhisperForConditionalGeneration" and os.environ.get("VLLM_WORKER_MULTIPROC_METHOD") != "spawn" ): logger.warning( "Whisper is known to have issues with " "forked workers. If startup is hanging, " "try setting 'VLLM_WORKER_MULTIPROC_METHOD' " "to 'spawn'." ) if ( self.kv_events_config is not None and self.kv_events_config.enable_kv_cache_events and not self.cache_config.enable_prefix_caching ): logger.warning( "KV cache events are on, but prefix caching is not enabled. " "Use --enable-prefix-caching to enable." ) if ( self.kv_events_config is not None and self.kv_events_config.publisher != "null" and not self.kv_events_config.enable_kv_cache_events ): logger.warning( "KV cache events are disabled, " "but the scheduler is configured to publish them. " "Modify KVEventsConfig.enable_kv_cache_events " "to True to enable." ) current_platform.check_and_update_config(self) # If DCP, ensure the block size is right. if self.parallel_config.decode_context_parallel_size > 1: if self.parallel_config.dcp_kv_cache_interleave_size > 1 and ( self.parallel_config.cp_kv_cache_interleave_size != self.parallel_config.dcp_kv_cache_interleave_size ): self.parallel_config.cp_kv_cache_interleave_size = ( self.parallel_config.dcp_kv_cache_interleave_size ) logger.warning_once( "cp_kv_cache_interleave_size is overridden by dcp_kv_cache" "_interleave_size. And dcp-kv-cache-interleave-size will be " "deprecated when PCP is fully supported." ) assert ( self.parallel_config.cp_kv_cache_interleave_size <= self.cache_config.block_size and self.cache_config.block_size % self.parallel_config.cp_kv_cache_interleave_size == 0 ), ( f"Block_size({self.cache_config.block_size}) should be greater " "than or equal to and divisible by cp_kv_cache_interleave_size " f"({self.parallel_config.cp_kv_cache_interleave_size})." ) # Do this after all the updates to compilation_config.mode effective_dp_size = ( self.parallel_config.data_parallel_size if self.model_config is None or self.model_config.is_moe else 1 ) self.compilation_config.set_splitting_ops_for_v1( all2all_backend=self.parallel_config.all2all_backend, data_parallel_size=effective_dp_size, ) if self.compilation_config.pass_config.enable_sp: # With pipeline parallelism or dynamo partitioning, # native rms norm tracing errors due to incorrect residual shape. # Use custom rms norm to unblock. In the future, # the pass will operate on higher-level IR to avoid the issue. # TODO: https://github.com/vllm-project/vllm/issues/27894 if self.compilation_config.mode != CompilationMode.VLLM_COMPILE: logger.warning( "Sequence parallelism is enabled, but running in wrong " "vllm compile mode: %s.", self.compilation_config.mode, ) is_fullgraph = ( self.compilation_config.use_inductor_graph_partition or len(self.compilation_config.splitting_ops) == 0 ) if self.parallel_config.pipeline_parallel_size > 1 or not is_fullgraph: if "-rms_norm" not in self.compilation_config.custom_ops: self.compilation_config.custom_ops.append("+rms_norm") else: regime = ( "Dynamo partition" if not is_fullgraph else "pipeline parallelism" ) logger.warning_once( "Sequence parallelism not supported with " "native rms_norm when using %s, " "this will likely lead to an error.", regime, ) # final check of cudagraph mode after all possible updates if current_platform.is_cuda_alike(): if ( self.compilation_config.cudagraph_mode.has_full_cudagraphs() and self.model_config is not None and not self.model_config.disable_cascade_attn and not self.compilation_config.cudagraph_mode.has_piecewise_cudagraphs() # noqa: E501 ): logger.warning_once( "No piecewise cudagraph for executing cascade attention." " Will fall back to eager execution if a batch runs " "into cascade attentions." ) if self.compilation_config.cudagraph_mode.requires_piecewise_compilation(): assert self.compilation_config.mode == CompilationMode.VLLM_COMPILE, ( "Compilation mode should be CompilationMode.VLLM_COMPILE " "when cudagraph_mode piecewise cudagraphs is used, " f"cudagraph_mode={self.compilation_config.cudagraph_mode}" ) from vllm.model_executor.layers.batch_invariant import vllm_is_batch_invariant if ( self.model_config and vllm_is_batch_invariant() and not self.model_config.disable_cascade_attn ): self.model_config.disable_cascade_attn = True logger.warning_once( "Disabling cascade attention when VLLM_BATCH_INVARIANT is enabled.", scope="local", ) if self.parallel_config.use_ubatching: a2a_backend = self.parallel_config.all2all_backend assert a2a_backend in [ "deepep_low_latency", "deepep_high_throughput", ], ( "Microbatching currently only supports the deepep_low_latency and " f"deepep_high_throughput all2all backend. {a2a_backend} is not " "supported. To fix use --all2all-backend=deepep_low_latency or " "--all2all-backend=deepep_high_throughput and install the DeepEP" " kernels." ) if not self.model_config.disable_cascade_attn: self.model_config.disable_cascade_attn = True logger.warning_once("Disabling cascade attention when DBO is enabled.") if not self.instance_id: self.instance_id = random_uuid()[:5] # Hybrid KV cache manager (HMA) runtime rules: # - Explicit enable (--no-disable-kv-cache-manager): error if runtime # disables it # - No preference: auto-disable for unsupported features (e.g. kv connector) # - Explicit disable (--disable-kv-cache-manager): always respect it need_disable_hybrid_kv_cache_manager = False # logger should only print warning message for hybrid models. As we # can't know whether the model is hybrid or not now, so we don't log # warning message here and will log it later. if not current_platform.support_hybrid_kv_cache(): # Hybrid KV cache manager is not supported on non-GPU platforms. need_disable_hybrid_kv_cache_manager = True if self.kv_events_config is not None: # Hybrid KV cache manager is not compatible with KV events. need_disable_hybrid_kv_cache_manager = True if ( self.model_config is not None and self.model_config.attention_chunk_size is not None ): if ( self.speculative_config is not None and self.speculative_config.use_eagle() ): # Hybrid KV cache manager is not yet supported with chunked # local attention + eagle. need_disable_hybrid_kv_cache_manager = True elif not envs.VLLM_ALLOW_CHUNKED_LOCAL_ATTN_WITH_HYBRID_KV_CACHE: logger.warning( "There is a latency regression when using chunked local" " attention with the hybrid KV cache manager. Disabling" " it, by default. To enable it, set the environment " "VLLM_ALLOW_CHUNKED_LOCAL_ATTN_WITH_HYBRID_KV_CACHE=1." ) # Hybrid KV cache manager is not yet supported with chunked # local attention. need_disable_hybrid_kv_cache_manager = True if self.scheduler_config.disable_hybrid_kv_cache_manager is None: # Default to disable HMA, but only if the user didn't express a preference. if self.kv_transfer_config is not None: # NOTE(Kuntai): turn HMA off for connector unless specifically enabled. need_disable_hybrid_kv_cache_manager = True logger.warning( "Turning off hybrid kv cache manager because " "`--kv-transfer-config` is set. This will reduce the " "performance of vLLM on LLMs with sliding window attention " "or Mamba attention. If you are a developer of kv connector" ", please consider supporting hybrid kv cache manager for " "your connector by making sure your connector is a subclass" " of `SupportsHMA` defined in kv_connector/v1/base.py and" " use --no-disable-hybrid-kv-cache-manager to start vLLM." ) self.scheduler_config.disable_hybrid_kv_cache_manager = ( need_disable_hybrid_kv_cache_manager ) elif ( self.scheduler_config.disable_hybrid_kv_cache_manager is False and need_disable_hybrid_kv_cache_manager ): raise ValueError( "Hybrid KV cache manager was explicitly enabled but is not " "supported in this configuration. Consider omitting the " "--no-disable-hybrid-kv-cache-manager flag to let vLLM decide" " automatically." ) if self.scheduler_config.disable_hybrid_kv_cache_manager is None: # Default to enable HMA if not explicitly disabled by user or logic above. self.scheduler_config.disable_hybrid_kv_cache_manager = False if self.cache_config.mamba_cache_mode == "align": assert ( self.cache_config.block_size <= self.scheduler_config.max_num_batched_tokens ), ( "In Mamba cache align mode, block_size " f"({self.cache_config.block_size}) must be <= " "max_num_batched_tokens " f"({self.scheduler_config.max_num_batched_tokens})." ) if self.scheduler_config.long_prefill_token_threshold > 0: assert ( self.scheduler_config.long_prefill_token_threshold >= self.cache_config.block_size ) assert not self.scheduler_config.disable_chunked_mm_input, ( "Chunked MM input is required because we need the flexibility to " "schedule a multiple of block_size tokens even if they are in the " "middle of a mm input" ) if self.compilation_config.debug_dump_path: self.compilation_config.debug_dump_path = ( self.compilation_config.debug_dump_path.absolute().expanduser() ) if envs.VLLM_DEBUG_DUMP_PATH is not None: env_path = Path(envs.VLLM_DEBUG_DUMP_PATH).absolute().expanduser() if self.compilation_config.debug_dump_path: logger.warning( "Config-specified debug dump path is overridden" " by VLLM_DEBUG_DUMP_PATH to %s", env_path, ) self.compilation_config.debug_dump_path = env_path def has_blocked_weights(): if self.quant_config is not None: if hasattr(self.quant_config, "weight_block_size"): return self.quant_config.weight_block_size is not None elif hasattr(self.quant_config, "has_blocked_weights"): return self.quant_config.has_blocked_weights() return False # Enable quant_fp8 CUDA ops (TODO disable in follow up) # On H100 the CUDA kernel is faster than # native implementation # https://github.com/vllm-project/vllm/issues/25094 if has_blocked_weights(): custom_ops = self.compilation_config.custom_ops if "-quant_fp8" not in custom_ops: custom_ops.append("+quant_fp8") # Handle the KV connector configs self._post_init_kv_transfer_config() def update_sizes_for_sequence_parallelism(self, possible_sizes: list) -> list: # remove the sizes that not multiple of tp_size when # enable sequence parallelism removed_sizes = [ size for size in possible_sizes if size % self.parallel_config.tensor_parallel_size != 0 ] if removed_sizes: logger.warning( "Batch sizes %s are removed because they are not " "multiple of tp_size %d when " "sequence parallelism is enabled", removed_sizes, self.parallel_config.tensor_parallel_size, ) return [ size for size in possible_sizes if size % self.parallel_config.tensor_parallel_size == 0 ] def _set_max_num_scheduled_tokens(self): """ In most cases, the scheduler may schedule a batch with as many tokens as the worker is configured to handle. However for some speculative decoding methods, the drafter model may insert additional slots into the batch when drafting. To account for this, we need to decrease the max_num_scheduled_tokens by an upper bound on the number of slots that can be added. """ if self.speculative_config is not None: scheduled_token_delta = ( self.speculative_config.max_num_new_slots_for_drafting * self.scheduler_config.max_num_seqs ) max_num_batched_tokens = self.scheduler_config.max_num_batched_tokens if self.scheduler_config.max_num_scheduled_tokens is None: self.scheduler_config.max_num_scheduled_tokens = ( max_num_batched_tokens - scheduled_token_delta ) max_num_scheduled_tokens = self.scheduler_config.max_num_scheduled_tokens if max_num_batched_tokens < max_num_scheduled_tokens + ( self.speculative_config.max_num_new_slots_for_drafting * self.scheduler_config.max_num_seqs ): raise ValueError( f"VllmConfig received max_num_scheduled_tokens but it does not have" " enough slots to support the speculative decoding settings." f" It should be greater by at least {scheduled_token_delta}, but" f" got {max_num_batched_tokens=} and {max_num_scheduled_tokens=}." ) def _set_cudagraph_sizes(self): """ vLLM defines the default candidate list of batch sizes for CUDA graph capture as: ```python max_graph_size = min(max_num_seqs * 2, 512) # 1, 2, 4, then multiples of 8 up to 256 and then multiples of 16 # up to max_graph_size cudagraph_capture_sizes = [1, 2, 4] + list(range(8, 256, 8)) + list( range(256, max_graph_size + 1, 16)) In the end, `vllm_config.compilation_config.cudagraph_capture_sizes` will be the final sizes to capture cudagraph (in ascending order). These sizes are used to capture and reuse CUDA graphs for performance-critical paths (e.g., decoding). Capturing enables significantly faster kernel dispatch by avoiding Python overhead. The list is then filtered based on `max_num_batched_tokens` (e.g., 8192 on most GPUs), which controls the total allowed number of tokens in a batch. Since each sequence may have a variable number of tokens, the maximum usable batch size will depend on actual sequence lengths. Example: With `max_num_batched_tokens = 8192`, and typical sequences averaging ~32 tokens, most practical batch sizes fall below 256. However, the system will still allow capture sizes up to 512 if shape and memory permit. Note: If users explicitly specify cudagraph capture sizes in the compilation config, those will override this default logic. At runtime: - If batch size <= one of the `cudagraph_capture_sizes`, the closest padded CUDA graph will be used. - If batch size > largest `cudagraph_capture_sizes`, cudagraph will not be used. """ if ( self.model_config is not None and not self.model_config.enforce_eager and self.compilation_config.cudagraph_mode != CUDAGraphMode.NONE ): # determine the initial max_cudagraph_capture_size max_cudagraph_capture_size = ( self.compilation_config.max_cudagraph_capture_size ) if max_cudagraph_capture_size is None: decode_query_len = 1 if ( self.speculative_config and self.speculative_config.num_speculative_tokens ): decode_query_len += self.speculative_config.num_speculative_tokens max_cudagraph_capture_size = min( self.scheduler_config.max_num_seqs * decode_query_len * 2, 512 ) max_num_tokens = self.scheduler_config.max_num_batched_tokens max_cudagraph_capture_size = min(max_num_tokens, max_cudagraph_capture_size) assert max_cudagraph_capture_size >= 1, ( "Maximum cudagraph size should be greater than or equal to 1 " "when using cuda graph." ) # determine the cudagraph_capture_sizes if self.compilation_config.cudagraph_capture_sizes is not None: assert len(self.compilation_config.cudagraph_capture_sizes) > 0, ( "cudagraph_capture_sizes should contain at least one element " "when using cuda graph." ) # de-duplicate the sizes provided by the config dedup_sizes = list(set(self.compilation_config.cudagraph_capture_sizes)) cudagraph_capture_sizes = [ i for i in dedup_sizes if i <= max_num_tokens ] # sort to make sure the sizes are in ascending order cudagraph_capture_sizes.sort() else: if self.performance_mode == "interactivity": # Fine-grained CUDA graphs at small batch sizes # for minimal padding overhead interactivity_max = min(max_cudagraph_capture_size, 32) cudagraph_capture_sizes = list(range(1, interactivity_max + 1)) else: cudagraph_capture_sizes = [ i for i in [1, 2, 4] if i <= max_cudagraph_capture_size ] if max_cudagraph_capture_size >= 8: # Step size 8 for small batch sizes, up to 256(not included) cudagraph_capture_sizes += list( range(8, min(max_cudagraph_capture_size + 1, 256), 8) ) if max_cudagraph_capture_size >= 256: # Step size 16 for larger batch sizes cudagraph_capture_sizes += list( range(256, max_cudagraph_capture_size + 1, 16) ) # de-duplicate and sort the sizes cudagraph_capture_sizes = sorted(set(cudagraph_capture_sizes)) if ( self.parallel_config.tensor_parallel_size > 1 and self.compilation_config.pass_config.enable_sp ): cudagraph_capture_sizes = self.update_sizes_for_sequence_parallelism( cudagraph_capture_sizes ) # user-specific compilation_config.max_cudagraph_capture_size get # truncated to valid_max_size when they are inconsistent. valid_max_size = ( cudagraph_capture_sizes[-1] if cudagraph_capture_sizes else 0 ) if ( self.compilation_config.max_cudagraph_capture_size is not None and self.compilation_config.max_cudagraph_capture_size != valid_max_size ): # raise error only when both two flags are user-specified # and they are inconsistent with each other if self.compilation_config.cudagraph_capture_sizes is not None: raise ValueError( "customized max_cudagraph_capture_size" f"(={self.compilation_config.max_cudagraph_capture_size}) " "should be consistent with the max value of " f"cudagraph_capture_sizes(={valid_max_size})" ) logger.warning( "Truncating max_cudagraph_capture_size to %d", valid_max_size, ) # always set the final max_cudagraph_capture_size self.compilation_config.max_cudagraph_capture_size = valid_max_size if self.compilation_config.cudagraph_capture_sizes is not None and len( cudagraph_capture_sizes ) < len(self.compilation_config.cudagraph_capture_sizes): # If users have specified capture sizes, we only need to # compare the lens before and after modification since the modified # list is only the subset of the original list. logger.warning( ( "cudagraph_capture_sizes specified in compilation_config" " %s is overridden by config %s" ), self.compilation_config.cudagraph_capture_sizes, cudagraph_capture_sizes, ) # always write back the final sizes self.compilation_config.cudagraph_capture_sizes = cudagraph_capture_sizes else: # no cudagraph in use self.compilation_config.max_cudagraph_capture_size = 0 self.compilation_config.cudagraph_capture_sizes = [] # complete the remaining process. self.compilation_config.post_init_cudagraph_sizes() def _set_compile_ranges(self): """ Set the compile ranges for the compilation config. """ compilation_config = self.compilation_config computed_compile_ranges_split_points = [] # The upper bound of the compile ranges is the max_num_batched_tokens. compile_range_end = self.scheduler_config.max_num_batched_tokens if compile_range_end is not None: computed_compile_ranges_split_points.append(compile_range_end) # Add the compile ranges for flashinfer if compilation_config.pass_config.fuse_allreduce_rms: tp_size = self.parallel_config.tensor_parallel_size max_size = compilation_config.pass_config.flashinfer_max_size(tp_size) if max_size is not None: max_token_num = max_size // ( self.model_config.get_hidden_size() * self.model_config.dtype.itemsize ) if compile_range_end is not None and max_token_num < compile_range_end: computed_compile_ranges_split_points.append(max_token_num) else: logger.debug( "Max num batched tokens below allreduce-rms fusion threshold, " "allreduce-rms fusion will be enabled for all num_tokens." ) # Add the compile ranges for sequence parallelism if compilation_config.pass_config.enable_sp: pass_config = compilation_config.pass_config # Calculate min_token_num if not explicitly provided # User override works regardless of hidden_size if pass_config.sp_min_token_num is None: from vllm.compilation.passes.fusion.sequence_parallelism import ( get_sequence_parallelism_threshold, ) tp_size = self.parallel_config.tensor_parallel_size hidden_size = self.model_config.get_hidden_size() element_size = self.model_config.dtype.itemsize pass_config.sp_min_token_num = get_sequence_parallelism_threshold( hidden_size, tp_size, element_size ) min_token_num = pass_config.sp_min_token_num max_num_batched_tokens = self.scheduler_config.max_num_batched_tokens if min_token_num is not None and ( max_num_batched_tokens is not None and min_token_num < max_num_batched_tokens and min_token_num > 1 ): # Add split point at min_token_num - 1 to ensure SP applies # starting from min_token_num # This creates ranges: [1, min-1] (no SP), [min, max] (SP applies) computed_compile_ranges_split_points.append(min_token_num - 1) if compilation_config.pass_config.fuse_rope_kvcache: max_token_num = ( compilation_config.pass_config.rope_kvcache_fusion_max_token_num ) if max_token_num is not None: if compile_range_end is not None and max_token_num < compile_range_end: computed_compile_ranges_split_points.append(max_token_num) else: logger.debug( "Max num batched tokens below rope+kvcache fusion threshold, " "rope+kvcache fusion enabled for num_tokens <= %d.", compile_range_end, ) if compilation_config.compile_ranges_split_points is not None: for x in compilation_config.compile_ranges_split_points: assert isinstance(x, int) assert x > 0, f"Invalid compile range split point: {x}" if compile_range_end is not None and x < compile_range_end and x > 1: computed_compile_ranges_split_points.append(x) compilation_config.compile_ranges_split_points = sorted( computed_compile_ranges_split_points ) def try_verify_and_update_config(self): if self.model_config is None: return # Avoid running try_verify_and_update_config multiple times if getattr(self.model_config, "config_updated", False): return self.model_config.config_updated = True architecture = self.model_config.architecture if architecture is None: return from vllm.model_executor.models.config import ( MODELS_CONFIG_MAP, HybridAttentionMambaModelConfig, ) cls = MODELS_CONFIG_MAP.get(architecture, None) if cls is not None: cls.verify_and_update_config(self) if self.model_config.is_hybrid: HybridAttentionMambaModelConfig.verify_and_update_config(self) if self.model_config.convert_type == "classify": # Maybe convert ForCausalLM into ForSequenceClassification model. from vllm.model_executor.models.adapters import SequenceClassificationConfig SequenceClassificationConfig.verify_and_update_config(self) if hasattr(self.model_config, "model_weights") and is_runai_obj_uri( self.model_config.model_weights ): if self.load_config.load_format == "auto": logger.info( "Detected Run:ai model config. " "Overriding `load_format` to 'runai_streamer'" ) self.load_config.load_format = "runai_streamer" elif self.load_config.load_format not in ( "runai_streamer", "runai_streamer_sharded", ): raise ValueError( f"To load a model from S3, 'load_format' " f"must be 'runai_streamer' or 'runai_streamer_sharded', " f"but got '{self.load_config.load_format}'. " f"Model: {self.model_config.model}" ) def compile_debug_dump_path(self) -> Path | None: """Returns a rank-aware path for dumping torch.compile debug information. """ if self.compilation_config.debug_dump_path is None: return None tp_rank = self.parallel_config.rank dp_rank = self.parallel_config.data_parallel_index append_path = f"rank_{tp_rank}_dp_{dp_rank}" path = self.compilation_config.debug_dump_path / append_path return path def __str__(self): return ( f"model={self.model_config.model!r}, " f"speculative_config={self.speculative_config!r}, " f"tokenizer={self.model_config.tokenizer!r}, " f"skip_tokenizer_init={self.model_config.skip_tokenizer_init}, " f"tokenizer_mode={self.model_config.tokenizer_mode}, " f"revision={self.model_config.revision}, " f"tokenizer_revision={self.model_config.tokenizer_revision}, " f"trust_remote_code={self.model_config.trust_remote_code}, " f"dtype={self.model_config.dtype}, " f"max_seq_len={self.model_config.max_model_len}, " f"download_dir={self.load_config.download_dir!r}, " f"load_format={self.load_config.load_format}, " f"tensor_parallel_size={self.parallel_config.tensor_parallel_size}, " # noqa f"pipeline_parallel_size={self.parallel_config.pipeline_parallel_size}, " # noqa f"data_parallel_size={self.parallel_config.data_parallel_size}, " # noqa f"disable_custom_all_reduce={self.parallel_config.disable_custom_all_reduce}, " # noqa f"quantization={self.model_config.quantization}, " f"enforce_eager={self.model_config.enforce_eager}, " f"enable_return_routed_experts={self.model_config.enable_return_routed_experts}, " # noqa f"kv_cache_dtype={self.cache_config.cache_dtype}, " f"device_config={self.device_config.device}, " f"structured_outputs_config={self.structured_outputs_config!r}, " f"observability_config={self.observability_config!r}, " f"seed={self.model_config.seed}, " f"served_model_name={self.model_config.served_model_name}, " f"enable_prefix_caching={self.cache_config.enable_prefix_caching}, " f"enable_chunked_prefill={self.scheduler_config.enable_chunked_prefill}, " # noqa f"pooler_config={self.model_config.pooler_config!r}, " f"compilation_config={self.compilation_config!r}" ) @model_validator(mode="after") def validate_mamba_block_size(self) -> "VllmConfig": if self.model_config is None: return self mamba_block_size_is_set = ( self.cache_config.mamba_block_size is not None and self.cache_config.mamba_block_size != self.model_config.max_model_len ) if mamba_block_size_is_set and not self.cache_config.enable_prefix_caching: raise ValueError( "--mamba-block-size can only be set with --enable-prefix-caching" ) return self _current_vllm_config: VllmConfig | None = None _current_prefix: str | None = None @contextmanager def set_current_vllm_config( vllm_config: VllmConfig, check_compile=False, prefix: str | None = None ): """ Temporarily set the current vLLM config. Used during model initialization. We save the current vLLM config in a global variable, so that all modules can access it, e.g. custom ops can access the vLLM config to determine how to dispatch. """ global _current_vllm_config, _current_prefix old_vllm_config = _current_vllm_config old_prefix = _current_prefix from vllm.compilation.counter import compilation_counter num_models_seen = compilation_counter.num_models_seen try: # Clear the compilation config cache when context changes. # This is needed since the old config may have been accessed # and cached before the new config is set. get_cached_compilation_config.cache_clear() _current_vllm_config = vllm_config _current_prefix = prefix yield except Exception: raise else: if check_compile: vllm_config.compilation_config.custom_op_log_check() if ( check_compile and vllm_config.compilation_config.mode == CompilationMode.VLLM_COMPILE and compilation_counter.num_models_seen == num_models_seen ): # If the model supports compilation, # compilation_counter.num_models_seen should be increased # by at least 1. # If it is not increased, it means the model does not support # compilation (does not have @support_torch_compile decorator). logger.warning( "`torch.compile` is turned on, but the model %s" " does not support it. Please open an issue on GitHub" " if you want it to be supported.", vllm_config.model_config.model, ) finally: _current_vllm_config = old_vllm_config _current_prefix = old_prefix # Clear the compilation config cache when context changes get_cached_compilation_config.cache_clear() @lru_cache(maxsize=1) def get_cached_compilation_config(): """Cache config to avoid repeated calls to get_current_vllm_config()""" return get_current_vllm_config().compilation_config def get_current_vllm_config() -> VllmConfig: if _current_vllm_config is None: raise AssertionError( "Current vLLM config is not set. This typically means " "get_current_vllm_config() was called outside of a " "set_current_vllm_config() context, or a CustomOp was instantiated " "at module import time or model forward time when config is not set. " "For tests that directly test custom ops/modules, use the " "'default_vllm_config' pytest fixture from tests/conftest.py." ) return _current_vllm_config def get_current_vllm_config_or_none() -> VllmConfig | None: return _current_vllm_config T = TypeVar("T") def get_layers_from_vllm_config( vllm_config: VllmConfig, layer_type: type[T], layer_names: list[str] | None = None, ) -> dict[str, T]: """ Get layers from the vLLM config. Args: vllm_config: The vLLM config. layer_type: The type of the layer to get. layer_names: The names of the layers to get. If None, return all layers. """ if layer_names is None: layer_names = list(vllm_config.compilation_config.static_forward_context.keys()) forward_context = vllm_config.compilation_config.static_forward_context return { layer_name: forward_context[layer_name] for layer_name in layer_names if isinstance(forward_context[layer_name], layer_type) }
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/config/vllm.py", "license": "Apache License 2.0", "lines": 1568, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/utils_/test_gc_utils.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from dataclasses import dataclass from typing import Any from vllm.utils.gc_utils import ( GCDebugConfig, _compute_detailed_type, _compute_top_gc_collected_objects, ) @dataclass class Normal: v: int @dataclass class ListWrapper: vs: list[int] def __len__(self) -> int: return len(self.vs) def test_compute_detailed_type(): assert ( _compute_detailed_type(Normal(v=8)) == "<class 'tests.utils_.test_gc_utils.Normal'>" ) assert _compute_detailed_type([1, 2, 3]) == "<class 'list'>(size:3)" assert _compute_detailed_type({4, 5}) == "<class 'set'>(size:2)" assert _compute_detailed_type({6: 7}) == "<class 'dict'>(size:1)" assert ( _compute_detailed_type(ListWrapper(vs=[])) == "<class 'tests.utils_.test_gc_utils.ListWrapper'>(size:0)" ) def test_compute_top_gc_collected_objects(): objects: list[Any] = [ [1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12], {13, 14}, {15: 16, 17: 18}, Normal(v=19), Normal(v=20), Normal(v=21), ] assert _compute_top_gc_collected_objects(objects, top=-1) == "" assert _compute_top_gc_collected_objects(objects, top=0) == "" assert ( _compute_top_gc_collected_objects(objects, top=1) == " 4:<class 'list'>(size:3)" ) assert _compute_top_gc_collected_objects(objects, top=2) == "\n".join( [ " 4:<class 'list'>(size:3)", " 3:<class 'tests.utils_.test_gc_utils.Normal'>", ] ) assert _compute_top_gc_collected_objects(objects, top=3) == "\n".join( [ " 4:<class 'list'>(size:3)", " 3:<class 'tests.utils_.test_gc_utils.Normal'>", " 1:<class 'set'>(size:2)", ] ) def test_gc_debug_config(): assert not GCDebugConfig(None).enabled assert not GCDebugConfig("").enabled assert not GCDebugConfig("0").enabled config = GCDebugConfig("1") assert config.enabled assert config.top_objects == -1 config = GCDebugConfig('{"top_objects":5}') assert config.enabled assert config.top_objects == 5
{ "repo_id": "vllm-project/vllm", "file_path": "tests/utils_/test_gc_utils.py", "license": "Apache License 2.0", "lines": 70, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/utils/gc_utils.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import gc import json import time from collections import Counter from contextlib import suppress from typing import Any import vllm.envs as envs from vllm.logger import init_logger logger = init_logger(__name__) class GCDebugConfig: """ Config for GC Debugger. - 0: disable GC debugger - 1: enable GC debugger with gc.collect elapsed times - '{"top_objects":5}': enable GC debugger with top 5 collected objects """ def __init__(self, gc_debug_conf: str | None = None) -> None: self.enabled: bool = False self.top_objects: int = -1 if not gc_debug_conf or gc_debug_conf == "0": pass elif gc_debug_conf == "1": self.enabled = True else: try: json_conf = json.loads(gc_debug_conf) self.enabled = True self.top_objects = json_conf.get("top_objects", -1) except Exception: self.enabled = False logger.error("Failed to parse VLLM_GC_DEBUG(%s)", envs.VLLM_GC_DEBUG) logger.debug("GC Debug Config. %s", str(self)) def __repr__(self) -> str: return f"enabled:{self.enabled},top_objects:{self.top_objects}" class GCDebugger: """ Debugger for GC which logs helpful information for GC understanding. To enable, you should call maybe_attach_gc_debug_callback in the process. """ def __init__(self, config: GCDebugConfig) -> None: self.config = config # Start time in micro second of this GC cycle self.start_time_ns: int = time.monotonic_ns() self.num_objects: int = 0 # If config.top_objects is positive, # compute top collected objects by object types self.gc_top_collected_objects: str = "" def handle(self, phase: str, info: dict[str, int]) -> None: """ Handles a GC event (e.g. GC start or GC finish) """ generation = info.get("generation") if generation is None: return if phase == "start": # Before GC started, record GC start time # and top collected objects self.start_time_ns = time.monotonic_ns() objects = gc.get_objects(generation) self.num_objects = len(objects) self.gc_top_collected_objects = _compute_top_gc_collected_objects( objects, self.config.top_objects ) elif phase == "stop": # After GC finished, Record GC elapsed time and # optionally top collected objects elpased_ms = (time.monotonic_ns() - self.start_time_ns) / 1e6 logger.info( "GC took %.3fms to complete. " "Collected %s objects (out of %d) in GC generation %d.%s", elpased_ms, str(info.get("collected", "?")), self.num_objects, generation, ( f" Top collected objects: \n{self.gc_top_collected_objects}" if self.gc_top_collected_objects else "" ), ) def freeze_gc_heap() -> None: """ Freeze all objects tracked by the garbage collector. It should be invoked after server init / warmup, to reduce GC overhead from static objects during serving time. """ # Ensure all static objects are pushed down to the oldest generation for # freeze gc.collect(0) gc.collect(1) gc.collect(2) # Freeze all GC tracked objects gc.freeze() def maybe_attach_gc_debug_callback() -> None: """ Attached a callback for GC debug when VLLM_GC_DEBUG is enabled. """ config = GCDebugConfig(envs.VLLM_GC_DEBUG) if config.enabled: debugger: GCDebugger = GCDebugger(config) def gc_callback(phase: str, info: dict[str, int]) -> None: debugger.handle(phase, info) gc.callbacks.append(gc_callback) def _compute_detailed_type(o: Any) -> str: """ Detailed object type. TODO(Jialin): Further enhance the detailed type with element types for easier debugging. We tried but occasionally it would run into signals which kills the engine. """ size_str: str = "" # Object doesn't support len() - this can happen with type objects # or other objects that don't implement __len__ properly with suppress(Exception): size_str = f"(size:{len(o)})" return f"{str(type(o))}{size_str}" def _compute_top_gc_collected_objects(objects: list[Any], top: int) -> str: """ Group collected objects by types. """ if top <= 0: return "" object_types = [_compute_detailed_type(o) for o in objects] return "\n".join( f"{count:>5}:{object_type}" for object_type, count in Counter(object_types).most_common(top) )
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/utils/gc_utils.py", "license": "Apache License 2.0", "lines": 129, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/v1/worker/test_utils.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import torch from vllm.v1.worker.utils import bind_kv_cache def test_bind_kv_cache(default_vllm_config): from vllm.model_executor.layers.attention import Attention ctx = { "layers.0.self_attn": Attention(32, 128, 0.1, prefix="layers.0.self_attn"), "layers.1.self_attn": Attention(32, 128, 0.1, prefix="layers.1.self_attn"), "layers.2.self_attn": Attention(32, 128, 0.1, prefix="layers.2.self_attn"), "layers.3.self_attn": Attention(32, 128, 0.1, prefix="layers.3.self_attn"), } kv_cache = { "layers.0.self_attn": torch.zeros((1,)), "layers.1.self_attn": torch.zeros((1,)), "layers.2.self_attn": torch.zeros((1,)), "layers.3.self_attn": torch.zeros((1,)), } runner_kv_caches: list[torch.Tensor] = [] bind_kv_cache(kv_cache, ctx, runner_kv_caches) assert ctx["layers.0.self_attn"].kv_cache[0] is kv_cache["layers.0.self_attn"] assert ctx["layers.1.self_attn"].kv_cache[0] is kv_cache["layers.1.self_attn"] assert ctx["layers.2.self_attn"].kv_cache[0] is kv_cache["layers.2.self_attn"] assert ctx["layers.3.self_attn"].kv_cache[0] is kv_cache["layers.3.self_attn"] assert runner_kv_caches[0] is kv_cache["layers.0.self_attn"] assert runner_kv_caches[1] is kv_cache["layers.1.self_attn"] assert runner_kv_caches[2] is kv_cache["layers.2.self_attn"] assert runner_kv_caches[3] is kv_cache["layers.3.self_attn"] def test_bind_kv_cache_non_attention(default_vllm_config): from vllm.model_executor.layers.attention import Attention # example from Jamba PP=2 ctx = { "model.layers.20.attn": Attention(32, 128, 0.1, prefix="model.layers.20.attn"), "model.layers.28.attn": Attention(32, 128, 0.1, prefix="model.layers.28.attn"), } kv_cache = { "model.layers.20.attn": torch.zeros((1,)), "model.layers.28.attn": torch.zeros((1,)), } runner_kv_caches: list[torch.Tensor] = [] bind_kv_cache(kv_cache, ctx, runner_kv_caches) assert ctx["model.layers.20.attn"].kv_cache[0] is kv_cache["model.layers.20.attn"] assert ctx["model.layers.28.attn"].kv_cache[0] is kv_cache["model.layers.28.attn"] assert runner_kv_caches[0] is kv_cache["model.layers.20.attn"] assert runner_kv_caches[1] is kv_cache["model.layers.28.attn"] def test_bind_kv_cache_draft_model(default_vllm_config): from vllm.model_executor.layers.attention import Attention layer_names = [ "model.layers.0.attn", "model.layers.1.attn", "draft_model.layers.0.attn", "draft_model.layers.1.attn", ] ctx = { layer_name: Attention(32, 128, 0.1, prefix=layer_name) for layer_name in layer_names } kv_cache = {layer_name: torch.zeros((1,)) for layer_name in layer_names} runner_kv_caches: list[torch.Tensor] = [] bind_kv_cache(kv_cache, ctx, runner_kv_caches) assert ctx["model.layers.0.attn"].kv_cache[0] is kv_cache["model.layers.0.attn"] assert ctx["model.layers.1.attn"].kv_cache[0] is kv_cache["model.layers.1.attn"] assert ( ctx["draft_model.layers.0.attn"].kv_cache[0] is kv_cache["draft_model.layers.0.attn"] ) assert ( ctx["draft_model.layers.1.attn"].kv_cache[0] is kv_cache["draft_model.layers.1.attn"] ) # caches are ordered by layer_index, interleaving target and draft model assert runner_kv_caches[0] is kv_cache["model.layers.0.attn"] assert runner_kv_caches[1] is kv_cache["draft_model.layers.0.attn"] assert runner_kv_caches[2] is kv_cache["model.layers.1.attn"] assert runner_kv_caches[3] is kv_cache["draft_model.layers.1.attn"]
{ "repo_id": "vllm-project/vllm", "file_path": "tests/v1/worker/test_utils.py", "license": "Apache License 2.0", "lines": 75, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/model_executor/layers/batch_invariant.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import os from collections.abc import Callable from typing import Any import torch from vllm.logger import init_logger from vllm.platforms import current_platform from vllm.triton_utils import tl, triton from vllm.utils.platform_utils import num_compute_units from vllm.utils.torch_utils import is_torch_equal_or_newer from vllm.v1.attention.backends.registry import AttentionBackendEnum logger = init_logger(__name__) def _matmul_launch_metadata( grid: Callable[..., Any], kernel: Any, args: dict[str, Any] ) -> dict[str, Any]: ret = {} m, n, k = args["M"], args["N"], args["K"] ret["name"] = f"{kernel.name} [M={m}, N={n}, K={k}]" if "tiles_per_update" in args: ret["name"] = ( f"{kernel.name} [M={m}, N={n}, K={k}, " f"tiles_per_update={args['tiles_per_update']:02}]" ) if "c_ptr" in args: bytes_per_elem = args["c_ptr"].element_size() else: bytes_per_elem = 1 if args["FP8_OUTPUT"] else 2 ret[f"flops{bytes_per_elem * 8}"] = 2.0 * m * n * k ret["bytes"] = bytes_per_elem * (m * k + n * k + m * n) return ret @triton.jit def _compute_pid(tile_id, num_pid_in_group, num_pid_m, GROUP_SIZE_M, NUM_SMS): group_id = tile_id // num_pid_in_group first_pid_m = group_id * GROUP_SIZE_M group_size_m = min(num_pid_m - first_pid_m, GROUP_SIZE_M) pid_m = first_pid_m + (tile_id % group_size_m) pid_n = (tile_id % num_pid_in_group) // group_size_m return pid_m, pid_n @triton.jit(launch_metadata=_matmul_launch_metadata) def matmul_kernel_persistent( a_ptr, b_ptr, c_ptr, # bias_ptr, M, N, K, # stride_am, stride_ak, stride_bk, stride_bn, stride_cm, stride_cn, BLOCK_SIZE_M: tl.constexpr, # BLOCK_SIZE_N: tl.constexpr, # BLOCK_SIZE_K: tl.constexpr, # GROUP_SIZE_M: tl.constexpr, # NUM_SMS: tl.constexpr, # A_LARGE: tl.constexpr, B_LARGE: tl.constexpr, C_LARGE: tl.constexpr, HAS_BIAS: tl.constexpr, ): start_pid = tl.program_id(axis=0) num_pid_m = tl.cdiv(M, BLOCK_SIZE_M) num_pid_n = tl.cdiv(N, BLOCK_SIZE_N) k_tiles = tl.cdiv(K, BLOCK_SIZE_K) num_tiles = num_pid_m * num_pid_n tile_id_c = start_pid - NUM_SMS offs_k_for_mask = tl.arange(0, BLOCK_SIZE_K) num_pid_in_group = GROUP_SIZE_M * num_pid_n for tile_id in tl.range(start_pid, num_tiles, NUM_SMS, flatten=True): pid_m, pid_n = _compute_pid( tile_id, num_pid_in_group, num_pid_m, GROUP_SIZE_M, NUM_SMS ) start_m = pid_m * BLOCK_SIZE_M start_n = pid_n * BLOCK_SIZE_N offs_am = start_m + tl.arange(0, BLOCK_SIZE_M) offs_bn = start_n + tl.arange(0, BLOCK_SIZE_N) if A_LARGE: offs_am = offs_am.to(tl.int64) if B_LARGE: offs_bn = offs_bn.to(tl.int64) offs_am = tl.where(offs_am < M, offs_am, 0) offs_bn = tl.where(offs_bn < N, offs_bn, 0) offs_am = tl.max_contiguous(tl.multiple_of(offs_am, BLOCK_SIZE_M), BLOCK_SIZE_M) offs_bn = tl.max_contiguous(tl.multiple_of(offs_bn, BLOCK_SIZE_N), BLOCK_SIZE_N) accumulator = tl.zeros((BLOCK_SIZE_M, BLOCK_SIZE_N), dtype=tl.float32) for ki in range(k_tiles): if A_LARGE or B_LARGE: offs_k = ki * BLOCK_SIZE_K + tl.arange(0, BLOCK_SIZE_K).to(tl.int64) else: offs_k = ki * BLOCK_SIZE_K + tl.arange(0, BLOCK_SIZE_K) a_ptrs = a_ptr + ( offs_am[:, None] * stride_am + offs_k[None, :] * stride_ak ) b_ptrs = b_ptr + ( offs_k[:, None] * stride_bk + offs_bn[None, :] * stride_bn ) a = tl.load( a_ptrs, mask=offs_k_for_mask[None, :] < K - ki * BLOCK_SIZE_K, other=0.0 ) b = tl.load( b_ptrs, mask=offs_k_for_mask[:, None] < K - ki * BLOCK_SIZE_K, other=0.0 ) accumulator = tl.dot(a, b, accumulator) tile_id_c += NUM_SMS pid_m, pid_n = _compute_pid( tile_id_c, num_pid_in_group, num_pid_m, GROUP_SIZE_M, NUM_SMS ) offs_cm = pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M) offs_cn = pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N) if C_LARGE: offs_cm = offs_cm.to(tl.int64) offs_cn = offs_cn.to(tl.int64) c_ptrs = c_ptr + stride_cm * offs_cm[:, None] + stride_cn * offs_cn[None, :] c_mask = (offs_cm[:, None] < M) & (offs_cn[None, :] < N) if HAS_BIAS: bias_ptrs = bias_ptr + offs_cn bias = tl.load(bias_ptrs, mask=offs_cn < N, other=0.0).to(tl.float32) accumulator += bias c = accumulator.to(c_ptr.dtype.element_ty) tl.store(c_ptrs, c, mask=c_mask) def matmul_persistent( a: torch.Tensor, b: torch.Tensor, bias: torch.Tensor | None = None ): # Check constraints. assert a.shape[1] == b.shape[0], "Incompatible dimensions" assert a.dtype == b.dtype, "Incompatible dtypes" assert bias is None or bias.dim() == 1, ( "Currently assuming bias is 1D, let Horace know if you run into this" ) NUM_SMS = num_compute_units(a.device.index) M, K = a.shape K, N = b.shape dtype = a.dtype # Allocates output. c = torch.empty((M, N), device=a.device, dtype=dtype) # 1D launch kernel where each block gets its own program. def grid(META): return ( min( NUM_SMS, triton.cdiv(M, META["BLOCK_SIZE_M"]) * triton.cdiv(N, META["BLOCK_SIZE_N"]), ), ) configs = { torch.bfloat16: { "BLOCK_SIZE_M": 128, "BLOCK_SIZE_N": 128, "BLOCK_SIZE_K": 64, "GROUP_SIZE_M": 8, "num_stages": 3, "num_warps": 8, }, torch.float16: { "BLOCK_SIZE_M": 128, "BLOCK_SIZE_N": 256, "BLOCK_SIZE_K": 64, "GROUP_SIZE_M": 8, "num_stages": 3, "num_warps": 8, }, torch.float32: { "BLOCK_SIZE_M": 128, "BLOCK_SIZE_N": 128, "BLOCK_SIZE_K": 32, "GROUP_SIZE_M": 8, "num_stages": 3, "num_warps": 8, }, } # print(a.device, b.device, c.device) matmul_kernel_persistent[grid]( a, b, c, # bias, M, N, K, # a.stride(0), a.stride(1), # b.stride(0), b.stride(1), # c.stride(0), c.stride(1), # NUM_SMS=NUM_SMS, # A_LARGE=a.numel() > 2**31, B_LARGE=b.numel() > 2**31, C_LARGE=c.numel() > 2**31, HAS_BIAS=bias is not None, **configs[dtype], ) return c @triton.jit def bmm_kernel( a_ptr, # (*, ) pointer to A, (B, M, K) b_ptr, # (*, ) pointer to B, (B, K, N) c_ptr, # (*, ) pointer to C, (B, M, N) B, # int, batch size M, # int, output rows N, # int, output cols K, # int, reduction dim stride_ab, stride_am, stride_ak, stride_bb, stride_bk, stride_bn, stride_cb, stride_cm, stride_cn, BLOCK_SIZE_M: tl.constexpr, BLOCK_SIZE_N: tl.constexpr, BLOCK_SIZE_K: tl.constexpr, A_LARGE: tl.constexpr, B_LARGE: tl.constexpr, C_LARGE: tl.constexpr, ): """Batched GEMM: (B, M, K) x (B, K, N) -> (B, M, N) Each program computes one (batch_idx, tile_m, tile_n) tile, accumulating along K in a fixed order to preserve batch invariance. """ pid_b = tl.program_id(0) pid = tl.program_id(1) if pid_b >= B: return # number of tiles along M / N num_pid_m = tl.cdiv(M, BLOCK_SIZE_M) num_pid_n = tl.cdiv(N, BLOCK_SIZE_N) pid_m = pid // num_pid_n pid_n = pid % num_pid_n if pid_m >= num_pid_m or pid_n >= num_pid_n: return # offs_m / offs_n: raw global row/col indices for this tile offs_m = pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M) offs_n = pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N) # masks for valid logical rows/cols within (M, N) mask_m = offs_m < M # [BLOCK_SIZE_M] mask_n = offs_n < N # [BLOCK_SIZE_N] if A_LARGE or B_LARGE or C_LARGE: offs_m = offs_m.to(tl.int64) offs_n = offs_n.to(tl.int64) offs_m = tl.where(mask_m, offs_m, 0) offs_n = tl.where(mask_n, offs_n, 0) # hint for triton contiguous memory offs_m = tl.max_contiguous(tl.multiple_of(offs_m, BLOCK_SIZE_M), BLOCK_SIZE_M) offs_n = tl.max_contiguous(tl.multiple_of(offs_n, BLOCK_SIZE_N), BLOCK_SIZE_N) # base pointers for current batch, shape-wise: # a_batch_ptr points to A[pid_b, 0, 0] # b_batch_ptr points to B[pid_b, 0, 0] # c_batch_ptr points to C[pid_b, 0, 0] a_batch_ptr = a_ptr + pid_b * stride_ab b_batch_ptr = b_ptr + pid_b * stride_bb c_batch_ptr = c_ptr + pid_b * stride_cb accumulator = tl.zeros((BLOCK_SIZE_M, BLOCK_SIZE_N), dtype=tl.float32) # number of K-blocks this tile iterates over k_tiles = tl.cdiv(K, BLOCK_SIZE_K) offs_k_mask = tl.arange(0, BLOCK_SIZE_K) for ki in range(k_tiles): if A_LARGE or B_LARGE: # offs_k: [BLOCK_SIZE_K], global K indices offs_k = ki * BLOCK_SIZE_K + tl.arange(0, BLOCK_SIZE_K).to(tl.int64) else: offs_k = ki * BLOCK_SIZE_K + tl.arange(0, BLOCK_SIZE_K) # a_ptrs: [BLOCK_SIZE_M, BLOCK_SIZE_K] # element (i, j) points to A[pid_b, offs_m[i], offs_k[j]] a_ptrs = a_batch_ptr + ( offs_m[:, None] * stride_am + offs_k[None, :] * stride_ak ) # b_ptrs: [BLOCK_SIZE_K, BLOCK_SIZE_N] # element (i, j) points to B[pid_b, offs_k[i], offs_n[j]] b_ptrs = b_batch_ptr + ( offs_k[:, None] * stride_bk + offs_n[None, :] * stride_bn ) # valid K lanes for this block k_valid = offs_k_mask < (K - ki * BLOCK_SIZE_K) # A mask within (M, K): [BLOCK_SIZE_M, BLOCK_SIZE_K] a_mask = mask_m[:, None] & k_valid[None, :] # B mask within (K, N): [BLOCK_SIZE_K, BLOCK_SIZE_N] b_mask = k_valid[:, None] & mask_n[None, :] # a: [BLOCK_SIZE_M, BLOCK_SIZE_K] from A[offs_m, offs_k] a = tl.load( a_ptrs, mask=a_mask, other=0.0, ) # b: [BLOCK_SIZE_K, BLOCK_SIZE_N] from B[offs_k, offs_n] b = tl.load( b_ptrs, mask=b_mask, other=0.0, ) accumulator = tl.dot(a, b, accumulator) # c_m / c_n: [BLOCK_SIZE_M] / [BLOCK_SIZE_N], row/col indices for C c_m = offs_m c_n = offs_n if C_LARGE: c_m = c_m.to(tl.int64) c_n = c_n.to(tl.int64) # c_ptrs: [BLOCK_SIZE_M, BLOCK_SIZE_N] # element (i, j) points to C[pid_b, c_m[i], c_n[j]] c_ptrs = c_batch_ptr + stride_cm * c_m[:, None] + stride_cn * c_n[None, :] # mask out elements that fall outside logical (M, N) range c_mask = mask_m[:, None] & mask_n[None, :] # cast FP32 accumulator back to original dtype of C c = accumulator.to(c_ptr.dtype.element_ty) tl.store(c_ptrs, c, mask=c_mask) @triton.jit def _log_softmax_kernel( input_ptr, output_ptr, input_row_stride, output_row_stride, n_cols, BLOCK_SIZE: tl.constexpr, ): """ Compute log_softmax along the last dimension of a 2D tensor. Each block handles one row of the input tensor. """ # Get the row index for this block row_idx = tl.program_id(0).to(tl.int64) # Compute base pointers for input and output rows row_start_ptr = input_ptr + row_idx * input_row_stride output_row_start_ptr = output_ptr + row_idx * output_row_stride # Step 1: Find maximum value in the row for numerical stability max_val = -float("inf") for col_offset in range(0, n_cols, BLOCK_SIZE): col_idx = col_offset + tl.arange(0, BLOCK_SIZE) mask = col_idx < n_cols # Load values vals = tl.load(row_start_ptr + col_idx, mask=mask, other=-float("inf")) # Update maximum max_val = tl.max(tl.maximum(vals, max_val)) # Step 2: Compute sum of exp(x - max_val) sum_exp = 0.0 for col_offset in range(0, n_cols, BLOCK_SIZE): col_idx = col_offset + tl.arange(0, BLOCK_SIZE) mask = col_idx < n_cols # Load values vals = tl.load(row_start_ptr + col_idx, mask=mask, other=0.0) # Compute exp(x - max_val) and accumulate exp_vals = tl.exp(vals - max_val) sum_exp += tl.sum(tl.where(mask, exp_vals, 0.0)) # Compute log(sum_exp) log_sum_exp = tl.log(sum_exp) # Step 3: Compute final log_softmax values: x - max_val - log_sum_exp for col_offset in range(0, n_cols, BLOCK_SIZE): col_idx = col_offset + tl.arange(0, BLOCK_SIZE) mask = col_idx < n_cols # Load values vals = tl.load(row_start_ptr + col_idx, mask=mask) # Compute log_softmax output = vals - max_val - log_sum_exp # Store results tl.store(output_row_start_ptr + col_idx, output, mask=mask) def log_softmax(input: torch.Tensor, dim: int = -1) -> torch.Tensor: """ Compute log_softmax using Triton kernel. Args: input: Input tensor dim: Dimension along which to compute log_softmax (only -1 or last dim supported) >> Stashed changes Returns: Tensor with log_softmax applied along the specified dimension """ if dim != -1 and dim != input.ndim - 1: raise ValueError( "This implementation only supports log_softmax along the last dimension" ) # Flatten all dimensions except the last one original_shape = input.shape input_2d = input.reshape(-1, input.shape[-1]) input_2d = input_2d.contiguous() n_rows, n_cols = input_2d.shape # Allocate output tensor output = torch.empty_like(input_2d) # Choose block size based on the number of columns BLOCK_SIZE = 1024 # Launch kernel with one block per row grid = (n_rows,) _log_softmax_kernel[grid]( input_2d, output, input_2d.stride(0), output.stride(0), n_cols, BLOCK_SIZE=BLOCK_SIZE, ) # Reshape output back to original shape return output.reshape(original_shape) @triton.jit def mean_kernel( input_ptr, output_ptr, input_stride0, input_stride1, input_stride2, output_stride0, output_stride1, M, # size before reduction dim N, # size of reduction dim K, # size after reduction dim BLOCK_SIZE: tl.constexpr, ): """ Kernel for computing mean along a single dimension. Input is viewed as (M, N, K) where N is the dimension being reduced. """ # Program ID gives us which output element we're computing pid = tl.program_id(0) # Compute output indices m_idx = pid // K k_idx = pid % K # Bounds check if m_idx >= M or k_idx >= K: return # Accumulate sum across reduction dimension acc = 0.0 for n_start in range(0, N, BLOCK_SIZE): n_offsets = n_start + tl.arange(0, BLOCK_SIZE) mask = n_offsets < N # Calculate input indices input_idx = ( m_idx * input_stride0 + n_offsets * input_stride1 + k_idx * input_stride2 ) # Load and accumulate vals = tl.load(input_ptr + input_idx, mask=mask, other=0.0) acc += tl.sum(vals) # Compute mean and store mean_val = acc / N output_idx = m_idx * output_stride0 + k_idx * output_stride1 tl.store(output_ptr + output_idx, mean_val) def mean_dim( input: torch.Tensor, dim: int, keepdim: bool = False, dtype: torch.dtype | None = None, ) -> torch.Tensor: """ Triton implementation of torch.mean with single dimension reduction. Args: input: Input tensor dim: Single dimension along which to compute mean keepdim: Whether to keep the reduced dimension dtype: Output dtype. If None, uses input dtype (or float32 for integer inputs) Returns: Tensor with mean values along specified dimension """ # Validate inputs assert -input.ndim <= dim < input.ndim, ( f"Invalid dimension {dim} for tensor with {input.ndim} dimensions" ) # Handle negative dim if dim < 0: dim = dim + input.ndim # Handle dtype if dtype is None: if input.dtype in [torch.int8, torch.int16, torch.int32, torch.int64]: dtype = torch.float32 else: dtype = input.dtype # Convert input to appropriate dtype if needed if input.dtype != dtype: input = input.to(dtype) # Get input shape and strides shape = list(input.shape) # Calculate dimensions for kernel M = 1 for i in range(dim): M *= shape[i] N = shape[dim] K = 1 for i in range(dim + 1, len(shape)): K *= shape[i] # Reshape input to 3D view (M, N, K) input_3d = input.reshape(M, N, K) # Create output shape if keepdim: output_shape = shape.copy() output_shape[dim] = 1 else: output_shape = shape[:dim] + shape[dim + 1 :] # Create output tensor output = torch.empty(output_shape, dtype=dtype, device=input.device) # Reshape output for kernel output_2d = output.reshape(M, 1, K).squeeze(1) if keepdim else output.reshape(M, K) # Launch kernel grid = (M * K,) BLOCK_SIZE = 1024 mean_kernel[grid]( input_3d, output_2d, input_3d.stride(0), input_3d.stride(1), input_3d.stride(2), output_2d.stride(0), output_2d.stride(1) if output_2d.ndim > 1 else 0, M, N, K, BLOCK_SIZE, ) return output def mm_batch_invariant(a, b): return matmul_persistent(a, b) def matmul_batch_invariant(a, b, *, out=None): # torch.matmul can handle various dimensions # For 2D x 2D, it's the same as mm if a.ndim == 2 and b.ndim == 2: result = matmul_persistent(a, b) if out is not None: out.copy_(result) return out return result elif a.ndim == 3 and b.ndim == 3: # Handle batched case like bmm return bmm_batch_invariant(a, b, out=out) elif a.ndim == 3 and b.ndim == 2: # Handle 3D x 2D: common for linear layers # (batch, seq, hidden) @ (hidden, out) -> (batch, seq, out) # Reshape to 2D, do mm, reshape back batch, seq, hidden = a.shape a_2d = a.reshape(-1, hidden) result_2d = matmul_persistent(a_2d, b) result = result_2d.reshape(batch, seq, -1) if out is not None: out.copy_(result) return out return result elif a.ndim == 2 and b.ndim == 3: # Handle 2D x 3D: (M, K) @ (B, K, N) -> (B, M, N) # By broadcasting `a` to 3D, we can reuse the batched matrix # multiplication logic. a_expanded = a.unsqueeze(0).expand(b.shape[0], -1, -1) return bmm_batch_invariant(a_expanded, b, out=out) elif a.ndim == 4 and b.ndim == 4: # Handle 4D attention tensors: [batch, heads, seq, dim] # Reshape to 3D, process, reshape back batch, heads, seq_a, dim_a = a.shape _, _, dim_b, seq_b = b.shape # Reshape to [batch*heads, seq_a, dim_a] a_3d = a.reshape(batch * heads, seq_a, dim_a) b_3d = b.reshape(batch * heads, dim_b, seq_b) # Do batched matmul result_3d = bmm_batch_invariant(a_3d, b_3d) # Reshape back to [batch, heads, seq_a, seq_b] result = result_3d.reshape(batch, heads, seq_a, seq_b) if out is not None: out.copy_(result) return out return result else: raise ValueError( f"matmul_batch_invariant currently only supports 2D x 2D, 3D x 3D, " f"3D x 2D, 2D x 3D, and 4D x 4D, " f"got shapes {a.shape} and {b.shape}" ) def bmm_batch_invariant(a, b, *, out=None): # Batched matrix multiply: (B, M, K) x (B, K, N) -> (B, M, N) if not (a.ndim == 3 and b.ndim == 3): raise ValueError( f"bmm_batch_invariant expects 3D tensors, " f"got shapes {a.shape} and {b.shape}" ) if a.shape[0] != b.shape[0]: raise ValueError( f"Batch dimensions of tensors must match, " f"but got {a.shape[0]} and {b.shape[0]}." ) if a.shape[2] != b.shape[1]: raise ValueError( f"Incompatible inner dimensions for matmul: got {a.shape} and {b.shape}." ) if a.dtype != b.dtype: raise ValueError(f"Incompatible dtypes: got {a.dtype} and {b.dtype}.") B, M, K = a.shape _, _, N = b.shape dtype = a.dtype if out is None: c = torch.empty((B, M, N), device=a.device, dtype=dtype) else: assert out.shape == (B, M, N), "out tensor has incorrect shape" assert out.dtype == dtype and out.device == a.device, "out tensor mismatch" c = out configs = { torch.bfloat16: { "BLOCK_SIZE_M": 128, "BLOCK_SIZE_N": 128, "BLOCK_SIZE_K": 64, "num_stages": 3, "num_warps": 8, }, torch.float16: { "BLOCK_SIZE_M": 128, "BLOCK_SIZE_N": 256, "BLOCK_SIZE_K": 64, "num_stages": 3, "num_warps": 8, }, torch.float32: { "BLOCK_SIZE_M": 128, "BLOCK_SIZE_N": 128, "BLOCK_SIZE_K": 32, "num_stages": 3, "num_warps": 8, }, } cfg = configs[dtype] # grid = (B, num_tiles_per_matrix) grid = ( B, triton.cdiv(M, cfg["BLOCK_SIZE_M"]) * triton.cdiv(N, cfg["BLOCK_SIZE_N"]), ) bmm_kernel[grid]( a, b, c, B, M, N, K, a.stride(0), a.stride(1), a.stride(2), b.stride(0), b.stride(1), b.stride(2), c.stride(0), c.stride(1), c.stride(2), A_LARGE=a.numel() > 2**31, B_LARGE=b.numel() > 2**31, C_LARGE=c.numel() > 2**31, **cfg, ) return c def addmm_batch_invariant(bias, a, b): return matmul_persistent(a, b, bias=bias) def _log_softmax_batch_invariant(input, dim, _half_to_float): assert not _half_to_float, "not implemented" return log_softmax(input, dim=dim) def softmax_batch_invariant(input, dim, dtype=None): # Compute softmax in a deterministic way # First subtract max for numerical stability (standard practice) input_max = torch.amax(input, dim=dim, keepdim=True) input = input - input_max exp_x = torch.exp(input) sum_exp_x = torch.sum(exp_x, dim=dim, keepdim=True) return exp_x / sum_exp_x def mean_batch_invariant(input, dim, keepdim=False, dtype: torch.dtype | None = None): assert dtype is None or dtype == torch.float32, f"unsupported dtype: {dtype}" result = input.to(torch.float32) if len(dim) == 0: dim = [i for i in range(len(input.shape))] # Sort dimensions to reduce from largest to smallest to handle shifting dims # during iterative reduction. sorted_dims = sorted([d % input.ndim for d in dim], reverse=True) # Iteratively apply a deterministic mean. for d in sorted_dims: result = mean_dim(result, dim=d, keepdim=True) if not keepdim: # Squeeze the reduced dimensions. for d in sorted_dims: result = result.squeeze(d) return result @triton.jit def _rms_norm_kernel( input_ptr, weight_ptr, output_ptr, input_row_stride, output_row_stride, n_cols, eps, BLOCK_SIZE: tl.constexpr, ): """ Compute RMS normalization along the last dimension of a 2D tensor. RMS Norm: y = x / sqrt(mean(x^2) + eps) * weight Each block handles one row of the input tensor. """ row_idx = tl.program_id(0).to(tl.int64) row_start_ptr = input_ptr + row_idx * input_row_stride output_row_start_ptr = output_ptr + row_idx * output_row_stride # Step 1: Compute sum of squares in float32 to avoid overflow sum_sq = tl.zeros([1], dtype=tl.float32) for col_offset in range(0, n_cols, BLOCK_SIZE): col_idx = col_offset + tl.arange(0, BLOCK_SIZE) mask = col_idx < n_cols vals = tl.load(row_start_ptr + col_idx, mask=mask, other=0.0) # Convert to float32 for accumulation to prevent overflow vals_f32 = vals.to(tl.float32) sq_vals = vals_f32 * vals_f32 sum_sq += tl.sum(tl.where(mask, sq_vals, 0.0)) # Step 2: Compute RMS (root mean square) in float32 mean_sq = sum_sq / n_cols rms = tl.sqrt(mean_sq + eps) inv_rms = 1.0 / rms # Step 3: Normalize and apply weight for col_offset in range(0, n_cols, BLOCK_SIZE): col_idx = col_offset + tl.arange(0, BLOCK_SIZE) mask = col_idx < n_cols vals = tl.load(row_start_ptr + col_idx, mask=mask, other=0.0) weight = tl.load(weight_ptr + col_idx, mask=mask, other=1.0) # Compute in float32 then convert back to input dtype vals_f32 = vals.to(tl.float32) weight_f32 = weight.to(tl.float32) output_f32 = vals_f32 * inv_rms * weight_f32 output = output_f32.to(vals.dtype) tl.store(output_row_start_ptr + col_idx, output, mask=mask) def rms_norm( input: torch.Tensor, weight: torch.Tensor, eps: float = 1e-6 ) -> torch.Tensor: """ Compute RMS normalization using Triton kernel. RMS Norm normalizes the input by the root mean square and scales by weight: output = input / sqrt(mean(input^2) + eps) * weight Args: input: Input tensor of shape (..., hidden_size) weight: Weight tensor of shape (hidden_size,) eps: Small constant for numerical stability Returns: Tensor with RMS normalization applied along the last dimension """ assert weight.dim() == 1, "Weight must be 1-dimensional" assert input.shape[-1] == weight.shape[0], ( f"Input last dimension ({input.shape[-1]}) must match " f"weight dimension ({weight.shape[0]})" ) # Flatten all dimensions except the last one original_shape = input.shape input_2d = input.reshape(-1, input.shape[-1]) input_2d = input_2d.contiguous() weight = weight.contiguous() n_rows, n_cols = input_2d.shape output = torch.empty_like(input_2d) BLOCK_SIZE = 1024 grid = (n_rows,) _rms_norm_kernel[grid]( input_2d, weight, output, input_2d.stride(0), output.stride(0), n_cols, eps, BLOCK_SIZE=BLOCK_SIZE, ) return output.reshape(original_shape) def rms_norm_batch_invariant( input: torch.Tensor, weight: torch.Tensor, eps: float = 1e-6 ) -> torch.Tensor: """ Batch-invariant wrapper for RMS normalization. This function provides a deterministic, batch-invariant implementation of RMS normalization for use with the batch_invariant mode. Args: input: Input tensor of shape (..., hidden_size) weight: Weight tensor of shape (hidden_size,) eps: Small constant for numerical stability Returns: RMS normalized tensor """ return rms_norm(input, weight, eps=eps) def linear_batch_invariant(input, weight, bias=None): output = matmul_batch_invariant(input, weight.t()) if bias is not None: output = output + bias return output _batch_invariant_MODE = False _batch_invariant_LIB = None _original_torch_bmm = None _original_fp16_reduction_precision = None _original_bf16_reduction_precision = None _original_cublas_workspace_cfg = None _original_cublaslt_workspace_size = None def enable_batch_invariant_mode(): global _batch_invariant_MODE, _batch_invariant_LIB, _original_torch_bmm global _original_fp16_reduction_precision, _original_bf16_reduction_precision global _original_cublas_workspace_cfg, _original_cublaslt_workspace_size if _batch_invariant_MODE: return _batch_invariant_MODE = True _batch_invariant_LIB = torch.library.Library("aten", "IMPL") if ( current_platform.is_device_capability_family(100) or current_platform.is_device_capability(80) or current_platform.is_device_capability(89) ): # For PyTorch 2.9, B200 uses GEMV for bs=1 # Requires https://github.com/pytorch/pytorch/pull/166735 _batch_invariant_LIB.impl("aten::mm", mm_batch_invariant, "CUDA") _batch_invariant_LIB.impl("aten::addmm", addmm_batch_invariant, "CUDA") _batch_invariant_LIB.impl("aten::matmul", matmul_batch_invariant, "CUDA") _batch_invariant_LIB.impl("aten::linear", linear_batch_invariant, "CUDA") else: # Only source of batch invariance for Hopper is split-k, can disable through # cuBLAS workspace config _original_cublas_workspace_cfg = os.environ.get("CUBLAS_WORKSPACE_CONFIG", None) _original_cublaslt_workspace_size = os.environ.get( "CUBLASLT_WORKSPACE_SIZE", None ) os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":16:8" os.environ["CUBLASLT_WORKSPACE_SIZE"] = "1" _batch_invariant_LIB.impl( "aten::_log_softmax", _log_softmax_batch_invariant, "CUDA" ) _batch_invariant_LIB.impl("aten::softmax", softmax_batch_invariant, "CUDA") _batch_invariant_LIB.impl("aten::_softmax", softmax_batch_invariant, "CUDA") _batch_invariant_LIB.impl("aten::mean.dim", mean_batch_invariant, "CUDA") # Also monkeypatch torch.bmm directly as a fallback _batch_invariant_LIB.impl("aten::bmm", bmm_batch_invariant, "CUDA") _original_torch_bmm = torch.bmm torch.bmm = bmm_batch_invariant _original_bf16_reduction_precision = ( torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction ) _original_fp16_reduction_precision = ( torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction ) reduced_precision_val = ( (False, False) if is_torch_equal_or_newer("2.10.0") else False ) torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction = ( reduced_precision_val ) torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction = ( reduced_precision_val ) torch.backends.cuda.preferred_blas_library(backend="cublaslt") def _read_vllm_batch_invariant() -> bool: val = os.getenv("VLLM_BATCH_INVARIANT", "0") try: return int(val) != 0 except ValueError: return False VLLM_BATCH_INVARIANT: bool = _read_vllm_batch_invariant() def vllm_is_batch_invariant() -> bool: return VLLM_BATCH_INVARIANT def override_envs_for_invariance( attention_backend: AttentionBackendEnum | None, ): decode_invariant_backends = [ AttentionBackendEnum.FLASH_ATTN, # best supported backend AttentionBackendEnum.TRITON_ATTN, ] supported_backends = decode_invariant_backends + [ # FlashInfer temporarily disabled due to invariant CTA sizes. # See FlashInfer issue #2424 # AttentionBackendEnum.FLASHINFER, AttentionBackendEnum.FLASH_ATTN_MLA, AttentionBackendEnum.TRITON_MLA, # Not yet supported MLA backends # AttentionBackendEnum.FLASHMLA, # AttentionBackendEnum.FLEX_ATTENTION, # IMA issue # AttentionBackendEnum.FLASHINFER_MLA, # PR #28967 ] if attention_backend not in supported_backends: supported_names = [b.name for b in supported_backends] backend_name = attention_backend.name if attention_backend else None error = ( "VLLM batch_invariant mode requires an attention backend in " f"{supported_names}, but got '{backend_name}'. " "Please use --attention-backend or attention_config to set " "one of the supported backends before enabling batch_invariant." ) raise RuntimeError(error) if attention_backend not in decode_invariant_backends: warning = ( "You are using a non-decode-invariant form of batch invariance. " "This will not be invariant between prefill and decode." ) logger.warning_once(warning, scope="local") os.environ["VLLM_ALLREDUCE_USE_SYMM_MEM"] = "0" os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":4096:8" # NCCL determinism settings os.environ["NCCL_LAUNCH_MODE"] = "GROUP" os.environ["NCCL_COLLNET_ENABLE"] = "0" os.environ["NCCL_NVLS_ENABLE"] = "0" os.environ["NCCL_P2P_NET_DISABLE"] = "1" os.environ["NCCL_MIN_NCHANNELS"] = "1" os.environ["NCCL_MAX_NCHANNELS"] = "1" os.environ["NCCL_PROTO"] = "Simple" os.environ["NCCL_ALGO"] = "allreduce:tree" os.environ["NCCL_NTHREADS"] = "1" os.environ["NCCL_SOCKET_NTHREADS"] = "1" # torch.compile settings os.environ["VLLM_USE_AOT_COMPILE"] = "0" def init_batch_invariance( attention_backend: AttentionBackendEnum | None, ): # this will hit all the csrc overrides as well if vllm_is_batch_invariant(): override_envs_for_invariance(attention_backend) enable_batch_invariant_mode() # Disable TF32 for batch invariance - it causes non-deterministic rounding torch.backends.cuda.matmul.fp32_precision = "ieee" torch.backends.cudnn.conv.fp32_precision = "ieee" torch.backends.cudnn.rnn.fp32_precision = "ieee"
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/model_executor/layers/batch_invariant.py", "license": "Apache License 2.0", "lines": 906, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/v1/spec_decode/test_mtp.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from unittest import mock import pytest import torch from tests.v1.attention.utils import ( BatchSpec, create_common_attn_metadata, create_standard_kv_cache_spec, try_get_attention_backend, ) from vllm.config import ( CacheConfig, DeviceConfig, ModelConfig, ParallelConfig, SchedulerConfig, SpeculativeConfig, VllmConfig, ) from vllm.config.load import LoadConfig from vllm.model_executor.models.llama import LlamaForCausalLM from vllm.platforms import current_platform from vllm.v1.attention.backends.registry import AttentionBackendEnum from vllm.v1.spec_decode.eagle import EagleProposer mimo_7b_dir = "XiaomiMiMo/MiMo-7B-Base" def _create_mtp_proposer(num_speculative_tokens: int) -> EagleProposer: """Create an MTP proposer with unified model configuration.""" model_config = ModelConfig( model=mimo_7b_dir, runner="generate", max_model_len=100, trust_remote_code=True ) speculative_config = SpeculativeConfig( target_model_config=model_config, target_parallel_config=ParallelConfig(), model=mimo_7b_dir, method="mtp", num_speculative_tokens=num_speculative_tokens, ) vllm_config = VllmConfig( model_config=model_config, cache_config=CacheConfig(), speculative_config=speculative_config, device_config=DeviceConfig(device=current_platform.device_type), parallel_config=ParallelConfig(), load_config=LoadConfig(), scheduler_config=SchedulerConfig( max_model_len=model_config.max_model_len, is_encoder_decoder=model_config.is_encoder_decoder, ), ) return EagleProposer(vllm_config=vllm_config, device=current_platform.device_type) @mock.patch("vllm.v1.spec_decode.eagle.get_pp_group") @mock.patch("vllm.v1.spec_decode.eagle.get_layers_from_vllm_config") @mock.patch("vllm.v1.spec_decode.eagle.get_model") def test_mtp_load_model_unified(mock_get_model, mock_get_layers, mock_get_pp_group): """Test MTP-specific model loading with unified model approach.""" # Setup mocks mock_model = mock.MagicMock() mock_model.model.embed_tokens.weight.shape = (131072, 4096) mock_get_model.return_value = mock_model # MTP does not have its own embed_tokens or lm_head # so it should share them with the target model mock_model.has_own_embed_tokens = False mock_model.has_own_lm_head = False target_attn_layers = {"target_attn_1": mock.MagicMock()} all_attn_layers = {**target_attn_layers, "draft_attn_1": mock.MagicMock()} target_indexer_layers: dict = {} all_indexer_layers: dict = {} mock_get_layers.side_effect = [ target_attn_layers, target_indexer_layers, all_attn_layers, all_indexer_layers, ] mock_pp_group = mock.MagicMock() mock_pp_group.world_size = 1 mock_get_pp_group.return_value = mock_pp_group # Create target model class _TargetModelStub(LlamaForCausalLM): model: mock.MagicMock lm_head: mock.MagicMock target_model = mock.create_autospec(_TargetModelStub, instance=True) target_model.model = mock.MagicMock() target_model.model.embed_tokens.weight.shape = (131072, 4096) target_model.lm_head = mock.MagicMock() # Create MTP proposer proposer = _create_mtp_proposer(num_speculative_tokens=4) proposer.load_model(target_model) # Verify MTP-specific behavior: # Model is loaded mock_get_model.assert_called_once() # MTP shares lm_head with target model assert proposer.model.lm_head == target_model.lm_head # MTP shares embed_tokens with target model assert proposer.model.model.embed_tokens == target_model.model.embed_tokens @pytest.mark.parametrize("num_speculative_tokens", [1]) def test_mtp_propose(num_speculative_tokens, monkeypatch): """Test that MTP's forward method returns hidden states directly""" device = torch.device(current_platform.device_type) batch_size = 2 seq_lens = [5, 3] total_tokens = sum(seq_lens) vocab_size = 100 proposer = _create_mtp_proposer(num_speculative_tokens) hidden_size = proposer.hidden_size # Mock the MTP model to verify it returns hidden states directly model_mock = mock.MagicMock() # MTP returns hidden states directly if num_speculative_tokens == 1: model_mock.return_value = torch.zeros(total_tokens, hidden_size, device=device) else: # Multiple forward passes for multi-token speculation forward_returns = [] for i in range(num_speculative_tokens): if i == 0: h_states = torch.zeros(total_tokens, hidden_size, device=device) else: h_states = torch.zeros(batch_size, hidden_size, device=device) forward_returns.append(h_states) model_mock.side_effect = forward_returns # Mock compute_logits def create_deterministic_logits(batch_size, vocab_size, token_offset): logits = torch.full((batch_size, vocab_size), -100.0, device=device) logits[:, token_offset] = 100.0 return logits if num_speculative_tokens == 1: model_mock.compute_logits.return_value = create_deterministic_logits( batch_size, vocab_size, 42 ) else: logits_returns = [ create_deterministic_logits(batch_size, vocab_size, 42 + i) for i in range(num_speculative_tokens) ] model_mock.compute_logits.side_effect = logits_returns proposer.model = model_mock proposer.attn_layer_names = ["layer.0"] # Prepare inputs batch_spec = BatchSpec(seq_lens=seq_lens, query_lens=seq_lens) common_attn_metadata = create_common_attn_metadata( batch_spec, block_size=16, device=device ) target_token_ids = torch.randint(0, vocab_size, (total_tokens,), device=device) target_positions = torch.cat( [ torch.arange(seq_lens[0], device=device), torch.arange(seq_lens[1], device=device), ] ) target_hidden_states = torch.randn(total_tokens, hidden_size, device=device) next_token_ids = torch.randint( 0, vocab_size, (batch_size,), dtype=torch.int32, device=device ) sampling_metadata = mock.MagicMock() # Setup attention metadata attn_metadata_builder_cls, _ = try_get_attention_backend( AttentionBackendEnum.FLASH_ATTN ) attn_metadata_builder = attn_metadata_builder_cls( kv_cache_spec=create_standard_kv_cache_spec(proposer.vllm_config), layer_names=proposer.attn_layer_names, vllm_config=proposer.vllm_config, device=device, ) proposer.runner = mock.MagicMock() proposer.attn_metadata_builder = attn_metadata_builder # Run propose result = proposer.propose( target_token_ids=target_token_ids, target_positions=target_positions, target_hidden_states=target_hidden_states, next_token_ids=next_token_ids, token_indices_to_sample=None, common_attn_metadata=common_attn_metadata, sampling_metadata=sampling_metadata, ) # Verify the model was called correctly assert model_mock.called # Verify output shape assert result.shape == (batch_size, num_speculative_tokens)
{ "repo_id": "vllm-project/vllm", "file_path": "tests/v1/spec_decode/test_mtp.py", "license": "Apache License 2.0", "lines": 180, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/quantization/test_blackwell_moe.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import json import os from typing import Any import pytest from tests.utils import RemoteOpenAIServer from vllm.platforms import current_platform if not current_platform.is_device_capability_family(100): pytest.skip( "This test only runs on Blackwell GPUs (SM10x).", allow_module_level=True ) @pytest.fixture(scope="module", autouse=True) def set_test_environment(): """Sets environment variables required for this test module.""" # Make sure TRTLLM attention is available os.environ["VLLM_HAS_FLASHINFER_CUBIN"] = "1" # Set compilation threads to 16 to speed up startup os.environ["FLASHINFER_NVCC_THREADS"] = "16" # Overide the backbone layers to 4 for faster startup HF_OVERRIDE_TEXT = { "num_layers": 4, "num_hidden_layers": 4, } HF_OVERRIDE_MM = { "text_config": {"num_layers": 4, "num_hidden_layers": 4}, } def can_initialize( model: str, hf_overrides: dict[str, Any] | None = None, extra_args: list[str] | None = None, ): # Server arguments extra_args = extra_args if extra_args is not None else [] server_args = [ "--max-model-len", "2048", "--max-num-batched-tokens", "256", "--load-format", "dummy", "--trust-remote-code", "--limit-mm-per-prompt", json.dumps({"image": 0}), *extra_args, ] # Launch server and make a simple request with RemoteOpenAIServer( model, server_args, max_wait_seconds=1500, # Due to FlashInfer compile override_hf_configs=hf_overrides, ) as server: client = server.get_client() # Make a simple request to verify the server works completion = client.completions.create( model=model, prompt=["Hello, World!"], temperature=0, max_tokens=2, ) print(completion) assert completion.choices[0].text is not None ## Llama4 ## @pytest.mark.skip( reason=( "RuntimeError: run_moe() Expected a value of type " "'Optional[List[Tensor]]' for argument '_9' but instead found type " "'list'." ) ) def test_llama4_fp8_tensor_moe_flashinfer_cutlass(monkeypatch: pytest.MonkeyPatch): can_initialize( "nvidia/Llama-4-Scout-17B-16E-Instruct-FP8", hf_overrides=HF_OVERRIDE_MM, extra_args=["--moe-backend=flashinfer_cutlass"], ) def test_llama4_fp8_tensor_moe_flashinfer_trtllm(monkeypatch: pytest.MonkeyPatch): can_initialize( "nvidia/Llama-4-Scout-17B-16E-Instruct-FP8", hf_overrides=HF_OVERRIDE_MM, extra_args=["--moe-backend=flashinfer_trtllm"], ) def test_llama4_nvfp4_moe_flashinfer_cutlass(monkeypatch: pytest.MonkeyPatch): can_initialize( "nvidia/Llama-4-Scout-17B-16E-Instruct-FP4", hf_overrides=HF_OVERRIDE_MM, extra_args=["--moe-backend=flashinfer_cutlass"], ) def test_llama4_nvfp4_moe_flashinfer_trtllm(monkeypatch: pytest.MonkeyPatch): can_initialize( "nvidia/Llama-4-Scout-17B-16E-Instruct-FP4", hf_overrides=HF_OVERRIDE_MM, extra_args=["--moe-backend=flashinfer_trtllm"], ) ## DeepSeekV3 ## def test_deepseek_fp8_block_moe_deep_gemm(monkeypatch: pytest.MonkeyPatch): can_initialize( "deepseek-ai/DeepSeek-V3.1", hf_overrides=HF_OVERRIDE_TEXT, extra_args=["--moe-backend=deep_gemm"], ) @pytest.mark.skip( reason=( "Known issue: lack of kernel support. " "Expected failure: assert self.block_quant is None" ) ) def test_deepseek_fp8_block_moe_flashinfer_cutlass(monkeypatch: pytest.MonkeyPatch): can_initialize( "deepseek-ai/DeepSeek-V3.1", hf_overrides=HF_OVERRIDE_TEXT, extra_args=["--moe-backend=flashinfer_cutlass"], ) def test_deepseek_fp8_block_moe_flashinfer_trtllm(monkeypatch: pytest.MonkeyPatch): can_initialize( "deepseek-ai/DeepSeek-V3.1", hf_overrides=HF_OVERRIDE_TEXT, extra_args=["--moe-backend=flashinfer_trtllm"], ) def test_deepseek_nvfp4_moe_flashinfer_cutlass(monkeypatch: pytest.MonkeyPatch): can_initialize( "nvidia/DeepSeek-R1-0528-FP4-v2", hf_overrides=HF_OVERRIDE_TEXT, extra_args=["--moe-backend=flashinfer_cutlass"], ) def test_deepseek_nvfp4_moe_flashinfer_trtllm(monkeypatch: pytest.MonkeyPatch): can_initialize( "nvidia/DeepSeek-R1-0528-FP4-v2", hf_overrides=HF_OVERRIDE_TEXT, extra_args=["--moe-backend=flashinfer_trtllm"], ) ## GPT-OSS ## def test_gptoss_mxfp4bf16_moe_flashinfer(monkeypatch: pytest.MonkeyPatch): monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_MXFP4_BF16", "1") can_initialize("openai/gpt-oss-20b", hf_overrides=HF_OVERRIDE_TEXT) def test_gptoss_mxfp4mxfp8_moe_flashinfer_cutlass(monkeypatch: pytest.MonkeyPatch): monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8_CUTLASS", "1") can_initialize("openai/gpt-oss-20b", hf_overrides=HF_OVERRIDE_TEXT) def test_gptoss_mxfp4mxfp8_moe_flashinfer_trtllm(monkeypatch: pytest.MonkeyPatch): monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8", "1") can_initialize("openai/gpt-oss-20b", hf_overrides=HF_OVERRIDE_TEXT) def test_gptoss_eager(monkeypatch: pytest.MonkeyPatch): can_initialize( "openai/gpt-oss-20b", hf_overrides=HF_OVERRIDE_TEXT, extra_args=["--enforce-eager"], ) ## Qwen3 Next ## def test_qwen3_next_bf16_moe_flashinfer_trtllm(monkeypatch: pytest.MonkeyPatch): can_initialize( "Qwen/Qwen3-Next-80B-A3B-Instruct", hf_overrides=HF_OVERRIDE_TEXT, extra_args=["--moe-backend=flashinfer_trtllm"], )
{ "repo_id": "vllm-project/vllm", "file_path": "tests/quantization/test_blackwell_moe.py", "license": "Apache License 2.0", "lines": 155, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/reasoning/test_glm4_moe_reasoning_parser.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import pytest from transformers import AutoTokenizer from tests.reasoning.utils import run_reasoning_extraction from vllm.reasoning import ReasoningParser, ReasoningParserManager parser_name = "glm45" start_token = "<think>" end_token = "</think>" REASONING_MODEL_NAME = "zai-org/GLM-4.5" @pytest.fixture(scope="module") def glm45_tokenizer(): return AutoTokenizer.from_pretrained(REASONING_MODEL_NAME) WITH_THINK = { "output": "<think>This is a reasoning section</think>This is the rest", "reasoning": "This is a reasoning section", "content": "This is the rest", "is_reasoning_end": True, } WITH_THINK_STREAM = { "output": "<think>This is a reasoning section</think>This is the rest", "reasoning": "This is a reasoning section", "content": "This is the rest", "is_reasoning_end": True, } WITHOUT_THINK = { "output": "This is the rest", "reasoning": None, "content": "This is the rest", "is_reasoning_end": False, } WITHOUT_THINK_STREAM = { "output": "This is the rest", "reasoning": None, "content": "This is the rest", "is_reasoning_end": False, } COMPLETE_REASONING = { "output": "<think>This is a reasoning section</think>", "reasoning": "This is a reasoning section", "content": None, "is_reasoning_end": True, } MULTILINE_REASONING = { "output": "<think>This is a reasoning\nsection</think>This is the rest\nThat", "reasoning": "This is a reasoning\nsection", "content": "This is the rest\nThat", "is_reasoning_end": True, } ONLY_OPEN_TAG = { "output": "<think>This is a reasoning section", "reasoning": None, "content": "<think>This is a reasoning section", "is_reasoning_end": False, } ONLY_OPEN_TAG_STREAM = { "output": "<think>This is a reasoning section", "reasoning": "This is a reasoning section", "content": None, "is_reasoning_end": False, } TEST_CASES = [ pytest.param( False, WITH_THINK, id="with_think", ), pytest.param( True, WITH_THINK_STREAM, id="with_think_stream", ), pytest.param( False, WITHOUT_THINK, id="without_think", ), pytest.param( True, WITHOUT_THINK_STREAM, id="without_think_stream", ), pytest.param( False, COMPLETE_REASONING, id="complete_reasoning", ), pytest.param( True, COMPLETE_REASONING, id="complete_reasoning_stream", ), pytest.param( False, MULTILINE_REASONING, id="multiline_reasoning", ), pytest.param( True, MULTILINE_REASONING, id="multiline_reasoning_stream", ), pytest.param( False, ONLY_OPEN_TAG, id="only_open_tag", ), pytest.param( True, ONLY_OPEN_TAG_STREAM, id="only_open_tag_stream", ), ] STILL_REASONING_PROMPT = """[gMASK]<sop><|system|> You are a helpful assistant.<|user|> What is the capital of France?<|assistant|> <think>The user is asking for the capital of""" DONE_REASONING_PROMPT = """[gMASK]<sop><|system|> You are a helpful assistant.<|user|> What is the capital of France?<|assistant|> <think>The user is asking for the capital of France.</think> The capital of France is Paris.""" MULTI_TURN_STILL_REASONING_PROMPT = """[gMASK]<sop><|system|> You are a helpful assistant.<|user|> What is the capital of France?<|assistant|> <think></think> The capital of France is Paris.<|user|> What about Chile?<|assistant|> <think>The user is asking for the capital of""" MULTI_TURN_DONE_REASONING_PROMPT = """[gMASK]<sop><|system|> You are a helpful assistant.<|user|> What is the capital of France?<|assistant|> <think></think> The capital of France is Paris.<|user|> What about Chile?<|assistant|> <think>The user is asking for the capital of Chile.</think> The capital of Chile is Santiago.""" REASONING_END_TEST_CASES = [ pytest.param(STILL_REASONING_PROMPT, False, id="still_reasoning"), pytest.param(DONE_REASONING_PROMPT, True, id="done_reasoning"), pytest.param( MULTI_TURN_STILL_REASONING_PROMPT, False, id="multi_turn_still_reasoning" ), pytest.param( MULTI_TURN_DONE_REASONING_PROMPT, True, id="multi_turn_done_reasoning" ), ] @pytest.mark.parametrize("streaming, param_dict", TEST_CASES) def test_reasoning( streaming: bool, param_dict: dict, glm45_tokenizer, ): output = glm45_tokenizer.tokenize(param_dict["output"]) output_tokens: list[str] = [ glm45_tokenizer.convert_tokens_to_string([token]) for token in output ] parser: ReasoningParser = ReasoningParserManager.get_reasoning_parser(parser_name)( glm45_tokenizer ) reasoning, content = run_reasoning_extraction( parser, output_tokens, streaming=streaming ) assert reasoning == param_dict["reasoning"] assert content == param_dict["content"] output_ids = glm45_tokenizer.convert_tokens_to_ids(output) is_reasoning_end = parser.is_reasoning_end(output_ids) assert is_reasoning_end == param_dict["is_reasoning_end"] @pytest.mark.parametrize("prompt, is_reasoning_end", REASONING_END_TEST_CASES) def test_is_reasoning_end_full_prompt( prompt: str, is_reasoning_end: bool, glm45_tokenizer ): parser: ReasoningParser = ReasoningParserManager.get_reasoning_parser(parser_name)( glm45_tokenizer ) tokens = glm45_tokenizer.tokenize(prompt) token_ids = glm45_tokenizer.convert_tokens_to_ids(tokens) check_is_reasoning_end = parser.is_reasoning_end(token_ids) assert check_is_reasoning_end == is_reasoning_end
{ "repo_id": "vllm-project/vllm", "file_path": "tests/reasoning/test_glm4_moe_reasoning_parser.py", "license": "Apache License 2.0", "lines": 179, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/models/multimodal/generation/test_qwen2_5_vl.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import pytest from vllm.multimodal.video import sample_frames_from_video from ....conftest import VIDEO_ASSETS models = ["Qwen/Qwen2.5-VL-3B-Instruct"] target_dtype = "bfloat16" VIDEO_PLACEHOLDER = "<|vision_start|><|video_pad|><|vision_end|>" def qwen2_5_vl_chat_template(*query): return f"<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n<|im_start|>user\n{''.join(query)}<|im_end|><|im_start|>assistant\n" # noqa: E501 VIDEO_PROMPTS = VIDEO_ASSETS.prompts( { "baby_reading": qwen2_5_vl_chat_template( VIDEO_PLACEHOLDER, "Describe this video with a short sentence ", "(no more than 20 words)", ), } ) @pytest.mark.core_model @pytest.mark.parametrize("model", models) @pytest.mark.parametrize("video_pruning_rate", [0.0, 0.75]) @pytest.mark.parametrize("num_frames", [16]) @pytest.mark.parametrize("dtype", [target_dtype]) @pytest.mark.parametrize("max_tokens", [128]) @pytest.mark.parametrize("use_bytecode_hook", [True, False]) def test_qwen2_5_vl_evs_functionality( vllm_runner, video_assets, model, video_pruning_rate: float, num_frames: int, dtype: str, max_tokens: int, use_bytecode_hook: bool, monkeypatch, ) -> None: """Test EVS (Efficient Video Sampling) functionality with different pruning rates. """ # Set the environment variable for this test monkeypatch.setenv("VLLM_USE_BYTECODE_HOOK", "1" if use_bytecode_hook else "0") # Sample frames from video assets sampled_vids = [ sample_frames_from_video(asset.np_ndarrays, num_frames) for asset in video_assets ] prompts = [VIDEO_PROMPTS[0]] videos = [sampled_vids[0]] # Initialize model with EVS configuration with vllm_runner( model, runner="generate", max_model_len=4000, dtype=dtype, limit_mm_per_prompt={"video": 1}, video_pruning_rate=video_pruning_rate, ) as vllm_model: # Generate output - this should not crash outputs = vllm_model.generate_greedy(prompts, max_tokens, videos=videos) # Basic validation that we got a response assert len(outputs) == 1 output_ids, output_text = outputs[0] # Ensure we got some output assert len(output_ids) > 0 assert len(output_text) > 0 # Ensure the output is a string assert isinstance(output_text, str) @pytest.mark.core_model @pytest.mark.parametrize("model", models) @pytest.mark.parametrize("video_pruning_rate", [0.0, 0.75]) @pytest.mark.parametrize("num_frames", [16]) @pytest.mark.parametrize("dtype", [target_dtype]) @pytest.mark.parametrize("max_tokens", [128]) @pytest.mark.parametrize("use_bytecode_hook", [True, False]) def test_qwen2_5_vl_evs_batched_videos( vllm_runner, video_assets, model, video_pruning_rate: float, num_frames: int, dtype: str, max_tokens: int, use_bytecode_hook: bool, monkeypatch, ) -> None: """Test EVS functionality with batched videos. This test validates that: 1. The model handles batched video inputs correctly with EVS 2. Both pruning configurations work with multiple videos 3. The model doesn't crash when processing multiple videos simultaneously """ # Set the environment variable for this test monkeypatch.setenv("VLLM_USE_BYTECODE_HOOK", "1" if use_bytecode_hook else "0") # Sample frames from video assets sampled_vids = [ sample_frames_from_video(asset.np_ndarrays, num_frames) for asset in video_assets ] # Test batched videos prompts = [VIDEO_PROMPTS[0], VIDEO_PROMPTS[0]] videos = [sampled_vids[0], sampled_vids[0]] # Use same video twice for testing # Initialize model with EVS configuration with vllm_runner( model, runner="generate", max_model_len=4000, max_num_seqs=2, dtype=dtype, limit_mm_per_prompt={"video": 2}, tensor_parallel_size=1, video_pruning_rate=video_pruning_rate, ) as vllm_model: # Generate output - this should not crash outputs = vllm_model.generate_greedy(prompts, max_tokens, videos=videos) # Basic validation that we got responses for both videos assert len(outputs) == 2 for output_ids, output_text in outputs: # Ensure we got some output for each video assert len(output_ids) > 0 assert len(output_text) > 0 # Ensure the output is a string assert isinstance(output_text, str)
{ "repo_id": "vllm-project/vllm", "file_path": "tests/models/multimodal/generation/test_qwen2_5_vl.py", "license": "Apache License 2.0", "lines": 123, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/multimodal/evs.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project # SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. import typing import torch def compute_retained_tokens_count( tokens_per_frame: int, num_frames: int, q: float ) -> int: """ Compute the number of retained tokens for a given video. Method ensures that we retain all the tokens from the first frame regardless of the pruning rate. Args: tokens_per_frame: The number of tokens per frame. num_frames: The total number of frames. q: The pruning rate. Returns: The number of retained tokens. """ total_tokens = tokens_per_frame * num_frames evs_num_tokens = int(total_tokens * (1 - q)) min_num_tokens = tokens_per_frame return max(min_num_tokens, evs_num_tokens) def compute_retention_mask( video_embeds: torch.Tensor, video_size_thw: torch.LongTensor | tuple[int, int, int], spatial_merge_size: int, q: float, ) -> torch.Tensor: """ Computes the retention mask for input video embeddings. Args: video_embeds (`torch.Tensor`): The input video embeddings of shape `(T * H * W // spatial_merge_size ^ 2, hidden_size)` video_size_thw (`torch.LongTensor` of shape `(3)`): The temporal, height and width of video. spatial_merge_size: Size reduction for rows & cols dimensions. q: (`float`): Pruning rate factor [0,1) Returns: `torch.Tensor`: The retention mask for the video embeddings of `(T * H * W // spatial_merge_size ^ 2)` shape. """ T, H, W = map(int, video_size_thw) # Use reshape instead of einops to avoid graph breaks video_embeds = video_embeds.reshape( T, H // spatial_merge_size, W // spatial_merge_size, video_embeds.size(-1), ) tokens_per_frame = (H // spatial_merge_size) * (W // spatial_merge_size) # Core EVS similarity = torch.nn.functional.cosine_similarity( video_embeds[1:, ...], video_embeds[:-1, ...], dim=-1 ) dissimilarity = 1 - similarity # Always ensure we include all tokens from the first frame dissimilarity = torch.cat( [255 * torch.ones_like(video_embeds[:1, :, :, 0]), dissimilarity], dim=0 ) dissimilarity_flat = dissimilarity.view(-1) order = torch.argsort(dissimilarity_flat, dim=-1, descending=True, stable=True) retain_num_tokens = compute_retained_tokens_count( tokens_per_frame=tokens_per_frame, num_frames=T, q=q ) topk_indices = order[:retain_num_tokens] retention_mask = torch.zeros_like(dissimilarity_flat, dtype=torch.bool) retention_mask[topk_indices] = True retention_mask = retention_mask.reshape(dissimilarity.size()) mask = retention_mask.view(-1) # "T H W -> (T H W)" return mask def compute_mrope_for_media( video_size_thw: torch.LongTensor, spatial_merge_size: int, tokens_per_second: float = 1.0, video_second_per_grid: float = 1.0, ) -> torch.Tensor: """ Computes the mrope for video embeddings based on the grid dimensions. Computed mrope positions match original qwen 2.5 implementation, but positions are built for media being the first element in sequence. Args: video_size_thw: Media size (num frames, rows, cols) spatial_merge_size: Size reduction for rows & cols dimensions. tokens_per_second: Number of tokens per second. video_second_per_grid: Number of seconds per video. Returns: Tensor of shape `(T * H * W, 4)` where last dimension represents mrope positions [0:3), while the last channel contains value of llm_grid_w repeated for all positions. """ llm_grid_t = video_size_thw[0] llm_grid_h = video_size_thw[1] // spatial_merge_size llm_grid_w = video_size_thw[2] // spatial_merge_size t_index = ( ( torch.arange(llm_grid_t) .view(-1, 1) .expand(-1, llm_grid_h * llm_grid_w) .mul(tokens_per_second * video_second_per_grid) ) .long() .flatten() ) h_index = ( torch.arange(llm_grid_h) .view(1, -1, 1) .expand(llm_grid_t, -1, llm_grid_w) .flatten() ) w_index = ( torch.arange(llm_grid_w) .view(1, 1, -1) .expand(llm_grid_t, llm_grid_h, -1) .flatten() ) llm_grid_w = ( torch.tensor([llm_grid_w]) .view(1, 1, 1) .expand(llm_grid_t, llm_grid_h, llm_grid_w) .flatten() ) positions = torch.stack([t_index, h_index, w_index, llm_grid_w], dim=1) return positions def recompute_mrope_positions( input_ids: torch.LongTensor, multimodal_positions: list[torch.Tensor], mrope_positions: torch.LongTensor, num_computed_tokens: int, vision_start_token_id: int, image_token_id: int, video_token_id: int, ) -> tuple[torch.LongTensor, int]: """ Update part of input mrope positions. Original mrope_positions are computed incorrectly, so once we prune media tokens we should reflect this in the mrope positions for the LLM. This method supports chunked prefill approach where multimodal_embeddings are passed to LLM in chunks, so input multimodal_embeddings may contain zero, some or even some part of all multimodal_embeddings for a given prompt. Each multimodal_positions has 4 extra channels (First 3 channels corresponds to original 3 mrope positions, last channel is the maximum width of the media repeated). Provided multimodal_positions do not reflect location of media position in sequence - they are computed like the media is in the 0-th position in the sequence. Method works as follows: it recomputes mrope_positions starting from the `num_computed_tokens` for `total_len_of_multimodal_embeddings` and then shifts all text tokens that goes after total_len_of_multimodal_embeddings. It also handles case when multimodal_embeddings is partial (e.g. one media is split into two prefill stages) Args: input_ids: (N,) All input tokens of the prompt (entire sequence). multimodal_positions: List of mrope positions for each media. mrope_positions: Existing mrope positions (4, N) for entire sequence. num_computed_tokens: A number of computed tokens so far. vision_start_token_id: Token indicating start of vision media. image_token_id: Image token id video_token_id: Video token id Returns: Tuple of (mrope_positions, mrope_position_delta). """ # Tensors positions: torch.LongTensor = typing.cast( torch.LongTensor, mrope_positions.clone() ) # (3, N) N = input_ids.numel() image_mask = input_ids.eq(image_token_id) video_mask = input_ids.eq(video_token_id) media_mask = image_mask | video_mask text_mask = ~media_mask # Early exit: no media in this chunk if len(multimodal_positions) == 0: delta = int((positions.max().item() + 1) - N) if positions.numel() else -N return positions, delta total_mm_tokens = torch.count_nonzero(media_mask) seen_mm_tokens = torch.count_nonzero(media_mask[:num_computed_tokens]) # Early exit: we've updated positions for all media tokens # (and consequently - for all remaining text tokens) if seen_mm_tokens == total_mm_tokens: delta = int((positions.max().item() + 1) - N) if positions.numel() else -N return positions, delta vision_start_indices = (input_ids == vision_start_token_id).nonzero(as_tuple=True)[ 0 ] for mm_pos in multimodal_positions: # Each mm_pos can be a complete embedding for single media # or it can be a part of a single media (due to chunked prefill) # Cases to cover # - Current prefill chunk has no vision start indexes at all # - Vision start token appeared in previous prefill round # - Regular case seen_vision_start_indices = vision_start_indices[ vision_start_indices < num_computed_tokens ] if len(seen_vision_start_indices): # If we have encountered some vision start indexes, # then we should check the condition: # | --- prefill 1 ------| ---- prefill 2 ----- | # | TTTTTTTTTSVVVVVVVVVV|VVVVVVTTTTTTTTTTTTTTTT| last_vision_start_token = seen_vision_start_indices[-1] seem_mm_tokens_before_last_vision_start = torch.count_nonzero( media_mask[:last_vision_start_token] ) in_the_middle_of_media = ( seen_mm_tokens > seem_mm_tokens_before_last_vision_start ) if in_the_middle_of_media: mm_embeddings_seen = ( seen_mm_tokens - seem_mm_tokens_before_last_vision_start ) global_mm_start = last_vision_start_token else: # We have completed previous mm_embedding part and # ready to start a new one next_vision_start_token = vision_start_indices[ vision_start_indices >= num_computed_tokens ][0] mm_embeddings_seen = 0 global_mm_start = next_vision_start_token else: # If there were no vision start indexes so far, # let's find first vision start index next_vision_start_token = vision_start_indices[ vision_start_indices >= num_computed_tokens ][0] mm_embeddings_seen = 0 global_mm_start = next_vision_start_token # Offset right after vision_start_token base = positions[-1, global_mm_start] + 1 local_start = global_mm_start + 1 + mm_embeddings_seen local_end = local_start + mm_pos.shape[1] positions[:, local_start:local_end] = mm_pos[0:3] + base # mm_pos[3, 0] is the max width of the media offset = mm_pos[3, 0] + base text_pos_sum = torch.cumsum(text_mask[local_end:].long(), dim=0) positions[:, local_end:N] = text_pos_sum + offset - 1 # Include distance to the next vision start token num_computed_tokens += mm_pos.shape[1] mrope_positions_delta = (positions.max() + 1 - N).item() return positions, mrope_positions_delta
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/multimodal/evs.py", "license": "Apache License 2.0", "lines": 247, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/model_executor/models/longcat_flash.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project # Apache License, Version 2.0: # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # # MIT License: # Permission is hereby granted, free of charge, to any person obtaining a copy # of this software and associated documentation files (the "Software"), to deal # in the Software without restriction, including without limitation the rights # to use, copy, modify, merge, publish, distribute, sublicense, and/or sell # copies of the Software, and to permit persons to whom the Software is # furnished to do so, subject to the following conditions: # # The above copyright notice and this permission notice shall be included in all # copies or substantial portions of the Software. # # THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR # IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, # FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE # AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER # LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE # SOFTWARE. """Inference-only Flash model compatible with HuggingFace weights.""" import typing from collections.abc import Callable, Iterable from itertools import islice import torch from torch import nn from transformers import PretrainedConfig from vllm.compilation.decorators import support_torch_compile from vllm.config import CacheConfig, VllmConfig from vllm.distributed import get_pp_group from vllm.logger import init_logger from vllm.model_executor.layers.activation import SiluAndMul from vllm.model_executor.layers.fused_moe import FusedMoE, ZeroExpertFusedMoE from vllm.model_executor.layers.layernorm import RMSNorm from vllm.model_executor.layers.linear import ( MergedColumnParallelLinear, ReplicatedLinear, RowParallelLinear, ) from vllm.model_executor.layers.logits_processor import LogitsProcessor from vllm.model_executor.layers.quantization import QuantizationConfig from vllm.model_executor.layers.quantization.utils.int8_utils import block_dequant from vllm.model_executor.layers.vocab_parallel_embedding import ( ParallelLMHead, VocabParallelEmbedding, ) from vllm.model_executor.model_loader.weight_utils import default_weight_loader from vllm.model_executor.models.deepseek_v2 import DeepseekV2MLAAttention from vllm.sequence import IntermediateTensors from .interfaces import SupportsLoRA, SupportsPP from .utils import ( PPMissingLayer, is_pp_missing_parameter, make_empty_intermediate_tensors_factory, make_layers, maybe_prefix, ) logger = init_logger(__name__) class FlashConfig(PretrainedConfig): """Flash model configuration.""" model_type = "longcat_flash" keys_to_ignore_at_inference = ["past_key_values"] def __init__( self, vocab_size=131072, hidden_size=4096, intermediate_size=8192, num_layers=28, num_hidden_layers=None, num_attention_heads=96, num_key_value_heads=128, ep_size=1, kv_lora_rank=512, q_lora_rank=1536, qk_rope_head_dim=64, v_head_dim=128, qk_nope_head_dim=128, num_experts_per_tok=None, norm_topk_prob=False, max_position_embeddings=8192, initializer_range=0.02, rms_norm_eps=1e-05, use_cache=True, pad_token_id=None, bos_token_id=100000, eos_token_id=100001, pretraining_tp=1, tie_word_embeddings=False, rope_parameters=None, attention_bias=False, attention_dropout=0.0, mla_scale_q_lora=False, mla_scale_kv_lora=False, dtype="bfloat16", params_dtype="bfloat16", router_dtype="float32", router_bias=False, topk_method=None, routed_scaling_factor=1.0, zero_expert_num=0, zero_expert_type=None, nextn_use_scmoe=False, **kwargs, ): super().__init__( pad_token_id=pad_token_id, bos_token_id=bos_token_id, eos_token_id=eos_token_id, tie_word_embeddings=tie_word_embeddings, dtype=dtype, params_dtype=params_dtype, router_dtype=router_dtype, topk_method=topk_method, router_bias=router_bias, nextn_use_scmoe=nextn_use_scmoe, **kwargs, ) self.vocab_size = vocab_size self.max_position_embeddings = max_position_embeddings self.hidden_size = hidden_size self.num_hidden_layers = ( num_hidden_layers if num_hidden_layers is not None else num_layers ) self.num_attention_heads = num_attention_heads self.ep_size = ep_size self.kv_lora_rank = kv_lora_rank self.q_lora_rank = q_lora_rank self.qk_rope_head_dim = qk_rope_head_dim self.v_head_dim = v_head_dim self.qk_nope_head_dim = qk_nope_head_dim self.num_experts_per_tok = num_experts_per_tok self.norm_topk_prob = norm_topk_prob # for backward compatibility if num_key_value_heads is None: num_key_value_heads = num_attention_heads self.num_key_value_heads = num_key_value_heads self.initializer_range = initializer_range self.rms_norm_eps = rms_norm_eps self.pretraining_tp = pretraining_tp self.use_cache = use_cache # Try to set `rope_scaling` if available, otherwise use `rope_parameters` rope_scaling = kwargs.pop("rope_scaling", None) rope_parameters = rope_scaling or rope_parameters or {"rope_type": "default"} rope_theta = kwargs.pop("rope_theta", 1000000.0) if "rope_theta" not in rope_parameters: rope_parameters["rope_theta"] = rope_theta self.rope_parameters = rope_parameters self.attention_bias = attention_bias self.attention_dropout = attention_dropout self.mla_scale_q_lora = mla_scale_q_lora self.mla_scale_kv_lora = mla_scale_kv_lora self.zero_expert_num = zero_expert_num self.zero_expert_type = zero_expert_type self.routed_scaling_factor = routed_scaling_factor self.hidden_act = "silu" self.intermediate_size = ( self.ffn_hidden_size if hasattr(self, "ffn_hidden_size") else intermediate_size ) if hasattr(self, "moe_intermediate_size"): self.moe_intermediate_size = self.moe_intermediate_size elif hasattr(self, "expert_ffn_hidden_size"): self.moe_intermediate_size = self.expert_ffn_hidden_size else: self.moe_intermediate_size = self.intermediate_size class FlashMLP(nn.Module): """Flash MLP layer.""" def __init__( self, hidden_size: int, intermediate_size: int, hidden_act: str, quant_config: QuantizationConfig | None = None, reduce_results: bool = True, prefix: str = "", ) -> None: super().__init__() self.gate_up_proj = MergedColumnParallelLinear( hidden_size, [intermediate_size] * 2, bias=False, quant_config=quant_config, prefix=f"{prefix}.gate_up_proj", ) self.down_proj = RowParallelLinear( intermediate_size, hidden_size, bias=False, quant_config=quant_config, reduce_results=reduce_results, prefix=f"{prefix}.down_proj", ) 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: torch.Tensor) -> torch.Tensor: if x.numel() == 0: return x gate_up, _ = self.gate_up_proj(x) x = self.act_fn(gate_up) x, _ = self.down_proj(x) return x class LongcatRouter(nn.Module): def __init__( self, config: FlashConfig, zero_expert_num: int, rounter_params_dtype: torch.dtype, prefix: str = "", ): super().__init__() self.n_routed_experts = ( config.n_routed_experts if hasattr(config, "n_routed_experts") else config.num_experts[0] ) self.n_routed_experts = self.n_routed_experts + zero_expert_num self.classifier = ReplicatedLinear( config.hidden_size, self.n_routed_experts, bias=config.router_bias, params_dtype=rounter_params_dtype, quant_config=None, prefix=f"{prefix}.classifier", ) self.e_score_correction_bias = nn.Parameter( torch.zeros((self.n_routed_experts), dtype=rounter_params_dtype) ) def forward(self, hidden_states): logits, _ = self.classifier(hidden_states) return logits class LongcatMoe(nn.Module): def __init__( self, config: FlashConfig, num_experts: int, top_k: int, hidden_size: int, intermediate_size: int, params_dtype: torch.dtype | None = None, quant_config: QuantizationConfig | None = None, prefix: str = "", enable_eplb: bool = False, ): super().__init__() self.hidden_size = hidden_size # Gate always runs at half / full precision for now. self.rounter_params_dtype = params_dtype if config.router_dtype == "float32": self.rounter_params_dtype = torch.float32 self.router = LongcatRouter( config=config, zero_expert_num=config.zero_expert_num, rounter_params_dtype=self.rounter_params_dtype, prefix=f"{prefix}.gate", ) assert config.zero_expert_num is not None assert config.zero_expert_type is not None self.experts = ZeroExpertFusedMoE( zero_expert_num=config.zero_expert_num, zero_expert_type=config.zero_expert_type, router=self.router, num_experts=num_experts, top_k=top_k, hidden_size=hidden_size, intermediate_size=intermediate_size, reduce_results=True, params_dtype=params_dtype, renormalize=False, quant_config=quant_config, prefix=f"{prefix}.experts", enable_eplb=enable_eplb, routed_scaling_factor=config.routed_scaling_factor, router_logits_dtype=self.rounter_params_dtype, ) def forward(self, hidden_states: torch.Tensor) -> torch.Tensor: num_tokens, hidden_dim = hidden_states.shape hidden_states = hidden_states.view(-1, hidden_dim) # Align to FusedMoE padded hidden size to avoid dim mismatch padded_hidden = self.experts.hidden_size if hidden_dim < padded_hidden: hidden_states_padded = torch.nn.functional.pad( hidden_states, (0, padded_hidden - hidden_dim), mode="constant", value=0.0, ) else: hidden_states_padded = hidden_states router_logits_full = self.router( hidden_states_padded.to(self.rounter_params_dtype) ) # ZeroExpertFusedMoE handles routing memoization and zero expert computation # internally. Pass full router_logits (including zero experts) so that # zero experts can be properly identified in routing. final_hidden_states = self.experts( hidden_states=hidden_states_padded, router_logits=router_logits_full, # Full logits (includes zero experts) ) # Crop back to original hidden dimension if padded earlier if padded_hidden != hidden_dim: final_hidden_states = final_hidden_states[..., :hidden_dim] return final_hidden_states.view(num_tokens, hidden_dim) class FlashDecoderLayer(nn.Module): """Flash decoder layer with dual attention and MLP structure.""" def __init__( self, vllm_config: VllmConfig, config: FlashConfig, cache_config: CacheConfig | None = None, quant_config: QuantizationConfig | None = None, prefix: str = "", enable_eplb: bool = False, ) -> None: super().__init__() self.layer_idx = int(prefix.split(sep=".")[-1]) self.hidden_size = config.hidden_size max_position_embeddings = getattr(config, "max_position_embeddings", 8192) # Dual attention structure self.self_attn = nn.ModuleList( [ DeepseekV2MLAAttention( vllm_config=vllm_config, config=config, hidden_size=self.hidden_size, num_heads=config.num_attention_heads, qk_nope_head_dim=config.qk_nope_head_dim, qk_rope_head_dim=config.qk_rope_head_dim, v_head_dim=config.v_head_dim, q_lora_rank=( config.q_lora_rank if hasattr(config, "q_lora_rank") else None ), kv_lora_rank=config.kv_lora_rank, max_position_embeddings=max_position_embeddings, cache_config=cache_config, quant_config=None if "self_attn" in getattr(config, "disable_quant_module", []) else quant_config, prefix=f"{prefix}.self_attn.{i}", ) for i in range(2) ] ) self.input_layernorm = nn.ModuleList( [RMSNorm(config.hidden_size, eps=config.rms_norm_eps) for i in range(2)] ) self.post_attention_layernorm = nn.ModuleList( [RMSNorm(config.hidden_size, eps=config.rms_norm_eps) for i in range(2)] ) # Dual MLP structure self.mlps = nn.ModuleList( [ FlashMLP( hidden_size=self.hidden_size, intermediate_size=config.intermediate_size, hidden_act=config.hidden_act, quant_config=None if "mlps" in getattr(config, "disable_quant_module", []) else quant_config, prefix=f"{prefix}.mlps.{i}", ) for i in range(2) ] ) self.mlp = LongcatMoe( config=config, num_experts=config.n_routed_experts if hasattr(config, "n_routed_experts") else config.num_experts[self.layer_idx], top_k=config.moe_topk if hasattr(config, "moe_topk") else config.num_experts_per_tok, hidden_size=config.hidden_size, intermediate_size=config.moe_intermediate_size, quant_config=quant_config, prefix=(f"{prefix}.mlp"), ) def forward( self, positions: torch.Tensor, hidden_states: torch.Tensor, residual: torch.Tensor | None, ) -> tuple[torch.Tensor, torch.Tensor]: if residual is None: residual = hidden_states hidden_states = self.input_layernorm[0](hidden_states) else: hidden_states, residual = self.input_layernorm[0](hidden_states, residual) hidden_states = self.self_attn[0]( positions=positions, hidden_states=hidden_states, llama_4_scaling=None, ) hidden_states, residual = self.post_attention_layernorm[0]( hidden_states, residual ) # moe hidden_states_copy = hidden_states.clone() moe_hidden_states = self.mlp(hidden_states_copy) # first mlp hidden_states = self.mlps[0](hidden_states) hidden_states, residual = self.input_layernorm[1](hidden_states, residual) # second_attn hidden_states = self.self_attn[1]( positions=positions, hidden_states=hidden_states, llama_4_scaling=None, ) hidden_states, residual = self.post_attention_layernorm[1]( hidden_states, residual ) # second_mlp hidden_states = self.mlps[1](hidden_states) hidden_states = hidden_states + moe_hidden_states return hidden_states, residual @support_torch_compile class FlashModel(nn.Module): """Flash model.""" def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): super().__init__() config = FlashConfig(**vllm_config.model_config.hf_config.__dict__) cache_config = vllm_config.cache_config quant_config = vllm_config.quant_config self.config = config self.vocab_size = config.vocab_size if get_pp_group().is_first_rank: self.embed_tokens = VocabParallelEmbedding( config.vocab_size, config.hidden_size, prefix=maybe_prefix(prefix, "embed_tokens"), ) else: self.embed_tokens = PPMissingLayer() self.start_layer, self.end_layer, self.layers = make_layers( config.num_hidden_layers, lambda prefix: FlashDecoderLayer( vllm_config, config, cache_config=cache_config, quant_config=quant_config, prefix=prefix, ), prefix=f"{prefix}.layers", ) if get_pp_group().is_last_rank: self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps) else: self.norm = PPMissingLayer() self.make_empty_intermediate_tensors = make_empty_intermediate_tensors_factory( ["hidden_states", "residual"], config.hidden_size ) def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor: return self.embed_tokens(input_ids) def forward( self, input_ids: torch.Tensor | None, positions: torch.Tensor, intermediate_tensors: IntermediateTensors | None = None, inputs_embeds: torch.Tensor | None = None, ) -> torch.Tensor | IntermediateTensors: if get_pp_group().is_first_rank: if inputs_embeds is not None: hidden_states = inputs_embeds else: hidden_states = self.embed_input_ids(input_ids) residual = None else: assert intermediate_tensors is not None hidden_states = intermediate_tensors["hidden_states"] residual = intermediate_tensors["residual"] for layer in islice(self.layers, self.start_layer, self.end_layer): hidden_states, residual = layer( positions, hidden_states, residual, ) if not get_pp_group().is_last_rank: return IntermediateTensors( {"hidden_states": hidden_states, "residual": residual} ) hidden_states, _ = self.norm(hidden_states, residual) return hidden_states class LongcatFlashForCausalLM(nn.Module, SupportsLoRA, SupportsPP): """Flash model for causal language modeling.""" packed_modules_mapping = { "qkv_proj": [ "q_proj", "k_proj", "v_proj", ], "gate_up_proj": [ "gate_proj", "up_proj", ], } def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): super().__init__() config = FlashConfig(**vllm_config.model_config.hf_config.__dict__) quant_config = vllm_config.quant_config self.config = config config.intermediate_size = ( config.ffn_hidden_size if hasattr(config, "ffn_hidden_size") else config.intermediate_size ) self.quant_config = quant_config self.model = FlashModel( vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model") ) if get_pp_group().is_last_rank: self.lm_head = ParallelLMHead( config.vocab_size, config.hidden_size, quant_config=quant_config, prefix=maybe_prefix(prefix, "lm_head"), ) else: self.lm_head = PPMissingLayer() self.logits_processor = LogitsProcessor(config.vocab_size) self.make_empty_intermediate_tensors = ( self.model.make_empty_intermediate_tensors ) def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor: return self.model.embed_input_ids(input_ids) def forward( self, input_ids: torch.Tensor | None, positions: torch.Tensor, intermediate_tensors: IntermediateTensors | None = None, inputs_embeds: torch.Tensor | None = None, ) -> torch.Tensor | IntermediateTensors: hidden_states = self.model( input_ids, positions, intermediate_tensors, inputs_embeds ) return hidden_states def compute_logits( self, hidden_states: torch.Tensor, ) -> torch.Tensor | None: logits = self.logits_processor(self.lm_head, hidden_states) return logits def get_expert_mapping(self) -> list[tuple[str, str, int, str]]: # Params for weights, fp8 weight scales, fp8 activation scales # (param_name, weight_name, expert_id, shard_id) return FusedMoE.make_expert_params_mapping( self, ckpt_gate_proj_name="gate_proj", ckpt_down_proj_name="down_proj", ckpt_up_proj_name="up_proj", num_experts=self.config.n_routed_experts if hasattr(self.config, "n_routed_experts") else self.config.num_experts[0], ) def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: stacked_params_mapping = [ ("fused_qkv_a_proj", "q_a_proj", 0), ("fused_qkv_a_proj", "kv_a_proj_with_mqa", 1), (".gate_up_proj", ".gate_proj", 0), (".gate_up_proj", ".up_proj", 1), ] expert_params_mapping = self.get_expert_mapping() loaded_params: set[str] = set() params_dict = dict(self.named_parameters()) for name, loaded_weight in weights: 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 if "mlp" in name and "mlps" not in name: continue name = name.replace(weight_name, param_name) # Skip loading extra bias for GPTQ models. if ( name.endswith(".bias") or name.endswith("_bias") ) and name not in params_dict: continue # Skip mtp if ".mtp." in name: continue if is_pp_missing_parameter(name, self): continue param = params_dict[name] weight_loader = param.weight_loader weight_loader(param, loaded_weight, shard_id) break else: is_expert_weight = False for mapping in expert_params_mapping: param_name, weight_name, expert_id, shard_id = mapping if weight_name not in name: continue is_expert_weight = True name_mapped = name.replace(weight_name, param_name) # Skip mtp if ".mtp." in name_mapped: continue if ( name_mapped.endswith(".bias") or name_mapped.endswith("_bias") ) and name not in params_dict: continue if is_pp_missing_parameter(name, self): continue param = params_dict[name_mapped] weight_loader = param.weight_loader weight_loader = typing.cast( Callable[..., bool], param.weight_loader ) success = weight_loader( param, loaded_weight, name_mapped, shard_id=shard_id, expert_id=expert_id, return_success=True, ) if success: name = name_mapped break else: if is_expert_weight: # We've checked that this is an expert weight # However it's not mapped locally to this rank # So we simply skip it continue # Skip loading extra bias for GPTQ models. if name.endswith(".bias") and name not in params_dict: continue # Skip loading kv_scale from ckpts towards new design. if name.endswith(".kv_scale") and name not in params_dict: continue # Skip mtp if ".mtp." in name: continue if name is None: continue if is_pp_missing_parameter(name, self): continue param = params_dict[name] weight_loader = getattr( param, "weight_loader", default_weight_loader ) weight_loader(param, loaded_weight) loaded_params.add(name) for layer_id in range(self.config.num_hidden_layers): for i in range(2): if isinstance(self.model.layers[layer_id], PPMissingLayer): continue self_attn = self.model.layers[layer_id].self_attn[i] if hasattr( self.quant_config, "weight_block_size" ) and self_attn.kv_b_proj.weight.dtype in ( torch.float8_e4m3fn, torch.float8_e4m3fnuz, ): weight_block_size = self.quant_config.weight_block_size if weight_block_size is not None: assert hasattr(self_attn.kv_b_proj, "weight_scale_inv") dtype = torch.get_default_dtype() w = block_dequant( self_attn.kv_b_proj.weight, self_attn.kv_b_proj.weight_scale_inv, weight_block_size, ).to(dtype) else: w = self_attn.kv_b_proj.weight w_kc, w_vc = w.unflatten( 0, (-1, self_attn.qk_nope_head_dim + self_attn.v_head_dim) ).split([self_attn.qk_nope_head_dim, self_attn.v_head_dim], dim=1) self_attn.w_kc = w_kc.transpose(1, 2).contiguous().transpose(1, 2) self_attn.w_vc = w_vc.contiguous().transpose(1, 2) if self.config.mla_scale_q_lora: self_attn.q_a_layernorm.weight.data *= ( self.config.hidden_size / self.config.q_lora_rank ) ** 0.5 if self.config.mla_scale_kv_lora: self_attn.kv_a_layernorm.weight.data *= ( self.config.hidden_size / self.config.kv_lora_rank ) ** 0.5 return loaded_params
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/model_executor/models/longcat_flash.py", "license": "Apache License 2.0", "lines": 696, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/model_executor/models/longcat_flash_mtp.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project # Adapted from # https://github.com/vllm-project/vllm/blob/v0.7.3/vllm/model_executor/models/deepseek_mtp.py from collections.abc import Iterable import torch import torch.nn as nn from transformers import PretrainedConfig from vllm.config import VllmConfig from vllm.model_executor.layers.layernorm import RMSNorm from vllm.model_executor.layers.linear import ReplicatedLinear from vllm.model_executor.layers.logits_processor import LogitsProcessor from vllm.model_executor.layers.quantization import QuantizationConfig from vllm.model_executor.layers.quantization.utils.int8_utils import block_dequant from vllm.model_executor.layers.vocab_parallel_embedding import ( ParallelLMHead, VocabParallelEmbedding, ) from vllm.model_executor.model_loader.weight_utils import default_weight_loader from vllm.model_executor.models.longcat_flash import FlashConfig from vllm.sequence import IntermediateTensors from .deepseek_v2 import DeepseekV2DecoderLayer from .utils import maybe_prefix class LongCatMultiTokenPredictorLayer(nn.Module): def __init__( self, config: PretrainedConfig, prefix: str, vllm_config: VllmConfig, quant_config: QuantizationConfig | None = None, ) -> None: super().__init__() self.enorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps) self.hnorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps) self.eh_proj = ReplicatedLinear( 2 * config.hidden_size, config.hidden_size, bias=False, quant_config=quant_config, prefix="eh_proj", ) self.mtp_block = DeepseekV2DecoderLayer(vllm_config, prefix) self.final_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps) def forward( self, input_ids: torch.Tensor, positions: torch.Tensor, previous_hidden_states: torch.Tensor, inputs_embeds: torch.Tensor | None = None, spec_step_index: int = 0, ) -> torch.Tensor: assert inputs_embeds is not None inputs_embeds = self.enorm(inputs_embeds) previous_hidden_states = self.hnorm(previous_hidden_states) hidden_states, _ = self.eh_proj( torch.cat([inputs_embeds, previous_hidden_states], dim=-1) ) hidden_states, residual = self.mtp_block( positions=positions, hidden_states=hidden_states, residual=None ) hidden_states, _ = self.final_layernorm(hidden_states, residual) return hidden_states class LongCatMultiTokenPredictor(nn.Module): def __init__( self, *, vllm_config: VllmConfig, quant_config: QuantizationConfig | None = None, prefix: str = "", ): super().__init__() config = FlashConfig(**vllm_config.model_config.hf_config.__dict__) vllm_config.model_config.hf_config.intermediate_size = config.intermediate_size self.mtp_start_layer_idx = config.num_hidden_layers * 2 self.num_mtp_layers = 1 self.layers = torch.nn.ModuleDict( { str(idx): LongCatMultiTokenPredictorLayer( config, prefix=f"{prefix}.layers.{idx}", vllm_config=vllm_config, quant_config=quant_config, ) for idx in range( self.mtp_start_layer_idx, self.mtp_start_layer_idx + self.num_mtp_layers, ) } ) self.embed_tokens = VocabParallelEmbedding( config.vocab_size, config.hidden_size, ) def forward( self, input_ids: torch.Tensor, positions: torch.Tensor, previous_hidden_states: torch.Tensor, inputs_embeds: torch.Tensor | None = None, spec_step_idx: int = 0, ) -> torch.Tensor: if inputs_embeds is None: inputs_embeds = self.embed_tokens(input_ids) current_step_idx = spec_step_idx % self.num_mtp_layers return self.layers[str(self.mtp_start_layer_idx + current_step_idx)]( input_ids, positions, previous_hidden_states, inputs_embeds, current_step_idx, ) class LongCatFlashMTP(nn.Module): def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): super().__init__() # LongCat MTP without MoE layers vllm_config.model_config.hf_config.n_routed_experts = None self.config = FlashConfig(**vllm_config.model_config.hf_config.__dict__) self.quant_config = ( None if "mtp" in getattr(self.config, "disable_quant_module", []) else vllm_config.quant_config ) self.model = LongCatMultiTokenPredictor( vllm_config=vllm_config, quant_config=self.quant_config, prefix=maybe_prefix(prefix, "model"), ) self.lm_head = ParallelLMHead( self.config.vocab_size, self.config.hidden_size, quant_config=self.quant_config, prefix=maybe_prefix(prefix, "lm_head"), ) self.logits_processor = LogitsProcessor(self.config.vocab_size) def forward( self, input_ids: torch.Tensor | None, positions: torch.Tensor, hidden_states: torch.Tensor, intermediate_tensors: IntermediateTensors | None = None, inputs_embeds: torch.Tensor | None = None, spec_step_idx: int = 0, ) -> torch.Tensor: hidden_states = self.model( input_ids, positions, hidden_states, inputs_embeds, spec_step_idx ) return hidden_states def compute_logits( self, hidden_states: torch.Tensor, spec_step_idx: int = 0, ) -> torch.Tensor | None: logits = self.logits_processor(self.lm_head, hidden_states) return logits def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: stacked_params_mapping = [ ("gate_up_proj", "gate_proj", 0), ("gate_up_proj", "up_proj", 1), ("fused_qkv_a_proj", "q_a_proj", 0), ("fused_qkv_a_proj", "kv_a_proj_with_mqa", 1), ] new_to_old_names_mapping = { "model.mtp.embed_tokens.weight": "model.layers.0.embed_tokens.weight", "model.mtp.layers.0.eh_proj.weight": "eh_proj.weight", "model.mtp.layers.0.eh_proj.weight_scale_inv": "eh_proj.weight_scale_inv", "model.mtp.layers.0.enorm.m.weight": "enorm.weight", "model.mtp.layers.0.hnorm.m.weight": "hnorm.weight", "model.mtp.layers.0.input_layernorm.weight": "model.layers.0.input_layernorm.weight", # noqa: E501 "model.mtp.layers.0.post_attention_layernorm.weight": "model.layers.0.post_attention_layernorm.weight", # noqa: E501 "model.mtp.layers.0.self_attn.kv_a_layernorm.weight": "model.layers.0.self_attn.kv_a_layernorm.weight", # noqa: E501 "model.mtp.layers.0.self_attn.kv_a_proj_with_mqa.weight": "model.layers.0.self_attn.kv_a_proj_with_mqa.weight", # noqa: E501 "model.mtp.layers.0.self_attn.kv_a_proj_with_mqa.weight_scale_inv": "model.layers.0.self_attn.kv_a_proj_with_mqa.weight_scale_inv", # noqa: E501 "model.mtp.layers.0.self_attn.kv_b_proj.weight": "model.layers.0.self_attn.kv_b_proj.weight", # noqa: E501 "model.mtp.layers.0.self_attn.kv_b_proj.weight_scale_inv": "model.layers.0.self_attn.kv_b_proj.weight_scale_inv", # noqa: E501 "model.mtp.layers.0.self_attn.o_proj.weight": "model.layers.0.self_attn.o_proj.weight", # noqa: E501 "model.mtp.layers.0.self_attn.o_proj.weight_scale_inv": "model.layers.0.self_attn.o_proj.weight_scale_inv", # noqa: E501 "model.mtp.layers.0.self_attn.q_a_layernorm.weight": "model.layers.0.self_attn.q_a_layernorm.weight", # noqa: E501 "model.mtp.layers.0.self_attn.q_a_proj.weight": "model.layers.0.self_attn.q_a_proj.weight", # noqa: E501 "model.mtp.layers.0.self_attn.q_a_proj.weight_scale_inv": "model.layers.0.self_attn.q_a_proj.weight_scale_inv", # noqa: E501 "model.mtp.layers.0.self_attn.q_b_proj.weight": "model.layers.0.self_attn.q_b_proj.weight", # noqa: E501 "model.mtp.layers.0.self_attn.q_b_proj.weight_scale_inv": "model.layers.0.self_attn.q_b_proj.weight_scale_inv", # noqa: E501 "model.mtp.layers.0.transformer_layer.mlp.down_proj.weight": "model.layers.0.mlp.down_proj.weight", # noqa: E501 "model.mtp.layers.0.transformer_layer.mlp.down_proj.weight_scale_inv": "model.layers.0.mlp.down_proj.weight_scale_inv", # noqa: E501 "model.mtp.layers.0.transformer_layer.mlp.gate_proj.weight": "model.layers.0.mlp.gate_proj.weight", # noqa: E501 "model.mtp.layers.0.transformer_layer.mlp.gate_proj.weight_scale_inv": "model.layers.0.mlp.gate_proj.weight_scale_inv", # noqa: E501 "model.mtp.layers.0.transformer_layer.mlp.up_proj.weight": "model.layers.0.mlp.up_proj.weight", # noqa: E501 "model.mtp.layers.0.transformer_layer.mlp.up_proj.weight_scale_inv": "model.layers.0.mlp.up_proj.weight_scale_inv", # noqa: E501 "model.mtp.norm.weight": "final_layernorm.weight", } params_dict = dict(self.named_parameters()) loaded_params: set[str] = set() for name, loaded_weight in weights: if "rotary_emb.inv_freq" in name: continue spec_layer = self.get_spec_layer_idx_from_weight_name(self.config, name) if spec_layer is None: continue name = self._rewrite_spec_layer_name( spec_layer, name, new_to_old_names_mapping ) for param_name, weight_name, shard_id in stacked_params_mapping: # Skip non-stacked layers and experts (experts handled below). if weight_name not in name: continue # We have mlp.experts[0].gate_proj in the checkpoint. # Since we handle the experts below in expert_params_mapping, # we need to skip here BEFORE we update the name, otherwise # name will be updated to mlp.experts[0].gate_up_proj, which # will then be updated below in expert_params_mapping # for mlp.experts[0].gate_gate_up_proj, which breaks load. if ("mlp.experts." in name) and name not in params_dict: continue name = name.replace(weight_name, param_name) # QKV fusion is optional, fall back to normal # weight loading if it's not enabled if (param_name == "fused_qkv_a_proj") and name not in params_dict: continue # 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 # According to DeepSeek-V3 Technical Report, MTP modules # shares embedding layer. We only load the first weights. if ( spec_layer != self.model.mtp_start_layer_idx and ".layers" not in name ): continue param = params_dict[name] weight_loader = getattr(param, "weight_loader", default_weight_loader) weight_loader(param, loaded_weight) loaded_params.add(name) spec_layer_id = self.config.num_hidden_layers * 2 self_attn = self.model.layers[str(spec_layer_id)].mtp_block.self_attn if hasattr( self.quant_config, "weight_block_size" ) and self_attn.kv_b_proj.weight.dtype in ( torch.float8_e4m3fn, torch.float8_e4m3fnuz, ): weight_block_size = self.quant_config.weight_block_size if weight_block_size is not None: dtype = torch.get_default_dtype() w = block_dequant( self_attn.kv_b_proj.weight, self_attn.kv_b_proj.weight_scale_inv, weight_block_size, ).to(dtype) else: w = self_attn.kv_b_proj.weight else: w = self_attn.kv_b_proj.weight w_kc, w_vc = w.unflatten( 0, (-1, self_attn.qk_nope_head_dim + self_attn.v_head_dim) ).split([self_attn.qk_nope_head_dim, self_attn.v_head_dim], dim=1) self_attn.w_kc = w_kc.transpose(1, 2).contiguous().transpose(1, 2) self_attn.w_vc = w_vc.contiguous().transpose(1, 2) if self.config.mla_scale_q_lora: self_attn.q_a_layernorm.weight.data *= ( self.config.hidden_size / self.config.q_lora_rank ) ** 0.5 if self.config.mla_scale_kv_lora: self_attn.kv_a_layernorm.weight.data *= ( self.config.hidden_size / self.config.kv_lora_rank ) ** 0.5 return loaded_params def _rewrite_spec_layer_name( self, spec_layer: int, name: str, new_to_old_names_mapping: dict ) -> str: """ Rewrite the weight name to match the format of the original model. Add .mtp_block for modules in transformer layer block for spec layer and rename shared layer weights to be top level. """ if name in new_to_old_names_mapping: name = new_to_old_names_mapping[name] spec_layer_weight_names = [ "embed_tokens", "enorm", "hnorm", "eh_proj", "shared_head", ] if ( name.startswith("enorm") or name.startswith("hnorm") or name.startswith("eh_proj") or name.startswith("final_layernorm") ): name = "model.layers." + str(spec_layer) + "." + name shared_weight_names = ["embed_tokens"] spec_layer_weight = False shared_weight = False for weight_name in spec_layer_weight_names: if weight_name in name: spec_layer_weight = True if weight_name in shared_weight_names: shared_weight = True break if not spec_layer_weight: # treat rest weights as weights for transformer layer block name = name.replace( "model.layers.0.", f"model.layers.{spec_layer}.mtp_block." ) elif shared_weight: # treat shared weights as top level weights name = name.replace("model.layers.0.", "model.") return name def get_spec_layer_idx_from_weight_name( self, config: PretrainedConfig, weight_name: str ) -> int | None: if "model.mtp" in weight_name: return config.num_hidden_layers * 2 return None
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/model_executor/models/longcat_flash_mtp.py", "license": "Apache License 2.0", "lines": 321, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/distributed/device_communicators/mnnvl_compat.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from typing import Any import torch.distributed as dist from flashinfer.comm.mnnvl import CommBackend as CommBackend from vllm.utils.flashinfer import has_flashinfer_all2all assert has_flashinfer_all2all(), "Flashinfer alltoallv module cannot be found" class CustomCommunicator(CommBackend): def __init__(self, group): self._group = group def Get_rank(self) -> int: return self._group.rank() def Get_size(self) -> int: return self._group.size() def allgather(self, data: int): gathered = [None] * self.Get_size() dist.all_gather_object(gathered, data, group=self._group) return gathered # NOTE(rob): CommBackend is an abstract class, and bcast/barrier # are unimplemented on vLLM side. If we need to utilize these # methods in the future, can create a concrete implementation. def bcast(self, data: Any, root: int) -> Any: raise NotImplementedError def barrier(self) -> None: raise NotImplementedError def Split(self, color: int, key: int) -> "CustomCommunicator": return self
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/distributed/device_communicators/mnnvl_compat.py", "license": "Apache License 2.0", "lines": 27, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/config/device.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from dataclasses import field from typing import Any, Literal import torch from pydantic import ConfigDict, SkipValidation from vllm.config.utils import config from vllm.utils.hashing import safe_hash Device = Literal["auto", "cuda", "cpu", "tpu", "xpu"] @config(config=ConfigDict(arbitrary_types_allowed=True)) class DeviceConfig: """Configuration for the device to use for vLLM execution.""" device: SkipValidation[Device | torch.device | None] = "auto" """Device type for vLLM execution. This parameter is deprecated and will be removed in a future release. It will now be set automatically based on the current platform.""" device_type: str = field(init=False) """Device type from the current platform. This is set in `__post_init__`.""" def compute_hash(self) -> str: """ WARNING: Whenever a new field is added to this config, ensure that it is included in the factors list if it affects the computation graph. Provide a hash that uniquely identifies all the configs that affect the structure of the computation graph from input ids/embeddings to the final hidden states, excluding anything before input ids/embeddings and after the final hidden states. """ # no factors to consider. # the device/platform information will be summarized # by torch/vllm automatically. factors: list[Any] = [] hash_str = safe_hash(str(factors).encode(), usedforsecurity=False).hexdigest() return hash_str def __post_init__(self): if self.device == "auto": # Automated device type detection from vllm.platforms import current_platform self.device_type = current_platform.device_type if not self.device_type: raise RuntimeError( "Failed to infer device type, please set " "the environment variable `VLLM_LOGGING_LEVEL=DEBUG` " "to turn on verbose logging to help debug the issue." ) else: # Device type is assigned explicitly if isinstance(self.device, str): self.device_type = self.device elif isinstance(self.device, torch.device): self.device_type = self.device.type # Some device types require processing inputs on CPU if self.device_type in ["tpu"]: self.device = None else: # Set device with device type self.device = torch.device(self.device_type)
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/config/device.py", "license": "Apache License 2.0", "lines": 61, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/config/observability.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from functools import cached_property from typing import Any, Literal, cast from packaging.version import parse from pydantic import Field, field_validator, model_validator from vllm import version from vllm.config.utils import config from vllm.utils.hashing import safe_hash DetailedTraceModules = Literal["model", "worker", "all"] @config class ObservabilityConfig: """Configuration for observability - metrics and tracing.""" show_hidden_metrics_for_version: str | None = None """Enable deprecated Prometheus metrics that have been hidden since the specified version. For example, if a previously deprecated metric has been hidden since the v0.7.0 release, you use `--show-hidden-metrics-for-version=0.7` as a temporary escape hatch while you migrate to new metrics. The metric is likely to be removed completely in an upcoming release.""" @cached_property def show_hidden_metrics(self) -> bool: """Check if the hidden metrics should be shown.""" if self.show_hidden_metrics_for_version is None: return False return version._prev_minor_version_was(self.show_hidden_metrics_for_version) otlp_traces_endpoint: str | None = None """Target URL to which OpenTelemetry traces will be sent.""" collect_detailed_traces: list[DetailedTraceModules] | None = None """It makes sense to set this only if `--otlp-traces-endpoint` is set. If set, it will collect detailed traces for the specified modules. This involves use of possibly costly and or blocking operations and hence might have a performance impact. Note that collecting detailed timing information for each request can be expensive.""" kv_cache_metrics: bool = False """Enable KV cache residency metrics (lifetime, idle time, reuse gaps). Uses sampling to minimize overhead. Requires log stats to be enabled (i.e., --disable-log-stats not set).""" kv_cache_metrics_sample: float = Field(default=0.01, gt=0, le=1) """Sampling rate for KV cache metrics (0.0, 1.0]. Default 0.01 = 1% of blocks.""" cudagraph_metrics: bool = False """Enable CUDA graph metrics (number of padded/unpadded tokens, runtime cudagraph dispatch modes, and their observed frequencies at every logging interval).""" enable_layerwise_nvtx_tracing: bool = False """Enable layerwise NVTX tracing. This traces the execution of each layer or module in the model and attach informations such as input/output shapes to nvtx range markers. Noted that this doesn't work with CUDA graphs enabled.""" enable_mfu_metrics: bool = False """Enable Model FLOPs Utilization (MFU) metrics.""" enable_mm_processor_stats: bool = False """Enable collection of timing statistics for multimodal processor operations. This is for internal use only (e.g., benchmarks) and is not exposed as a CLI argument.""" enable_logging_iteration_details: bool = False """Enable detailed logging of iteration details. If set, vllm EngineCore will log iteration details This includes number of context/generation requests and tokens and the elapsed cpu time for the iteration.""" @cached_property def collect_model_forward_time(self) -> bool: """Whether to collect model forward time for the request.""" return self.collect_detailed_traces is not None and ( "model" in self.collect_detailed_traces or "all" in self.collect_detailed_traces ) @cached_property def collect_model_execute_time(self) -> bool: """Whether to collect model execute time for the request.""" return self.collect_detailed_traces is not None and ( "worker" in self.collect_detailed_traces or "all" in self.collect_detailed_traces ) def compute_hash(self) -> str: """ WARNING: Whenever a new field is added to this config, ensure that it is included in the factors list if it affects the computation graph. Provide a hash that uniquely identifies all the configs that affect the structure of the computation graph from input ids/embeddings to the final hidden states, excluding anything before input ids/embeddings and after the final hidden states. """ # no factors to consider. # this config will not affect the computation graph. factors: list[Any] = [] hash_str = safe_hash(str(factors).encode(), usedforsecurity=False).hexdigest() return hash_str @field_validator("show_hidden_metrics_for_version") @classmethod def _validate_show_hidden_metrics_for_version(cls, value: str | None) -> str | None: if value is not None: # Raises an exception if the string is not a valid version. parse(value) return value @field_validator("otlp_traces_endpoint") @classmethod def _validate_otlp_traces_endpoint(cls, value: str | None) -> str | None: if value is not None: from vllm.tracing import is_tracing_available, otel_import_error_traceback if not is_tracing_available(): raise ValueError( "OpenTelemetry is not available. Unable to configure " "'otlp_traces_endpoint'. Ensure OpenTelemetry packages are " f"installed. Original error:\n{otel_import_error_traceback}" ) return value @field_validator("collect_detailed_traces") @classmethod def _validate_collect_detailed_traces( cls, value: list[DetailedTraceModules] | None ) -> list[DetailedTraceModules] | None: """Handle the legacy case where users might provide a comma-separated string instead of a list of strings.""" if value is not None and len(value) == 1 and "," in value[0]: value = cast(list[DetailedTraceModules], value[0].split(",")) return value @model_validator(mode="after") def _validate_tracing_config(self): if self.collect_detailed_traces and not self.otlp_traces_endpoint: raise ValueError( "collect_detailed_traces requires `--otlp-traces-endpoint` to be set." ) return self
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/config/observability.py", "license": "Apache License 2.0", "lines": 125, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/config/speech_to_text.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from vllm.config.utils import config @config class SpeechToTextConfig: """Configuration for speech-to-text models.""" sample_rate: float = 16_000 """Sample rate (Hz) to resample input audio to. Most speech models expect 16kHz audio input. The input audio will be automatically resampled to this rate before processing.""" max_audio_clip_s: int | None = 30 """Maximum duration in seconds for a single audio clip without chunking. Audio longer than this will be split into smaller chunks if `allow_audio_chunking` evaluates to True, otherwise it will be rejected. `None` means audio duration can be unlimited and won't be chunked.""" overlap_chunk_second: int = 1 """Overlap duration in seconds between consecutive audio chunks when splitting long audio. This helps maintain context across chunk boundaries and improves transcription quality at split points.""" min_energy_split_window_size: int | None = 1600 """Window size in samples for finding low-energy (quiet) regions to split audio chunks. The algorithm looks for the quietest moment within this window to minimize cutting through speech. Default 1600 samples ≈ 100ms at 16kHz. If None, no chunking will be done.""" @property def allow_audio_chunking(self) -> bool: return ( self.min_energy_split_window_size is not None and self.max_audio_clip_s is not None )
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/config/speech_to_text.py", "license": "Apache License 2.0", "lines": 30, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/reasoning/test_base_thinking_reasoning_parser.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import pytest from transformers import AutoTokenizer from tests.reasoning.utils import run_reasoning_extraction from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionRequest from vllm.reasoning.basic_parsers import BaseThinkingReasoningParser # Create a concrete test implementation of BaseThinkingReasoningParser class TestThinkingReasoningParser(BaseThinkingReasoningParser): """Test implementation of BaseThinkingReasoningParser.""" @property def start_token(self) -> str: return "<test:think>" @property def end_token(self) -> str: return "</test:think>" class TestThinkingReasoningParserAlt(BaseThinkingReasoningParser): """Alternative test implementation with different tokens.""" @property def start_token(self) -> str: return "<alt:start>" @property def end_token(self) -> str: return "<alt:end>" # Use a test model REASONING_MODEL_NAME = "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B" @pytest.fixture(scope="module") def test_tokenizer(): tokenizer = AutoTokenizer.from_pretrained(REASONING_MODEL_NAME) # Add custom test tokens test_tokens = ["<test:think>", "</test:think>", "<alt:start>", "<alt:end>"] existing_tokens = set(tokenizer.get_vocab().keys()) new_tokens = [token for token in test_tokens if token not in existing_tokens] if new_tokens: tokenizer.add_tokens(new_tokens) return tokenizer class TestBaseThinkingReasoningParserInit: """ Test initialization and basic properties of BaseThinkingReasoningParser. """ def test_successful_initialization(self, test_tokenizer): """Test successful initialization with valid tokens.""" parser = TestThinkingReasoningParser(test_tokenizer) assert parser.start_token == "<test:think>" assert parser.end_token == "</test:think>" assert parser.start_token_id is not None assert parser.end_token_id is not None def test_initialization_with_missing_tokenizer(self): """Test that initialization fails without tokenizer.""" with pytest.raises(ValueError, match="model tokenizer must be passed"): TestThinkingReasoningParser(None) def test_initialization_with_missing_tokens(self, test_tokenizer): """Test that initialization fails when tokens are not in vocabulary.""" # Create a parser with tokens not in vocabulary class MissingTokenParser(BaseThinkingReasoningParser): @property def start_token(self) -> str: return "<missing:start>" @property def end_token(self) -> str: return "<missing:end>" with pytest.raises( RuntimeError, match="could not locate think start/end tokens" ): MissingTokenParser(test_tokenizer) def test_initialization_with_empty_tokens(self, test_tokenizer): """Test that initialization fails with empty token strings.""" class EmptyTokenParser(BaseThinkingReasoningParser): @property def start_token(self) -> str: return "" @property def end_token(self) -> str: return "" with pytest.raises( ValueError, match="start_token and end_token must be defined" ): EmptyTokenParser(test_tokenizer) class TestBaseThinkingReasoningParserMethods: """Test the methods of BaseThinkingReasoningParser.""" def test_is_reasoning_end(self, test_tokenizer): """Test the is_reasoning_end method.""" parser = TestThinkingReasoningParser(test_tokenizer) end_token_id = parser.end_token_id start_token_id = parser.start_token_id # Test with end token present assert parser.is_reasoning_end([1, 2, end_token_id, 4]) is True # Test without end token assert parser.is_reasoning_end([1, 2, 3, 4]) is False # Test with empty list assert parser.is_reasoning_end([]) is False # Test with interleaved thinking assert parser.is_reasoning_end([1, start_token_id, 2, end_token_id]) is True assert parser.is_reasoning_end([1, start_token_id, 2, 3]) is False assert ( parser.is_reasoning_end( [1, start_token_id, 2, end_token_id, 2, 2, start_token_id] ) is False ) def test_is_reasoning_end_streaming(self, test_tokenizer): """Test the is_reasoning_end_streaming method.""" parser = TestThinkingReasoningParser(test_tokenizer) end_token_id = parser.end_token_id start_token_id = parser.start_token_id assert ( parser.is_reasoning_end_streaming([1, 2, end_token_id], [end_token_id]) is True ) assert parser.is_reasoning_end_streaming([1, 2, 3, 4], [4]) is False assert parser.is_reasoning_end_streaming([], []) is False assert ( parser.is_reasoning_end_streaming( [1, start_token_id, 2, end_token_id], [end_token_id] ) is True ) assert ( parser.is_reasoning_end_streaming([1, start_token_id, 2, 3], [3]) is False ) assert ( parser.is_reasoning_end_streaming( [1, start_token_id, 2, end_token_id, 2, start_token_id, 2], [2], ) is False ) assert ( parser.is_reasoning_end_streaming( [1, start_token_id, 2, end_token_id, 2, 2], [2] ) is False ) def test_count_reasoning_tokens(self, test_tokenizer): """Count tokens between start/end markers.""" parser = TestThinkingReasoningParser(test_tokenizer) start = parser.start_token_id end = parser.end_token_id token_ids = [0, start, 11, 12, end, 99] assert parser.count_reasoning_tokens(token_ids) == 2 def test_count_reasoning_tokens_nested(self, test_tokenizer): """Ensure nested thinking spans count all inner tokens safely.""" parser = TestThinkingReasoningParser(test_tokenizer) s = parser.start_token_id e = parser.end_token_id token_ids = [s, 1, s, 2, e, 3, e] # Tokens 1,2,3 are inside reasoning (depth>0) => 3 tokens assert parser.count_reasoning_tokens(token_ids) == 3 def test_extract_content_ids(self, test_tokenizer): """Test the extract_content_ids method.""" parser = TestThinkingReasoningParser(test_tokenizer) end_token_id = parser.end_token_id # Test with end token in the middle input_ids = [1, 2, end_token_id, 4, 5] content_ids = parser.extract_content_ids(input_ids) assert content_ids == [4, 5] # Test with end token at the end input_ids = [1, 2, 3, end_token_id] content_ids = parser.extract_content_ids(input_ids) assert content_ids == [] # Test without end token input_ids = [1, 2, 3, 4] content_ids = parser.extract_content_ids(input_ids) assert content_ids == [] # Test with end token as last element (should not extract) input_ids = [1, 2, 3, end_token_id] content_ids = parser.extract_content_ids(input_ids) assert content_ids == [] class TestBaseThinkingReasoningParserExtraction: """Test reasoning content extraction methods.""" def test_extract_reasoning_with_both_tokens(self, test_tokenizer): """Test extraction when both start and end tokens are present.""" parser = TestThinkingReasoningParser(test_tokenizer) request = ChatCompletionRequest(messages=[], model="test-model") model_output = "<test:think>This is reasoning</test:think>This is content" reasoning, content = parser.extract_reasoning(model_output, request) assert reasoning == "This is reasoning" assert content == "This is content" def test_extract_reasoning_only_end_token(self, test_tokenizer): """Test extraction when only end token is present.""" parser = TestThinkingReasoningParser(test_tokenizer) request = ChatCompletionRequest(messages=[], model="test-model") model_output = "This is reasoning</test:think>This is content" reasoning, content = parser.extract_reasoning(model_output, request) assert reasoning == "This is reasoning" assert content == "This is content" def test_extract_reasoning_no_end_token(self, test_tokenizer): """Test extraction when no end token is present.""" parser = TestThinkingReasoningParser(test_tokenizer) request = ChatCompletionRequest(messages=[], model="test-model") model_output = "This is just content" reasoning, content = parser.extract_reasoning(model_output, request) assert reasoning == "This is just content" assert content is None def test_extract_reasoning_empty_output(self, test_tokenizer): """Test extraction with empty output.""" parser = TestThinkingReasoningParser(test_tokenizer) request = ChatCompletionRequest(messages=[], model="test-model") model_output = "" reasoning, content = parser.extract_reasoning(model_output, request) assert reasoning == "" assert content is None def test_extract_reasoning_only_tokens(self, test_tokenizer): """Test extraction with only tokens and no content.""" parser = TestThinkingReasoningParser(test_tokenizer) request = ChatCompletionRequest(messages=[], model="test-model") model_output = "<test:think></test:think>" reasoning, content = parser.extract_reasoning(model_output, request) assert reasoning == "" assert content is None class TestBaseThinkingReasoningParserStreaming: """Test streaming functionality of BaseThinkingReasoningParser.""" @pytest.mark.parametrize("streaming", [True, False]) def test_simple_reasoning_extraction(self, test_tokenizer, streaming): """ Test basic reasoning extraction in both streaming and non-streaming modes. """ parser = TestThinkingReasoningParser(test_tokenizer) model_output = [ "<test:think>", "Some ", "reasoning ", "content", "</test:think>", "Final ", "answer", ] reasoning, content = run_reasoning_extraction( parser, model_output, streaming=streaming ) assert reasoning == "Some reasoning content" assert content == "Final answer" def test_streaming_with_incremental_deltas(self, test_tokenizer): """Test streaming processing with small incremental deltas.""" parser = TestThinkingReasoningParser(test_tokenizer) deltas = [ "<test:think>", "Some ", "reasoning ", "content", "</test:think>", "Final ", "answer", ] reasoning, content = run_reasoning_extraction(parser, deltas, streaming=True) assert reasoning == "Some reasoning content" assert content == "Final answer" def test_streaming_with_start_token(self, test_tokenizer): """Test streaming with start token included.""" parser = TestThinkingReasoningParser(test_tokenizer) deltas = [ "<test:think>", "Some ", "reasoning", "</test:think>", "Answer", ] reasoning, content = run_reasoning_extraction(parser, deltas, streaming=True) assert reasoning == "Some reasoning" assert content == "Answer" def test_streaming_no_end_token(self, test_tokenizer): """Test streaming when no end token is encountered.""" parser = TestThinkingReasoningParser(test_tokenizer) deltas = [ "<test:think>", "Some ", "reasoning ", "without ", "end", ] reasoning, content = run_reasoning_extraction(parser, deltas, streaming=True) assert reasoning == "Some reasoning without end" assert content is None def test_streaming_only_end_token(self, test_tokenizer): """Test streaming when only end token appears.""" parser = TestThinkingReasoningParser(test_tokenizer) deltas = [ "<test:think>", "Reasoning ", "content", "</test:think>", "Final", ] reasoning, content = run_reasoning_extraction(parser, deltas, streaming=True) assert reasoning == "Reasoning content" assert content == "Final" class TestBaseThinkingReasoningParserMultipleImplementations: """ Test that multiple implementations of BaseThinkingReasoningParser work correctly. """ def test_different_token_implementations(self, test_tokenizer): """ Test that different implementations with different tokens work independently. """ parser1 = TestThinkingReasoningParser(test_tokenizer) parser2 = TestThinkingReasoningParserAlt(test_tokenizer) # Test parser1 model_output1 = "Reasoning1</test:think>Content1" reasoning1, content1 = run_reasoning_extraction(parser1, [model_output1]) assert reasoning1 == "Reasoning1" assert content1 == "Content1" # Test parser2 model_output2 = "Reasoning2<alt:end>Content2" reasoning2, content2 = run_reasoning_extraction(parser2, [model_output2]) assert reasoning2 == "Reasoning2" assert content2 == "Content2" # Verify tokens are different assert parser1.start_token != parser2.start_token assert parser1.end_token != parser2.end_token assert parser1.start_token_id != parser2.start_token_id assert parser1.end_token_id != parser2.end_token_id class TestBaseThinkingReasoningParserEdgeCases: """Test edge cases and error conditions.""" def test_multiple_end_tokens(self, test_tokenizer): """Test behavior with multiple end tokens.""" parser = TestThinkingReasoningParser(test_tokenizer) model_output = "First</test:think>Middle</test:think>Last" reasoning, content = run_reasoning_extraction(parser, [model_output]) # Should stop at first end token assert reasoning == "First" assert content == "Middle</test:think>Last" def test_nested_tokens(self, test_tokenizer): """Test behavior with nested-like token patterns.""" parser = TestThinkingReasoningParser(test_tokenizer) model_output = "<test:think>Outer<test:think>Inner</test:think>Content" reasoning, content = run_reasoning_extraction(parser, [model_output]) # Should process normally, start from first start token assert reasoning == "Outer<test:think>Inner" assert content == "Content" def test_malformed_tokens(self, test_tokenizer): """Test behavior with malformed token-like strings.""" parser = TestThinkingReasoningParser(test_tokenizer) model_output = "<test:thinking>Not a real token</test:thinking>Content" reasoning, content = run_reasoning_extraction(parser, [model_output]) # Should treat as regular content since tokens don't match exactly assert reasoning == ("<test:thinking>Not a real token</test:thinking>Content") assert content is None
{ "repo_id": "vllm-project/vllm", "file_path": "tests/reasoning/test_base_thinking_reasoning_parser.py", "license": "Apache License 2.0", "lines": 341, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/reasoning/test_seedoss_reasoning_parser.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from typing import Any, cast import pytest from transformers import AutoTokenizer from tests.reasoning.utils import run_reasoning_extraction from vllm.reasoning import ReasoningParser, ReasoningParserManager parser_name = "seed_oss" start_token = "<seed:think>" end_token = "</seed:think>" # Use a test model that contains our custom tokens REASONING_MODEL_NAME = "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B" @pytest.fixture(scope="module") def seedoss_tokenizer(): tokenizer = AutoTokenizer.from_pretrained(REASONING_MODEL_NAME) # Add custom SeedOSS tokens if they don't exist if start_token not in tokenizer.get_vocab(): tokenizer.add_tokens([start_token, end_token]) return tokenizer SIMPLE_REASONING: dict[str, Any] = { "output": "This is a reasoning section</seed:think>This is the rest", "reasoning": "This is a reasoning section", "content": "This is the rest", "is_reasoning_end": True, } COMPLETE_REASONING: dict[str, Any] = { "output": "This is a reasoning section</seed:think>", "reasoning": "This is a reasoning section", "content": None, "is_reasoning_end": True, } NO_CONTENT: dict[str, Any] = { "output": "This is content", "reasoning": "This is content", "content": None, "is_reasoning_end": False, } NO_REASONING_STREAMING: dict[str, Any] = { "output": "This is a reasoning section", "reasoning": "This is a reasoning section", "content": None, "is_reasoning_end": False, } MULTIPLE_LINES: dict[str, Any] = { "output": "This\nThat</seed:think>This is the rest\nThat", "reasoning": "This\nThat", "content": "This is the rest\nThat", "is_reasoning_end": True, } WITH_START_TOKEN: dict[str, Any] = { "output": ("<seed:think>This is a reasoning section</seed:think>This is the rest"), "reasoning": "This is a reasoning section", "content": "This is the rest", "is_reasoning_end": True, } ONLY_END_TOKEN: dict[str, Any] = { "output": "Some reasoning</seed:think>This is the rest", "reasoning": "Some reasoning", "content": "This is the rest", "is_reasoning_end": True, } NO_TOKENS: dict[str, Any] = { "output": "This is just content without any reasoning tokens", "reasoning": "This is just content without any reasoning tokens", "content": None, "is_reasoning_end": False, } def test_seedoss_reasoning_parser_creation(seedoss_tokenizer): """Test that the SeedOSS reasoning parser can be created and registered.""" parser_cls = ReasoningParserManager.get_reasoning_parser(parser_name) parser = parser_cls(seedoss_tokenizer) assert isinstance(parser, ReasoningParser) assert parser.start_token == start_token assert parser.end_token == end_token @pytest.mark.parametrize("streaming", [True, False]) def test_simple_reasoning(seedoss_tokenizer, streaming): """Test basic reasoning extraction with both tokens.""" parser_cls = ReasoningParserManager.get_reasoning_parser(parser_name) parser = parser_cls(seedoss_tokenizer) reasoning, content = run_reasoning_extraction( parser, [cast(str, SIMPLE_REASONING["output"])], streaming=streaming ) assert reasoning == SIMPLE_REASONING["reasoning"] assert content == SIMPLE_REASONING["content"] @pytest.mark.parametrize("streaming", [True, False]) def test_complete_reasoning(seedoss_tokenizer, streaming): """Test reasoning extraction when there's no content after reasoning.""" parser_cls = ReasoningParserManager.get_reasoning_parser(parser_name) parser = parser_cls(seedoss_tokenizer) reasoning, content = run_reasoning_extraction( parser, [cast(str, COMPLETE_REASONING["output"])], streaming=streaming ) assert reasoning == COMPLETE_REASONING["reasoning"] assert content == COMPLETE_REASONING["content"] @pytest.mark.parametrize("streaming", [True, False]) def test_no_content(seedoss_tokenizer, streaming): """Test when there's no end token - everything is reasoning content.""" parser_cls = ReasoningParserManager.get_reasoning_parser(parser_name) parser = parser_cls(seedoss_tokenizer) reasoning, content = run_reasoning_extraction( parser, [cast(str, NO_CONTENT["output"])], streaming=streaming ) assert reasoning == NO_CONTENT["reasoning"] assert content == NO_CONTENT["content"] @pytest.mark.parametrize("streaming", [True, False]) def test_multiple_lines(seedoss_tokenizer, streaming): """Test reasoning extraction with multiline content.""" parser_cls = ReasoningParserManager.get_reasoning_parser(parser_name) parser = parser_cls(seedoss_tokenizer) reasoning, content = run_reasoning_extraction( parser, [cast(str, MULTIPLE_LINES["output"])], streaming=streaming ) assert reasoning == MULTIPLE_LINES["reasoning"] assert content == MULTIPLE_LINES["content"] @pytest.mark.parametrize("streaming", [True, False]) def test_with_start_token(seedoss_tokenizer, streaming): """Test reasoning extraction with both start and end tokens.""" parser_cls = ReasoningParserManager.get_reasoning_parser(parser_name) parser = parser_cls(seedoss_tokenizer) reasoning, content = run_reasoning_extraction( parser, [cast(str, WITH_START_TOKEN["output"])], streaming=streaming ) assert reasoning == WITH_START_TOKEN["reasoning"] assert content == WITH_START_TOKEN["content"] @pytest.mark.parametrize("streaming", [True, False]) def test_only_end_token(seedoss_tokenizer, streaming): """ Test reasoning extraction with only end token (SeedOSS typical behavior). """ parser_cls = ReasoningParserManager.get_reasoning_parser(parser_name) parser = parser_cls(seedoss_tokenizer) reasoning, content = run_reasoning_extraction( parser, [cast(str, ONLY_END_TOKEN["output"])], streaming=streaming ) assert reasoning == ONLY_END_TOKEN["reasoning"] assert content == ONLY_END_TOKEN["content"] @pytest.mark.parametrize("streaming", [True, False]) def test_no_tokens(seedoss_tokenizer, streaming): """Test when there are no reasoning tokens at all.""" parser_cls = ReasoningParserManager.get_reasoning_parser(parser_name) parser = parser_cls(seedoss_tokenizer) reasoning, content = run_reasoning_extraction( parser, [cast(str, NO_TOKENS["output"])], streaming=streaming ) assert reasoning == NO_TOKENS["reasoning"] assert content == NO_TOKENS["content"] def test_is_reasoning_end(seedoss_tokenizer): """Test the is_reasoning_end method.""" parser_cls = ReasoningParserManager.get_reasoning_parser(parser_name) parser = parser_cls(seedoss_tokenizer) # Test with end token present end_token_id = parser.end_token_id assert parser.is_reasoning_end([1, 2, end_token_id, 4]) is True # Test without end token assert parser.is_reasoning_end([1, 2, 3, 4]) is False def test_extract_content_ids(seedoss_tokenizer): """Test the extract_content_ids method.""" parser_cls = ReasoningParserManager.get_reasoning_parser(parser_name) parser = parser_cls(seedoss_tokenizer) end_token_id = parser.end_token_id # Test with end token in the middle input_ids = [1, 2, end_token_id, 4, 5] content_ids = parser.extract_content_ids(input_ids) assert content_ids == [4, 5] # Test with end token at the end input_ids = [1, 2, 3, end_token_id] content_ids = parser.extract_content_ids(input_ids) assert content_ids == [] # Test without end token input_ids = [1, 2, 3, 4] content_ids = parser.extract_content_ids(input_ids) assert content_ids == [] def test_streaming_delta_processing(seedoss_tokenizer): """Test streaming processing with small deltas.""" parser_cls = ReasoningParserManager.get_reasoning_parser(parser_name) parser = parser_cls(seedoss_tokenizer) # Test streaming with incremental tokens deltas = ["Some ", "reasoning ", "content", "</seed:think>", "Final ", "answer"] reasoning, content = run_reasoning_extraction(parser, deltas, streaming=True) assert reasoning == "Some reasoning content" assert content == "Final answer"
{ "repo_id": "vllm-project/vllm", "file_path": "tests/reasoning/test_seedoss_reasoning_parser.py", "license": "Apache License 2.0", "lines": 182, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/reasoning/basic_parsers.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from abc import abstractmethod from collections.abc import Iterable, Sequence from itertools import islice from typing import TYPE_CHECKING, Any from vllm.entrypoints.openai.engine.protocol import DeltaMessage from vllm.reasoning.abs_reasoning_parsers import ReasoningParser from vllm.tokenizers import TokenizerLike if TYPE_CHECKING: from vllm.entrypoints.openai.chat_completion.protocol import ( ChatCompletionRequest, ) from vllm.entrypoints.openai.responses.protocol import ( ResponsesRequest, ) else: ChatCompletionRequest = Any ResponsesRequest = Any class BaseThinkingReasoningParser(ReasoningParser): """ Base class for reasoning parsers that use thinking tokens. This class provides common functionality for parsers that use start and end tokens to delimit reasoning content ( e.g., <think>...</think>, <seed:think>...</seed:think>). Subclasses must implement the start and end tokens via abstract properties. """ @property @abstractmethod def start_token(self) -> str: """The token that starts reasoning content.""" raise NotImplementedError @property @abstractmethod def end_token(self) -> str: """The token that ends reasoning content.""" raise NotImplementedError def __init__(self, tokenizer: TokenizerLike, *args, **kwargs): super().__init__(tokenizer, *args, **kwargs) if not self.model_tokenizer: raise ValueError( "The model tokenizer must be passed to the ReasoningParser " "constructor during construction." ) if not self.start_token or not self.end_token: raise ValueError("start_token and end_token must be defined in subclasses") self.start_token_id = self.vocab.get(self.start_token) self.end_token_id = self.vocab.get(self.end_token) if self.start_token_id is None or self.end_token_id is None: raise RuntimeError( f"{self.__class__.__name__} reasoning parser could not locate " "think start/end tokens in the tokenizer!" ) def is_reasoning_end(self, input_ids: Sequence[int]) -> bool: start_token_id = self.start_token_id end_token_id = self.end_token_id for i in range(len(input_ids) - 1, -1, -1): if input_ids[i] == start_token_id: return False if input_ids[i] == end_token_id: return True return False def is_reasoning_end_streaming( self, input_ids: Sequence[int], delta_ids: Iterable[int] ) -> bool: end_token_id = self.end_token_id return end_token_id in delta_ids def extract_content_ids(self, input_ids: list[int]) -> list[int]: """ Extract the content after the end tokens """ if self.end_token_id not in islice(input_ids, 0, max(0, len(input_ids) - 1)): return [] else: return input_ids[input_ids.index(self.end_token_id) + 1 :] def extract_reasoning_streaming( self, previous_text: str, current_text: str, delta_text: str, previous_token_ids: Sequence[int], current_token_ids: Sequence[int], delta_token_ids: Sequence[int], ) -> DeltaMessage | None: """ Extract reasoning content from a delta message. Handles streaming output where previous + delta = current. Uses token IDs for faster processing. """ # Skip single special tokens if len(delta_token_ids) == 1 and ( delta_token_ids[0] in [self.start_token_id, self.end_token_id] ): return None # Check if start token is present in previous or delta. # Keep compatibility with models that don't generate start tokens. if self.start_token_id in previous_token_ids: if self.end_token_id in delta_token_ids: # start token in previous, end token in delta, # extract reasoning content end_index = delta_text.find(self.end_token) reasoning = delta_text[:end_index] content = delta_text[end_index + len(self.end_token) :] return DeltaMessage( reasoning=reasoning, content=content if content else None ) elif self.end_token_id in previous_token_ids: # start token in previous, end token in previous, # reasoning content continues return DeltaMessage(content=delta_text) else: # start token in previous, no end token in previous or delta, # reasoning content continues return DeltaMessage(reasoning=delta_text) elif self.start_token_id in delta_token_ids: if self.end_token_id in delta_token_ids: # start token in delta, end token in delta, # extract reasoning content start_index = delta_text.find(self.start_token) end_index = delta_text.find(self.end_token) reasoning = delta_text[start_index + len(self.start_token) : end_index] content = delta_text[end_index + len(self.end_token) :] return DeltaMessage( reasoning=reasoning, content=content if content else None ) else: # start token in delta, no end token in delta, # reasoning content continues return DeltaMessage(reasoning=delta_text) else: # not find thinking start token return DeltaMessage(content=delta_text) def extract_reasoning( self, model_output: str, request: ChatCompletionRequest | ResponsesRequest ) -> tuple[str | None, str | None]: """ Extract reasoning content from the model output. This is the base implementation that works for most models. Subclasses can override this method for specific behavior. """ # Check if the start token is present in the model output, remove it # if it is present. model_output_parts = model_output.partition(self.start_token) model_output = ( model_output_parts[2] if model_output_parts[1] else model_output_parts[0] ) # For models that may not generate start token, # assume the reasoning content is always at the start. if self.end_token not in model_output: return model_output, None else: reasoning, _, content = model_output.partition(self.end_token) # If generation stops right after end-of-think, return null content final_content = content or None return reasoning, final_content def count_reasoning_tokens(self, token_ids: Sequence[int]) -> int: """Count tokens that fall within start/end thinking markers. Uses a depth counter so nested spans are handled safely and stray end tokens do not drive the counter negative. """ count = 0 depth = 0 for token_id in token_ids: if token_id == self.start_token_id: depth += 1 continue if token_id == self.end_token_id: if depth > 0: depth -= 1 continue if depth > 0: count += 1 return count
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/reasoning/basic_parsers.py", "license": "Apache License 2.0", "lines": 174, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/reasoning/seedoss_reasoning_parser.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from vllm.reasoning.basic_parsers import BaseThinkingReasoningParser class SeedOSSReasoningParser(BaseThinkingReasoningParser): """ Reasoning parser for SeedOSS model. The SeedOSS model uses <seed:think>...</seed:think> tokens to denote reasoning content text. This parser extracts the reasoning content from the model output. Similar to DeepSeek R1, it supports cases where the model doesn't generate the start token. """ @property def start_token(self) -> str: """The token that starts reasoning content.""" return "<seed:think>" @property def end_token(self) -> str: """The token that ends reasoning content.""" return "</seed:think>"
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/reasoning/seedoss_reasoning_parser.py", "license": "Apache License 2.0", "lines": 20, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:benchmarks/kernels/benchmark_cutlass_moe_fp8.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """ Benchmark the performance of the cutlass_moe_fp8 kernel vs the triton_moe kernel. Both kernels take in fp8 quantized weights and 16-bit activations, but use different quantization strategies and backends. """ import torch import vllm.model_executor.layers.fused_moe.modular_kernel as mk from tests.kernels.moe.utils import make_dummy_moe_config from vllm import _custom_ops as ops from vllm.model_executor.layers.fused_moe.activation import MoEActivation from vllm.model_executor.layers.fused_moe.config import fp8_w8a8_moe_quant_config from vllm.model_executor.layers.fused_moe.cutlass_moe import CutlassExpertsFp8 from vllm.model_executor.layers.fused_moe.fused_moe import fused_experts, fused_topk from vllm.model_executor.layers.fused_moe.prepare_finalize import ( MoEPrepareAndFinalizeNoEP, ) from vllm.platforms import current_platform from vllm.utils.argparse_utils import FlexibleArgumentParser from vllm.v1.worker.workspace import init_workspace_manager # Weight shapes for different models: [num_experts, topk, hidden_size, # intermediate_size] WEIGHT_SHAPES_MOE = { "mixtral-8x7b": [ [8, 2, 4096, 14336], ], "deepseek-v2": [ [160, 6, 5120, 12288], ], "custom-small": [ [8, 2, 2048, 7168], ], "glm45-fp8": [ [128, 8, 4096, 1408], ], "Llama-4-Maverick-17B-128E-Instruct-FP8": [ [128, 1, 5120, 8192], ], } DEFAULT_MODELS = [ "mixtral-8x7b", ] DEFAULT_BATCH_SIZES = [4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048] DEFAULT_TP_SIZES = [1] PER_ACT_TOKEN_OPTS = [False, True] PER_OUT_CH_OPTS = [False, True] FP8_DTYPE = current_platform.fp8_dtype() def bench_run( results: list, model: str, num_experts: int, topk: int, per_act_token: bool, per_out_ch: bool, mkn: tuple[int, int, int], ): init_workspace_manager(torch.cuda.current_device()) (m, k, n) = mkn dtype = torch.half device = "cuda" # Create input activations a = torch.randn((m, k), device=device, dtype=dtype) / 10 # Create weights w1 = torch.randn((num_experts, 2 * n, k), device=device, dtype=dtype) / 10 w2 = torch.randn((num_experts, k, n), device=device, dtype=dtype) / 10 # Create FP8 quantized weights and scales for both kernels w1_fp8q = torch.empty((num_experts, 2 * n, k), device=device, dtype=FP8_DTYPE) w2_fp8q = torch.empty((num_experts, k, n), device=device, dtype=FP8_DTYPE) # Create scales based on quantization strategy if per_out_ch: # Per-channel quantization w1_scale = torch.empty( (num_experts, 2 * n, 1), device=device, dtype=torch.float32 ) w2_scale = torch.empty((num_experts, k, 1), device=device, dtype=torch.float32) else: # Per-tensor quantization w1_scale = torch.empty((num_experts, 1, 1), device=device, dtype=torch.float32) w2_scale = torch.empty((num_experts, 1, 1), device=device, dtype=torch.float32) # Quantize weights for expert in range(num_experts): if per_out_ch: # Per-channel quantization - not yet implemented properly # For now, fall back to per-tensor quantization w1_fp8q[expert], w1_scale_temp = ops.scaled_fp8_quant(w1[expert]) w2_fp8q[expert], w2_scale_temp = ops.scaled_fp8_quant(w2[expert]) # Expand scalar scales to the expected per-channel shape w1_scale[expert] = w1_scale_temp.expand(2 * n, 1) w2_scale[expert] = w2_scale_temp.expand(k, 1) else: # Per-tensor quantization w1_fp8q[expert], w1_scale_temp = ops.scaled_fp8_quant(w1[expert]) w2_fp8q[expert], w2_scale_temp = ops.scaled_fp8_quant(w2[expert]) # Store scalar scales in [1, 1] tensors w1_scale[expert, 0, 0] = w1_scale_temp w2_scale[expert, 0, 0] = w2_scale_temp # Prepare weights for CUTLASS (no transpose needed) w1_fp8q_cutlass = w1_fp8q # Keep original [E, 2N, K] w2_fp8q_cutlass = w2_fp8q # Keep original [E, K, N] # Create router scores and get topk score = torch.randn((m, num_experts), device=device, dtype=dtype) topk_weights, topk_ids, _ = fused_topk(a, score, topk, renormalize=False) # WORKAROUND: CUTLASS MoE FP8 has issues with per-token quantization # Force per-tensor quantization for all cases to match working e2e setup a1_scale = torch.full((), 1e-2, device=device, dtype=torch.float32) a2_scale = torch.full((), 1e-2, device=device, dtype=torch.float32) # Force per-tensor quantization for all cases per_act_token = False # Pre-create quantization config to avoid creating it inside CUDA graph quant_config = fp8_w8a8_moe_quant_config( w1_scale=w1_scale, w2_scale=w2_scale, a1_scale=a1_scale, a2_scale=a2_scale, per_act_token_quant=per_act_token, per_out_ch_quant=per_out_ch, ) fn = mk.FusedMoEModularKernel( MoEPrepareAndFinalizeNoEP(), CutlassExpertsFp8( moe_config=make_dummy_moe_config( num_experts=num_experts, hidden_dim=k, intermediate_size_per_partition=n, in_dtype=a.dtype, ), quant_config=quant_config, ), ) # Create CUDA graphs for CUTLASS (match benchmark_moe.py pattern exactly) cutlass_stream = torch.cuda.Stream() cutlass_graph = torch.cuda.CUDAGraph() with torch.cuda.graph(cutlass_graph, stream=cutlass_stream): # Capture 10 invocations like benchmark_moe.py for _ in range(10): fn( a, w1_fp8q_cutlass, w2_fp8q_cutlass, topk_weights, topk_ids, activation=MoEActivation.SILU, global_num_experts=num_experts, ) torch.cuda.synchronize() # Create CUDA graphs for Triton (match benchmark_moe.py pattern exactly) triton_stream = torch.cuda.Stream() triton_graph = torch.cuda.CUDAGraph() with torch.cuda.graph(triton_graph, stream=triton_stream): # Capture 10 invocations like benchmark_moe.py for _ in range(10): fused_experts( a, w1_fp8q, w2_fp8q, topk_weights, topk_ids, quant_config=quant_config, ) torch.cuda.synchronize() def bench_cuda_graph(graph, num_warmup=5, num_iters=100): """Benchmark CUDA graph using events like benchmark_moe.py""" # Warmup for _ in range(num_warmup): graph.replay() torch.cuda.synchronize() # Timing start_event = torch.Event(enable_timing=True) end_event = torch.Event(enable_timing=True) latencies = [] for _ in range(num_iters): torch.cuda.synchronize() start_event.record() graph.replay() end_event.record() end_event.synchronize() latencies.append(start_event.elapsed_time(end_event)) # Divide by 10 since graph contains 10 calls return sum(latencies) / (num_iters * 10) # Benchmark parameters num_warmup = 5 num_iters = 100 # Benchmark only CUDA graphs (more reliable and faster) # Benchmark Triton MoE with CUDA graphs triton_graph_time = bench_cuda_graph( triton_graph, num_warmup=num_warmup, num_iters=num_iters ) # Benchmark CUTLASS MoE with CUDA graphs cutlass_graph_time = bench_cuda_graph( cutlass_graph, num_warmup=num_warmup, num_iters=num_iters ) # Convert ms to us and return results triton_time_us = triton_graph_time * 1000 cutlass_time_us = cutlass_graph_time * 1000 return { "batch_size": m, "triton_time_us": triton_time_us, "cutlass_time_us": cutlass_time_us, } def main(args): # Initialize workspace manager (required for CUTLASS MoE kernels) device = torch.device("cuda:0") init_workspace_manager(device) print("Benchmarking models:") for i, model in enumerate(args.models): print(f"[{i}] {model}") all_results = [] for model in args.models: for tp in args.tp_sizes: for layer in WEIGHT_SHAPES_MOE[model]: num_experts = layer[0] topk = layer[1] size_k = layer[2] size_n = layer[3] // tp if len(args.limit_k) > 0 and size_k not in args.limit_k: continue if len(args.limit_n) > 0 and size_n not in args.limit_n: continue for per_act_token in args.per_act_token_opts: for per_out_ch in args.per_out_ch_opts: print( f"\n=== {model}, experts={num_experts}, topk={topk}," f"per_act={per_act_token}, per_out_ch={per_out_ch} ===" ) config_results = [] for size_m in args.batch_sizes: mkn = (size_m, size_k, size_n) result = bench_run( [], # Not used anymore model, num_experts, topk, per_act_token, per_out_ch, mkn, ) if result: config_results.append(result) # Print results table for this configuration if config_results: print( f"\n{'Batch Size':<12}" f"{'Triton (us)':<15}" f"{'CUTLASS (us)':<15}" ) print("-" * 45) for result in config_results: print( f"{result['batch_size']:<12}" f"{result['triton_time_us']:<15.2f}" f"{result['cutlass_time_us']:<15.2f}" ) all_results.extend(config_results) print(f"\nTotal benchmarks completed: {len(all_results)}") if __name__ == "__main__": parser = FlexibleArgumentParser( description="""Benchmark CUTLASS FP8 MOE vs Triton FP8 FUSED MOE across specified models/shapes/batches Example usage: python benchmark_cutlass_moe_fp8.py \ --model "Llama-4-Maverick-17B-128E-Instruct-FP8" \ --tp-sizes 8 \ --batch-size 2 4 8 \ --per-act-token-opts false \ --per-out-ch-opts false """ ) parser.add_argument( "--models", nargs="+", type=str, default=DEFAULT_MODELS, choices=WEIGHT_SHAPES_MOE.keys(), ) parser.add_argument("--tp-sizes", nargs="+", type=int, default=DEFAULT_TP_SIZES) parser.add_argument( "--batch-sizes", nargs="+", type=int, default=DEFAULT_BATCH_SIZES ) parser.add_argument("--limit-k", nargs="+", type=int, default=[]) parser.add_argument("--limit-n", nargs="+", type=int, default=[]) parser.add_argument( "--per-act-token-opts", nargs="+", type=lambda x: x.lower() == "true", default=[False, True], help="Per-activation token quantization options (true/false)", ) parser.add_argument( "--per-out-ch-opts", nargs="+", type=lambda x: x.lower() == "true", default=[False, True], help="Per-output channel quantization options (true/false)", ) args = parser.parse_args() main(args)
{ "repo_id": "vllm-project/vllm", "file_path": "benchmarks/kernels/benchmark_cutlass_moe_fp8.py", "license": "Apache License 2.0", "lines": 297, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/v1/worker/test_worker_memory_snapshot.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import multiprocessing as mp import os import tempfile from multiprocessing.queues import Queue from unittest.mock import patch import pytest import torch from vllm.config import set_current_vllm_config from vllm.engine.arg_utils import EngineArgs from vllm.utils.mem_utils import MemorySnapshot from vllm.v1.worker.gpu_worker import Worker, init_worker_distributed_environment # Global queue to track operation order across processes _QUEUE: Queue | None = None def track_operation(operation: str, rank: int): """Track when an operation happens and its rank.""" if _QUEUE is not None: _QUEUE.put((operation, rank)) def make_operation_tracker(operation_name: str, original_func): """Create a mock function that tracks when an operation is called. Args: operation_name: Name to use when tracking this operation original_func: The original function to wrap Returns: A wrapper function that tracks the operation and calls the original """ def wrapper(*args, **kwargs): rank = int(os.environ.get("RANK", "-1")) track_operation(operation_name, rank) return original_func(*args, **kwargs) return wrapper def worker_process( rank: int, world_size: int, distributed_init_method: str, queue: Queue, error_queue: Queue, ): """Worker process that initializes a GPU worker with proper tracking.""" global _QUEUE _QUEUE = queue try: # Set environment variables os.environ["RANK"] = str(rank) os.environ["LOCAL_RANK"] = str(rank) os.environ["WORLD_SIZE"] = str(world_size) # Create vLLM config with small model vllm_config = EngineArgs( model="facebook/opt-125m", tensor_parallel_size=2, load_format="dummy" ).create_engine_config() # Create worker worker = Worker( vllm_config=vllm_config, local_rank=rank, rank=rank, distributed_init_method=distributed_init_method, ) # Get original functions before patching original_init_worker = init_worker_distributed_environment original_memory_snapshot_init = MemorySnapshot.__init__ original_all_reduce = torch.distributed.all_reduce # Apply minimal patches to track operation order init_patch = patch( "vllm.v1.worker.gpu_worker.init_worker_distributed_environment", side_effect=make_operation_tracker( "init_distributed", original_init_worker ), ) memory_patch = patch.object( MemorySnapshot, "__init__", make_operation_tracker("memory_snapshot", original_memory_snapshot_init), ) all_reduce_patch = patch( "torch.distributed.all_reduce", side_effect=make_operation_tracker("nccl_all_reduce", original_all_reduce), ) with ( init_patch, memory_patch, all_reduce_patch, set_current_vllm_config(vllm_config), ): # Initialize device (this is where we test the order) worker.init_device() # Load model to ensure everything works worker.load_model() # Signal success queue.put(("success", rank)) except Exception as e: error_queue.put((rank, str(e), type(e).__name__)) raise @pytest.mark.skipif( torch.cuda.device_count() < 2, reason="Need at least 2 GPUs for tensor parallelism" ) def test_init_distributed_is_called_before_memory_snapshot(): """Test that distributed env is setup before memory snapshot. This test makes sure during worker initialization, the initial memory snapshot is taken after distributed env is setup to include all the buffers allocated by distributed env. """ world_size = 2 # Create a temporary file for distributed init with tempfile.NamedTemporaryFile(delete=False) as f: distributed_init_method = f"file://{f.name}" # Create queues for inter-process communication ctx = mp.get_context("spawn") operation_queue = ctx.Queue() error_queue = ctx.Queue() # Start worker processes processes = [] for rank in range(world_size): p = ctx.Process( target=worker_process, args=( rank, world_size, distributed_init_method, operation_queue, error_queue, ), ) p.start() processes.append(p) # Wait for all processes to complete for p in processes: p.join(timeout=60) # 60 second timeout # Check for errors errors = [] while not error_queue.empty(): rank, error_msg, error_type = error_queue.get() errors.append(f"Rank {rank}: {error_type}: {error_msg}") if errors: pytest.fail("Worker processes failed:\n" + "\n".join(errors)) # Collect all operations from the queue operations = [] while not operation_queue.empty(): operations.append(operation_queue.get()) # Verify we got operations from both ranks print(f"Collected operations: {operations}") # Check operations for each rank for rank in range(world_size): rank_ops = [op for op, r in operations if r == rank] print(f"\nRank {rank} operations: {rank_ops}") # Raises ValueError if the operation is not found init_distributed = rank_ops.index("init_distributed") nccl_all_reduce = rank_ops.index("nccl_all_reduce") memory_snapshot = rank_ops.index("memory_snapshot") # Verify order: init_distributed should happen before memory_snapshot assert init_distributed < nccl_all_reduce < memory_snapshot, ( f"Rank {rank}: init_distributed (index {init_distributed}) " f"must happen before nccl_all_reduce (index {nccl_all_reduce}) " f"and memory_snapshot (index {memory_snapshot})" ) # Clean up os.unlink(distributed_init_method.replace("file://", ""))
{ "repo_id": "vllm-project/vllm", "file_path": "tests/v1/worker/test_worker_memory_snapshot.py", "license": "Apache License 2.0", "lines": 157, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/distributed/test_nccl_symm_mem_allreduce.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import random import typing import pytest import torch import torch.distributed as dist import torch.multiprocessing as mp import vllm.envs as envs from tests.utils import ensure_current_vllm_config from vllm.distributed import cleanup_dist_env_and_memory from vllm.distributed.device_communicators.cuda_communicator import CudaCommunicator from vllm.distributed.device_communicators.pynccl import register_nccl_symmetric_ops from vllm.distributed.device_communicators.pynccl_allocator import ( get_nccl_mem_pool, is_symmetric_memory_enabled, ) from vllm.distributed.parallel_state import ( get_tp_group, init_distributed_environment, initialize_model_parallel, ) from vllm.platforms import current_platform from vllm.utils.system_utils import update_environment_variables torch.manual_seed(42) random.seed(44) test_size_elements = 4 * 1024 * 1024 def nccl_symm_mem_allreduce_worker(local_rank: int, world_size: int): monkeypatch = pytest.MonkeyPatch() with monkeypatch.context() as m: m.delenv("CUDA_VISIBLE_DEVICES", raising=False) dtype = torch.bfloat16 device = torch.device(f"cuda:{local_rank}") torch.cuda.set_device(device) torch.set_default_device(device) torch.set_default_dtype(dtype) update_environment_variables( { "RANK": str(local_rank), "LOCAL_RANK": str(local_rank), "WORLD_SIZE": str(world_size), "MASTER_ADDR": "localhost", "MASTER_PORT": "12345", } ) init_distributed_environment() with ensure_current_vllm_config(): initialize_model_parallel(tensor_model_parallel_size=world_size) cuda_communicator = typing.cast( CudaCommunicator, get_tp_group().device_communicator ) pynccl_comm = cuda_communicator.pynccl_comm if get_nccl_mem_pool() is None: pytest.skip( "NCCL allocator compilation failed (probably missing NCCL headers)." ) if not is_symmetric_memory_enabled(): pytest.skip("NCCL symmetric memory allreduce is disabled.") register_nccl_symmetric_ops(pynccl_comm) input = torch.randint(1, 23, (test_size_elements,), dtype=dtype, device=device) input_clone = input.clone() output = torch.ops.vllm.all_reduce_symmetric_with_copy(input) assert output is not None group = get_tp_group().device_group dist.all_reduce(input_clone, group=group) torch.testing.assert_close(output, input_clone, atol=2.5, rtol=0.1) @pytest.mark.skipif( not current_platform.is_cuda(), reason="NCCLSymmMemAllreduce is only available for CUDA platforms.", ) @pytest.mark.parametrize("world_size", [2]) @pytest.mark.skipif(envs.VLLM_TARGET_DEVICE not in ["cuda"], reason="Only test on CUDA") def test_nccl_symm_mem_allreduce(monkeypatch: pytest.MonkeyPatch, world_size): if world_size > torch.cuda.device_count(): pytest.skip("Not enough GPUs to run the test.") # Enable SymmMemCommunicator monkeypatch.setenv("VLLM_USE_NCCL_SYMM_MEM", "1") monkeypatch.setenv("NCCL_NVLS_ENABLE", "1") monkeypatch.setenv("NCCL_CUMEM_ENABLE", "1") mp.spawn(nccl_symm_mem_allreduce_worker, args=(world_size,), nprocs=world_size) cleanup_dist_env_and_memory()
{ "repo_id": "vllm-project/vllm", "file_path": "tests/distributed/test_nccl_symm_mem_allreduce.py", "license": "Apache License 2.0", "lines": 81, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/distributed/device_communicators/pynccl_allocator.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import atexit import contextlib import tempfile from typing import Any import torch from packaging import version from torch.cuda.memory import CUDAPluggableAllocator from torch.utils.cpp_extension import load_inline from vllm import envs from vllm.distributed.device_communicators.pynccl import PyNcclCommunicator from vllm.logger import init_logger from vllm.platforms import current_platform from vllm.utils.nccl import find_nccl_include_paths logger = init_logger(__name__) nccl_allocator_source = """ #include <nccl.h> extern "C" { void* nccl_alloc_plug(size_t size, int device, void* stream) { void* ptr; ncclResult_t err = ncclMemAlloc(&ptr, size); return ptr; } void nccl_free_plug(void* ptr, size_t size, int device, void* stream) { ncclResult_t err = ncclMemFree(ptr); } } """ _allocator = None _allocator_wrapper = None _mem_pool = None _registered_base_addrs = set() _graph_pool_id = None _nccl_allocator_failed_to_compile = False _cached_pool_snapshot = None def is_symmetric_memory_enabled(): global _nccl_allocator_failed_to_compile return envs.VLLM_USE_NCCL_SYMM_MEM and not _nccl_allocator_failed_to_compile def is_symmetric_memory_tensor(tensor: torch.Tensor): if not is_symmetric_memory_enabled() or _cached_pool_snapshot is None: return False for segment in _cached_pool_snapshot: for block in segment["blocks"]: if block["address"] == tensor.untyped_storage().data_ptr(): return True return False def set_graph_pool_id(graph_pool_id: Any) -> None: global _graph_pool_id _graph_pool_id = graph_pool_id def compile_nccl_allocator(): global _allocator, _allocator_wrapper, _nccl_allocator_failed_to_compile if not current_platform.is_cuda(): _nccl_allocator_failed_to_compile = True return try: out_dir = tempfile.gettempdir() nccl_allocator_libname = "nccl_allocator" nccl_include_paths = find_nccl_include_paths() load_inline( name=nccl_allocator_libname, cpp_sources=nccl_allocator_source, with_cuda=True, extra_ldflags=["-lnccl"], verbose=envs.VLLM_LOGGING_LEVEL == "DEBUG", is_python_module=False, build_directory=out_dir, extra_include_paths=nccl_include_paths, ) _allocator_wrapper = CUDAPluggableAllocator( f"{out_dir}/{nccl_allocator_libname}.so", "nccl_alloc_plug", "nccl_free_plug", ) _allocator = _allocator_wrapper.allocator() except Exception as e: _nccl_allocator_failed_to_compile = True logger.warning( "Failed to compile NCCL memory allocator. " "Symmetric memory will be disabled. " "This is expected if NCCL headers are not available. " "optionally set VLLM_NCCL_INCLUDE_PATH to point to a directory " "containing the NCCL header. " "Error: %s", str(e), ) def get_nccl_mem_pool(): global _mem_pool, _nccl_allocator_failed_to_compile if _mem_pool is None and not _nccl_allocator_failed_to_compile: compile_nccl_allocator() if _allocator is not None: _mem_pool = torch.cuda.MemPool(_allocator) return _mem_pool def _cleanup_nccl_mem_pool(): global _mem_pool _mem_pool = None def _cleanup_nccl_allocator_wrapper(): global _allocator_wrapper _allocator_wrapper = None atexit.register(_cleanup_nccl_mem_pool) atexit.register(_cleanup_nccl_allocator_wrapper) class nccl_symm_mem_context: def __init__( self, pynccl_comm: PyNcclCommunicator, disabled: bool = False, ): self.disabled = ( disabled or not is_symmetric_memory_enabled() or pynccl_comm.world_size == 1 or not current_platform.is_cuda() or get_nccl_mem_pool() is None or version.parse(torch.__version__) < version.parse("2.8.0.a0") ) if self.disabled: self.pynccl_comm: PyNcclCommunicator | None = None self._mem_pool_ctx: contextlib.AbstractContextManager[Any] = ( contextlib.nullcontext() ) self.is_graph_capture = None self.device = None else: self.pynccl_comm = pynccl_comm self._mem_pool_ctx = torch.cuda.use_mem_pool(get_nccl_mem_pool()) self.is_graph_capture = torch.cuda.is_current_stream_capturing() self.device = torch.cuda.current_device() def __enter__(self): if self.disabled: return self assert self.pynccl_comm is not None, ( "Symmetric memory requires pynccl to be initialized" ) assert self.pynccl_comm.nccl_version >= 22703, ( "NCCL version 2.27.3 or higher is required for NCCL symmetric memory" ) if self.is_graph_capture: assert _graph_pool_id is not None, ( "graph_pool_id is not set under graph capture" ) # Pause graph memory pool to use symmetric memory with cuda graph torch._C._cuda_endAllocateToPool(self.device, _graph_pool_id) self._mem_pool_ctx.__enter__() return self def __exit__(self, exc_type, exc_val, exc_tb): if self.disabled: return global _cached_pool_snapshot global _registered_base_addrs self._mem_pool_ctx.__exit__(exc_type, exc_val, exc_tb) _pool = get_nccl_mem_pool() assert _pool is not None _cached_pool_snapshot = _pool.snapshot() assert self.pynccl_comm is not None for segment in _cached_pool_snapshot: if segment["address"] not in _registered_base_addrs: self.pynccl_comm.register_comm_window_raw( segment["address"], segment["total_size"] ) _registered_base_addrs.add(segment["address"]) if self.is_graph_capture: torch._C._cuda_beginAllocateCurrentThreadToPool(self.device, _graph_pool_id)
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/distributed/device_communicators/pynccl_allocator.py", "license": "Apache License 2.0", "lines": 162, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/v1/kv_connector/unit/test_kv_connector_lifecyle.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from vllm.distributed.kv_transfer.kv_connector.v1.example_connector import ( # noqa: E501 ExampleConnectorMetadata, ) from vllm.distributed.kv_transfer.kv_transfer_state import ( ensure_kv_transfer_initialized, get_kv_transfer_group, ) from vllm.v1.core.sched.output import CachedRequestData, SchedulerOutput from vllm.v1.worker.kv_connector_model_runner_mixin import KVConnectorModelRunnerMixin # Importing utils registers TestExampleConnector with the factory from .utils import create_vllm_config def _make_empty_scheduler_output(): return SchedulerOutput( scheduled_new_reqs=[], scheduled_cached_reqs=CachedRequestData.make_empty(), num_scheduled_tokens={}, total_num_scheduled_tokens=0, scheduled_spec_decode_tokens={}, scheduled_encoder_inputs={}, num_common_prefix_blocks=[], finished_req_ids=set(), free_encoder_mm_hashes=[], kv_connector_metadata=ExampleConnectorMetadata(), ) def test_kv_connector_mixin_clears_metadata(): vllm_config = create_vllm_config() vllm_config.kv_transfer_config.kv_connector = "TestExampleConnector" vllm_config.kv_transfer_config.kv_role = "kv_both" vllm_config.kv_transfer_config.kv_connector_extra_config["name"] = "unit" # Initialize the global connector instance ensure_kv_transfer_initialized(vllm_config) try: # Minimal scheduler output with empty metadata; mixin should still # bind/clear metadata even if no loads happen scheduler_output = _make_empty_scheduler_output() # Invoke the no-forward path which uses the mixin context manager KVConnectorModelRunnerMixin.kv_connector_no_forward( scheduler_output, vllm_config ) # Verify clear_connector_metadata was called on the connector connector = get_kv_transfer_group() assert connector._connector_metadata is None # Test connector wrapper records method calls assert connector.call_record.get("bind_connector_metadata", 0) == 1 assert connector.call_record.get("clear_connector_metadata", 0) == 1 finally: # Ensure we clean up the global connector between tests KVConnectorModelRunnerMixin.ensure_kv_transfer_shutdown()
{ "repo_id": "vllm-project/vllm", "file_path": "tests/v1/kv_connector/unit/test_kv_connector_lifecyle.py", "license": "Apache License 2.0", "lines": 50, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/test_envs.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import os from unittest.mock import patch import pytest import vllm.envs as envs from vllm.envs import ( disable_envs_cache, enable_envs_cache, env_list_with_choices, env_set_with_choices, env_with_choices, environment_variables, ) def test_getattr_without_cache(monkeypatch: pytest.MonkeyPatch): assert envs.VLLM_HOST_IP == "" assert envs.VLLM_PORT is None monkeypatch.setenv("VLLM_HOST_IP", "1.1.1.1") monkeypatch.setenv("VLLM_PORT", "1234") assert envs.VLLM_HOST_IP == "1.1.1.1" assert envs.VLLM_PORT == 1234 # __getattr__ is not decorated with functools.cache assert not hasattr(envs.__getattr__, "cache_info") def test_getattr_with_cache(monkeypatch: pytest.MonkeyPatch): monkeypatch.setenv("VLLM_HOST_IP", "1.1.1.1") monkeypatch.setenv("VLLM_PORT", "1234") # __getattr__ is not decorated with functools.cache assert not hasattr(envs.__getattr__, "cache_info") # Enable envs cache and ignore ongoing environment changes enable_envs_cache() # __getattr__ is decorated with functools.cache assert hasattr(envs.__getattr__, "cache_info") start_hits = envs.__getattr__.cache_info().hits # 2 more hits due to VLLM_HOST_IP and VLLM_PORT accesses assert envs.VLLM_HOST_IP == "1.1.1.1" assert envs.VLLM_PORT == 1234 assert envs.__getattr__.cache_info().hits == start_hits + 2 # All environment variables are cached for environment_variable in environment_variables: envs.__getattr__(environment_variable) assert envs.__getattr__.cache_info().hits == start_hits + 2 + len( environment_variables ) # Reset envs.__getattr__ back to none-cached version to # avoid affecting other tests envs.__getattr__ = envs.__getattr__.__wrapped__ def test_getattr_with_reset(monkeypatch: pytest.MonkeyPatch) -> None: monkeypatch.setenv("VLLM_HOST_IP", "1.1.1.1") # __getattr__ is not decorated with functools.cache assert not hasattr(envs.__getattr__, "cache_info") # Enable envs cache and ignore ongoing environment changes enable_envs_cache() assert envs.VLLM_HOST_IP == "1.1.1.1" # With cache enabled, the environment variable value is cached and unchanged monkeypatch.setenv("VLLM_HOST_IP", "2.2.2.2") assert envs.VLLM_HOST_IP == "1.1.1.1" disable_envs_cache() assert envs.VLLM_HOST_IP == "2.2.2.2" # After cache disabled, the environment variable value would be synced # with os.environ monkeypatch.setenv("VLLM_HOST_IP", "3.3.3.3") assert envs.VLLM_HOST_IP == "3.3.3.3" def test_is_envs_cache_enabled() -> None: assert not envs._is_envs_cache_enabled() enable_envs_cache() assert envs._is_envs_cache_enabled() # Only wrap one-layer of cache, so we only need to # call disable once to reset. enable_envs_cache() enable_envs_cache() enable_envs_cache() disable_envs_cache() assert not envs._is_envs_cache_enabled() disable_envs_cache() assert not envs._is_envs_cache_enabled() class TestEnvWithChoices: """Test cases for env_with_choices function.""" def test_default_value_returned_when_env_not_set(self): """Test default is returned when env var is not set.""" env_func = env_with_choices( "NONEXISTENT_ENV", "default", ["option1", "option2"] ) assert env_func() == "default" def test_none_default_returned_when_env_not_set(self): """Test that None is returned when env not set and default is None.""" env_func = env_with_choices("NONEXISTENT_ENV", None, ["option1", "option2"]) assert env_func() is None def test_valid_value_returned_case_sensitive(self): """Test that valid value is returned in case sensitive mode.""" with patch.dict(os.environ, {"TEST_ENV": "option1"}): env_func = env_with_choices( "TEST_ENV", "default", ["option1", "option2"], case_sensitive=True ) assert env_func() == "option1" def test_valid_lowercase_value_returned_case_insensitive(self): """Test that lowercase value is accepted in case insensitive mode.""" with patch.dict(os.environ, {"TEST_ENV": "option1"}): env_func = env_with_choices( "TEST_ENV", "default", ["OPTION1", "OPTION2"], case_sensitive=False ) assert env_func() == "option1" def test_valid_uppercase_value_returned_case_insensitive(self): """Test that uppercase value is accepted in case insensitive mode.""" with patch.dict(os.environ, {"TEST_ENV": "OPTION1"}): env_func = env_with_choices( "TEST_ENV", "default", ["option1", "option2"], case_sensitive=False ) assert env_func() == "OPTION1" def test_invalid_value_raises_error_case_sensitive(self): """Test that invalid value raises ValueError in case sensitive mode.""" with patch.dict(os.environ, {"TEST_ENV": "invalid"}): env_func = env_with_choices( "TEST_ENV", "default", ["option1", "option2"], case_sensitive=True ) with pytest.raises( ValueError, match="Invalid value 'invalid' for TEST_ENV" ): env_func() def test_case_mismatch_raises_error_case_sensitive(self): """Test that case mismatch raises ValueError in case sensitive mode.""" with patch.dict(os.environ, {"TEST_ENV": "OPTION1"}): env_func = env_with_choices( "TEST_ENV", "default", ["option1", "option2"], case_sensitive=True ) with pytest.raises( ValueError, match="Invalid value 'OPTION1' for TEST_ENV" ): env_func() def test_invalid_value_raises_error_case_insensitive(self): """Test that invalid value raises ValueError when case insensitive.""" with patch.dict(os.environ, {"TEST_ENV": "invalid"}): env_func = env_with_choices( "TEST_ENV", "default", ["option1", "option2"], case_sensitive=False ) with pytest.raises( ValueError, match="Invalid value 'invalid' for TEST_ENV" ): env_func() def test_callable_choices_resolved_correctly(self): """Test that callable choices are resolved correctly.""" def get_choices(): return ["dynamic1", "dynamic2"] with patch.dict(os.environ, {"TEST_ENV": "dynamic1"}): env_func = env_with_choices("TEST_ENV", "default", get_choices) assert env_func() == "dynamic1" def test_callable_choices_with_invalid_value(self): """Test that callable choices raise error for invalid values.""" def get_choices(): return ["dynamic1", "dynamic2"] with patch.dict(os.environ, {"TEST_ENV": "invalid"}): env_func = env_with_choices("TEST_ENV", "default", get_choices) with pytest.raises( ValueError, match="Invalid value 'invalid' for TEST_ENV" ): env_func() class TestEnvListWithChoices: """Test cases for env_list_with_choices function.""" def test_default_list_returned_when_env_not_set(self): """Test that default list is returned when env var is not set.""" env_func = env_list_with_choices( "NONEXISTENT_ENV", ["default1", "default2"], ["option1", "option2"] ) assert env_func() == ["default1", "default2"] def test_empty_default_list_returned_when_env_not_set(self): """Test that empty default list is returned when env not set.""" env_func = env_list_with_choices("NONEXISTENT_ENV", [], ["option1", "option2"]) assert env_func() == [] def test_single_valid_value_parsed_correctly(self): """Test that single valid value is parsed correctly.""" with patch.dict(os.environ, {"TEST_ENV": "option1"}): env_func = env_list_with_choices("TEST_ENV", [], ["option1", "option2"]) assert env_func() == ["option1"] def test_multiple_valid_values_parsed_correctly(self): """Test that multiple valid values are parsed correctly.""" with patch.dict(os.environ, {"TEST_ENV": "option1,option2"}): env_func = env_list_with_choices("TEST_ENV", [], ["option1", "option2"]) assert env_func() == ["option1", "option2"] def test_values_with_whitespace_trimmed(self): """Test that values with whitespace are trimmed correctly.""" with patch.dict(os.environ, {"TEST_ENV": " option1 , option2 "}): env_func = env_list_with_choices("TEST_ENV", [], ["option1", "option2"]) assert env_func() == ["option1", "option2"] def test_empty_values_filtered_out(self): """Test that empty values are filtered out.""" with patch.dict(os.environ, {"TEST_ENV": "option1,,option2,"}): env_func = env_list_with_choices("TEST_ENV", [], ["option1", "option2"]) assert env_func() == ["option1", "option2"] def test_empty_string_returns_default(self): """Test that empty string returns default.""" with patch.dict(os.environ, {"TEST_ENV": ""}): env_func = env_list_with_choices( "TEST_ENV", ["default"], ["option1", "option2"] ) assert env_func() == ["default"] def test_only_commas_returns_default(self): """Test that string with only commas returns default.""" with patch.dict(os.environ, {"TEST_ENV": ",,,"}): env_func = env_list_with_choices( "TEST_ENV", ["default"], ["option1", "option2"] ) assert env_func() == ["default"] def test_case_sensitive_validation(self): """Test case sensitive validation.""" with patch.dict(os.environ, {"TEST_ENV": "option1,OPTION2"}): env_func = env_list_with_choices( "TEST_ENV", [], ["option1", "option2"], case_sensitive=True ) with pytest.raises(ValueError, match="Invalid value 'OPTION2' in TEST_ENV"): env_func() def test_case_insensitive_validation(self): """Test case insensitive validation.""" with patch.dict(os.environ, {"TEST_ENV": "OPTION1,option2"}): env_func = env_list_with_choices( "TEST_ENV", [], ["option1", "option2"], case_sensitive=False ) assert env_func() == ["OPTION1", "option2"] def test_invalid_value_in_list_raises_error(self): """Test that invalid value in list raises ValueError.""" with patch.dict(os.environ, {"TEST_ENV": "option1,invalid,option2"}): env_func = env_list_with_choices("TEST_ENV", [], ["option1", "option2"]) with pytest.raises(ValueError, match="Invalid value 'invalid' in TEST_ENV"): env_func() def test_callable_choices_resolved_correctly(self): """Test that callable choices are resolved correctly.""" def get_choices(): return ["dynamic1", "dynamic2"] with patch.dict(os.environ, {"TEST_ENV": "dynamic1,dynamic2"}): env_func = env_list_with_choices("TEST_ENV", [], get_choices) assert env_func() == ["dynamic1", "dynamic2"] def test_callable_choices_with_invalid_value(self): """Test that callable choices raise error for invalid values.""" def get_choices(): return ["dynamic1", "dynamic2"] with patch.dict(os.environ, {"TEST_ENV": "dynamic1,invalid"}): env_func = env_list_with_choices("TEST_ENV", [], get_choices) with pytest.raises(ValueError, match="Invalid value 'invalid' in TEST_ENV"): env_func() def test_duplicate_values_preserved(self): """Test that duplicate values in the list are preserved.""" with patch.dict(os.environ, {"TEST_ENV": "option1,option1,option2"}): env_func = env_list_with_choices("TEST_ENV", [], ["option1", "option2"]) assert env_func() == ["option1", "option1", "option2"] class TestEnvSetWithChoices: """Test cases for env_set_with_choices function.""" def test_default_list_returned_when_env_not_set(self): """Test that default list is returned when env var is not set.""" env_func = env_set_with_choices( "NONEXISTENT_ENV", ["default1", "default2"], ["option1", "option2"] ) assert env_func() == {"default1", "default2"} def test_empty_default_list_returned_when_env_not_set(self): """Test that empty default list is returned when env not set.""" env_func = env_set_with_choices("NONEXISTENT_ENV", [], ["option1", "option2"]) assert env_func() == set() def test_single_valid_value_parsed_correctly(self): """Test that single valid value is parsed correctly.""" with patch.dict(os.environ, {"TEST_ENV": "option1"}): env_func = env_set_with_choices("TEST_ENV", [], ["option1", "option2"]) assert env_func() == {"option1"} def test_multiple_valid_values_parsed_correctly(self): """Test that multiple valid values are parsed correctly.""" with patch.dict(os.environ, {"TEST_ENV": "option1,option2"}): env_func = env_set_with_choices("TEST_ENV", [], ["option1", "option2"]) assert env_func() == {"option1", "option2"} def test_values_with_whitespace_trimmed(self): """Test that values with whitespace are trimmed correctly.""" with patch.dict(os.environ, {"TEST_ENV": " option1 , option2 "}): env_func = env_set_with_choices("TEST_ENV", [], ["option1", "option2"]) assert env_func() == {"option1", "option2"} def test_empty_values_filtered_out(self): """Test that empty values are filtered out.""" with patch.dict(os.environ, {"TEST_ENV": "option1,,option2,"}): env_func = env_set_with_choices("TEST_ENV", [], ["option1", "option2"]) assert env_func() == {"option1", "option2"} def test_empty_string_returns_default(self): """Test that empty string returns default.""" with patch.dict(os.environ, {"TEST_ENV": ""}): env_func = env_set_with_choices( "TEST_ENV", ["default"], ["option1", "option2"] ) assert env_func() == {"default"} def test_only_commas_returns_default(self): """Test that string with only commas returns default.""" with patch.dict(os.environ, {"TEST_ENV": ",,,"}): env_func = env_set_with_choices( "TEST_ENV", ["default"], ["option1", "option2"] ) assert env_func() == {"default"} def test_case_sensitive_validation(self): """Test case sensitive validation.""" with patch.dict(os.environ, {"TEST_ENV": "option1,OPTION2"}): env_func = env_set_with_choices( "TEST_ENV", [], ["option1", "option2"], case_sensitive=True ) with pytest.raises(ValueError, match="Invalid value 'OPTION2' in TEST_ENV"): env_func() def test_case_insensitive_validation(self): """Test case insensitive validation.""" with patch.dict(os.environ, {"TEST_ENV": "OPTION1,option2"}): env_func = env_set_with_choices( "TEST_ENV", [], ["option1", "option2"], case_sensitive=False ) assert env_func() == {"OPTION1", "option2"} def test_invalid_value_in_list_raises_error(self): """Test that invalid value in list raises ValueError.""" with patch.dict(os.environ, {"TEST_ENV": "option1,invalid,option2"}): env_func = env_set_with_choices("TEST_ENV", [], ["option1", "option2"]) with pytest.raises(ValueError, match="Invalid value 'invalid' in TEST_ENV"): env_func() def test_callable_choices_resolved_correctly(self): """Test that callable choices are resolved correctly.""" def get_choices(): return ["dynamic1", "dynamic2"] with patch.dict(os.environ, {"TEST_ENV": "dynamic1,dynamic2"}): env_func = env_set_with_choices("TEST_ENV", [], get_choices) assert env_func() == {"dynamic1", "dynamic2"} def test_callable_choices_with_invalid_value(self): """Test that callable choices raise error for invalid values.""" def get_choices(): return ["dynamic1", "dynamic2"] with patch.dict(os.environ, {"TEST_ENV": "dynamic1,invalid"}): env_func = env_set_with_choices("TEST_ENV", [], get_choices) with pytest.raises(ValueError, match="Invalid value 'invalid' in TEST_ENV"): env_func() def test_duplicate_values_deduped(self): """Test that duplicate values in the list are deduped.""" with patch.dict(os.environ, {"TEST_ENV": "option1,option1,option2"}): env_func = env_set_with_choices("TEST_ENV", [], ["option1", "option2"]) assert env_func() == {"option1", "option2"} class TestVllmConfigureLogging: """Test cases for VLLM_CONFIGURE_LOGGING environment variable.""" def test_configure_logging_defaults_to_true(self): """Test that VLLM_CONFIGURE_LOGGING defaults to True when not set.""" # Ensure the env var is not set with patch.dict(os.environ, {}, clear=False): if "VLLM_CONFIGURE_LOGGING" in os.environ: del os.environ["VLLM_CONFIGURE_LOGGING"] # Clear cache if it exists if hasattr(envs.__getattr__, "cache_clear"): envs.__getattr__.cache_clear() result = envs.VLLM_CONFIGURE_LOGGING assert result is True assert isinstance(result, bool) def test_configure_logging_with_zero_string(self): """Test that VLLM_CONFIGURE_LOGGING='0' evaluates to False.""" with patch.dict(os.environ, {"VLLM_CONFIGURE_LOGGING": "0"}): # Clear cache if it exists if hasattr(envs.__getattr__, "cache_clear"): envs.__getattr__.cache_clear() result = envs.VLLM_CONFIGURE_LOGGING assert result is False assert isinstance(result, bool) def test_configure_logging_with_one_string(self): """Test that VLLM_CONFIGURE_LOGGING='1' evaluates to True.""" with patch.dict(os.environ, {"VLLM_CONFIGURE_LOGGING": "1"}): # Clear cache if it exists if hasattr(envs.__getattr__, "cache_clear"): envs.__getattr__.cache_clear() result = envs.VLLM_CONFIGURE_LOGGING assert result is True assert isinstance(result, bool) def test_configure_logging_with_invalid_value_raises_error(self): """Test that invalid VLLM_CONFIGURE_LOGGING value raises ValueError.""" with patch.dict(os.environ, {"VLLM_CONFIGURE_LOGGING": "invalid"}): # Clear cache if it exists if hasattr(envs.__getattr__, "cache_clear"): envs.__getattr__.cache_clear() with pytest.raises(ValueError, match="invalid literal for int"): _ = envs.VLLM_CONFIGURE_LOGGING
{ "repo_id": "vllm-project/vllm", "file_path": "tests/test_envs.py", "license": "Apache License 2.0", "lines": 369, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/v1/kv_offload/test_cpu_offloading.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import socket import time import msgspec import msgspec.msgpack import pytest import zmq from tqdm import tqdm from vllm import LLM, SamplingParams, TokensPrompt from vllm.config import KVEventsConfig, KVTransferConfig from vllm.distributed.kv_events import BlockStored, KVEventBatch from vllm.platforms import current_platform CPU_BLOCK_SIZES = [48] ATTN_BACKENDS = [] if current_platform.is_cuda(): ATTN_BACKENDS = ["FLASH_ATTN", "FLASHINFER", "TRITON_ATTN"] elif current_platform.is_rocm(): ATTN_BACKENDS = ["TRITON_ATTN"] class MockSubscriber: """Helper class to receive and verify published events""" def __init__( self, endpoint: str, topic: str, ): self.ctx = zmq.Context.instance() self.topic_bytes = topic.encode("utf-8") # Set up subscriber socket self.sub = self.ctx.socket(zmq.SUB) self.sub.setsockopt(zmq.SUBSCRIBE, self.topic_bytes) self.sub.connect(endpoint) self.decoder = msgspec.msgpack.Decoder(type=KVEventBatch) def get_new_cpu_stored_events(self) -> list[BlockStored]: cpu_stored_events: list[BlockStored] = [] poller = zmq.Poller() poller.register(self.sub, zmq.POLLIN) timeout = 1000 # 1 second while True: events = dict(poller.poll(timeout)) if events.get(self.sub) != zmq.POLLIN: return cpu_stored_events topic_bytes, _, payload = self.sub.recv_multipart() assert topic_bytes == self.topic_bytes event_batch = self.decoder.decode(payload) assert isinstance(event_batch, KVEventBatch) for event in event_batch.events: if isinstance(event, BlockStored) and event.medium == "CPU": cpu_stored_events.append(event) timeout = 100 def close(self): """Clean up resources""" self.sub.close() def _latency_test(llm: LLM, subscriber: MockSubscriber): sampling_params = SamplingParams(max_tokens=1) num_times_cpu_better_than_cold = 0 num_tests = 10 total_cold_time = 0.0 total_gpu_hit_time = 0.0 total_cpu_hit_time = 0.0 prompt_token_ids = [0] * 10001 for i in tqdm(range(num_tests), desc="Running tests"): prompt_token_ids[0] = i prompts = [TokensPrompt(prompt_token_ids=prompt_token_ids)] # run generation - this should trigger saving KV cache start_time = time.time() llm.generate(prompts, sampling_params, use_tqdm=False) cold_time = time.time() - start_time total_cold_time += cold_time # run generation again - should hit the GPU prefix cache start_time = time.time() llm.generate(prompts, sampling_params, use_tqdm=False) gpu_hit_time = time.time() - start_time total_gpu_hit_time += gpu_hit_time # reset prefix cache to avoid GPU hit. llm.reset_prefix_cache() assert subscriber.get_new_cpu_stored_events() # run generation again - this should trigger loading from CPU start_time = time.time() llm.generate(prompts, sampling_params, use_tqdm=False) cpu_hit_time = time.time() - start_time total_cpu_hit_time += cpu_hit_time if cpu_hit_time < cold_time: num_times_cpu_better_than_cold += 1 print("Average times:") print(f" Cold: {total_cold_time * 1000 / num_tests:.2f}ms") print(f" GPU hit: {total_gpu_hit_time * 1000 / num_tests:.2f}ms") print(f" CPU hit: {total_cpu_hit_time * 1000 / num_tests:.2f}ms") assert num_times_cpu_better_than_cold >= 0.8 * num_tests def _accuracy_test(llm: LLM, subscriber: MockSubscriber): sampling_params = SamplingParams(max_tokens=1) cpu_block_size = ( llm.llm_engine.vllm_config.kv_transfer_config.kv_connector_extra_config[ "block_size" ] ) subscriber.get_new_cpu_stored_events() # prepend prompt to be cpu block aligned prompt = "Let's count to 10. One, two, three, four," while ( len(llm.generate(prompt, use_tqdm=False)[0].prompt_token_ids) % cpu_block_size != 0 ): prompt = ". " + prompt assert subscriber.get_new_cpu_stored_events() test_count = 100 success_count = 0 for i in range(test_count): if ( llm.generate(prompt, sampling_params, use_tqdm=False)[0].outputs[0].text == " five" ): success_count += 1 assert success_count >= 0.5 * test_count @pytest.mark.parametrize("cpu_block_size", CPU_BLOCK_SIZES) @pytest.mark.parametrize("attn_backend", ATTN_BACKENDS) def test_cpu_offloading(cpu_block_size: int, attn_backend: str) -> None: """ Tests OffloadingConnector with CPUOffloadingSpec. """ # configure OffloadingConnector (spec_name=CPUOffloadingSpec by default) kv_transfer_config = KVTransferConfig( kv_connector="OffloadingConnector", kv_role="kv_both", kv_connector_extra_config={ "cpu_bytes_to_use": 500 << 20, "block_size": cpu_block_size, }, ) port: int with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s: s.bind(("0.0.0.0", 0)) port = s.getsockname()[1] events_endpoint = f"tcp://*:{port}" kv_events_config = KVEventsConfig( enable_kv_cache_events=True, publisher="zmq", endpoint=events_endpoint, topic="test", ) llm = LLM( model="meta-llama/Llama-3.2-1B-Instruct", gpu_memory_utilization=0.5, kv_events_config=kv_events_config, kv_transfer_config=kv_transfer_config, attention_config={"backend": attn_backend}, ) events_endpoint = events_endpoint.replace("*", "127.0.0.1") subscriber = MockSubscriber(events_endpoint, topic=kv_events_config.topic) try: _latency_test(llm, subscriber) _accuracy_test(llm, subscriber) finally: subscriber.close() del llm
{ "repo_id": "vllm-project/vllm", "file_path": "tests/v1/kv_offload/test_cpu_offloading.py", "license": "Apache License 2.0", "lines": 155, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/v1/kv_offload/cpu.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from collections.abc import Iterator import torch from vllm.config import VllmConfig from vllm.platforms import current_platform from vllm.v1.attention.backend import AttentionBackend from vllm.v1.kv_cache_interface import KVCacheConfig from vllm.v1.kv_offload.abstract import LoadStoreSpec, OffloadingManager from vllm.v1.kv_offload.arc_manager import ARCOffloadingManager from vllm.v1.kv_offload.backends.cpu import CPUBackend from vllm.v1.kv_offload.lru_manager import LRUOffloadingManager from vllm.v1.kv_offload.mediums import CPULoadStoreSpec, GPULoadStoreSpec from vllm.v1.kv_offload.spec import OffloadingSpec from vllm.v1.kv_offload.worker.cpu_gpu import CpuGpuOffloadingHandlers from vllm.v1.kv_offload.worker.worker import OffloadingHandler class CPUOffloadingSpec(OffloadingSpec): def __init__(self, vllm_config: VllmConfig, kv_cache_config: KVCacheConfig): super().__init__(vllm_config, kv_cache_config) cpu_bytes_to_use = self.extra_config.get("cpu_bytes_to_use") if not cpu_bytes_to_use: raise Exception( "cpu_bytes_to_use must be specified in kv_connector_extra_config" ) # calculate kv_bytes_per_offloaded_block assert kv_cache_config is not None page_sizes = { kv_cache_group.kv_cache_spec.page_size_bytes for kv_cache_group in kv_cache_config.kv_cache_groups } assert len(page_sizes) == 1 page_size_bytes = page_sizes.pop() kv_bytes_per_block = ( page_size_bytes * len(kv_cache_config.kv_cache_tensors) * vllm_config.parallel_config.world_size ) kv_bytes_per_offloaded_block = kv_bytes_per_block * ( self.offloaded_block_size // self.gpu_block_size ) self.num_blocks = ( int(cpu_bytes_to_use) // kv_bytes_per_offloaded_block if kv_bytes_per_offloaded_block > 0 else 0 ) # scheduler-side self._manager: OffloadingManager | None = None # worker-side self._handlers: CpuGpuOffloadingHandlers | None = None self.eviction_policy: str = self.extra_config.get("eviction_policy", "lru") def get_manager(self) -> OffloadingManager: if not self._manager: kv_events_config = self.vllm_config.kv_events_config enable_events = ( kv_events_config is not None and kv_events_config.enable_kv_cache_events ) backend = CPUBackend( block_size=self.offloaded_block_size, num_blocks=self.num_blocks ) if self.eviction_policy == "lru": self._manager = LRUOffloadingManager( backend=backend, enable_events=enable_events ) elif self.eviction_policy == "arc": self._manager = ARCOffloadingManager( backend=backend, enable_events=enable_events ) else: raise ValueError( f"Unknown eviction policy: {self.eviction_policy}. " f"Supported policies: lru, arc" ) return self._manager def get_handlers( self, kv_caches: dict[str, torch.Tensor], attn_backends: dict[str, type[AttentionBackend]], ) -> Iterator[tuple[type[LoadStoreSpec], type[LoadStoreSpec], OffloadingHandler]]: if not self._handlers: if not current_platform.is_cuda_alike(): raise Exception( "CPU Offloading is currently only supported on CUDA-alike GPUs" ) self._handlers = CpuGpuOffloadingHandlers( attn_backends=attn_backends, gpu_block_size=self.gpu_block_size, cpu_block_size=self.offloaded_block_size, num_cpu_blocks=self.num_blocks, gpu_caches=kv_caches, ) assert self._handlers is not None yield GPULoadStoreSpec, CPULoadStoreSpec, self._handlers.gpu_to_cpu_handler yield CPULoadStoreSpec, GPULoadStoreSpec, self._handlers.cpu_to_gpu_handler
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/kv_offload/cpu.py", "license": "Apache License 2.0", "lines": 93, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:examples/offline_inference/torchrun_dp_example.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """ experimental support for data-parallel inference with torchrun Note the data load balancing and distribution is done out of the vllm engine, no internal lb supported in external_launcher mode. To run this example: ```bash $ torchrun --nproc-per-node=2 examples/offline_inference/torchrun_dp_example.py ``` With custom parallelism settings: ```bash $ torchrun --nproc-per-node=8 examples/offline_inference/torchrun_dp_example.py \ --tp-size=2 --pp-size=1 --dp-size=4 --enable-ep ``` """ import argparse from vllm import LLM, SamplingParams def parse_args(): parser = argparse.ArgumentParser( description="Data-parallel inference with torchrun" ) parser.add_argument( "--tp-size", type=int, default=1, help="Tensor parallel size (default: 1)", ) parser.add_argument( "--pp-size", type=int, default=1, help="Pipeline parallel size (default: 1)", ) parser.add_argument( "--dp-size", type=int, default=2, help="Data parallel size (default: 2)", ) parser.add_argument( "--enable-ep", action="store_true", help="Enable expert parallel (default: False)", ) parser.add_argument( "--model", type=str, default="microsoft/Phi-mini-MoE-instruct", help="Model name or path (default: microsoft/Phi-mini-MoE-instruct)", ) parser.add_argument( "--max-model-len", type=int, default=4096, help="Maximum model length (default: 4096)", ) parser.add_argument( "--gpu-memory-utilization", type=float, default=0.6, help="GPU memory utilization (default: 0.6)", ) parser.add_argument( "--seed", type=int, default=1, help="Random seed (default: 1)", ) return parser.parse_args() args = parse_args() # Create prompts, the same across all ranks prompts = [ "Hello, my name is", "The president of the United States is", "The capital of France is", "The future of AI is", ] # Create sampling parameters, the same across all ranks sampling_params = SamplingParams(temperature=0.8, top_p=0.95) # Use `distributed_executor_backend="external_launcher"` so that # this llm engine/instance only creates one worker. # it is important to set an explicit seed to make sure that # all ranks have the same random seed, so that sampling can be # deterministic across ranks. llm = LLM( model=args.model, tensor_parallel_size=args.tp_size, data_parallel_size=args.dp_size, pipeline_parallel_size=args.pp_size, enable_expert_parallel=args.enable_ep, distributed_executor_backend="external_launcher", max_model_len=args.max_model_len, gpu_memory_utilization=args.gpu_memory_utilization, seed=args.seed, ) dp_rank = llm.llm_engine.vllm_config.parallel_config.data_parallel_rank dp_size = llm.llm_engine.vllm_config.parallel_config.data_parallel_size prompts = [ f"{idx}.{prompt}" for idx, prompt in enumerate(prompts) if idx % dp_size == dp_rank ] outputs = llm.generate(prompts, sampling_params) for output in outputs: prompt = output.prompt generated_text = output.outputs[0].text print( f"DP Rank: {dp_rank} Prompt: {prompt!r}\nGenerated text: {generated_text!r}\n" ) """ Further tips: 1. to communicate control messages across all ranks, use the cpu group, a PyTorch ProcessGroup with GLOO backend. ```python from vllm.distributed.parallel_state import get_world_group cpu_group = get_world_group().cpu_group torch_rank = dist.get_rank(group=cpu_group) if torch_rank == 0: # do something for rank 0, e.g. saving the results to disk. ``` 2. to communicate data across all ranks, use the model's device group, a PyTorch ProcessGroup with NCCL backend. ```python from vllm.distributed.parallel_state import get_world_group device_group = get_world_group().device_group ``` 3. to access the model directly in every rank, use the following code: ```python llm.llm_engine.model_executor.driver_worker.worker.model_runner.model ``` """
{ "repo_id": "vllm-project/vllm", "file_path": "examples/offline_inference/torchrun_dp_example.py", "license": "Apache License 2.0", "lines": 130, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/distributed/test_torchrun_example_moe.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project # unit test for `examples/offline_inference/torchrun_example.py` import os import random import torch.distributed as dist from vllm import LLM, SamplingParams from vllm.distributed.parallel_state import get_tp_group, get_world_group dist.init_process_group(backend="gloo") # Create prompts prompts = [ "Hello, my name is", "The president of the United States is", "The capital of France is", "The future of AI is", ] * 10 dp_size = int(os.getenv("DP_SIZE", "1")) dp_rank = int(os.getenv("DP_RANK", "0")) if dp_size > 1: # distribute the prompts across the data parallel ranks prompts = [prompt for idx, prompt in enumerate(prompts) if idx % dp_size == dp_rank] sampling_params = SamplingParams(temperature=0.8, top_p=0.95) # set different `gpu_memory_utilization` and `swap_space` for different ranks, # to test if all ranks agree on the same kv cache configuration. llm = LLM( model="microsoft/Phi-mini-MoE-instruct", tensor_parallel_size=int(os.getenv("TP_SIZE", "1")), pipeline_parallel_size=int(os.getenv("PP_SIZE", "1")), enable_expert_parallel=int(os.getenv("ENABLE_EP", "0")) == 1, distributed_executor_backend="external_launcher", gpu_memory_utilization=random.uniform(0.7, 0.9), swap_space=random.randint(1, 4), seed=0, ) outputs = llm.generate(prompts, sampling_params) group = get_world_group() if dp_size == 1 else get_tp_group() cpu_group = group.cpu_group group_rank = dist.get_rank(group=cpu_group) def test_consistent_across_ranks(obj): if group_rank == 0: dist.broadcast_object_list([obj], src=group.ranks[0], group=cpu_group) else: container = [None] dist.broadcast_object_list(container, src=group.ranks[0], group=cpu_group) assert container[0] == obj test_consistent_across_ranks(llm.llm_engine.vllm_config.cache_config.num_cpu_blocks) test_consistent_across_ranks(llm.llm_engine.vllm_config.cache_config.num_gpu_blocks) # make sure we can access the model parameters from the calling process # of the `LLM` instance. params = list( llm.llm_engine.model_executor.driver_worker.worker.model_runner.model.parameters() ) test_consistent_across_ranks(len(params)) # all ranks should have the same outputs for output in outputs: prompt = output.prompt generated_text = output.outputs[0].text test_consistent_across_ranks(prompt) test_consistent_across_ranks(generated_text) print(f"Rank {group_rank}, Prompt: {prompt!r}, Generated text: {generated_text!r}")
{ "repo_id": "vllm-project/vllm", "file_path": "tests/distributed/test_torchrun_example_moe.py", "license": "Apache License 2.0", "lines": 60, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tools/pre_commit/mypy.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """ Run mypy on changed files. This script is designed to be used as a pre-commit hook. It runs mypy on files that have been changed. It groups files into different mypy calls based on their directory to avoid import following issues. Usage: python tools/pre_commit/mypy.py <ci> <python_version> <changed_files...> Args: ci: "1" if running in CI, "0" otherwise. In CI, follow_imports is set to "silent" for the main group of files. python_version: Python version to use (e.g., "3.10") or "local" to use the local Python version. changed_files: List of changed files to check. """ import subprocess import sys import regex as re # After fixing errors resulting from changing follow_imports # from "skip" to "silent", remove its directory from SEPARATE_GROUPS. SEPARATE_GROUPS = [ "tests", # v0 related "vllm/lora", "vllm/model_executor", ] # TODO(woosuk): Include the code from Megatron and HuggingFace. EXCLUDE = [ "vllm/model_executor/models", "vllm/model_executor/layers/fla/ops", # Ignore triton kernels in ops. "vllm/v1/attention/ops", # TODO: Remove these entries after fixing mypy errors. "vllm/benchmarks", "vllm/config", "vllm/device_allocator", "vllm/reasoning", "vllm/tool_parser", ] def group_files(changed_files: list[str]) -> dict[str, list[str]]: """ Group changed files into different mypy calls. Args: changed_files: List of changed files. Returns: A dictionary mapping file group names to lists of changed files. """ exclude_pattern = re.compile(f"^{'|'.join(EXCLUDE)}.*") file_groups = {"": []} file_groups.update({k: [] for k in SEPARATE_GROUPS}) for changed_file in changed_files: # Skip files which should be ignored completely if exclude_pattern.match(changed_file): continue # Group files by mypy call for directory in SEPARATE_GROUPS: if re.match(f"^{directory}.*", changed_file): file_groups[directory].append(changed_file) break else: if changed_file.startswith("vllm/"): file_groups[""].append(changed_file) return file_groups def mypy( targets: list[str], python_version: str | None, follow_imports: str | None, file_group: str, ) -> int: """ Run mypy on the given targets. Args: targets: List of files or directories to check. python_version: Python version to use (e.g., "3.10") or None to use the default mypy version. follow_imports: Value for the --follow-imports option or None to use the default mypy behavior. file_group: The file group name for logging purposes. Returns: The return code from mypy. """ args = ["mypy"] if python_version is not None: args += ["--python-version", python_version] if follow_imports is not None: args += ["--follow-imports", follow_imports] print(f"$ {' '.join(args)} {file_group}") return subprocess.run(args + targets, check=False).returncode def main(): ci = sys.argv[1] == "1" python_version = sys.argv[2] file_groups = group_files(sys.argv[3:]) if python_version == "local": python_version = f"{sys.version_info.major}.{sys.version_info.minor}" returncode = 0 for file_group, changed_files in file_groups.items(): follow_imports = None if ci and file_group == "" else "skip" if changed_files: returncode |= mypy( changed_files, python_version, follow_imports, file_group ) return returncode if __name__ == "__main__": sys.exit(main())
{ "repo_id": "vllm-project/vllm", "file_path": "tools/pre_commit/mypy.py", "license": "Apache License 2.0", "lines": 105, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/model_executor/models/dots_ocr.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from collections.abc import Iterable, Mapping from typing import Annotated, Literal, TypeAlias import torch import torch.nn as nn from torch.nn import LayerNorm from transformers.models.qwen2_vl import Qwen2VLProcessor from vllm.config import VllmConfig from vllm.config.multimodal import BaseDummyOptions from vllm.distributed import utils as dist_utils from vllm.distributed.parallel_state import ( get_tensor_model_parallel_rank, get_tensor_model_parallel_world_size, ) from vllm.model_executor.layers.activation import SiluAndMul from vllm.model_executor.layers.attention import ( MMEncoderAttention, ) from vllm.model_executor.layers.conv import Conv2dLayer from vllm.model_executor.layers.layernorm import RMSNorm from vllm.model_executor.layers.linear import ( ColumnParallelLinear, MergedColumnParallelLinear, QKVParallelLinear, RowParallelLinear, ) from vllm.model_executor.layers.quantization import QuantizationConfig from vllm.model_executor.layers.rotary_embedding.common import ( ApplyRotaryEmb, ) from vllm.model_executor.model_loader.weight_utils import default_weight_loader from vllm.model_executor.models.interfaces import ( MultiModalEmbeddings, SupportsLoRA, SupportsMultiModal, SupportsPP, ) from vllm.model_executor.models.module_mapping import MultiModelKeys from vllm.model_executor.models.qwen2 import Qwen2ForCausalLM from vllm.model_executor.models.qwen2_vl import ( Qwen2VisionAttention, Qwen2VLDummyInputsBuilder, Qwen2VLMultiModalProcessor, Qwen2VLProcessingInfo, ) from vllm.model_executor.models.utils import ( AutoWeightsLoader, WeightsMapper, init_vllm_registered_model, maybe_prefix, ) from vllm.model_executor.models.vision import get_vit_attn_backend from vllm.multimodal import MULTIMODAL_REGISTRY from vllm.multimodal.inputs import MultiModalDataDict from vllm.sequence import IntermediateTensors from vllm.transformers_utils.configs.dotsocr import DotsOCRConfig, DotsVisionConfig from vllm.utils.tensor_schema import TensorSchema, TensorShape from vllm.v1.attention.backends.registry import AttentionBackendEnum from .vision import is_vit_use_data_parallel, run_dp_sharded_mrope_vision_model IMAGE_TOKEN = "<|imgpad|>" class DotsOCRImagePixelInputs(TensorSchema): """ Dimensions: - np: The total number of patches over each image over each prompt in the batch - ni: Number of images - cps: Number of channels * patch_size * patch_size """ type: Literal["pixel_values"] pixel_values: Annotated[torch.Tensor, TensorShape("np", "cps")] image_grid_thw: Annotated[torch.Tensor, TensorShape("ni", 3)] class DotsOCRImageEmbeddingInputs(TensorSchema): """ Dimensions: - nf: Number of image features - hs: Hidden size - ni: Number of images """ type: Literal["image_embeds"] image_embeds: Annotated[torch.Tensor, TensorShape("nf", "hs")] image_grid_thw: Annotated[torch.Tensor, TensorShape("ni", 3)] DotsOCRImageInputs: TypeAlias = DotsOCRImagePixelInputs | DotsOCRImageEmbeddingInputs class DotsOCRDummyInputsBuilder(Qwen2VLDummyInputsBuilder): def get_dummy_text(self, mm_counts: Mapping[str, int]) -> str: num_images = mm_counts.get("image", 0) return IMAGE_TOKEN * num_images def get_dummy_mm_data( self, seq_len: int, mm_counts: Mapping[str, int], mm_options: Mapping[str, BaseDummyOptions], ) -> MultiModalDataDict: num_images = mm_counts.get("image", 0) target_width, target_height = self.info.get_image_size_with_most_features() image_overrides = mm_options.get("image") return { "image": self._get_dummy_images( width=target_width, height=target_height, num_images=num_images, overrides=image_overrides, ), } class DotsOCRProcessingInfo(Qwen2VLProcessingInfo): def get_hf_config(self) -> DotsOCRConfig: config = self.ctx.get_hf_config() if not config.__class__.__name__ == "DotsOCRConfig": raise TypeError(f"Expected DotsOCRConfig, got {type(config)}") if hasattr(config, "vision_config") and isinstance(config.vision_config, dict): config.vision_config = DotsVisionConfig(**config.vision_config) return config def get_supported_mm_limits(self) -> Mapping[str, int | None]: return {"image": None} def get_mm_max_tokens_per_item( self, seq_len: int, mm_counts: Mapping[str, int], ) -> Mapping[str, int]: max_image_tokens = self.get_max_image_tokens() return {"image": max_image_tokens} def get_hf_processor( self, **kwargs: object, ) -> Qwen2VLProcessor: self.get_tokenizer().image_token = IMAGE_TOKEN # Ensure image token is set processor = self.ctx.get_hf_processor( Qwen2VLProcessor, **kwargs, ) processor.image_token = IMAGE_TOKEN processor.video_token = "<|video_pad|>" return processor class VisionRotaryEmbedding(nn.Module): def __init__(self, dim: int, theta: float = 10000.0) -> None: super().__init__() inv_freq = 1.0 / (theta ** (torch.arange(0, dim, 2, dtype=torch.float) / dim)) self.register_buffer("inv_freq", inv_freq, persistent=False) def forward(self, seqlen: int) -> torch.Tensor: seq = torch.arange( seqlen, device=self.inv_freq.device, dtype=self.inv_freq.dtype ) freqs = torch.outer(seq, self.inv_freq) return freqs class PatchMerger(nn.Module): def __init__( self, dim: int, context_dim: int, spatial_merge_size: int = 2, pre_norm="layernorm", prefix: str = "", ) -> None: super().__init__() use_data_parallel = is_vit_use_data_parallel() self.hidden_size = context_dim * (spatial_merge_size**2) self.pre_norm = pre_norm if self.pre_norm == "layernorm": self.ln_q = LayerNorm(context_dim, eps=1e-6) elif self.pre_norm == "rmsnorm": self.ln_q = RMSNorm(context_dim, eps=1e-6) self.mlp = nn.Sequential( ColumnParallelLinear( self.hidden_size, self.hidden_size, bias=True, return_bias=False, prefix=f"{prefix}.0", disable_tp=use_data_parallel, ), nn.GELU(), RowParallelLinear( self.hidden_size, dim, bias=True, return_bias=False, prefix=f"{prefix}.2", disable_tp=use_data_parallel, ), ) def forward(self, x: torch.Tensor) -> torch.Tensor: if self.pre_norm: x = self.mlp(self.ln_q(x).view(-1, self.hidden_size)) else: x = self.mlp(x.view(-1, self.hidden_size)) return x class DotsVisionAttention(nn.Module): def __init__( self, config, dim: int, num_heads: int = 16, bias: bool = True, *, quant_config: QuantizationConfig | None = None, prefix: str = "", ) -> None: super().__init__() use_data_parallel = is_vit_use_data_parallel() self.embed_dim = dim self.tp_size = ( 1 if use_data_parallel else get_tensor_model_parallel_world_size() ) self.tp_rank = 0 if use_data_parallel else get_tensor_model_parallel_rank() self.hidden_size_per_attention_head = dist_utils.divide(dim, num_heads) self.num_attention_heads_per_partition = dist_utils.divide( num_heads, self.tp_size ) # qkv/proj follow Qwen2-VL style; bias controlled by arg self.qkv = QKVParallelLinear( hidden_size=dim, head_size=self.hidden_size_per_attention_head, total_num_heads=num_heads, bias=bias, quant_config=quant_config, prefix=f"{prefix}.qkv", disable_tp=use_data_parallel, ) self.proj = RowParallelLinear( input_size=dim, output_size=dim, bias=bias, quant_config=quant_config, prefix=f"{prefix}.proj", disable_tp=use_data_parallel, ) self.attn = MMEncoderAttention( num_heads=self.num_attention_heads_per_partition, head_size=self.hidden_size_per_attention_head, scale=self.hidden_size_per_attention_head**-0.5, prefix=f"{prefix}.attn", ) self.apply_rotary_emb = ApplyRotaryEmb( enforce_enable=True, enable_fp32_compute=True, ) def forward( self, hidden_states: torch.Tensor, cu_seqlens: torch.Tensor, rotary_pos_emb: torch.Tensor | None = None, *, max_seqlen: torch.Tensor | None = None, ) -> torch.Tensor: # [S, C] -> [S, B=1, C] x = hidden_states.unsqueeze(1) x, _ = self.qkv(x) q, k, v = Qwen2VisionAttention.split_qkv(self, x) bs = q.shape[1] # [S,B,H,D] -> [B,S,H,D] q = q.permute(1, 0, 2, 3).contiguous() k = k.permute(1, 0, 2, 3).contiguous() v = v.permute(1, 0, 2, 3).contiguous() if rotary_pos_emb is not None: qk_concat = torch.cat([q, k], dim=0) qk_rotated = self.apply_rotary_emb( qk_concat, rotary_pos_emb.cos(), rotary_pos_emb.sin(), ) q, k = torch.chunk(qk_rotated, 2, dim=0) context_layer = self.attn( query=q, key=k, value=v, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen, ) # [B,S,H,D] -> [S,B,H*D] -> [S, C] context_layer = context_layer.permute(1, 0, 2, 3).contiguous() context_layer = context_layer.view(context_layer.shape[0], bs, -1) out, _ = self.proj(context_layer) return out.squeeze(1) class DotsSwiGLUFFN(nn.Module): def __init__( self, config, *, quant_config: QuantizationConfig | None = None, prefix: str = "", ): super().__init__() hidden_features = config.intermediate_size in_features = config.embed_dim bias = config.use_bias use_data_parallel = is_vit_use_data_parallel() # Referenced aimv2.py AIMv2SwiGLUFFN self.fc13 = MergedColumnParallelLinear( in_features, [hidden_features] * 2, bias=bias, quant_config=quant_config, prefix=f"{prefix}.fc13", disable_tp=use_data_parallel, ) self.fc2 = RowParallelLinear( hidden_features, in_features, bias=bias, quant_config=quant_config, prefix=f"{prefix}.fc2", disable_tp=use_data_parallel, ) self.act_fn = SiluAndMul() def forward(self, x: torch.Tensor) -> torch.Tensor: x, _ = self.fc13(x) x = self.act_fn(x) x, _ = self.fc2(x) return x def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: stacked_params_mapping = [ ("fc13", "fc1", 0), ("fc13", "fc3", 1), ] params_dict = dict(self.named_parameters()) loaded_params: set[str] = set() for name, loaded_weight in weights: 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) loaded_params.add(name) return loaded_params class DotsPatchEmbed(nn.Module): def __init__(self, config): super().__init__() self.num_channels = config.num_channels self.patch_size = config.patch_size self.temporal_patch_size = config.temporal_patch_size self.embed_dim = config.embed_dim self.config = config self.proj = Conv2dLayer( config.num_channels, config.embed_dim, kernel_size=(config.patch_size, config.patch_size), stride=(config.patch_size, config.patch_size), ) self.norm = RMSNorm(config.embed_dim, eps=config.rms_norm_eps) def forward(self, x: torch.Tensor, grid_thw=None) -> torch.Tensor: x = x.view( -1, self.num_channels, self.temporal_patch_size, self.patch_size, self.patch_size, )[:, :, 0] x = self.proj(x).view(-1, self.embed_dim) x = self.norm(x) return x class DotsViTPreprocessor(nn.Module): def __init__(self, config): super().__init__() self.patch_h = config.patch_size self.patch_w = config.patch_size self.embed_dim = config.embed_dim self.config = config self.patchifier = DotsPatchEmbed(config) def forward(self, x: torch.Tensor, grid_thw=None) -> torch.Tensor: tokens = self.patchifier(x, grid_thw) return tokens class DotsVisionBlock(nn.Module): def __init__( self, config, *, quant_config: QuantizationConfig | None = None, prefix: str = "", ): super().__init__() self.attn = DotsVisionAttention( config, config.embed_dim, num_heads=config.num_attention_heads, bias=config.use_bias, quant_config=quant_config, prefix=f"{prefix}.attn", ) self.norm1 = RMSNorm(config.embed_dim, eps=config.rms_norm_eps) self.mlp = DotsSwiGLUFFN( config, quant_config=quant_config, prefix=f"{prefix}.mlp", ) self.norm2 = RMSNorm(config.embed_dim, eps=config.rms_norm_eps) def forward( self, hidden_states: torch.Tensor, *, cu_seqlens: torch.Tensor, rotary_pos_emb: torch.Tensor, max_seqlen: int | None = None, ) -> torch.Tensor: hidden_states = hidden_states + self.attn( self.norm1(hidden_states), cu_seqlens=cu_seqlens, rotary_pos_emb=rotary_pos_emb, max_seqlen=max_seqlen, ) hidden_states = hidden_states + self.mlp(self.norm2(hidden_states)) return hidden_states class DotsVisionTransformer(nn.Module): def __init__( self, config: DotsVisionConfig, quant_config: QuantizationConfig | None = None, *, num_hidden_layers_override: int | None = None, require_post_norm: bool | None = None, prefix: str = "", ) -> None: super().__init__() self.config = config self.spatial_merge_size = config.spatial_merge_size self.patch_embed = DotsViTPreprocessor(config) head_dim = config.embed_dim // config.num_attention_heads self.rotary_pos_emb = VisionRotaryEmbedding(head_dim // 2) self.attn_backend = get_vit_attn_backend( head_size=head_dim, dtype=torch.get_default_dtype(), ) self.out_hidden_size = config.hidden_size # Keep blocks for compatibility with other vision towers num_layers = ( config.num_hidden_layers if num_hidden_layers_override is None else num_hidden_layers_override ) self.blocks = nn.ModuleList( [ DotsVisionBlock( config, quant_config=quant_config, prefix=f"{prefix}.blocks.{i}", ) for i in range(num_layers) ] ) if require_post_norm is None: require_post_norm = len(self.blocks) == config.num_hidden_layers if require_post_norm and self.config.post_norm: self.post_trunk_norm = RMSNorm(config.embed_dim, eps=config.rms_norm_eps) else: self.post_trunk_norm = None self.merger = PatchMerger( dim=config.hidden_size, context_dim=config.embed_dim, spatial_merge_size=config.spatial_merge_size, ) @property def dtype(self) -> torch.dtype: return self.patch_embed.patchifier.proj.weight.dtype @property def device(self) -> torch.device: return self.patch_embed.patchifier.proj.weight.device def get_pos_ids_by_grid(self, grid_thw: list[list[int]]) -> list[torch.Tensor]: pos_ids = [] for t, h, w in grid_thw: hpos_ids = torch.arange(h).unsqueeze(1).expand(-1, w) hpos_ids = hpos_ids.reshape( h // self.spatial_merge_size, self.spatial_merge_size, w // self.spatial_merge_size, self.spatial_merge_size, ) hpos_ids = hpos_ids.permute(0, 2, 1, 3) hpos_ids = hpos_ids.flatten() wpos_ids = torch.arange(w).unsqueeze(0).expand(h, -1) wpos_ids = wpos_ids.reshape( h // self.spatial_merge_size, self.spatial_merge_size, w // self.spatial_merge_size, self.spatial_merge_size, ) wpos_ids = wpos_ids.permute(0, 2, 1, 3) wpos_ids = wpos_ids.flatten() pos_ids.append(torch.stack([hpos_ids, wpos_ids], dim=-1).repeat(t, 1)) return pos_ids def rot_pos_emb(self, grid_thw: list[list[int]]) -> torch.Tensor: pos_ids = self.get_pos_ids_by_grid(grid_thw) pos_ids = torch.cat(pos_ids, dim=0) max_grid_size = max(max(h, w) for _, h, w in grid_thw) rotary_pos_emb_full = self.rotary_pos_emb(max_grid_size) rotary_pos_emb = rotary_pos_emb_full[pos_ids].flatten(1) return rotary_pos_emb def compute_attn_mask_seqlen(self, cu_seqlens: torch.Tensor) -> int | None: max_seqlen = None if self.attn_backend in { AttentionBackendEnum.FLASH_ATTN, AttentionBackendEnum.ROCM_AITER_FA, AttentionBackendEnum.TRITON_ATTN, }: max_seqlen = (cu_seqlens[1:] - cu_seqlens[:-1]).max() return max_seqlen def forward( self, hidden_states: torch.Tensor, grid_thw: list[list[int]] ) -> torch.Tensor: rotary_pos_emb = self.rot_pos_emb(grid_thw) # Convert grid_thw to tensor (always expecting list format now) grid_thw = torch.tensor(grid_thw, device=hidden_states.device, dtype=torch.long) hidden_states = hidden_states.to(self.dtype) hidden_states = self.patch_embed(hidden_states, grid_thw) cu_seqlens = torch.repeat_interleave( grid_thw[:, 1] * grid_thw[:, 2], grid_thw[:, 0] ).cumsum( dim=0, dtype=grid_thw.dtype if torch.jit.is_tracing() else torch.int32, ) cu_seqlens = torch.cat([cu_seqlens.new_zeros(1), cu_seqlens]) max_seqlen = self.compute_attn_mask_seqlen(cu_seqlens) for blk in self.blocks: hidden_states = blk( hidden_states, cu_seqlens=cu_seqlens, rotary_pos_emb=rotary_pos_emb, max_seqlen=max_seqlen, ) if self.post_trunk_norm is not None: hidden_states = self.post_trunk_norm(hidden_states) hidden_states = self.merger(hidden_states) return hidden_states @MULTIMODAL_REGISTRY.register_processor( Qwen2VLMultiModalProcessor, info=DotsOCRProcessingInfo, dummy_inputs=DotsOCRDummyInputsBuilder, ) class DotsOCRForCausalLM(nn.Module, SupportsMultiModal, SupportsPP, SupportsLoRA): hf_to_vllm_mapper = WeightsMapper( orig_to_new_substr={ ".attn.qkv_proj.": ".attn.qkv.", ".attn.out_proj.": ".attn.proj.", }, orig_to_new_prefix={ "lm_head.": "language_model.lm_head.", "model.": "language_model.model.", }, ) packed_modules_mapping = { "qkv_proj": [ "q_proj", "k_proj", "v_proj", ], "gate_up_proj": [ "gate_proj", "up_proj", ], ".attn.qkv": [".attn.qkv"], "fc13": ["fc1", "fc3"], } supports_encoder_tp_data = True @classmethod def get_placeholder_str(cls, modality: str, i: int) -> str | None: if modality.startswith("image"): return "<|img|><|imgpad|><|endofimg|>" def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): super().__init__() self.config: DotsOCRConfig = vllm_config.model_config.hf_config self.quant_config = vllm_config.quant_config multimodal_config = vllm_config.model_config.multimodal_config self.use_data_parallel = multimodal_config.mm_encoder_tp_mode == "data" if isinstance(self.config.vision_config, dict): vision_config = DotsVisionConfig(**self.config.vision_config) self.config.vision_config = vision_config else: vision_config = self.config.vision_config with self._mark_tower_model(vllm_config, "image"): self.vision_tower = DotsVisionTransformer( vision_config, quant_config=self.quant_config, prefix=maybe_prefix(prefix, "vision_tower"), ) with self._mark_language_model(vllm_config): self.language_model: Qwen2ForCausalLM = init_vllm_registered_model( vllm_config=vllm_config, hf_config=self.config, prefix=maybe_prefix(prefix, "language_model"), architectures=["Qwen2ForCausalLM"], ) self.make_empty_intermediate_tensors = ( self.language_model.make_empty_intermediate_tensors ) def _parse_and_validate_image_input( self, **kwargs: object ) -> DotsOCRImageInputs | None: pixel_values = kwargs.pop("pixel_values", None) image_embeds = kwargs.pop("image_embeds", None) image_grid_thw = kwargs.pop("image_grid_thw", None) if pixel_values is None and image_embeds is None: return None if pixel_values is not None: return DotsOCRImagePixelInputs( type="pixel_values", pixel_values=pixel_values, image_grid_thw=image_grid_thw, ) if image_embeds is not None: return DotsOCRImageEmbeddingInputs( type="image_embeds", image_embeds=image_embeds, image_grid_thw=image_grid_thw, ) def _process_image_input( self, image_input: DotsOCRImageInputs ) -> tuple[torch.Tensor, ...]: grid_thw = image_input["image_grid_thw"] assert grid_thw.ndim == 2 grid_thw_list = grid_thw.tolist() if image_input["type"] == "image_embeds": image_embeds = image_input["image_embeds"].type(self.vision_tower.dtype) else: pixel_values = image_input["pixel_values"].type(self.vision_tower.dtype) if self.use_data_parallel: return run_dp_sharded_mrope_vision_model( self.vision_tower, pixel_values, grid_thw_list, rope_type="rope_3d", ) else: image_embeds = self.vision_tower(pixel_values, grid_thw_list)[ :, : self.config.hidden_size ] # Split concatenated embeddings for each image item. merge_size = self.vision_tower.spatial_merge_size sizes = ( torch.tensor(grid_thw_list, dtype=torch.long).prod(-1) // (merge_size * merge_size) ).tolist() return image_embeds.split(sizes) def get_num_mm_encoder_tokens(self, num_image_tokens: int) -> int: merge_size = self.vision_tower.spatial_merge_size return num_image_tokens * (merge_size**2) def get_num_mm_connector_tokens(self, num_vision_tokens: int) -> int: merge_size = self.vision_tower.spatial_merge_size return num_vision_tokens // (merge_size**2) def embed_multimodal(self, **kwargs: object) -> MultiModalEmbeddings: image_input = self._parse_and_validate_image_input(**kwargs) if image_input is None: return [] vision_embeddings = self._process_image_input(image_input) return vision_embeddings def forward( self, input_ids: torch.Tensor | None, positions: torch.Tensor, intermediate_tensors: IntermediateTensors | None = None, inputs_embeds: torch.Tensor | None = None, **kwargs, ) -> torch.Tensor | IntermediateTensors: if intermediate_tensors is not None: inputs_embeds = None hidden_states = self.language_model( input_ids=input_ids, positions=positions, intermediate_tensors=intermediate_tensors, inputs_embeds=inputs_embeds, ) return hidden_states def compute_logits( self, hidden_states: torch.Tensor, ) -> torch.Tensor | None: return self.language_model.compute_logits(hidden_states) def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: loader = AutoWeightsLoader(self) return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper) def get_mm_mapping(self) -> MultiModelKeys: """ Get the module prefix in multimodal models """ return MultiModelKeys.from_string_field( language_model="language_model", connector="vision_tower.merger", tower_model="vision_tower.", )
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/model_executor/models/dots_ocr.py", "license": "Apache License 2.0", "lines": 689, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/transformers_utils/configs/dotsocr.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from typing import Any from transformers.configuration_utils import PretrainedConfig from transformers.models.qwen2 import Qwen2Config class DotsVisionConfig(PretrainedConfig): model_type: str = "dots_vit" def __init__( self, embed_dim: int = 1536, # vision encoder embed size hidden_size: int = 1536, # after merger hidden size intermediate_size: int = 4224, num_hidden_layers: int = 42, num_attention_heads: int = 12, num_channels: int = 3, patch_size: int = 14, spatial_merge_size: int = 2, temporal_patch_size: int = 1, rms_norm_eps: float = 1e-5, use_bias: bool = False, attn_implementation="flash_attention_2", initializer_range=0.02, init_merger_std=0.02, is_causal=False, # ve causal forward post_norm=True, gradient_checkpointing=False, **kwargs: Any, ): super().__init__(**kwargs) self.embed_dim = embed_dim self.hidden_size = hidden_size self.intermediate_size = intermediate_size self.num_hidden_layers = num_hidden_layers self.num_attention_heads = num_attention_heads self.num_channels = num_channels self.patch_size = patch_size self.spatial_merge_size = spatial_merge_size self.temporal_patch_size = temporal_patch_size self.rms_norm_eps = rms_norm_eps self.use_bias = use_bias self.attn_implementation = attn_implementation self.initializer_range = initializer_range self.init_merger_std = init_merger_std self.is_causal = is_causal self.post_norm = post_norm self.gradient_checkpointing = gradient_checkpointing class DotsOCRConfig(Qwen2Config): model_type = "dots_ocr" def __init__( self, image_token_id=151665, video_token_id=151656, vision_config: dict | None = None, *args, **kwargs, ): super().__init__(*args, **kwargs) self.image_token_id = image_token_id self.video_token_id = video_token_id self.vision_config = DotsVisionConfig(**(vision_config or {})) def save_pretrained(self, save_directory, **kwargs): self._auto_class = None super().save_pretrained(save_directory, **kwargs)
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/transformers_utils/configs/dotsocr.py", "license": "Apache License 2.0", "lines": 63, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/multimodal/test_audio.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project # test_audio.py from unittest.mock import patch import numpy as np import pytest import torch from vllm.multimodal.audio import ( MONO_AUDIO_SPEC, PASSTHROUGH_AUDIO_SPEC, AudioResampler, AudioSpec, ChannelReduction, normalize_audio, resample_audio_librosa, resample_audio_scipy, split_audio, ) @pytest.fixture def dummy_audio(): return np.array([0.0, 0.1, 0.2, 0.3, 0.4], dtype=float) def test_resample_audio_librosa(dummy_audio): with patch("vllm.multimodal.audio.librosa.resample") as mock_resample: mock_resample.return_value = dummy_audio * 2 out = resample_audio_librosa(dummy_audio, orig_sr=44100, target_sr=22050) mock_resample.assert_called_once_with( dummy_audio, orig_sr=44100, target_sr=22050 ) assert np.all(out == dummy_audio * 2) def test_resample_audio_scipy(dummy_audio): out_down = resample_audio_scipy(dummy_audio, orig_sr=4, target_sr=2) out_up = resample_audio_scipy(dummy_audio, orig_sr=2, target_sr=4) out_same = resample_audio_scipy(dummy_audio, orig_sr=4, target_sr=4) assert len(out_down) == 3 assert len(out_up) == 10 assert np.all(out_same == dummy_audio) @pytest.mark.xfail(reason="resample_audio_scipy is buggy for non-integer ratios") def test_resample_audio_scipy_non_integer_ratio(dummy_audio): out = resample_audio_scipy(dummy_audio, orig_sr=5, target_sr=3) expected_len = int(round(len(dummy_audio) * 3 / 5)) assert len(out) == expected_len assert isinstance(out, np.ndarray) assert np.isfinite(out).all() def test_audio_resampler_librosa_calls_resample(dummy_audio): resampler = AudioResampler(target_sr=22050, method="librosa") with patch("vllm.multimodal.audio.resample_audio_librosa") as mock_resample: mock_resample.return_value = dummy_audio out = resampler.resample(dummy_audio, orig_sr=44100) mock_resample.assert_called_once_with( dummy_audio, orig_sr=44100, target_sr=22050 ) assert np.all(out == dummy_audio) def test_audio_resampler_scipy_calls_resample(dummy_audio): resampler = AudioResampler(target_sr=22050, method="scipy") with patch("vllm.multimodal.audio.resample_audio_scipy") as mock_resample: mock_resample.return_value = dummy_audio out = resampler.resample(dummy_audio, orig_sr=44100) mock_resample.assert_called_once_with( dummy_audio, orig_sr=44100, target_sr=22050 ) assert np.all(out == dummy_audio) def test_audio_resampler_invalid_method(dummy_audio): resampler = AudioResampler(target_sr=22050, method="invalid") with pytest.raises(ValueError): resampler.resample(dummy_audio, orig_sr=44100) def test_audio_resampler_no_target_sr(dummy_audio): resampler = AudioResampler(target_sr=None) with pytest.raises(RuntimeError): resampler.resample(dummy_audio, orig_sr=44100) # ============================================================ # Tests for normalize_audio function # ============================================================ class TestNormalizeAudio: """Tests for normalize_audio function with different specs.""" def test_passthrough_preserves_audio(self): """Passthrough spec should not modify audio.""" stereo = np.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]], dtype=np.float32) result = normalize_audio(stereo, PASSTHROUGH_AUDIO_SPEC) np.testing.assert_array_equal(result, stereo) def test_mono_spec_with_numpy_stereo(self): """Mono spec should reduce stereo numpy array to 1D.""" stereo = np.array([[1.0, 2.0], [-1.0, 0.0]], dtype=np.float32) result = normalize_audio(stereo, MONO_AUDIO_SPEC) assert result.ndim == 1 np.testing.assert_array_almost_equal(result, [0.0, 1.0]) def test_mono_spec_with_torch_stereo(self): """Mono spec should reduce stereo torch tensor to 1D.""" stereo = torch.tensor([[1.0, 2.0], [-1.0, 0.0]]) result = normalize_audio(stereo, MONO_AUDIO_SPEC) assert result.ndim == 1 torch.testing.assert_close(result, torch.tensor([0.0, 1.0])) def test_mono_passthrough_for_1d_numpy(self): """1D numpy array should pass through unchanged with mono spec.""" mono = np.array([1.0, 2.0, 3.0], dtype=np.float32) result = normalize_audio(mono, MONO_AUDIO_SPEC) assert result.ndim == 1 np.testing.assert_array_equal(result, mono) def test_mono_passthrough_for_1d_torch(self): """1D torch tensor should pass through unchanged with mono spec.""" mono = torch.tensor([1.0, 2.0, 3.0]) result = normalize_audio(mono, MONO_AUDIO_SPEC) assert result.ndim == 1 torch.testing.assert_close(result, mono) def test_first_channel_reduction(self): """FIRST reduction should take only the first channel.""" spec = AudioSpec(target_channels=1, channel_reduction=ChannelReduction.FIRST) stereo = np.array([[1.0, 2.0], [3.0, 4.0]], dtype=np.float32) result = normalize_audio(stereo, spec) np.testing.assert_array_equal(result, [1.0, 2.0]) def test_max_channel_reduction(self): """MAX reduction should take max across channels.""" spec = AudioSpec(target_channels=1, channel_reduction=ChannelReduction.MAX) stereo = np.array([[1.0, 4.0], [3.0, 2.0]], dtype=np.float32) result = normalize_audio(stereo, spec) np.testing.assert_array_equal(result, [3.0, 4.0]) def test_sum_channel_reduction(self): """SUM reduction should sum across channels.""" spec = AudioSpec(target_channels=1, channel_reduction=ChannelReduction.SUM) stereo = np.array([[1.0, 2.0], [3.0, 4.0]], dtype=np.float32) result = normalize_audio(stereo, spec) np.testing.assert_array_equal(result, [4.0, 6.0]) def test_invalid_3d_array_raises(self): """3D arrays should raise ValueError.""" audio_3d = np.random.randn(2, 3, 4).astype(np.float32) with pytest.raises(ValueError, match="Unsupported audio"): normalize_audio(audio_3d, MONO_AUDIO_SPEC) def test_channel_expansion_raises(self): """Expanding from mono to stereo should raise ValueError.""" mono = np.array([1.0, 2.0, 3.0], dtype=np.float32) spec = AudioSpec(target_channels=2) with pytest.raises(ValueError, match="Cannot expand"): normalize_audio(mono, spec) def test_time_channels_format_numpy(self): """Audio in (time, channels) format should be transposed to (channels, time). This handles the case where audio loaders like soundfile return (time, channels) format instead of (channels, time) like torchaudio. """ # Create audio in (time, channels) format: 1000 samples, 2 channels audio_time_channels = np.array( [[1.0, -1.0]] * 1000, # 1000 time steps, 2 channels dtype=np.float32, ) assert audio_time_channels.shape == (1000, 2) # (time, channels) result = normalize_audio(audio_time_channels, MONO_AUDIO_SPEC) # Should be reduced to mono 1D assert result.ndim == 1 assert result.shape == (1000,) # Mean of [1.0, -1.0] at each time step should be 0.0 np.testing.assert_array_almost_equal(result, np.zeros(1000)) def test_time_channels_format_torch(self): """Torch tensor in (time, channels) format should be transposed.""" # Create audio in (time, channels) format: 1000 samples, 2 channels audio_time_channels = torch.tensor( [[1.0, -1.0]] * 1000, # 1000 time steps, 2 channels ) assert audio_time_channels.shape == (1000, 2) # (time, channels) result = normalize_audio(audio_time_channels, MONO_AUDIO_SPEC) # Should be reduced to mono 1D assert result.ndim == 1 assert result.shape == (1000,) # Mean of [1.0, -1.0] at each time step should be 0.0 torch.testing.assert_close(result, torch.zeros(1000)) def test_channels_time_format_preserved(self): """Audio already in (channels, time) format should work correctly.""" # Create audio in standard (channels, time) format: 2 channels, 1000 samples audio_channels_time = np.array( [[1.0] * 1000, [-1.0] * 1000], # 2 channels, 1000 time steps dtype=np.float32, ) assert audio_channels_time.shape == (2, 1000) # (channels, time) result = normalize_audio(audio_channels_time, MONO_AUDIO_SPEC) # Should be reduced to mono 1D assert result.ndim == 1 assert result.shape == (1000,) # Mean of [1.0, -1.0] at each time step should be 0.0 np.testing.assert_array_almost_equal(result, np.zeros(1000)) def test_ambiguous_square_audio_numpy(self): """Square audio arrays (N, N) should use shape[0] > shape[1] heuristic. For a square array, shape[0] == shape[1], so no transpose happens and we assume (channels, time) format. """ # Create square audio: 4 channels, 4 samples audio_square = np.array( [ [1.0, 2.0, 3.0, 4.0], [5.0, 6.0, 7.0, 8.0], [9.0, 10.0, 11.0, 12.0], [13.0, 14.0, 15.0, 16.0], ], dtype=np.float32, ) assert audio_square.shape == (4, 4) result = normalize_audio(audio_square, MONO_AUDIO_SPEC) # Should be reduced to mono 1D with mean across channels (axis 0) assert result.ndim == 1 assert result.shape == (4,) # Mean across 4 channels: [1+5+9+13, 2+6+10+14, ...] / 4 expected = np.array([7.0, 8.0, 9.0, 10.0]) np.testing.assert_array_almost_equal(result, expected) # ============================================================ # Tests for MultiModalDataParser integration with target_channels # ============================================================ class TestMultiModalDataParserChannelNormalization: """Tests for MultiModalDataParser.target_channels integration. These tests verify that the target_channels parameter is properly used in the _parse_audio_data method to normalize audio channels. """ def test_parser_normalizes_stereo_to_mono(self): """Parser should normalize stereo to mono when target_channels=1.""" from vllm.multimodal.parse import MultiModalDataParser # Create parser with mono normalization enabled parser = MultiModalDataParser( target_sr=16000, target_channels=1, ) # Create stereo audio (simulating torchaudio output) stereo_audio = np.array( [[1.0, 1.0, 1.0], [-1.0, -1.0, -1.0]], # 2 channels, 3 samples dtype=np.float32, ) # Parse audio data result = parser._parse_audio_data((stereo_audio, 16000)) # Check that result is mono (1D) audio_item = result.get(0) assert audio_item.ndim == 1, f"Expected 1D mono audio, got {audio_item.ndim}D" assert audio_item.shape == (3,), f"Expected shape (3,), got {audio_item.shape}" # Channel average of [1, 1, 1] and [-1, -1, -1] should be [0, 0, 0] np.testing.assert_array_almost_equal(audio_item, np.zeros(3)) def test_parser_preserves_stereo_when_target_channels_none(self): """Parser should preserve stereo when target_channels=None.""" from vllm.multimodal.parse import MultiModalDataParser # Create parser without channel normalization parser = MultiModalDataParser( target_sr=16000, target_channels=None, ) # Create stereo audio stereo_audio = np.array( [[1.0, 1.0, 1.0], [-1.0, -1.0, -1.0]], dtype=np.float32, ) # Parse audio data result = parser._parse_audio_data((stereo_audio, 16000)) # Check that result preserves original shape (after resampling) audio_item = result.get(0) # When target_channels=None, stereo audio should be preserved assert audio_item.ndim == 2, f"Expected 2D stereo audio, got {audio_item.ndim}D" def test_parser_mono_passthrough_when_target_channels_1(self): """Parser should pass through mono audio unchanged when target_channels=1.""" from vllm.multimodal.parse import MultiModalDataParser # Create parser with mono normalization enabled parser = MultiModalDataParser( target_sr=16000, target_channels=1, ) # Create mono audio (already 1D) mono_audio = np.random.randn(16000).astype(np.float32) # Parse audio data result = parser._parse_audio_data((mono_audio, 16000)) # Check that result is still mono (1D) audio_item = result.get(0) assert audio_item.ndim == 1 assert audio_item.shape == (16000,) def test_parser_with_target_channels_2(self): """Parser should reduce 6-channel to 2-channel when target_channels=2.""" from vllm.multimodal.parse import MultiModalDataParser # Create parser with stereo target parser = MultiModalDataParser( target_sr=16000, target_channels=2, ) # Create 6-channel audio (5.1 surround) surround_audio = np.random.randn(6, 1000).astype(np.float32) # Parse audio data result = parser._parse_audio_data((surround_audio, 16000)) # Check that result is stereo (2 channels) audio_item = result.get(0) assert audio_item.ndim == 2 assert audio_item.shape[0] == 2 # 2 channels # ============================================================ # End-to-End Audio Pipeline Tests # ============================================================ class TestAudioPipelineE2E: """End-to-end tests for audio normalization in the full pipeline. These tests verify the complete flow from raw audio input through the MultiModalDataParser, simulating different audio loader formats. """ def test_stereo_audio_normalized_to_mono_e2e(self): """Full pipeline: stereo audio (torchaudio format) → mono output.""" from vllm.multimodal.parse import MultiModalDataParser # Simulate torchaudio output: (channels, time) format # Stereo audio with left channel = 1.0, right channel = -1.0 stereo_torchaudio = np.array( [[1.0] * 16000, [-1.0] * 16000], # 2 channels, 1 second at 16kHz dtype=np.float32, ) assert stereo_torchaudio.shape == (2, 16000) # Create parser with mono normalization (like Whisper models) parser = MultiModalDataParser( target_sr=16000, target_channels=1, ) # Process audio through the parser result = parser._parse_audio_data((stereo_torchaudio, 16000)) audio_output = result.get(0) # Verify output is mono 1D assert audio_output.ndim == 1, f"Expected 1D, got {audio_output.ndim}D" assert audio_output.shape == (16000,) # Verify channel averaging: mean of [1.0, -1.0] = 0.0 np.testing.assert_array_almost_equal(audio_output, np.zeros(16000), decimal=5) def test_soundfile_format_normalized_to_mono_e2e(self): """Full pipeline: soundfile format (time, channels) → mono output.""" from vllm.multimodal.parse import MultiModalDataParser # Simulate soundfile output: (time, channels) format # 16000 samples, 2 channels stereo_soundfile = np.array( [[0.5, -0.5]] * 16000, # Each row is [left, right] dtype=np.float32, ) assert stereo_soundfile.shape == (16000, 2) # Create parser with mono normalization parser = MultiModalDataParser( target_sr=16000, target_channels=1, ) # Process audio through the parser result = parser._parse_audio_data((stereo_soundfile, 16000)) audio_output = result.get(0) # Verify output is mono 1D assert audio_output.ndim == 1, f"Expected 1D, got {audio_output.ndim}D" assert audio_output.shape == (16000,) # Verify channel averaging: mean of [0.5, -0.5] = 0.0 np.testing.assert_array_almost_equal(audio_output, np.zeros(16000), decimal=5) def test_librosa_mono_passthrough_e2e(self): """Full pipeline: librosa mono format → preserved as mono.""" from vllm.multimodal.parse import MultiModalDataParser # Simulate librosa output: already mono (time,) format mono_librosa = np.random.randn(16000).astype(np.float32) assert mono_librosa.shape == (16000,) # Create parser with mono normalization parser = MultiModalDataParser( target_sr=16000, target_channels=1, ) # Process audio through the parser result = parser._parse_audio_data((mono_librosa, 16000)) audio_output = result.get(0) # Verify output is still mono 1D assert audio_output.ndim == 1 assert audio_output.shape == (16000,) # Verify audio content is preserved np.testing.assert_array_almost_equal(audio_output, mono_librosa) def test_multichannel_5_1_surround_to_mono_e2e(self): """Full pipeline: 5.1 surround (6 channels) → mono output.""" from vllm.multimodal.parse import MultiModalDataParser # Simulate 5.1 surround audio: 6 channels surround_audio = np.array( [ [1.0] * 8000, # Front Left [2.0] * 8000, # Front Right [3.0] * 8000, # Center [4.0] * 8000, # LFE (subwoofer) [5.0] * 8000, # Rear Left [6.0] * 8000, # Rear Right ], dtype=np.float32, ) assert surround_audio.shape == (6, 8000) # Create parser with mono normalization parser = MultiModalDataParser( target_sr=16000, target_channels=1, ) # Process audio through the parser result = parser._parse_audio_data((surround_audio, 16000)) audio_output = result.get(0) # Verify output is mono 1D assert audio_output.ndim == 1 # Verify channel averaging: mean of [1,2,3,4,5,6] = 3.5 expected_value = (1.0 + 2.0 + 3.0 + 4.0 + 5.0 + 6.0) / 6 np.testing.assert_array_almost_equal( audio_output, np.full(8000, expected_value), decimal=5 ) def test_torch_tensor_input_e2e(self): """Full pipeline: torch.Tensor stereo input → mono numpy output.""" from vllm.multimodal.parse import MultiModalDataParser # Simulate torch tensor input (from torchaudio) stereo_torch = torch.tensor( [[1.0] * 8000, [-1.0] * 8000], # 2 channels dtype=torch.float32, ) assert stereo_torch.shape == (2, 8000) # Create parser with mono normalization parser = MultiModalDataParser( target_sr=16000, target_channels=1, ) # Process audio through the parser # Note: Parser expects numpy, so we convert first (simulating real usage) result = parser._parse_audio_data((stereo_torch.numpy(), 16000)) audio_output = result.get(0) # Verify output is mono 1D numpy array assert audio_output.ndim == 1 assert isinstance(audio_output, np.ndarray) # Verify channel averaging np.testing.assert_array_almost_equal(audio_output, np.zeros(8000), decimal=5) def test_passthrough_preserves_stereo_e2e(self): """Full pipeline: stereo with target_channels=None → stereo preserved.""" from vllm.multimodal.parse import MultiModalDataParser # Stereo audio stereo_audio = np.array( [[1.0] * 8000, [-1.0] * 8000], dtype=np.float32, ) # Create parser WITHOUT mono normalization (passthrough) parser = MultiModalDataParser( target_sr=16000, target_channels=None, # Passthrough - no normalization ) # Process audio through the parser result = parser._parse_audio_data((stereo_audio, 16000)) audio_output = result.get(0) # Verify output preserves stereo (2D) assert audio_output.ndim == 2 assert audio_output.shape == (2, 8000) def test_resampling_with_channel_normalization_e2e(self): """Full pipeline: resample + channel normalize in single pass.""" from vllm.multimodal.parse import MultiModalDataParser # Stereo audio at 48kHz (common recording rate) stereo_48k = np.array( [[1.0] * 48000, [-1.0] * 48000], # 1 second at 48kHz dtype=np.float32, ) # Create parser with both resampling and mono normalization parser = MultiModalDataParser( target_sr=16000, # Resample to 16kHz target_channels=1, # Normalize to mono ) # Process audio through the parser result = parser._parse_audio_data((stereo_48k, 48000)) audio_output = result.get(0) # Verify output is mono 1D at target sample rate assert audio_output.ndim == 1 # After resampling from 48kHz to 16kHz, length should be ~16000 assert audio_output.shape[0] == 16000 def test_very_short_audio_e2e(self): """Full pipeline: very short audio (< 1 frame) handled correctly.""" from vllm.multimodal.parse import MultiModalDataParser # Very short stereo audio (10 samples) short_stereo = np.array( [[1.0] * 10, [-1.0] * 10], dtype=np.float32, ) parser = MultiModalDataParser( target_sr=16000, target_channels=1, ) result = parser._parse_audio_data((short_stereo, 16000)) audio_output = result.get(0) # Should still produce mono output assert audio_output.ndim == 1 assert audio_output.shape == (10,) np.testing.assert_array_almost_equal(audio_output, np.zeros(10)) # ============================================================ # Tests for Audio Chunking Utilities # ============================================================ class TestAudioChunking: """Tests for split_audio and find_split_point utilities in vllm.multimodal.audio.""" def test_split_audio_short_clip(self): """Audio shorter than max_clip_duration_s should not be split.""" # 10 seconds of audio at 16kHz audio = np.linspace(-1.0, 1.0, 160000, dtype=np.float32) chunks = split_audio( audio_data=audio, sample_rate=16000, max_clip_duration_s=30.0, overlap_duration_s=1.0, min_energy_window_size=1600, ) assert len(chunks) == 1 np.testing.assert_array_equal(chunks[0], audio) def test_split_audio_exact_length(self): """Audio exactly at max_clip_duration_s should not be split.""" # Exactly 30 seconds at 16kHz audio = np.linspace(-1.0, 1.0, 480000, dtype=np.float32) chunks = split_audio( audio_data=audio, sample_rate=16000, max_clip_duration_s=30.0, overlap_duration_s=1.0, min_energy_window_size=1600, ) assert len(chunks) == 1 np.testing.assert_array_equal(chunks[0], audio) def test_split_audio_long_clip(self): """Long audio should be split into multiple chunks.""" # 65 seconds of audio at 16kHz audio = np.linspace(-1.0, 1.0, 1040000, dtype=np.float32) chunks = split_audio( audio_data=audio, sample_rate=16000, max_clip_duration_s=30.0, overlap_duration_s=1.0, min_energy_window_size=1600, ) assert len(chunks) > 1 # First sample preserved assert chunks[0][0] == audio[0] # Last sample preserved assert chunks[-1][-1] == audio[-1] def test_split_audio_chunks_have_correct_length(self): """Each chunk (except last) should be approximately max_clip_duration_s.""" # 65 seconds of audio at 16kHz audio = np.linspace(-1.0, 1.0, 1040000, dtype=np.float32) chunks = split_audio( audio_data=audio, sample_rate=16000, max_clip_duration_s=30.0, overlap_duration_s=1.0, min_energy_window_size=1600, ) max_samples = int(30.0 * 16000) overlap_samples = int(1.0 * 16000) for chunk in chunks[:-1]: assert chunk.shape[0] >= max_samples - overlap_samples assert chunk.shape[0] <= max_samples def test_find_split_point_finds_quiet_region(self): """find_split_point should identify low-energy regions.""" from vllm.multimodal.audio import find_split_point # Create audio with a quiet section in the middle segment = np.ones(32000, dtype=np.float32) # Insert quiet region at sample 16000-17600 (100ms) segment[16000:17600] = 0.01 split_idx = find_split_point( wav=segment, start_idx=0, end_idx=32000, min_energy_window=1600, ) # Split should be in or near the quiet region assert 16000 <= split_idx <= 17600 def test_find_split_point_handles_uniform_audio(self): """find_split_point should handle uniform energy audio gracefully.""" from vllm.multimodal.audio import find_split_point segment = np.ones(32000, dtype=np.float32) * 0.5 split_idx = find_split_point( wav=segment, start_idx=0, end_idx=32000, min_energy_window=1600, ) assert 0 <= split_idx <= 32000 def test_find_split_point_silence(self): """find_split_point should prefer the quietest scanned window.""" from vllm.multimodal.audio import find_split_point # Deterministic signal: constant energy everywhere except silence. segment = np.ones(32000, dtype=np.float32) # Complete silence at 20000-21600. segment[20000:21600] = 0.0 split_idx = find_split_point( wav=segment, start_idx=16000, end_idx=28000, min_energy_window=1600, ) # Current implementation evaluates non-overlapping 1600-sample windows # from start_idx, so the quietest scanned window starts at 19200. assert split_idx == 19200 def test_split_audio_preserves_boundaries(self): """Verify first and last samples are preserved when chunking.""" audio = np.arange(1120000, dtype=np.float32) # 70s at 16kHz chunks = split_audio( audio_data=audio, sample_rate=16000, max_clip_duration_s=30.0, overlap_duration_s=1.0, min_energy_window_size=1600, ) assert chunks[0][0] == audio[0] assert chunks[-1][-1] == audio[-1] def test_split_audio_with_different_sample_rates(self): """Test chunking works with different sample rates.""" # 40 seconds at 8kHz audio_8k = np.linspace(-1.0, 1.0, 320000, dtype=np.float32) chunks = split_audio( audio_data=audio_8k, sample_rate=8000, max_clip_duration_s=30.0, overlap_duration_s=1.0, min_energy_window_size=800, ) assert len(chunks) >= 2 # 40 seconds at 48kHz audio_48k = np.linspace(-1.0, 1.0, 1920000, dtype=np.float32) chunks_48k = split_audio( audio_data=audio_48k, sample_rate=48000, max_clip_duration_s=30.0, overlap_duration_s=1.0, min_energy_window_size=4800, ) assert len(chunks_48k) >= 2
{ "repo_id": "vllm-project/vllm", "file_path": "tests/multimodal/test_audio.py", "license": "Apache License 2.0", "lines": 602, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/logging_utils/log_time.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """ Provides a timeslice logging decorator """ import functools import time def logtime(logger, msg=None): """ Logs the execution time of the decorated function. Always place it beneath other decorators. """ def _inner(func): @functools.wraps(func) def _wrapper(*args, **kwargs): start = time.perf_counter() result = func(*args, **kwargs) elapsed = time.perf_counter() - start prefix = ( f"Function '{func.__module__}.{func.__qualname__}'" if msg is None else msg ) logger.debug("%s: Elapsed time %.7f secs", prefix, elapsed) return result return _wrapper return _inner
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/logging_utils/log_time.py", "license": "Apache License 2.0", "lines": 27, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/v1/kv_connector/unit/test_offloading_connector.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import copy from collections.abc import Iterable, Iterator from dataclasses import dataclass from typing import Any from unittest.mock import MagicMock import pytest import torch from vllm import SamplingParams from vllm.config import KVTransferConfig, VllmConfig from vllm.distributed.kv_events import BlockRemoved, BlockStored from vllm.distributed.kv_transfer.kv_connector.v1 import KVConnectorRole from vllm.distributed.kv_transfer.kv_connector.v1.offloading_connector import ( OffloadingConnector, OffloadingConnectorMetadata, OffloadingConnectorStats, ) from vllm.forward_context import ForwardContext from vllm.utils.hashing import sha256 from vllm.v1.attention.backends.flash_attn import FlashAttentionBackend from vllm.v1.core.kv_cache_utils import ( BlockHash, get_request_block_hasher, init_none_hash, ) from vllm.v1.core.sched.scheduler import Scheduler from vllm.v1.kv_cache_interface import KVCacheConfig from vllm.v1.kv_offload.abstract import ( LoadStoreSpec, OffloadingEvent, OffloadingManager, PrepareStoreOutput, ) from vllm.v1.kv_offload.mediums import GPULoadStoreSpec from vllm.v1.kv_offload.spec import OffloadingSpec from vllm.v1.kv_offload.worker.worker import ( OffloadingHandler, TransferResult, TransferSpec, ) from vllm.v1.outputs import EMPTY_MODEL_RUNNER_OUTPUT, KVConnectorOutput from vllm.v1.request import Request, RequestStatus from .utils import ( EOS_TOKEN_ID, create_model_runner_output, create_scheduler, create_vllm_config, ) class MockLoadStoreSpec(LoadStoreSpec): def __init__(self, block_hashes: Iterable[BlockHash]): self.block_hashes: list[BlockHash] = list(block_hashes) @staticmethod def medium() -> str: return "Mock" def __repr__(self) -> str: return repr(self.block_hashes) class MockOffloadingHandler(OffloadingHandler): def __init__(self): self.transfer_specs: dict[int, TransferSpec] = {} self.completed_transfers: list[TransferResult] = [] self.waiting_jobs: set[int] = set() self.completed_jobs: list[int] = [] self.flushed_jobs: set[int] = set() def get_finished(self) -> list[TransferResult]: finished = self.completed_transfers self.completed_transfers = [] return finished def transfer_async(self, job_id: int, spec: TransferSpec) -> bool: self.transfer_specs[job_id] = spec self.waiting_jobs.add(job_id) return True def complete_jobs(self, job_ids: set[int]) -> None: for job_id in job_ids: if job_id in self.waiting_jobs: self.waiting_jobs.remove(job_id) self.completed_jobs.append(job_id) result = TransferResult( job_id=job_id, success=True, transfer_size=None, transfer_time=None, transfer_type=None, ) self.completed_transfers.append(result) def wait(self, job_ids: set[int]) -> None: self.flushed_jobs |= job_ids self.complete_jobs(job_ids) class MockOffloadingSpec(OffloadingSpec): def __init__(self, vllm_config: VllmConfig, kv_cache_config: KVCacheConfig): super().__init__(vllm_config, kv_cache_config) self.manager = MagicMock(spec=OffloadingManager) self.manager.lookup.return_value = 0 self.manager.prepare_load = lambda block_hashes: ( MockLoadStoreSpec(block_hashes) ) self.handler = MockOffloadingHandler() def get_manager(self) -> OffloadingManager: return self.manager def get_handlers( self, _, __ ) -> Iterator[tuple[type[LoadStoreSpec], type[LoadStoreSpec], OffloadingHandler]]: yield GPULoadStoreSpec, MockLoadStoreSpec, self.handler yield MockLoadStoreSpec, GPULoadStoreSpec, self.handler def complete_transfers(self): self.handler.complete_jobs(self.handler.waiting_jobs.copy()) def get_completed_transfers(self) -> list[TransferSpec]: specs = [ self.handler.transfer_specs[job_id] for job_id in self.handler.completed_jobs ] self.handler.completed_jobs.clear() return specs def get_flushed_transfers(self): specs = [ self.handler.transfer_specs[job_id] for job_id in self.handler.flushed_jobs ] self.handler.flushed_jobs.clear() return specs @dataclass class TransferSummary: gpu_block_indices: list[int] offload_addresses: list[Any] class RequestRunner: def __init__( self, offloaded_block_size: int, gpu_block_size: int, num_gpu_blocks: int ): self.offloaded_block_size: int = offloaded_block_size self.gpu_block_size: int = gpu_block_size self.num_gpu_blocks: int = num_gpu_blocks self.req_id: int = -1 vllm_config = create_vllm_config( block_size=gpu_block_size, max_num_batched_tokens=1000 ) vllm_config.kv_transfer_config = KVTransferConfig( kv_connector="OffloadingConnector", kv_role="kv_both", kv_connector_extra_config={ "spec_name": "MockOffloadingSpec", "spec_module_path": "tests.v1.kv_connector.unit.test_offloading_connector", # noqa: E501 "block_size": offloaded_block_size, }, ) self.scheduler: Scheduler = create_scheduler( vllm_config, num_blocks=num_gpu_blocks ) self.worker_connector = OffloadingConnector(vllm_config, KVConnectorRole.WORKER) # register worker kv_caches to enable OffloadingWorker creations self.worker_connector.register_cross_layers_kv_cache( kv_cache=torch.empty(0), attn_backend=FlashAttentionBackend, ) # extract connector of scheduler scheduler_connector = self.scheduler.connector assert scheduler_connector is not None assert isinstance(scheduler_connector, OffloadingConnector) self.scheduler_connector: OffloadingConnector = scheduler_connector # extract mocked OffloadingManager of scheduler connector connector_scheduler = scheduler_connector.connector_scheduler assert connector_scheduler is not None manager = connector_scheduler.manager assert isinstance(manager, MagicMock) self.manager: MagicMock = manager assert connector_scheduler.gpu_block_size == gpu_block_size assert connector_scheduler.offloaded_block_size == offloaded_block_size # extract OffloadingSpec of worker_connector connector_worker = self.worker_connector.connector_worker assert connector_worker is not None offloading_spec = connector_worker.spec assert isinstance(offloading_spec, MockOffloadingSpec) self.offloading_spec: MockOffloadingSpec = offloading_spec # mapping (offloading address) -> gpu_block_index self.offloaded: dict[Any, int] = {} self.completed_loads: list[TransferSummary] = [] self.completed_stores: list[TransferSummary] = [] self.flushed_gpu_block_indexes: set[int] = set() # maps {block_id: block_offset} self.gpu_block_index: dict[int, int] = {} init_none_hash(sha256) self._block_hasher = get_request_block_hasher(gpu_block_size, sha256) self._dummy_ctx: ForwardContext = ForwardContext( no_compile_layers={}, attn_metadata={}, virtual_engine=0, slot_mapping={}, ) def new_request(self, token_ids: list[int]): self.req_id += 1 sampling_params = SamplingParams(max_tokens=1000) sampling_params.update_from_generation_config({}, EOS_TOKEN_ID) req = Request( request_id=str(self.req_id), prompt_token_ids=token_ids, sampling_params=sampling_params, pooling_params=None, block_hasher=self._block_hasher, ) self.scheduler.add_request(req) def _parse_transfers(self): for transfer_spec in self.offloading_spec.get_flushed_transfers(): src_spec, dst_spec = transfer_spec assert isinstance(src_spec, GPULoadStoreSpec) for block_id in src_spec.block_ids: self.flushed_gpu_block_indexes.add( self.gpu_block_index[block_id.item()] ) block_size_factor = self.offloaded_block_size // self.gpu_block_size for transfer_spec in self.offloading_spec.get_completed_transfers(): src_spec, dst_spec = transfer_spec if isinstance(src_spec, GPULoadStoreSpec): store = True gpu_spec = src_spec offload_spec = dst_spec else: store = False gpu_spec = dst_spec offload_spec = src_spec assert isinstance(offload_spec, MockLoadStoreSpec) assert isinstance(gpu_spec, GPULoadStoreSpec) gpu_block_indices: list[int] = [] for block_id in gpu_spec.block_ids: gpu_block_indices.append(self.gpu_block_index[block_id.item()]) # list of (block_hash, sub_block_offset) offload_addresses: list[Any] = [] for block_hash in offload_spec.block_hashes: for sub_block_idx in range(block_size_factor): offload_addresses.append((block_hash, sub_block_idx)) if store: assert len(gpu_block_indices) == len(offload_addresses) self.completed_stores.append( TransferSummary(gpu_block_indices, offload_addresses) ) else: remainder_sub_block_count = len(offload_addresses) - len( gpu_block_indices ) assert remainder_sub_block_count >= 0 assert remainder_sub_block_count < block_size_factor offload_addresses = offload_addresses[remainder_sub_block_count:] self.completed_loads.append( TransferSummary(gpu_block_indices, offload_addresses) ) def _update_gpu_block_idx(self): for blocks in self.scheduler.kv_cache_manager.coordinator.single_type_managers[ 0 ].req_to_blocks.values(): for block_idx, block in enumerate(blocks): self.gpu_block_index[block.block_id] = block_idx def _run(self, decoded_tokens: list[int], complete_transfers: bool): """ Runs multiple engine (scheduler + worker) steps. Assumes a single request is running. Args: decoded_tokens: the tokens to yield at each step. complete_transfers: complete transfers immediately """ tokens_iter = iter(decoded_tokens) token_id = next(tokens_iter, None) while True: assert self.scheduler.requests scheduler_output = self.scheduler.schedule() self._update_gpu_block_idx() kv_connector_metadata = scheduler_output.kv_connector_metadata assert kv_connector_metadata is not None assert isinstance(kv_connector_metadata, OffloadingConnectorMetadata) if scheduler_output.preempted_req_ids: self.worker_connector.handle_preemptions( scheduler_output.preempted_req_ids ) self.worker_connector.bind_connector_metadata(kv_connector_metadata) self.worker_connector.start_load_kv(self._dummy_ctx) if scheduler_output.total_num_scheduled_tokens > 0: self.worker_connector.wait_for_save() if complete_transfers: self.offloading_spec.complete_transfers() finished_sending, finished_recving = self.worker_connector.get_finished( scheduler_output.finished_req_ids ) self.worker_connector.clear_connector_metadata() model_runner_output = create_model_runner_output( reqs=self.scheduler.running, finished_sending=finished_sending, finished_recving=finished_recving, token_id=token_id or 0, ) prev_token_id = token_id if self.scheduler.running: token_id = next(tokens_iter, None) self.scheduler.update_from_output(scheduler_output, model_runner_output) if ( prev_token_id == EOS_TOKEN_ID and prev_token_id != token_id and self.scheduler.requests ): # continue for one more step to allow offloading to kick off continue if token_id is None: break self._parse_transfers() # run one more step to update finished stored if EOS_TOKEN_ID in decoded_tokens: assert not self.scheduler.running while self.scheduler.requests: scheduler_output = self.scheduler.schedule() finished_sending, finished_recving = self.worker_connector.get_finished( scheduler_output.finished_req_ids ) assert not finished_recving model_runner_output = copy.deepcopy(EMPTY_MODEL_RUNNER_OUTPUT) model_runner_output.kv_connector_output = KVConnectorOutput( finished_sending=finished_sending ) self.scheduler.update_from_output(scheduler_output, model_runner_output) def run( self, decoded_tokens: list[int], complete_transfers: bool = True, expected_stored_gpu_block_indexes: tuple[int, ...] = (), expected_loaded_gpu_block_indexes: tuple[int, ...] = (), expected_flushed_gpu_block_indexes: tuple[int, ...] = (), ): """ Runs multiple engine (scheduler + worker) steps. Assumes a single request is running. Args: decoded_tokens: the tokens to yield at each step. complete_transfers: complete transfers immediately expected_stored_gpu_block_indexes: GPU block indexes that are expected to be written during the run. expected_loaded_gpu_block_indexes: GPU block indexes that are expected to be loaded during the run. expected_flushed_gpu_block_indexes: GPU block indexes that are expected to be flushed during the run. """ self.manager.reset_mock() self._run(decoded_tokens, complete_transfers) loaded_gpu_block_indexes: set[int] = set() for transfer in self.completed_loads: for gpu_block_idx, offloaded_address in zip( transfer.gpu_block_indices, transfer.offload_addresses ): loaded_gpu_block_indexes.add(gpu_block_idx) assert gpu_block_idx == self.offloaded[offloaded_address] assert set(expected_loaded_gpu_block_indexes) == loaded_gpu_block_indexes self.completed_loads.clear() stored_gpu_block_indexes: set[int] = set() for transfer in self.completed_stores: for gpu_block_idx, offloaded_address in zip( transfer.gpu_block_indices, transfer.offload_addresses ): stored_gpu_block_indexes.add(gpu_block_idx) self.offloaded[offloaded_address] = gpu_block_idx assert set(expected_stored_gpu_block_indexes) == stored_gpu_block_indexes self.completed_stores.clear() assert set(expected_flushed_gpu_block_indexes) == self.flushed_gpu_block_indexes self.flushed_gpu_block_indexes.clear() @pytest.fixture def request_runner(): runners = [] def runner_factory(offloaded_block_size, gpu_block_size, num_gpu_blocks): runner = RequestRunner( offloaded_block_size=offloaded_block_size, gpu_block_size=gpu_block_size, num_gpu_blocks=num_gpu_blocks, ) runners.append(runner) return runner yield runner_factory # pass factory to the test def generate_store_output(block_hashes: Iterable[BlockHash]): block_hashes = list(block_hashes) return PrepareStoreOutput( block_hashes_to_store=list(block_hashes), store_spec=MockLoadStoreSpec(block_hashes), block_hashes_evicted=[], ) def test_offloading_connector(request_runner): offloaded_block_size = 12 gpu_block_size = 4 num_gpu_blocks = 100 block_size_factor = offloaded_block_size // gpu_block_size runner = request_runner( offloaded_block_size=offloaded_block_size, gpu_block_size=gpu_block_size, num_gpu_blocks=num_gpu_blocks, ) # 3 blocks, store just the middle block (skip first and last) # blocks = [0, 1, 2], [3, 4, 5], [6, 7, 8] runner.new_request(token_ids=[0] * offloaded_block_size * 3) runner.manager.prepare_store.side_effect = ( lambda block_hashes: generate_store_output(list(block_hashes)[1:2]) ) runner.run(decoded_tokens=[0]) # add block missing 1 token -> no offload runner.run( decoded_tokens=[0] * (offloaded_block_size - 1), expected_stored_gpu_block_indexes=(3, 4, 5), ) runner.manager.prepare_store.assert_not_called() # +1 token -> single block, fail prepare_store runner.manager.prepare_store.side_effect = lambda block_hashes: None runner.run(decoded_tokens=[0]) runner.manager.prepare_store.assert_called() # 1 more block, now set block_hashes_to_store = [] runner.manager.prepare_store.side_effect = ( lambda block_hashes: generate_store_output([]) ) runner.run(decoded_tokens=[0] * offloaded_block_size) # 1 more block, now check touch was called with all 6 blocks runner.manager.prepare_store.side_effect = ( lambda block_hashes: generate_store_output(block_hashes) ) runner.run(decoded_tokens=[0] * offloaded_block_size) runner.manager.touch.assert_called() block_hashes1 = list(runner.manager.touch.call_args.args[0]) assert len(block_hashes1) == 6 # terminate request runner.run( decoded_tokens=[EOS_TOKEN_ID], expected_stored_gpu_block_indexes=(15, 16, 17), ) # create a new request differing only on the last token runner.new_request(token_ids=[0] * (offloaded_block_size * 6 - 1) + [1]) runner.run(decoded_tokens=[0]) runner.manager.touch.assert_called() block_hashes2 = list(runner.manager.touch.call_args.args[0]) assert len(block_hashes2) == 6 # verify hashes are the same, except for the last block assert block_hashes1[:5] == block_hashes2[:5] assert block_hashes1[5] != block_hashes2[5] # terminate request runner.run( decoded_tokens=[EOS_TOKEN_ID], expected_stored_gpu_block_indexes=tuple(range(6 * block_size_factor)), ) # full_block_tokens - num_computed_tokens < offloaded_block_size runner.new_request( token_ids=[0] * gpu_block_size + [1] * (offloaded_block_size - gpu_block_size) ) runner.manager.prepare_store.side_effect = ( lambda block_hashes: generate_store_output([]) ) runner.run(decoded_tokens=[EOS_TOKEN_ID]) runner.manager.lookup.assert_not_called() # single block lookup with no hits runner.new_request(token_ids=[1] * offloaded_block_size) runner.manager.prepare_store.side_effect = ( lambda block_hashes: generate_store_output([]) ) runner.run(decoded_tokens=[EOS_TOKEN_ID]) runner.manager.lookup.assert_called() assert len(list(runner.manager.lookup.call_args.args[0])) == 1 # single block lookup with a hit runner.scheduler.reset_prefix_cache() runner.new_request(token_ids=[0] * offloaded_block_size) runner.manager.prepare_store.side_effect = ( lambda block_hashes: generate_store_output([]) ) runner.manager.lookup.return_value = 1 runner.run( decoded_tokens=[EOS_TOKEN_ID], expected_loaded_gpu_block_indexes=(0, 1, 2) ) # single block lookup with a hit in a middle block runner.new_request( token_ids=[0] * offloaded_block_size * 2 + [1] * offloaded_block_size ) runner.manager.prepare_store.side_effect = ( lambda block_hashes: generate_store_output([]) ) runner.manager.lookup.return_value = 1 runner.run( decoded_tokens=[EOS_TOKEN_ID], expected_loaded_gpu_block_indexes=(3, 4, 5) ) # test take_events def to_hashes(int_hashes: list[int]) -> list[BlockHash]: return [BlockHash(str(i).encode()) for i in int_hashes] def take_events() -> Iterable[OffloadingEvent]: yield OffloadingEvent( block_hashes=to_hashes([1, 2, 3]), block_size=16, medium="A", removed=False ) yield OffloadingEvent( block_hashes=to_hashes([4, 5, 6]), block_size=32, medium="B", removed=True ) runner.manager.take_events.side_effect = take_events events = list(runner.scheduler_connector.take_events()) assert len(events) == 2 event = events[0] assert isinstance(event, BlockStored) assert event.block_hashes == to_hashes([1, 2, 3]) assert event.block_size == 16 assert event.medium == "A" assert event.token_ids == [] assert event.parent_block_hash is None assert event.lora_id is None assert event.lora_name is None event = events[1] assert isinstance(event, BlockRemoved) assert event.block_hashes == to_hashes([4, 5, 6]) assert event.medium == "B" def test_request_preemption(request_runner): offloaded_block_size = 12 gpu_block_size = 4 num_gpu_blocks = 100 runner = request_runner( offloaded_block_size=offloaded_block_size, gpu_block_size=gpu_block_size, num_gpu_blocks=num_gpu_blocks, ) free_block_queue = runner.scheduler.kv_cache_manager.block_pool.free_block_queue num_free_blocks_empty = free_block_queue.num_free_blocks # 2 blocks, store all, without flushing # blocks = [0, 1, 2], [3, 4, 5] runner.new_request(token_ids=[0] * offloaded_block_size * 2) runner.manager.prepare_store.side_effect = ( lambda block_hashes: generate_store_output(block_hashes) ) runner.run( decoded_tokens=[0], complete_transfers=False, ) # decode 2 more blocks - 1 gpu block, storing [6, 7, 8] (no flush) runner.manager.prepare_store.side_effect = ( lambda block_hashes: generate_store_output(block_hashes) ) runner.run( decoded_tokens=[0] * (2 * offloaded_block_size - gpu_block_size), complete_transfers=False, ) # simulate KV cache running out of space free_block_queue.num_free_blocks = 0 # request should be preempted now runner.run( decoded_tokens=[], complete_transfers=False, expected_flushed_gpu_block_indexes=(0, 1, 2, 3, 4, 5, 6, 7, 8), expected_stored_gpu_block_indexes=(0, 1, 2, 3, 4, 5, 6, 7, 8), ) # restore KV cache space and reset GPU prefix cache free_block_queue.num_free_blocks = num_free_blocks_empty runner.scheduler.reset_prefix_cache() # request should now return from preemption # re-load [0, ..., 8] from the CPU and store [9, 10, 11] runner.manager.lookup.return_value = 3 runner.manager.prepare_store.side_effect = ( lambda block_hashes: generate_store_output(block_hashes) ) runner.run( decoded_tokens=[0] * gpu_block_size, expected_loaded_gpu_block_indexes=(0, 1, 2, 3, 4, 5, 6, 7, 8), ) runner.run( decoded_tokens=[EOS_TOKEN_ID], expected_stored_gpu_block_indexes=(9, 10, 11), ) def test_concurrent_lookups_of_the_same_prefix(request_runner): offloaded_block_size = 12 gpu_block_size = 4 num_gpu_blocks = 100 runner = request_runner( offloaded_block_size=offloaded_block_size, gpu_block_size=gpu_block_size, num_gpu_blocks=num_gpu_blocks, ) # store 1 blocks runner.new_request(token_ids=[0] * offloaded_block_size) runner.manager.prepare_store.side_effect = ( lambda block_hashes: generate_store_output(block_hashes) ) runner.run( decoded_tokens=[EOS_TOKEN_ID], expected_stored_gpu_block_indexes=(0, 1, 2), ) # start a request to load the first block, but don't complete runner.scheduler.reset_prefix_cache() runner.new_request(token_ids=[0] * offloaded_block_size) runner.manager.lookup.return_value = 1 runner.run( decoded_tokens=[], complete_transfers=False, ) # request triggered a load transfer_jobs = list(runner.offloading_spec.handler.transfer_specs) assert transfer_jobs # start a new request to load the same first block runner.new_request(token_ids=[0] * offloaded_block_size) runner.manager.lookup.return_value = 1 runner.run( decoded_tokens=[], complete_transfers=False, ) # request did not trigger a load assert transfer_jobs == list(runner.offloading_spec.handler.transfer_specs) # complete transfers runner.manager.prepare_store.side_effect = ( lambda block_hashes: generate_store_output([]) ) runner.run( decoded_tokens=[EOS_TOKEN_ID], expected_loaded_gpu_block_indexes=(0, 1, 2), ) # second request will use the GPU prefix cache assert transfer_jobs == list(runner.offloading_spec.handler.transfer_specs) def test_abort_loading_requests(request_runner): offloaded_block_size = 12 gpu_block_size = 4 num_gpu_blocks = 100 runner = request_runner( offloaded_block_size=offloaded_block_size, gpu_block_size=gpu_block_size, num_gpu_blocks=num_gpu_blocks, ) # store 1 blocks runner.new_request(token_ids=[0] * offloaded_block_size) runner.manager.prepare_store.side_effect = ( lambda block_hashes: generate_store_output(block_hashes) ) runner.run( decoded_tokens=[EOS_TOKEN_ID], expected_stored_gpu_block_indexes=(0, 1, 2), ) # start a request to load the first block, but don't complete runner.scheduler.reset_prefix_cache() runner.new_request(token_ids=[0] * offloaded_block_size) runner.manager.lookup.return_value = 1 runner.run( decoded_tokens=[], complete_transfers=False, ) # request triggered a load transfer_jobs = list(runner.offloading_spec.handler.transfer_specs) assert transfer_jobs # abort request req_id = str(runner.req_id) runner.scheduler.finish_requests((req_id,), RequestStatus.FINISHED_ABORTED) # verify request is not deleted assert req_id in runner.scheduler.requests # complete loading request runner.run( decoded_tokens=[], expected_loaded_gpu_block_indexes=(0, 1, 2), ) # assert request is deleted assert req_id not in runner.scheduler.requests class TestOffloadingConnectorStats: """Tests for OffloadingConnector stats reconstruction and operations.""" def test_build_kv_connector_stats_with_none(self): """Test that build_kv_connector_stats returns empty stats when given None.""" stats = OffloadingConnector.build_kv_connector_stats(data=None) assert stats is not None assert isinstance(stats, OffloadingConnectorStats) assert len(stats.data) == 0 assert stats.is_empty() def test_build_kv_connector_stats_with_empty_dict(self): """Test that build_kv_connector_stats returns empty stats with empty dict.""" stats = OffloadingConnector.build_kv_connector_stats(data={}) assert stats is not None assert isinstance(stats, OffloadingConnectorStats) assert len(stats.data) == 0 assert stats.is_empty() def test_build_kv_connector_stats_reconstructs_offload_stats(self): """Test that OffloadingConnector stats are properly reconstructed with correct data.""" serialized_data = { "CPU_to_GPU": [ {"op_size": 16, "op_time": 1.0}, {"op_size": 8, "op_time": 0.5}, ], "GPU_to_CPU": [ {"op_size": 1, "op_time": 0.1}, {"op_size": 2, "op_time": 0.2}, ], } stats = OffloadingConnector.build_kv_connector_stats(data=serialized_data) offload_connector_stats = stats assert isinstance(offload_connector_stats, OffloadingConnectorStats) assert offload_connector_stats.data["CPU_to_GPU"] == [ {"op_size": 16, "op_time": 1.0}, {"op_size": 8, "op_time": 0.5}, ] assert offload_connector_stats.data["GPU_to_CPU"] == [ {"op_size": 1, "op_time": 0.1}, {"op_size": 2, "op_time": 0.2}, ] def test_aggregate_same_connector(self): """Test aggregating stats from the same connector type.""" stats1 = OffloadingConnectorStats( data={ "CPU_to_GPU": [ {"op_size": 16, "op_time": 1.0}, {"op_size": 8, "op_time": 0.5}, ], "GPU_to_CPU": [ {"op_size": 1, "op_time": 0.1}, {"op_size": 2, "op_time": 0.2}, ], } ) stats2 = OffloadingConnectorStats( data={ "CPU_to_GPU": [ {"op_size": 3, "op_time": 0.2}, {"op_size": 7, "op_time": 0.9}, ], "GPU_to_CPU": [{"op_size": 16, "op_time": 2}], } ) result = stats1.aggregate(stats2) assert result is stats1 # Should return self offload_connector_stats = result assert offload_connector_stats.data["CPU_to_GPU"] == [ {"op_size": 16, "op_time": 1.0}, {"op_size": 8, "op_time": 0.5}, {"op_size": 3, "op_time": 0.2}, {"op_size": 7, "op_time": 0.9}, ] assert offload_connector_stats.data["GPU_to_CPU"] == [ {"op_size": 1, "op_time": 0.1}, {"op_size": 2, "op_time": 0.2}, {"op_size": 16, "op_time": 2}, ] def test_reduce(self): """Test that reduce() correctly reduces all nested connector stats.""" stats = OffloadingConnectorStats( data={ "CPU_to_GPU": [ {"op_size": 16, "op_time": 1.0}, {"op_size": 8, "op_time": 0.5}, {"op_size": 3, "op_time": 0.2}, {"op_size": 7, "op_time": 0.9}, ], "GPU_to_CPU": [ {"op_size": 1, "op_time": 0.1}, {"op_size": 2, "op_time": 0.2}, {"op_size": 16, "op_time": 2}, ], } ) reduced = stats.reduce() assert isinstance(reduced, dict) # Check that the stats were reduced (should have aggregated values) assert "CPU_to_GPU_total_bytes" in reduced assert "CPU_to_GPU_total_time" in reduced assert "GPU_to_CPU_total_bytes" in reduced assert "GPU_to_CPU_total_time" in reduced assert reduced["CPU_to_GPU_total_bytes"] == 34 assert reduced["CPU_to_GPU_total_time"] == 2.6 assert reduced["GPU_to_CPU_total_time"] == 2.3 assert reduced["GPU_to_CPU_total_bytes"] == 19 def test_reset(self): """Test that reset() resets all nested connector stats.""" offload_connector_stats = OffloadingConnectorStats( data={ "CPU_to_GPU": [ {"op_size": 3, "op_time": 0.2}, {"op_size": 7, "op_time": 0.9}, ], "GPU_to_CPU": [{"op_size": 16, "op_time": 2}], } ) assert not offload_connector_stats.is_empty() offload_connector_stats.reset() # After reset, stats should be empty assert offload_connector_stats.is_empty() assert len(offload_connector_stats.data) == 0
{ "repo_id": "vllm-project/vllm", "file_path": "tests/v1/kv_connector/unit/test_offloading_connector.py", "license": "Apache License 2.0", "lines": 763, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from collections import defaultdict from collections.abc import Iterable from dataclasses import dataclass from itertools import islice from typing import Any import torch from vllm.config import VllmConfig, get_layers_from_vllm_config from vllm.distributed.kv_events import BlockRemoved, BlockStored, KVCacheEvent from vllm.distributed.kv_transfer.kv_connector.utils import yield_req_data from vllm.distributed.kv_transfer.kv_connector.v1 import ( KVConnectorBase_V1, KVConnectorRole, ) from vllm.distributed.kv_transfer.kv_connector.v1.base import KVConnectorMetadata from vllm.distributed.kv_transfer.kv_connector.v1.metrics import ( KVConnectorPromMetrics, KVConnectorStats, PromMetric, PromMetricT, ) from vllm.forward_context import ForwardContext from vllm.logger import init_logger from vllm.model_executor.layers.attention import Attention from vllm.v1.attention.backend import AttentionBackend, AttentionMetadata from vllm.v1.core.kv_cache_manager import KVCacheBlocks from vllm.v1.core.kv_cache_utils import BlockHash from vllm.v1.core.sched.output import SchedulerOutput from vllm.v1.kv_cache_interface import KVCacheConfig from vllm.v1.kv_offload.abstract import OffloadingManager from vllm.v1.kv_offload.factory import OffloadingSpecFactory from vllm.v1.kv_offload.mediums import GPULoadStoreSpec from vllm.v1.kv_offload.spec import OffloadingSpec from vllm.v1.kv_offload.worker.worker import ( OffloadingWorker, TransferSpec, TransferType, ) from vllm.v1.outputs import KVConnectorOutput from vllm.v1.request import Request ReqId = str logger = init_logger(__name__) @dataclass class OffloadingOperationMetrics: op_size: int op_time: float @dataclass class OffloadingConnectorStats(KVConnectorStats): def __post_init__(self): if not self.data: # Empty container init, no data is passed in. self.reset() def reset(self): self.data: dict[str, list[OffloadingOperationMetrics]] = {} def aggregate(self, other: KVConnectorStats) -> KVConnectorStats: if not other.is_empty(): for k, v in other.data.items(): if k not in self.data: self.data[k] = v else: accumulator = self.data[k] assert isinstance(accumulator, list) accumulator.extend(v) return self def reduce(self) -> dict[str, int | float]: """ Reduce the observations collected during a time interval to one or more representative values (eg avg/median/sum of the series). This is meant to be called by the logger to produce a summary of the stats for the last time interval. """ return_dict: dict[str, int | float] = {} for transfer_type, ops_list in self.data.items(): assert isinstance(ops_list, list) total_bytes = 0 total_time = 0.0 for op in ops_list: assert isinstance(op, dict) total_bytes += op["op_size"] total_time += op["op_time"] return_dict[f"{transfer_type}_total_bytes"] = total_bytes return_dict[f"{transfer_type}_total_time"] = total_time return return_dict def is_empty(self) -> bool: return not self.data def record_transfer(self, num_bytes: int, time: float, transfer_type: TransferType): src, dst = transfer_type transfer_type_key = src + "_to_" + dst op = OffloadingOperationMetrics(num_bytes, time) if transfer_type_key in self.data: self.data[transfer_type_key].append(op) else: self.data[transfer_type_key] = [op] @dataclass class OffloadingConnectorMetadata(KVConnectorMetadata): reqs_to_load: dict[ReqId, TransferSpec] reqs_to_store: dict[ReqId, TransferSpec] class OffloadingConnector(KVConnectorBase_V1): @property def prefer_cross_layer_blocks(self) -> bool: return True def __init__( self, vllm_config: VllmConfig, role: KVConnectorRole, kv_cache_config: KVCacheConfig | None = None, ): super().__init__(vllm_config, role, kv_cache_config) spec = OffloadingSpecFactory.create_spec(vllm_config, kv_cache_config) self.connector_scheduler: OffloadingConnectorScheduler | None = None self.connector_worker: OffloadingConnectorWorker | None = None if role == KVConnectorRole.SCHEDULER: self.connector_scheduler = OffloadingConnectorScheduler(spec) elif role == KVConnectorRole.WORKER: self.connector_worker = OffloadingConnectorWorker(spec) def register_kv_caches(self, kv_caches: dict[str, torch.Tensor]): assert self.connector_worker is not None self.connector_worker.register_kv_caches(kv_caches) def register_cross_layers_kv_cache( self, kv_cache: torch.Tensor, attn_backend: type[AttentionBackend] ): assert self.connector_worker is not None self.connector_worker.register_cross_layers_kv_cache(kv_cache, attn_backend) def handle_preemptions(self, preempted_req_ids: set[str]): assert self.connector_worker is not None self.connector_worker.handle_preemptions(preempted_req_ids) def start_load_kv(self, forward_context: "ForwardContext", **kwargs) -> None: assert self.connector_worker is not None assert isinstance(self._connector_metadata, OffloadingConnectorMetadata) self.connector_worker.start_kv_transfers(self._connector_metadata) def wait_for_layer_load(self, layer_name: str) -> None: pass def save_kv_layer( self, layer_name: str, kv_layer: torch.Tensor, attn_metadata: "AttentionMetadata", **kwargs, ) -> None: pass def wait_for_save(self): assert self.connector_worker is not None assert isinstance(self._connector_metadata, OffloadingConnectorMetadata) self.connector_worker.prepare_store_kv(self._connector_metadata) def get_finished(self, finished_req_ids: set[str]) -> tuple[set[str], set[str]]: assert self.connector_worker is not None return self.connector_worker.get_finished(finished_req_ids) def get_num_new_matched_tokens( self, request: "Request", num_computed_tokens: int ) -> tuple[int | None, bool]: assert self.connector_scheduler is not None return self.connector_scheduler.get_num_new_matched_tokens( request, num_computed_tokens ) def update_state_after_alloc( self, request: "Request", blocks: "KVCacheBlocks", num_external_tokens: int ): assert self.connector_scheduler is not None return self.connector_scheduler.update_state_after_alloc( request, blocks, num_external_tokens ) def build_connector_meta( self, scheduler_output: SchedulerOutput ) -> KVConnectorMetadata: assert self.connector_scheduler is not None return self.connector_scheduler.build_connector_meta(scheduler_output) def update_connector_output(self, connector_output: KVConnectorOutput): assert self.connector_scheduler is not None self.connector_scheduler.update_connector_output(connector_output) def request_finished( self, request: "Request", block_ids: list[int], ) -> tuple[bool, dict[str, Any] | None]: assert self.connector_scheduler is not None return self.connector_scheduler.request_finished(request, block_ids) def take_events(self) -> Iterable[KVCacheEvent]: assert self.connector_scheduler is not None return self.connector_scheduler.take_events() def get_kv_connector_stats(self) -> KVConnectorStats | None: if self.connector_worker is None: return None # We only emit stats from the worker-side return self.connector_worker.get_kv_connector_stats() @classmethod def build_kv_connector_stats( cls, data: dict[str, Any] | None = None ) -> KVConnectorStats | None: return ( OffloadingConnectorStats(data=data) if data is not None else OffloadingConnectorStats() ) @classmethod def build_prom_metrics( cls, vllm_config: VllmConfig, metric_types: dict[type[PromMetric], type[PromMetricT]], labelnames: list[str], per_engine_labelvalues: dict[int, list[object]], ) -> KVConnectorPromMetrics: return OffloadPromMetrics( vllm_config, metric_types, labelnames, per_engine_labelvalues ) class OffloadingConnectorScheduler: """Implementation of Scheduler side methods""" def __init__(self, spec: OffloadingSpec): self.gpu_block_size = spec.gpu_block_size self.offloaded_block_size = spec.offloaded_block_size self.block_size_factor = self.offloaded_block_size // self.gpu_block_size self.manager: OffloadingManager = spec.get_manager() self._requests: dict[ReqId, Request] = {} # list of GPU block IDs per request self._request_block_ids: dict[ReqId, list[int]] = {} # requests to load for the current scheduler step self._reqs_to_load: dict[ReqId, TransferSpec] = {} # request blocks are stored in order # index of next block (of size offloaded_block_size) to offload self._next_stored_block_idx: dict[ReqId, int] = {} # if GPU prefix caching is enabled, # track loaded blocks to avoid redundant loads self._blocks_being_loaded: set[BlockHash] | None = ( set() if spec.vllm_config.cache_config.enable_prefix_caching else None ) # request ID -> set(block hashes being stored/load) self._reqs_being_stored = defaultdict[ReqId, set[BlockHash]](set) self._reqs_being_loaded = defaultdict[ReqId, set[BlockHash]](set) def _get_block_hashes( self, req: Request, start_idx: int = 0, end_idx: int | None = None, ) -> Iterable[BlockHash]: return islice( req.block_hashes, self.block_size_factor * start_idx + self.block_size_factor - 1, self.block_size_factor * end_idx if end_idx else None, self.block_size_factor, ) def get_num_new_matched_tokens( self, request: Request, num_computed_tokens: int ) -> tuple[int | None, bool]: """ Get number of new tokens that can be loaded beyond the num_computed_tokens. Args: request (Request): the request object. num_computed_tokens (int): the number of locally computed tokens for this request Returns: A tuple with the following elements: - The number of tokens that can be loaded beyond what is already computed. If None, it means that the connector needs more time to determine the number of matched tokens, and the scheduler should query for this request again later. - `True` if tokens will be loaded asynchronously (between scheduler steps). """ num_blocks = request.num_tokens // self.offloaded_block_size assert len(request.block_hashes) // self.block_size_factor == num_blocks block_hashes = self._get_block_hashes(request) self.manager.touch(block_hashes) full_block_tokens = self.offloaded_block_size * num_blocks if full_block_tokens - num_computed_tokens < self.offloaded_block_size: # we can load less than a block, skip return 0, False start_block_idx = num_computed_tokens // self.offloaded_block_size hits = self.manager.lookup( self._get_block_hashes(request, start_idx=start_block_idx) ) if hits is None: # indicates a lookup that should be tried later return None, False if hits == 0: return 0, False num_hit_tokens = ( self.offloaded_block_size * (start_block_idx + hits) - num_computed_tokens ) logger.debug( "Request %s hit %s offloaded tokens after %s GPU hit tokens", request.request_id, num_hit_tokens, num_computed_tokens, ) if num_hit_tokens < self.offloaded_block_size: return 0, False if self._blocks_being_loaded: block_hashes = self._get_block_hashes( request, start_idx=start_block_idx, end_idx=start_block_idx + hits ) if any( block_hash in self._blocks_being_loaded for block_hash in block_hashes ): # hit blocks are being loaded, delay request logger.debug( "Delaying request %s since some of its blocks are already" " being loaded", request.request_id, ) return None, False return num_hit_tokens, True def update_state_after_alloc( self, request: Request, blocks: KVCacheBlocks, num_external_tokens: int ): self._requests[request.request_id] = request # the block ids are updated in _get_reqs_to_store self._request_block_ids[request.request_id] = [] if num_external_tokens == 0: return block_groups = blocks.get_block_ids() block_ids = block_groups[0] num_computed_gpu_blocks = sum( block.block_hash is not None for block in blocks.blocks[0] ) num_computed_tokens = num_computed_gpu_blocks * self.gpu_block_size full_block_tokens = num_computed_tokens + num_external_tokens assert full_block_tokens % self.offloaded_block_size == 0 num_pending_gpu_blocks = len(block_ids) - num_computed_gpu_blocks assert num_external_tokens == num_pending_gpu_blocks * self.gpu_block_size start_block_idx = num_computed_tokens // self.offloaded_block_size num_blocks = full_block_tokens // self.offloaded_block_size assert len(request.block_hashes) // self.block_size_factor >= num_blocks block_hashes = self._get_block_hashes( request, start_idx=start_block_idx, end_idx=num_blocks ) src_spec = self.manager.prepare_load(block_hashes) dst_spec = GPULoadStoreSpec(block_ids[num_computed_gpu_blocks:]) block_hashes = self._get_block_hashes( request, start_idx=start_block_idx, end_idx=num_blocks ) self._reqs_to_load[request.request_id] = (src_spec, dst_spec) req_blocks_being_loaded = self._reqs_being_loaded[request.request_id] req_blocks_being_loaded.update(block_hashes) self._next_stored_block_idx[request.request_id] = num_blocks if self._blocks_being_loaded is not None: self._blocks_being_loaded.update(req_blocks_being_loaded) def _get_reqs_to_store(self, scheduler_output: SchedulerOutput): reqs_to_store: dict[ReqId, TransferSpec] = {} # iterate over both new and cached requests for req_id, new_block_id_groups, preempted in yield_req_data(scheduler_output): if preempted: self._request_block_ids[req_id] = [] if new_block_id_groups: new_block_ids = new_block_id_groups[0] self._request_block_ids[req_id] += new_block_ids block_ids = self._request_block_ids[req_id] req = self._requests[req_id] new_tokens = scheduler_output.num_scheduled_tokens[req_id] total_tokens = req.num_computed_tokens + new_tokens num_blocks = total_tokens // self.offloaded_block_size start_block_idx = self._next_stored_block_idx.get(req_id, 0) num_new_blocks = num_blocks - start_block_idx if num_new_blocks <= 0: continue # NOTE: In async scheduling, placeholders may temporarily make # len(req.block_hashes) < num_blocks * self.block_size_factor. new_block_hashes = self._get_block_hashes( req, start_idx=start_block_idx, end_idx=num_blocks ) store_output = self.manager.prepare_store(new_block_hashes) if store_output is None: logger.warning( "Request %s: cannot store %s blocks", req_id, num_new_blocks ) continue self._next_stored_block_idx[req_id] = num_blocks if not store_output.block_hashes_to_store: continue block_hashes_to_store = set(store_output.block_hashes_to_store) block_hashes = self._get_block_hashes(req, end_idx=num_blocks) self.manager.touch(block_hashes) new_block_hashes = self._get_block_hashes( req, start_idx=start_block_idx, end_idx=num_blocks ) dst_spec = store_output.store_spec src_block_ids: list[int] = [] for idx, blk_hash in enumerate(new_block_hashes): if blk_hash not in block_hashes_to_store: continue offloaded_block_idx = start_block_idx + idx gpu_block_idx = offloaded_block_idx * self.block_size_factor for i in range(self.block_size_factor): src_block_ids.append(block_ids[gpu_block_idx + i]) src_spec = GPULoadStoreSpec(src_block_ids) reqs_to_store[req_id] = (src_spec, dst_spec) self._reqs_being_stored[req_id] |= block_hashes_to_store logger.debug( "Request %s offloading %s blocks starting from block #%d", req_id, len(block_hashes_to_store), start_block_idx, ) return reqs_to_store def build_connector_meta( self, scheduler_output: SchedulerOutput ) -> KVConnectorMetadata: meta = OffloadingConnectorMetadata( reqs_to_load=self._reqs_to_load, reqs_to_store=self._get_reqs_to_store(scheduler_output), ) self._reqs_to_load = {} # NOTE (orozery): we should move this logic to update_connector_output # once KVConnectorOutput allows us to report completed transfers for req_id in scheduler_output.preempted_req_ids or (): block_hashes = self._reqs_being_stored.get(req_id) if block_hashes: self.manager.complete_store(block_hashes) block_hashes.clear() return meta def update_connector_output(self, connector_output: KVConnectorOutput): """ Update KVConnector state from worker-side connectors output. Args: connector_output (KVConnectorOutput): the worker-side connectors output. """ for req_id in connector_output.finished_sending or []: block_hashes = self._reqs_being_stored.pop(req_id, None) if block_hashes: self.manager.complete_store(block_hashes) for req_id in connector_output.finished_recving or []: block_hashes = self._reqs_being_loaded.pop(req_id, None) if block_hashes: if self._blocks_being_loaded: self._blocks_being_loaded.difference_update(block_hashes) self.manager.complete_load(block_hashes) def request_finished( self, request: Request, block_ids: list[int], ) -> tuple[bool, dict[str, Any] | None]: """ Called when a request has finished, before its blocks are freed. Returns: True if the request is being saved/sent asynchronously and blocks should not be freed until the request_id is returned from get_finished(). Optional KVTransferParams to be included in the request outputs returned by the engine. """ req_id = request.request_id self._requests.pop(req_id, None) self._request_block_ids.pop(req_id, None) self._next_stored_block_idx.pop(req_id, None) request_being_stored = req_id in self._reqs_being_stored return request_being_stored, None def take_events(self) -> Iterable[KVCacheEvent]: """Take the KV cache events from the connector. Returns: A list of KV cache events. """ for event in self.manager.take_events(): if event.removed: yield BlockRemoved(block_hashes=event.block_hashes, medium=event.medium) else: yield BlockStored( block_hashes=event.block_hashes, parent_block_hash=None, token_ids=[], lora_id=None, block_size=event.block_size, medium=event.medium, lora_name=None, ) class OffloadingConnectorWorker: """Implementation of Worker side methods""" def __init__(self, spec: OffloadingSpec): self.spec = spec self.worker = OffloadingWorker() self._job_counter = 0 self.kv_connector_stats = OffloadingConnectorStats() # req_id -> (job_id, store) self._jobs: dict[int, tuple[ReqId, bool]] = {} # req_id -> active job IDs self._load_job: dict[ReqId, int] = {} # req_id -> set(active job IDs) self._store_jobs = defaultdict[ReqId, set[int]](set) # list of store jobs pending submission (job_id, transfer_spec) self._unsubmitted_store_jobs: list[tuple[int, TransferSpec]] = [] self._finished_reqs_waiting_for_store: set[ReqId] = set() def _generate_job_id(self) -> int: job_id = self._job_counter self._job_counter = job_id + 1 return job_id def _register_handlers( self, kv_caches: dict[str, torch.Tensor], attn_backends: dict[str, type[AttentionBackend]], ): for src_cls, dst_cls, handler in self.spec.get_handlers( kv_caches, attn_backends ): self.worker.register_handler(src_cls, dst_cls, handler) def register_kv_caches(self, kv_caches: dict[str, torch.Tensor]): layer_names = list(kv_caches.keys()) layers = get_layers_from_vllm_config( self.spec.vllm_config, Attention, layer_names ) attn_backends = { layer_name: layers[layer_name].get_attn_backend() for layer_name in layer_names } self._register_handlers(kv_caches, attn_backends) def register_cross_layers_kv_cache( self, kv_cache: torch.Tensor, attn_backend: type[AttentionBackend] ): cross_layer_name = "ALL_LAYERS" kv_caches = {cross_layer_name: kv_cache} attn_backends = {cross_layer_name: attn_backend} self._register_handlers(kv_caches, attn_backends) def handle_preemptions(self, preempted_req_ids: set[str]): for job_id, transfer_spec in self._unsubmitted_store_jobs: success = self.worker.transfer_async(job_id, transfer_spec) assert success self._unsubmitted_store_jobs.clear() for req_id in preempted_req_ids: job_ids = self._store_jobs.get(req_id) if job_ids: self.worker.wait(job_ids) def start_kv_transfers(self, metadata: OffloadingConnectorMetadata): for job_id, transfer_spec in self._unsubmitted_store_jobs: success = self.worker.transfer_async(job_id, transfer_spec) assert success self._unsubmitted_store_jobs.clear() for req_id, transfer_spec in metadata.reqs_to_load.items(): job_id = self._generate_job_id() self._jobs[job_id] = (req_id, False) assert req_id not in self._load_job self._load_job[req_id] = job_id success = self.worker.transfer_async(job_id, transfer_spec) assert success def prepare_store_kv(self, metadata: OffloadingConnectorMetadata): for req_id, transfer_spec in metadata.reqs_to_store.items(): job_id = self._generate_job_id() self._jobs[job_id] = (req_id, True) self._store_jobs[req_id].add(job_id) # NOTE(orozery): defer the store to the beginning of the next engine step, # so that offloading starts AFTER transfers related to token sampling, # thereby avoiding delays to token generation due to offloading. self._unsubmitted_store_jobs.append((job_id, transfer_spec)) def get_finished(self, finished_req_ids: set[str]) -> tuple[set[str], set[str]]: """ Notifies worker-side connector ids of requests that have finished generating tokens. Returns a list of request IDs that finished loading or storing. Returns: ids of requests that have finished asynchronous transfer tuple of (sending/saving ids, recving/loading ids). """ finished_sending = set() finished_recving = set() for transfer_result in self.worker.get_finished(): # we currently do not support job failures job_id = transfer_result.job_id assert transfer_result.success req_id, store = self._jobs.pop(job_id) if ( transfer_result.transfer_time and transfer_result.transfer_size is not None and transfer_result.transfer_type is not None ): self.kv_connector_stats.record_transfer( num_bytes=transfer_result.transfer_size, time=transfer_result.transfer_time, transfer_type=transfer_result.transfer_type, ) if store: req_jobs = self._store_jobs[req_id] req_jobs.remove(job_id) if req_jobs: continue if req_id in self._finished_reqs_waiting_for_store: self._finished_reqs_waiting_for_store.remove(req_id) finished_sending.add(req_id) del self._store_jobs[req_id] else: req_job = self._load_job[req_id] assert job_id == req_job del self._load_job[req_id] finished_recving.add(req_id) for req_id in finished_req_ids: pending_req_jobs = self._store_jobs.get(req_id) if pending_req_jobs: self._finished_reqs_waiting_for_store.add(req_id) elif pending_req_jobs is not None: finished_sending.add(req_id) del self._store_jobs[req_id] return finished_sending, finished_recving def get_kv_connector_stats(self) -> KVConnectorStats | None: """ Get the KV transfer stats for the connector. """ if self.kv_connector_stats.is_empty(): return None # Clear stats for next iteration kv_connector_stats = self.kv_connector_stats self.kv_connector_stats = OffloadingConnectorStats() return kv_connector_stats class OffloadPromMetrics(KVConnectorPromMetrics): def __init__( self, vllm_config: VllmConfig, metric_types: dict[type[PromMetric], type[PromMetricT]], labelnames: list[str], per_engine_labelvalues: dict[int, list[object]], ): super().__init__(vllm_config, metric_types, labelnames, per_engine_labelvalues) # (engine_idx, transfer_tupe) -> (metric with bounded labels) self.histogram_transfer_size: dict[tuple[int, str], PromMetricT] = {} self.counter_kv_bytes: dict[tuple[int, str], PromMetricT] = {} self.counter_kv_transfer_time: dict[tuple[int, str], PromMetricT] = {} buckets = [ # In bytes 1e6, 5e6, 10e6, 20e6, 40e6, 60e6, 80e6, 100e6, 150e6, 200e6, ] self._counter_kv_bytes = self._counter_cls( name="vllm:kv_offload_total_bytes", documentation="Number of bytes offloaded by KV connector", labelnames=labelnames + ["transfer_type"], ) self._counter_kv_transfer_time = self._counter_cls( name="vllm:kv_offload_total_time", documentation="Total time measured by all KV offloading operations", labelnames=labelnames + ["transfer_type"], ) self._histogram_transfer_size = self._histogram_cls( name="vllm:kv_offload_size", documentation="Histogram of KV offload transfer size, in bytes.", buckets=buckets[:], labelnames=labelnames + ["transfer_type"], ) def observe(self, transfer_stats_data: dict[str, Any], engine_idx: int = 0): """ Observe transfer statistics from the new data structure. transfer_stats_data is expected to be a dict where: - keys are transfer type strings (e.g., "cpu_to_gpu", "gpu_to_cpu") - values are lists of OffloadingOperationMetrics objects """ for transfer_type, ops in transfer_stats_data.items(): # Cache: if (engine_idx, transfer_type) not in self.histogram_transfer_size: self.histogram_transfer_size[(engine_idx, transfer_type)] = ( self._histogram_transfer_size.labels( *(self.per_engine_labelvalues[engine_idx] + [transfer_type]) ) ) self.counter_kv_bytes[(engine_idx, transfer_type)] = ( self._counter_kv_bytes.labels( *(self.per_engine_labelvalues[engine_idx] + [transfer_type]) ) ) self.counter_kv_transfer_time[(engine_idx, transfer_type)] = ( self._counter_kv_transfer_time.labels( *(self.per_engine_labelvalues[engine_idx] + [transfer_type]) ) ) # Process ops: assert isinstance(ops, list) for op in ops: # ops is a list of serialized OffloadingOperationMetrics assert isinstance(op, dict) # Observe size histogram self.histogram_transfer_size[(engine_idx, transfer_type)].observe( op["op_size"] ) # Increment byte and time counters self.counter_kv_bytes[(engine_idx, transfer_type)].inc(op["op_size"]) self.counter_kv_transfer_time[(engine_idx, transfer_type)].inc( op["op_time"] )
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py", "license": "Apache License 2.0", "lines": 678, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/v1/kv_offload/factory.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import importlib from collections.abc import Callable from typing import TYPE_CHECKING from vllm.logger import init_logger from vllm.v1.kv_offload.spec import OffloadingSpec if TYPE_CHECKING: from vllm.config import VllmConfig from vllm.v1.kv_cache_interface import KVCacheConfig logger = init_logger(__name__) class OffloadingSpecFactory: _registry: dict[str, Callable[[], type[OffloadingSpec]]] = {} @classmethod def register_spec(cls, name: str, module_path: str, class_name: str) -> None: """Register a spec with a lazy-loading module and class name.""" if name in cls._registry: raise ValueError(f"Connector '{name}' is already registered.") def loader() -> type[OffloadingSpec]: module = importlib.import_module(module_path) return getattr(module, class_name) cls._registry[name] = loader @classmethod def create_spec( cls, config: "VllmConfig", kv_cache_config: "KVCacheConfig | None", ) -> OffloadingSpec: kv_transfer_config = config.kv_transfer_config assert kv_transfer_config is not None extra_config = kv_transfer_config.kv_connector_extra_config spec_name = extra_config.get("spec_name", "CPUOffloadingSpec") if spec_name in cls._registry: spec_cls = cls._registry[spec_name]() else: spec_module_path = extra_config.get("spec_module_path") if spec_module_path is None: raise ValueError(f"Unsupported spec type: {spec_name}") spec_module = importlib.import_module(spec_module_path) spec_cls = getattr(spec_module, spec_name) assert issubclass(spec_cls, OffloadingSpec) logger.info("Creating offloading spec with name: %s", spec_name) return spec_cls(config, kv_cache_config) # Register various specs here. OffloadingSpecFactory.register_spec( "CPUOffloadingSpec", "vllm.v1.kv_offload.cpu", "CPUOffloadingSpec" )
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/kv_offload/factory.py", "license": "Apache License 2.0", "lines": 47, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/v1/kv_offload/spec.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from abc import ABC, abstractmethod from collections.abc import Iterator from typing import TYPE_CHECKING import torch from vllm.logger import init_logger from vllm.v1.attention.backend import AttentionBackend from vllm.v1.kv_offload.abstract import LoadStoreSpec, OffloadingManager from vllm.v1.kv_offload.worker.worker import OffloadingHandler if TYPE_CHECKING: from vllm.config import VllmConfig from vllm.v1.kv_cache_interface import KVCacheConfig logger = init_logger(__name__) class OffloadingSpec(ABC): """Spec for an offloading connector""" def __init__( self, vllm_config: "VllmConfig", kv_cache_config: "KVCacheConfig | None" ): logger.warning( "Initializing OffloadingSpec. This API is experimental and " "subject to change in the future as we iterate the design." ) self.vllm_config = vllm_config self.kv_cache_config = kv_cache_config kv_transfer_config = vllm_config.kv_transfer_config assert kv_transfer_config is not None self.extra_config = kv_transfer_config.kv_connector_extra_config self.gpu_block_size = vllm_config.cache_config.block_size self.offloaded_block_size = int( self.extra_config.get("block_size", self.gpu_block_size) ) assert self.offloaded_block_size % self.gpu_block_size == 0 @abstractmethod def get_manager(self) -> OffloadingManager: """ Get an OffloadingManager that will be used by the scheduler-side offloading connector to track offloaded blocks and manage evictions. """ pass @abstractmethod def get_handlers( self, kv_caches: dict[str, torch.Tensor], attn_backends: dict[str, type[AttentionBackend]], ) -> Iterator[tuple[type[LoadStoreSpec], type[LoadStoreSpec], OffloadingHandler]]: """ Get offloading handlers along with their respective src and dst types. Args: kv_caches: A dictionary of layer_name -> gpu_kv_cache tensor. attn_backends: A dictionary of layer_name -> AttentionBackend. Yields: Tuples of (src_type, dst_type, offloading_handler). """ pass
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/kv_offload/spec.py", "license": "Apache License 2.0", "lines": 56, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/v1/kv_offload/test_cpu_gpu.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import random import time import pytest import torch from vllm.platforms import current_platform from vllm.utils.torch_utils import set_random_seed from vllm.v1.attention.backends.flash_attn import FlashAttentionBackend from vllm.v1.kv_offload.mediums import CPULoadStoreSpec, GPULoadStoreSpec from vllm.v1.kv_offload.worker.cpu_gpu import CpuGpuOffloadingHandlers BACKENDS_TO_TEST = [FlashAttentionBackend] if not current_platform.is_rocm(): from vllm.v1.attention.backends.flashinfer import FlashInferBackend BACKENDS_TO_TEST.append(FlashInferBackend) from vllm.v1.attention.backends.mla.flashattn_mla import FlashAttnMLABackend BACKENDS_TO_TEST.append(FlashAttnMLABackend) NUM_GPU_BLOCKS = [64] NUM_CPU_BLOCKS = [256] KERNEL_BLOCK_SIZES = [16] LOGICAL_BLOCK_SIZES = [16, 32] LOGICAL_BLOCKS_PER_CPU_BLOCK = [1, 3] HEAD_SIZES = [64] NUM_HEADS = [8] NUM_LAYERS = [4] DTYPES = [torch.bfloat16] SEEDS = [0] CUDA_DEVICES = ["cuda:0"] NUM_MAPPINGS = [3] @pytest.mark.parametrize("gpu_to_cpu", [True, False]) @pytest.mark.parametrize("num_mappings", NUM_MAPPINGS) @pytest.mark.parametrize("head_size", HEAD_SIZES) @pytest.mark.parametrize("num_heads", NUM_HEADS) @pytest.mark.parametrize("kernel_block_size", KERNEL_BLOCK_SIZES) @pytest.mark.parametrize("logical_block_size", LOGICAL_BLOCK_SIZES) @pytest.mark.parametrize("logical_blocks_per_cpu_block", LOGICAL_BLOCKS_PER_CPU_BLOCK) @pytest.mark.parametrize("num_gpu_blocks", NUM_GPU_BLOCKS) @pytest.mark.parametrize("num_cpu_blocks", NUM_CPU_BLOCKS) @pytest.mark.parametrize("num_layers", NUM_LAYERS) @pytest.mark.parametrize("dtype", DTYPES) @pytest.mark.parametrize("seed", SEEDS) @pytest.mark.parametrize("device", CUDA_DEVICES) @torch.inference_mode() def test_transfer( default_vllm_config, gpu_to_cpu: bool, num_mappings: int, head_size: int, num_heads: int, kernel_block_size: int, logical_block_size: int, logical_blocks_per_cpu_block: int, num_gpu_blocks: int, num_cpu_blocks: int, num_layers: int, dtype: torch.dtype, seed: int, device: str, ) -> None: set_random_seed(seed) # create per-layer GPU KV caches based on available attn_backends attn_backends_list = BACKENDS_TO_TEST assert logical_block_size % kernel_block_size == 0 kernel_blocks_per_gpu_block = logical_block_size // kernel_block_size num_gpu_kernel_blocks = num_gpu_blocks * kernel_blocks_per_gpu_block gpu_caches = {} attn_backends = {} for i in range(num_layers): layer_name = f"layer {i}" attn_backend = attn_backends_list[i % len(attn_backends_list)] attn_backends[layer_name] = attn_backend gpu_cache_shape = attn_backend.get_kv_cache_shape( num_gpu_kernel_blocks, kernel_block_size, num_heads, head_size ) gpu_caches[layer_name] = torch.rand(gpu_cache_shape, dtype=dtype, device=device) # create handler cpu_block_size = logical_blocks_per_cpu_block * logical_block_size kernel_blocks_per_cpu_block = cpu_block_size // kernel_block_size handlers = CpuGpuOffloadingHandlers( attn_backends=attn_backends, gpu_block_size=logical_block_size, cpu_block_size=cpu_block_size, num_cpu_blocks=num_cpu_blocks, gpu_caches=gpu_caches, ) # select block mappings gpu_blocks = random.sample( range(num_gpu_blocks), num_mappings * logical_blocks_per_cpu_block ) cpu_blocks = random.sample(range(num_cpu_blocks), num_mappings) # convert gpu blocks to kernel block size gpu_blocks_in_kernel_block_size = [] for gpu_block in gpu_blocks: base_block_id = gpu_block * kernel_blocks_per_gpu_block for i in range(kernel_blocks_per_gpu_block): gpu_blocks_in_kernel_block_size.append(i + base_block_id) # convert cpu blocks to gpu block size cpu_blocks_in_kernel_block_size = [] for cpu_block in cpu_blocks: base_block_id = cpu_block * kernel_blocks_per_cpu_block for i in range(kernel_blocks_per_cpu_block): cpu_blocks_in_kernel_block_size.append(i + base_block_id) # maybe skip some GPU block to test reading from the middle of a CPU block if not gpu_to_cpu: gpu_blocks_to_skip = logical_blocks_per_cpu_block - 1 gpu_blocks = gpu_blocks[gpu_blocks_to_skip:] kernel_blocks_to_skip = gpu_blocks_to_skip * kernel_blocks_per_gpu_block gpu_blocks_in_kernel_block_size = gpu_blocks_in_kernel_block_size[ kernel_blocks_to_skip: ] cpu_blocks_in_kernel_block_size = cpu_blocks_in_kernel_block_size[ kernel_blocks_to_skip: ] # set transfer direction if gpu_to_cpu: handler = handlers.gpu_to_cpu_handler src_spec_class = GPULoadStoreSpec dst_spec_class = CPULoadStoreSpec src_blocks = gpu_blocks dst_blocks = cpu_blocks src_blocks_in_kernel_block_size = gpu_blocks_in_kernel_block_size dst_blocks_in_kernel_block_size = cpu_blocks_in_kernel_block_size dst_size_in_kernel_blocks = num_cpu_blocks * kernel_blocks_per_cpu_block else: handler = handlers.cpu_to_gpu_handler src_spec_class = CPULoadStoreSpec dst_spec_class = GPULoadStoreSpec src_blocks = cpu_blocks dst_blocks = gpu_blocks src_blocks_in_kernel_block_size = cpu_blocks_in_kernel_block_size dst_blocks_in_kernel_block_size = gpu_blocks_in_kernel_block_size dst_size_in_kernel_blocks = num_gpu_blocks * kernel_blocks_per_gpu_block # build dst -> src mapping dst_to_src = {} for src_block, dst_block in zip( src_blocks_in_kernel_block_size, dst_blocks_in_kernel_block_size ): dst_to_src[dst_block] = src_block # build transfer specs src_spec = src_spec_class(src_blocks) dst_spec = dst_spec_class(dst_blocks) # clone src and dst tensors before transfer orig_src_caches = [x.clone() for x in handler.src_tensors] orig_dst_caches = [x.clone() for x in handler.dst_tensors] # call transfer function start_time = time.time() assert handler.transfer_async(1, (src_spec, dst_spec)) assert set({x.job_id for x in handler._transfers}) == {1} # wait for transfer to complete end_time = time.time() + 10 while time.time() < end_time: finished = handler.get_finished() if finished: assert finished[0].job_id == 1 assert finished[0].success assert ( finished[0].transfer_type == ("GPU", "CPU") if gpu_to_cpu else ("CPU", "GPU") ) assert ( finished[0].transfer_size == handler.total_block_size_in_bytes * handler.dst_block_size_factor * len(dst_blocks) ) assert finished[0].transfer_time > 0 assert finished[0].transfer_time < (time.time() - start_time) break time.sleep(0.1) # verify src tensors did not change for orig_tensor, tensor in zip(orig_src_caches, handler.src_tensors): assert torch.equal(orig_tensor, tensor) # verify dst tensors for dst_block in range(dst_size_in_kernel_blocks): src_block_candidate = dst_to_src.get(dst_block) for src_cache, dst_cache, orig_dst_cache in zip( handler.src_tensors, handler.dst_tensors, orig_dst_caches, ): if src_block_candidate is not None: expected_value = src_cache[src_block_candidate] else: expected_value = orig_dst_cache[dst_block] torch.testing.assert_close(dst_cache[dst_block].cpu(), expected_value.cpu())
{ "repo_id": "vllm-project/vllm", "file_path": "tests/v1/kv_offload/test_cpu_gpu.py", "license": "Apache License 2.0", "lines": 186, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/v1/kv_offload/worker/cpu_gpu.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from collections import deque from dataclasses import dataclass import numpy as np import torch from vllm import _custom_ops as ops from vllm.logger import init_logger from vllm.utils.platform_utils import is_pin_memory_available from vllm.v1.attention.backend import AttentionBackend from vllm.v1.kv_offload.mediums import BlockIDsLoadStoreSpec from vllm.v1.kv_offload.worker.worker import ( OffloadingHandler, TransferResult, TransferSpec, ) logger = init_logger(__name__) @dataclass class Transfer: job_id: int stream: torch.cuda.Stream start_event: torch.Event end_event: torch.Event num_bytes: int def expand_block_ids( block_ids: np.ndarray, block_size_factor: int, output: np.ndarray, skip_count: int = 0, ): """ Convert a list of block IDs to a list of matching block ids, assuming each block is composed of actual block_size_factor blocks. Outputs to output tensor. The first skip_count blocks will be skipped. Note that skip_count must be less than block_size_factor. For example, if block_ids = [0, 1, 3] and block_size_factor = 4, then it yields [0, 1, 2, 3, 4, 5, 6, 7, 12, 13, 14, 15] since 0 maps to [0, 1, 2, 3] 1 maps to [4, 5, 6, 7] and 3 maps to [12, 13, 14, 15] """ assert skip_count < block_size_factor first_range = np.arange(skip_count, block_size_factor) full_range = np.arange(0, block_size_factor) output_idx = 0 for i, block_id in enumerate(block_ids): base_block_id = block_id * block_size_factor indices = first_range if i == 0 else full_range output_end_idx = output_idx + len(indices) output[output_idx:output_end_idx] = base_block_id + indices output_idx = output_end_idx class SingleDirectionOffloadingHandler(OffloadingHandler): """ SingleDirectionOffloadingHandler handles transfers for a single direction, either CPU->GPU or GPU->CPU. Transfers are guaranteed to be executed in order of their submission. Each transfer uses a unique CUDA stream, and its stream will start executing only after the streams of previous transfers have finished. """ def __init__( self, src_tensors: list[torch.Tensor], dst_tensors: list[torch.Tensor], src_block_size_factor: int, dst_block_size_factor: int, ): """ Initialize a SingleDirectionOffloadingHandler. Args: src_tensors: list of KV cache tensors to copy from. dst_tensors: list of KV cache tensors to copy to. Order should match src_tensors. src_block_size_factor: The number of kernel blocks per KV block in a source tensor. dst_block_size_factor: The number of kernel blocks per KV block in a destination tensor. """ assert len(src_tensors) == len(dst_tensors) self.src_tensors: list[torch.Tensor] = src_tensors self.dst_tensors: list[torch.Tensor] = dst_tensors min_block_size_factor = min(src_block_size_factor, dst_block_size_factor) self.src_block_size_factor: int = src_block_size_factor // min_block_size_factor self.dst_block_size_factor: int = dst_block_size_factor // min_block_size_factor self.block_size_in_bytes = [ tensor.element_size() * tensor.stride(0) * min_block_size_factor for tensor in src_tensors ] self.total_block_size_in_bytes = sum(self.block_size_in_bytes) assert len(src_tensors) > 0 self.gpu_to_cpu: bool = self.src_tensors[0].is_cuda self.transfer_type = ("GPU", "CPU") if self.gpu_to_cpu else ("CPU", "GPU") # job_id -> event self._transfer_events: dict[int, torch.Event] = {} # queue of transfers (job_id, stream, event) self._transfers: deque[Transfer] = deque() # list of CUDA streams available for re-use self._stream_pool: list[torch.cuda.Stream] = [] # list of CUDA events available for re-use self._event_pool: list[torch.Event] = [] def transfer_async(self, job_id: int, transfer_spec: TransferSpec) -> bool: src_spec, dst_spec = transfer_spec assert isinstance(src_spec, BlockIDsLoadStoreSpec) assert isinstance(dst_spec, BlockIDsLoadStoreSpec) src_blocks = src_spec.block_ids dst_blocks = dst_spec.block_ids assert src_blocks.ndim == 1 assert dst_blocks.ndim == 1 src_sub_block_count = src_blocks.size * self.src_block_size_factor dst_sub_block_count = dst_blocks.size * self.dst_block_size_factor src_sub_blocks_to_skip = -dst_blocks.size % self.src_block_size_factor assert dst_sub_block_count == src_sub_block_count - src_sub_blocks_to_skip src_to_dst = np.empty((dst_sub_block_count, 2), dtype=np.int64) expand_block_ids( src_blocks, self.src_block_size_factor, src_to_dst[:, 0], skip_count=src_sub_blocks_to_skip, ) expand_block_ids(dst_blocks, self.dst_block_size_factor, src_to_dst[:, 1]) src_to_dst_tensor = torch.from_numpy(src_to_dst) stream = self._stream_pool.pop() if self._stream_pool else torch.cuda.Stream() start_event = ( self._event_pool.pop() if self._event_pool else torch.Event(enable_timing=True) ) end_event = ( self._event_pool.pop() if self._event_pool else torch.Event(enable_timing=True) ) if self.gpu_to_cpu: # wait for model computation to finish before offloading stream.wait_stream(torch.cuda.current_stream()) if self._transfers: last_transfer: Transfer = self._transfers[-1] last_event = last_transfer.end_event # assure job will start only after the previous one completes stream.wait_event(last_event) with torch.cuda.stream(stream): start_event.record(stream) for src_tensor, dst_tensor, block_size_in_bytes in zip( self.src_tensors, self.dst_tensors, self.block_size_in_bytes, ): ops.swap_blocks( src_tensor, dst_tensor, block_size_in_bytes, src_to_dst_tensor, ) end_event.record(stream) self._transfer_events[job_id] = end_event self._transfers.append( Transfer( job_id=job_id, stream=stream, start_event=start_event, end_event=end_event, num_bytes=dst_sub_block_count * self.total_block_size_in_bytes, ) ) # success return True def get_finished(self) -> list[TransferResult]: results: list[TransferResult] = [] while self._transfers and self._transfers[0].end_event.query(): transfer = self._transfers.popleft() transfer_time = ( transfer.start_event.elapsed_time(transfer.end_event) * 1e-3 ) # elapsed_time is in miliseconds result = TransferResult( job_id=transfer.job_id, success=True, transfer_size=transfer.num_bytes, transfer_time=transfer_time, transfer_type=self.transfer_type, ) results.append(result) self._stream_pool.append(transfer.stream) self._event_pool.append(transfer.end_event) self._event_pool.append(transfer.start_event) del self._transfer_events[transfer.job_id] return results def wait(self, job_ids: set[int]): for job_id in job_ids: event = self._transfer_events.get(job_id) if event is not None: event.synchronize() class CpuGpuOffloadingHandlers: def __init__( self, gpu_block_size: int, cpu_block_size: int, num_cpu_blocks: int, gpu_caches: dict[str, torch.Tensor], attn_backends: dict[str, type[AttentionBackend]], ): assert gpu_caches assert cpu_block_size % gpu_block_size == 0 # find kernel block size and determine layout per each gpu tensor kernel_block_size: int | None = None # list of (gpu_tensor, split_k_and_v) parsed_gpu_tensors: list[tuple[torch.Tensor, bool]] = [] for layer_name, gpu_tensor in gpu_caches.items(): gpu_shape = gpu_tensor.shape attn_backend = attn_backends[layer_name] test_shape = attn_backend.get_kv_cache_shape( num_blocks=1234, block_size=16, num_kv_heads=8, head_size=256 ) has_layers_dim = False split_k_and_v = False if len(gpu_shape) != len(test_shape): # cross-layers tensor # shape is (num_blocks, ...) assert len(gpu_shape) == len(test_shape) + 1 has_layers_dim = True # prepend a dummy num_layers=80 to test_shape test_shape = (80,) + test_shape elif test_shape[0] != 1234: # shape should be (2, num_blocks, ...) assert test_shape[0] == 2 assert test_shape[1] == 1234 assert gpu_shape[0] == 2 split_k_and_v = True if has_layers_dim: # in the cross layers case, the registered kv cache tensor # shape matches the physical layout, whereas test_shape # is the logical layout. # To match them, we need to permute test_shape try: kv_cache_stride_order = attn_backend.get_kv_cache_stride_order( include_num_layers_dimension=has_layers_dim ) assert len(kv_cache_stride_order) == len(gpu_shape) except (AttributeError, NotImplementedError): kv_cache_stride_order = tuple(range(len(gpu_shape))) test_shape = tuple(test_shape[i] for i in kv_cache_stride_order) # find block_size (16) dimension index block_size_idx = test_shape.index(16) if kernel_block_size is not None: assert kernel_block_size == gpu_shape[block_size_idx] else: kernel_block_size = gpu_shape[block_size_idx] assert gpu_block_size % kernel_block_size == 0 parsed_gpu_tensors.append((gpu_tensor, split_k_and_v)) assert kernel_block_size is not None cpu_block_size_factor = cpu_block_size // kernel_block_size gpu_block_size_factor = gpu_block_size // kernel_block_size num_cpu_kernel_blocks = num_cpu_blocks * cpu_block_size_factor # allocate cpu tensors pin_memory = is_pin_memory_available() logger.info("Allocating %d CPU tensors...", len(parsed_gpu_tensors)) gpu_tensors: list[torch.Tensor] = [] cpu_tensors: list[torch.Tensor] = [] for gpu_tensor, split_k_and_v in parsed_gpu_tensors: cpu_shape = list(gpu_tensor.shape) cpu_shape[1 if split_k_and_v else 0] = num_cpu_kernel_blocks logger.debug("Allocating CPU tensor of shape %r", cpu_shape) cpu_tensor = torch.zeros( cpu_shape, dtype=gpu_tensor.dtype, device="cpu", pin_memory=pin_memory, ) gpu_tensors.extend(gpu_tensor.unbind(0) if split_k_and_v else [gpu_tensor]) cpu_tensors.extend(cpu_tensor.unbind(0) if split_k_and_v else [cpu_tensor]) self.gpu_to_cpu_handler = SingleDirectionOffloadingHandler( src_tensors=gpu_tensors, dst_tensors=cpu_tensors, src_block_size_factor=gpu_block_size_factor, dst_block_size_factor=cpu_block_size_factor, ) self.cpu_to_gpu_handler = SingleDirectionOffloadingHandler( src_tensors=cpu_tensors, dst_tensors=gpu_tensors, src_block_size_factor=cpu_block_size_factor, dst_block_size_factor=gpu_block_size_factor, )
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/kv_offload/worker/cpu_gpu.py", "license": "Apache License 2.0", "lines": 281, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/distributed/kv_transfer/kv_connector/v1/metrics.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from dataclasses import dataclass, field from typing import Any, TypeAlias, TypeVar from prometheus_client import Counter, Gauge, Histogram from vllm.config import KVTransferConfig, VllmConfig from vllm.distributed.kv_transfer.kv_connector.factory import KVConnectorFactory from vllm.logger import init_logger PromMetric: TypeAlias = Gauge | Counter | Histogram PromMetricT = TypeVar("PromMetricT", bound=PromMetric) logger = init_logger(__name__) @dataclass class KVConnectorStats: """ Base class for KV Connector Stats, a container for transfer performance metrics or otherwise important telemetry from the connector. All sub-classes need to be serializable as stats are sent from worker to logger process. """ data: dict[str, Any] = field(default_factory=dict) def reset(self): """Reset the stats, clear the state.""" raise NotImplementedError def aggregate(self, other: "KVConnectorStats") -> "KVConnectorStats": """ Aggregate stats with another `KVConnectorStats` object. """ raise NotImplementedError def reduce(self) -> dict[str, int | float]: """ Reduce the observations collected during a time interval to one or more representative values (eg avg/median/sum of the series). This is meant to be called by the logger to produce a summary of the stats for the last time interval. """ raise NotImplementedError def is_empty(self) -> bool: """Return True if the stats are empty.""" raise NotImplementedError class KVConnectorLogging: def __init__(self, kv_transfer_config: KVTransferConfig | None): # Instantiate the connector's stats class. if kv_transfer_config and kv_transfer_config.kv_connector: self.connector_cls = KVConnectorFactory.get_connector_class( kv_transfer_config ) self.reset() def reset(self): self.transfer_stats_accumulator: KVConnectorStats | None = None def observe(self, transfer_stats_data: dict[str, Any]): # Should not be called when a KVConnector is not configured. assert self.connector_cls is not None # Called periodically when connector syncs with the scheduler. # Note that this is not the same as the logging interval. # We expect transfer_stats_data to be aggregated across all workers and # consist of observations from a single connector or a MultiConnector. transfer_stats = self.connector_cls.build_kv_connector_stats( transfer_stats_data ) if transfer_stats is None: logger.warning_once( "The connector %s is collecting stats but " "does not implement the " "`build_kv_connector_stats` method. " "Stats will not be logged.", self.connector_cls, ) return if self.transfer_stats_accumulator is None: self.transfer_stats_accumulator = transfer_stats else: # Accumulate last interval stats. self.transfer_stats_accumulator = self.transfer_stats_accumulator.aggregate( transfer_stats ) def log(self, log_fn=logger.info): """Log transfer metrics periodically, similar to throughput logging""" if ( self.transfer_stats_accumulator and not self.transfer_stats_accumulator.is_empty() ): # Produce a single cumulative stats object for the last time # interval from the recorded observations. xfer_metrics = self.transfer_stats_accumulator.reduce() xfer_metrics_str = ", ".join(f"{k}={v}" for k, v in xfer_metrics.items()) log_fn("KV Transfer metrics: %s", xfer_metrics_str) # Reset metrics for next interval self.reset() class KVConnectorPromMetrics: """ A base class for per-connector Prometheus metric registration and recording. """ def __init__( self, vllm_config: VllmConfig, metric_types: dict[type[PromMetric], type[PromMetricT]], labelnames: list[str], per_engine_labelvalues: dict[int, list[object]], ): self._kv_transfer_config = vllm_config.kv_transfer_config self._gauge_cls = metric_types[Gauge] self._counter_cls = metric_types[Counter] self._histogram_cls = metric_types[Histogram] self._labelnames = labelnames self.per_engine_labelvalues = per_engine_labelvalues def make_per_engine(self, metric: PromMetric) -> dict[int, PromMetric]: """ Create a per-engine child of a prometheus_client.Metric with the appropriate labels set. The parent metric must be created using the labelnames list. """ return { idx: metric.labels(*labelvalues) for idx, labelvalues in self.per_engine_labelvalues.items() } def observe(self, transfer_stats_data: dict[str, Any], engine_idx: int = 0): """ Record the supplied transfer statistics to Prometheus metrics. These statistics are engine-specific, and should be recorded to a metric with the appropriate 'engine' label. These metric instances can be created using the make_per_engine() helper method. """ raise NotImplementedError class KVConnectorPrometheus: """ Support for registering per-connector Prometheus metrics, and recording transfer statistics to those metrics. Uses KVConnectorBase.build_prom_metrics(). """ _gauge_cls = Gauge _counter_cls = Counter _histogram_cls = Histogram def __init__( self, vllm_config: VllmConfig, labelnames: list[str], per_engine_labelvalues: dict[int, list[object]], ): self.prom_metrics: KVConnectorPromMetrics | None = None kv_transfer_config = vllm_config.kv_transfer_config if kv_transfer_config and kv_transfer_config.kv_connector: connector_cls = KVConnectorFactory.get_connector_class(kv_transfer_config) metric_types = { Gauge: self._gauge_cls, Counter: self._counter_cls, Histogram: self._histogram_cls, } self.prom_metrics = connector_cls.build_prom_metrics( vllm_config, metric_types, labelnames, per_engine_labelvalues, ) def observe(self, transfer_stats_data: dict[str, Any], engine_idx: int = 0): if self.prom_metrics is None: return self.prom_metrics.observe(transfer_stats_data, engine_idx)
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/distributed/kv_transfer/kv_connector/v1/metrics.py", "license": "Apache License 2.0", "lines": 158, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/config/pooler.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from typing import Any, Literal, get_args from vllm.config.utils import config from vllm.logger import init_logger from vllm.utils.hashing import safe_hash logger = init_logger(__name__) SequencePoolingType = Literal["CLS", "LAST", "MEAN"] SEQ_POOLING_TYPES: tuple[SequencePoolingType, ...] = get_args(SequencePoolingType) TokenPoolingType = Literal["ALL", "STEP"] TOK_POOLING_TYPES: tuple[TokenPoolingType, ...] = get_args(TokenPoolingType) @config class PoolerConfig: """Controls the behavior of output pooling in pooling models.""" pooling_type: SequencePoolingType | TokenPoolingType | None = None """ The pooling method used for pooling. If set, `seq_pooling_type` or `tok_pooling_type` are automatically populated with this field. Alternatively, users can set `seq_pooling_type` and `tok_pooling_type` explicitly. This field is mainly for user convenience. Internal code should always use `seq_pooling_type` or `tok_pooling_type` instead of `pooling_type`. """ seq_pooling_type: SequencePoolingType | None = None """ The pooling method used for sequence pooling. """ tok_pooling_type: TokenPoolingType | None = None """ The pooling method used for tokenwise pooling. """ use_activation: bool | None = None """ Whether to apply activation function to the pooler outputs. `None` uses the pooler's default, which is `True` in most cases. """ ## for embedding models dimensions: int | None = None """ Reduce the dimensions of embeddings if model support matryoshka representation. Defaults to None. """ enable_chunked_processing: bool = False """ Whether to enable chunked processing for long inputs that exceed the model's maximum position embeddings. When enabled, long inputs will be split into chunks, processed separately, and then aggregated using weighted averaging. This allows embedding models to handle arbitrarily long text without CUDA errors. Defaults to False. """ max_embed_len: int | None = None """ Maximum input length allowed for embedding generation. When set, allows inputs longer than max_embed_len to be accepted for embedding models. When an input exceeds max_embed_len, it will be handled according to the original max_model_len validation logic. Defaults to None (i.e. set to max_model_len). """ ## for classification models logit_bias: float | None = None """ If provided, apply classification logit biases. Defaults to None. """ ## for reward models step_tag_id: int | None = None """ If set, only the score corresponding to the `step_tag_id` in the generated sentence should be returned. Otherwise, the scores for all tokens are returned. """ returned_token_ids: list[int] | None = None """ A list of indices for the vocabulary dimensions to be extracted, such as the token IDs of `good_token` and `bad_token` in the `math-shepherd-mistral-7b-prm` model. """ def __post_init__(self) -> None: if pooling_type := self.pooling_type: if self.seq_pooling_type is not None: raise ValueError( "Cannot set both `pooling_type` and `seq_pooling_type`" ) if self.tok_pooling_type is not None: raise ValueError( "Cannot set both `pooling_type` and `tok_pooling_type`" ) if pooling_type in SEQ_POOLING_TYPES: logger.debug( "Resolved `pooling_type=%r` to `seq_pooling_type=%r`.", pooling_type, pooling_type, ) self.seq_pooling_type = pooling_type elif pooling_type in TOK_POOLING_TYPES: logger.debug( "Resolved `pooling_type=%r` to `tok_pooling_type=%r`.", pooling_type, pooling_type, ) self.tok_pooling_type = pooling_type else: raise NotImplementedError(pooling_type) def get_seq_pooling_type(self) -> SequencePoolingType: assert self.seq_pooling_type is not None, "Should be resolved by ModelConfig" return self.seq_pooling_type def get_tok_pooling_type(self) -> TokenPoolingType: assert self.tok_pooling_type is not None, "Should be resolved by ModelConfig" return self.tok_pooling_type def compute_hash(self) -> str: """ WARNING: Whenever a new field is added to this config, ensure that it is included in the factors list if it affects the computation graph. Provide a hash that uniquely identifies all the configs that affect the structure of the computation graph from input ids/embeddings to the final hidden states, excluding anything before input ids/embeddings and after the final hidden states. """ # no factors to consider. # this config will not affect the computation graph. factors: list[Any] = [] hash_str = safe_hash(str(factors).encode(), usedforsecurity=False).hexdigest() return hash_str
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/config/pooler.py", "license": "Apache License 2.0", "lines": 124, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/v1/kv_offload/backend.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import ctypes from abc import ABC, abstractmethod from collections.abc import Iterable from vllm.v1.core.kv_cache_utils import BlockHash from vllm.v1.kv_offload.abstract import LoadStoreSpec class BlockStatus(ctypes.Structure): """ Offloading status for a single block of KV data. Holds the following information: ref_cnt - the current number of transfers using this block as a source. A value of -1 indicates the block is not yet ready to be read. load_store_spec - backend-specific information on how to actually read/write the block. """ _fields_ = [("ref_cnt", ctypes.c_int32)] def __init__(self): super().__init__() # initialize block as "not ready" (ref_cnt = -1) self.ref_cnt = -1 @property def is_ready(self) -> bool: """ Returns whether the block is ready to be read. """ return self.ref_cnt >= 0 class Backend(ABC): """ An abstract class for allocating and returning specs for writing KV blocks to some backend. """ def __init__(self, block_size: int, medium: str): self.block_size = block_size self.medium = medium @abstractmethod def get_num_free_blocks(self): """ Returns the number of current number of blocks that can be allocated. """ pass @abstractmethod def allocate_blocks(self, block_hashes: list[BlockHash]) -> list[BlockStatus]: """ Allocate space for writing blocks. This method assumes there is enough space for allocation. It is unsafe to use without checking get_num_free_blocks beforehand. Args: block_hashes: the hashes identifying the blocks to be written. Returns: A list of BlockStatus for the allocated blocks. The ref_cnt of each returned item will be -1, meaning the block is not yet ready to be read. """ pass @abstractmethod def free(self, block: BlockStatus): """ Free a previously allocated block. You should only call this function with blocks returned by allocate_blocks, and only once per each block. Args: block: The block to be freed. """ pass def get_load_store_spec( self, block_hashes: Iterable[BlockHash], blocks: Iterable[BlockStatus] ) -> LoadStoreSpec: """ Get backend-specific information on how to read/write blocks. Args: block_hashes: the list of block hashes identifying the blocks. blocks: the list of blocks. Returns: A LoadStoreSpec that can be used by a worker to read/write the blocks. """ raise NotImplementedError
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/kv_offload/backend.py", "license": "Apache License 2.0", "lines": 78, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/v1/kv_offload/backends/cpu.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import ctypes from collections.abc import Iterable from vllm.v1.core.kv_cache_utils import BlockHash from vllm.v1.kv_offload.abstract import LoadStoreSpec from vllm.v1.kv_offload.backend import Backend, BlockStatus from vllm.v1.kv_offload.mediums import CPULoadStoreSpec class CPUBlockStatus(BlockStatus): _fields_ = BlockStatus._fields_ + [("block_id", ctypes.c_int64)] # type: ignore def __init__(self, block_id: int): super().__init__() self.block_id = block_id class CPUBackend(Backend): def __init__(self, block_size: int, num_blocks: int): super().__init__(block_size=block_size, medium=CPULoadStoreSpec.medium()) self.num_blocks: int = num_blocks self.num_allocated_blocks: int = 0 self.allocated_blocks_free_list: list[int] = [] def get_num_free_blocks(self): return ( len(self.allocated_blocks_free_list) + self.num_blocks - self.num_allocated_blocks ) def allocate_blocks(self, block_hashes: list[BlockHash]) -> list[BlockStatus]: num_fresh_blocks = min( len(block_hashes), self.num_blocks - self.num_allocated_blocks ) num_reused_blocks = len(block_hashes) - num_fresh_blocks assert len(self.allocated_blocks_free_list) >= num_reused_blocks # allocate fresh blocks blocks: list[BlockStatus] = [] for _ in range(num_fresh_blocks): blocks.append(CPUBlockStatus(self.num_allocated_blocks)) self.num_allocated_blocks += 1 # allocate reused blocks for _ in range(num_reused_blocks): block_id = self.allocated_blocks_free_list.pop() blocks.append(CPUBlockStatus(block_id)) return blocks def free(self, block: BlockStatus): assert isinstance(block, CPUBlockStatus) self.allocated_blocks_free_list.append(block.block_id) def get_load_store_spec( self, block_hashes: Iterable[BlockHash], blocks: Iterable[BlockStatus] ) -> LoadStoreSpec: return CPULoadStoreSpec([block.block_id for block in blocks])
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/kv_offload/backends/cpu.py", "license": "Apache License 2.0", "lines": 48, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/v1/kv_offload/lru_manager.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from collections import OrderedDict from collections.abc import Iterable from vllm.v1.core.kv_cache_utils import BlockHash from vllm.v1.kv_offload.abstract import ( LoadStoreSpec, OffloadingEvent, OffloadingManager, PrepareStoreOutput, ) from vllm.v1.kv_offload.backend import Backend, BlockStatus class LRUOffloadingManager(OffloadingManager): """ An OffloadingManager with a pluggable backend, which evicts blocks by LRU. """ def __init__(self, backend: Backend, enable_events: bool = False): self.backend: Backend = backend # block_hash -> BlockStatus self.blocks: OrderedDict[BlockHash, BlockStatus] = OrderedDict() self.events: list[OffloadingEvent] | None = [] if enable_events else None def lookup(self, block_hashes: Iterable[BlockHash]) -> int | None: hit_count = 0 for block_hash in block_hashes: block = self.blocks.get(block_hash) if block is None or not block.is_ready: break hit_count += 1 return hit_count def prepare_load(self, block_hashes: Iterable[BlockHash]) -> LoadStoreSpec: blocks = [] for block_hash in block_hashes: block = self.blocks[block_hash] assert block.is_ready block.ref_cnt += 1 blocks.append(block) return self.backend.get_load_store_spec(block_hashes, blocks) def touch(self, block_hashes: Iterable[BlockHash]): for block_hash in reversed(list(block_hashes)): if self.blocks.get(block_hash): self.blocks.move_to_end(block_hash) def complete_load(self, block_hashes: Iterable[BlockHash]): for block_hash in block_hashes: block = self.blocks[block_hash] assert block.ref_cnt > 0 block.ref_cnt -= 1 def prepare_store( self, block_hashes: Iterable[BlockHash] ) -> PrepareStoreOutput | None: # filter out blocks that are already stored block_hashes_to_store = [ block_hash for block_hash in block_hashes if block_hash not in self.blocks ] num_blocks_to_evict = ( len(block_hashes_to_store) - self.backend.get_num_free_blocks() ) # build list of blocks to evict to_evict = [] if num_blocks_to_evict > 0: for block_hash, block in self.blocks.items(): if block.ref_cnt == 0: to_evict.append(block_hash) num_blocks_to_evict -= 1 if num_blocks_to_evict == 0: break else: # we could not evict enough blocks return None # evict blocks for block_hash in to_evict: self.backend.free(self.blocks.pop(block_hash)) if to_evict and self.events is not None: self.events.append( OffloadingEvent( block_hashes=to_evict, block_size=self.backend.block_size, medium=self.backend.medium, removed=True, ) ) blocks = self.backend.allocate_blocks(block_hashes_to_store) assert len(blocks) == len(block_hashes_to_store) for block_hash, block in zip(block_hashes_to_store, blocks): self.blocks[block_hash] = block # build store specs for allocated blocks store_spec = self.backend.get_load_store_spec(block_hashes_to_store, blocks) return PrepareStoreOutput( block_hashes_to_store=block_hashes_to_store, store_spec=store_spec, block_hashes_evicted=to_evict, ) def complete_store(self, block_hashes: Iterable[BlockHash], success: bool = True): stored_block_hashes: list[BlockHash] = [] if success: for block_hash in block_hashes: block = self.blocks[block_hash] if not block.is_ready: block.ref_cnt = 0 stored_block_hashes.append(block_hash) else: for block_hash in block_hashes: block = self.blocks[block_hash] if not block.is_ready: self.backend.free(block) del self.blocks[block_hash] if stored_block_hashes and self.events is not None: self.events.append( OffloadingEvent( block_hashes=stored_block_hashes, block_size=self.backend.block_size, medium=self.backend.medium, removed=False, ) ) def take_events(self) -> Iterable[OffloadingEvent]: if self.events is not None: yield from self.events self.events.clear()
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/kv_offload/lru_manager.py", "license": "Apache License 2.0", "lines": 118, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/models/language/pooling/test_token_classification.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import random import numpy as np import pytest import torch from transformers import AutoModelForTokenClassification from tests.models.utils import softmax from vllm.platforms import current_platform @pytest.fixture(autouse=True) def seed_everything(): """Seed all random number generators for reproducibility.""" seed = 0 random.seed(seed) np.random.seed(seed) torch.manual_seed(seed) if torch.cuda.is_available(): torch.cuda.manual_seed_all(seed) torch.backends.cudnn.deterministic = True torch.backends.cudnn.benchmark = False yield @pytest.mark.parametrize("model", ["boltuix/NeuroBERT-NER"]) # The float32 is required for this tiny model to pass the test. @pytest.mark.parametrize("dtype", ["float"]) @torch.inference_mode def test_bert_models( hf_runner, vllm_runner, example_prompts, model: str, dtype: str, ) -> None: with vllm_runner(model, max_model_len=None, dtype=dtype) as vllm_model: vllm_outputs = vllm_model.token_classify(example_prompts) # Use eager attention on ROCm to avoid HF Transformers flash attention # accuracy issues: https://github.com/vllm-project/vllm/issues/30167 hf_model_kwargs = {} if current_platform.is_rocm(): hf_model_kwargs["attn_implementation"] = "eager" with hf_runner( model, dtype=dtype, auto_cls=AutoModelForTokenClassification, model_kwargs=hf_model_kwargs, ) as hf_model: tokenizer = hf_model.tokenizer hf_outputs = [] for prompt in example_prompts: inputs = tokenizer([prompt], return_tensors="pt") inputs = hf_model.wrap_device(inputs) output = hf_model.model(**inputs) hf_outputs.append(softmax(output.logits[0])) # check logits difference for hf_output, vllm_output in zip(hf_outputs, vllm_outputs): hf_output = hf_output.detach().clone().cpu().float() vllm_output = vllm_output.detach().clone().cpu().float() torch.testing.assert_close(hf_output, vllm_output, atol=3.2e-2, rtol=1e-3) @pytest.mark.parametrize("model", ["disham993/electrical-ner-ModernBERT-base"]) @pytest.mark.parametrize("dtype", ["float"]) @pytest.mark.flaky(reruns=3) @torch.inference_mode def test_modernbert_models( hf_runner, vllm_runner, example_prompts, model: str, dtype: str, ) -> None: # NOTE: https://github.com/vllm-project/vllm/pull/32403 # `disham993/electrical-ner-ModernBERT-base` is a randomly initialized # model, which can cause numerical precision variance and edge cases. # We use @flaky(reruns=3) to mitigate intermittent failures. print( f"\n[NOTE] Testing {model} (randomly initialized weights) - " "flaky tolerance enabled due to numerical precision variance." ) with vllm_runner(model, max_model_len=None, dtype=dtype) as vllm_model: vllm_outputs = vllm_model.token_classify(example_prompts) # Use eager attention on ROCm to avoid HF Transformers flash attention # accuracy issues: https://github.com/vllm-project/vllm/issues/30167 hf_model_kwargs = {} if current_platform.is_rocm(): hf_model_kwargs["attn_implementation"] = "eager" with hf_runner( model, dtype=dtype, auto_cls=AutoModelForTokenClassification, model_kwargs=hf_model_kwargs, ) as hf_model: tokenizer = hf_model.tokenizer hf_outputs = [] for prompt in example_prompts: inputs = tokenizer([prompt], return_tensors="pt") inputs = hf_model.wrap_device(inputs) output = hf_model.model(**inputs) hf_outputs.append(softmax(output.logits[0])) # check logits difference for hf_output, vllm_output in zip(hf_outputs, vllm_outputs): hf_output = hf_output.detach().clone().cpu().float() vllm_output = vllm_output.detach().clone().cpu().float() torch.testing.assert_close(hf_output, vllm_output, atol=3.2e-2, rtol=1e-3) @pytest.mark.parametrize("model", ["bd2lcco/Qwen3-0.6B-finetuned"]) @pytest.mark.parametrize("dtype", ["float"]) @torch.inference_mode def test_auto_conversion( hf_runner, vllm_runner, example_prompts, model: str, dtype: str, ) -> None: with vllm_runner(model, max_model_len=1024, dtype=dtype) as vllm_model: vllm_outputs = vllm_model.token_classify(example_prompts) with hf_runner( model, dtype=dtype, auto_cls=AutoModelForTokenClassification ) as hf_model: tokenizer = hf_model.tokenizer hf_outputs = [] for prompt in example_prompts: inputs = tokenizer([prompt], return_tensors="pt") inputs = hf_model.wrap_device(inputs) output = hf_model.model(**inputs) hf_outputs.append(softmax(output.logits[0])) # check logits difference for hf_output, vllm_output in zip(hf_outputs, vllm_outputs): hf_output = hf_output.detach().clone().cpu().float() vllm_output = vllm_output.detach().clone().cpu().float() assert torch.allclose(hf_output, vllm_output, atol=1e-2)
{ "repo_id": "vllm-project/vllm", "file_path": "tests/models/language/pooling/test_token_classification.py", "license": "Apache License 2.0", "lines": 128, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/config/structured_outputs.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from typing import Any, Literal from pydantic import model_validator from typing_extensions import Self from vllm.config.utils import config from vllm.utils.hashing import safe_hash StructuredOutputsBackend = Literal[ "auto", "xgrammar", "guidance", "outlines", "lm-format-enforcer" ] @config class StructuredOutputsConfig: """Dataclass which contains structured outputs config for the engine.""" backend: StructuredOutputsBackend = "auto" """Which engine will be used for structured outputs (e.g. JSON schema, regex, etc) by default. With "auto", we will make opinionated choices based on request contents and what the backend libraries currently support, so the behavior is subject to change in each release.""" disable_fallback: bool = False """If `True`, vLLM will not fallback to a different backend on error.""" disable_any_whitespace: bool = False """If `True`, json output will always be compact without any whitespace. If `False`, the model may generate whitespace between JSON fields, which is still valid JSON. This is only supported for xgrammar and guidance backends.""" disable_additional_properties: bool = False """If `True`, the `guidance` backend will not use `additionalProperties` in the JSON schema. This is only supported for the `guidance` backend and is used to better align its behaviour with `outlines` and `xgrammar`.""" reasoning_parser: str = "" """Select the reasoning parser depending on the model that you're using. This is used to parse the reasoning content into OpenAI API format.""" reasoning_parser_plugin: str = "" """Path to a dynamically reasoning parser plugin that can be dynamically loaded and registered.""" enable_in_reasoning: bool = False """Whether to use structured input for reasoning.""" def compute_hash(self) -> str: """ WARNING: Whenever a new field is added to this config, ensure that it is included in the factors list if it affects the computation graph. Provide a hash that uniquely identifies all the configs that affect the structure of the computation graph from input ids/embeddings to the final hidden states, excluding anything before input ids/embeddings and after the final hidden states. """ # no factors to consider. # this config will not affect the computation graph. factors: list[Any] = [] hash_str = safe_hash(str(factors).encode(), usedforsecurity=False).hexdigest() return hash_str @model_validator(mode="after") def _validate_structured_output_config(self) -> Self: if self.disable_any_whitespace and self.backend not in ("xgrammar", "guidance"): raise ValueError( "disable_any_whitespace is only supported for " "xgrammar and guidance backends." ) if self.disable_additional_properties and self.backend != "guidance": raise ValueError( "disable_additional_properties is only supported " "for the guidance backend." ) return self
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/config/structured_outputs.py", "license": "Apache License 2.0", "lines": 66, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import torch import vllm.model_executor.layers.fused_moe.modular_kernel as mk from vllm.model_executor.layers.fused_moe.activation import MoEActivation from vllm.model_executor.layers.fused_moe.config import ( FusedMoEConfig, FusedMoEParallelConfig, RoutingMethodType, ) from vllm.model_executor.layers.fused_moe.utils import moe_kernel_quantize_input from vllm.model_executor.layers.quantization.utils.fp8_utils import ( per_token_group_quant_fp8, ) from vllm.model_executor.layers.quantization.utils.quant_utils import ( QuantKey, kFp8Dynamic128Sym, kFp8Static128BlockSym, kFp8StaticTensorSym, ) from vllm.platforms import current_platform from vllm.utils.torch_utils import direct_register_custom_op # # Methods used by the oracle for kernel selection. # def _supports_current_device() -> bool: """Supports only Blackwell-family GPUs.""" p = current_platform return p.is_cuda() and p.is_device_capability_family(100) def _supports_no_act_and_mul() -> bool: """Supports non-gated MoE.""" return True def _supports_quant_scheme( weight_key: QuantKey | None, activation_key: QuantKey | None, ) -> bool: """Supports Fp8 per-tensor and Fp8 block.""" SUPPORTED_W_A = [ (kFp8Static128BlockSym, kFp8Dynamic128Sym), (kFp8StaticTensorSym, kFp8StaticTensorSym), ] return (weight_key, activation_key) in SUPPORTED_W_A def _supports_activation(activation: MoEActivation) -> bool: return activation in [MoEActivation.SILU, MoEActivation.RELU2_NO_MUL] def _supports_routing_method( weight_key: QuantKey | None, activation_key: QuantKey | None, routing_method: RoutingMethodType, ) -> bool: """Monolithic kernels need to express router support.""" # NOTE(dbari): TopK routing could also be enabled, but need to validate models # NOTE(dbari): Default is not implemented and should not be enabled until it is if (weight_key, activation_key) == (kFp8Static128BlockSym, kFp8Dynamic128Sym): # NOTE(rob): potentially allow others here. This is a conservative list. return routing_method in [ RoutingMethodType.DeepSeekV3, RoutingMethodType.Renormalize, RoutingMethodType.RenormalizeNaive, ] elif (weight_key, activation_key) == (kFp8StaticTensorSym, kFp8StaticTensorSym): # NOTE(dbari): as above, potentially allow others here. return routing_method in [ RoutingMethodType.DeepSeekV3, RoutingMethodType.Llama4, RoutingMethodType.Renormalize, RoutingMethodType.RenormalizeNaive, ] else: raise ValueError("Unsupported quantization scheme.") def _supports_routing_method_bf16( routing_method: RoutingMethodType, ) -> bool: return routing_method in [ RoutingMethodType.Default, RoutingMethodType.Renormalize, RoutingMethodType.DeepSeekV3, RoutingMethodType.Llama4, RoutingMethodType.RenormalizeNaive, ] def _supports_parallel_config(moe_parallel_config: FusedMoEParallelConfig) -> bool: """Supports TRTLLM Kernel does not support EPLB.""" return not moe_parallel_config.enable_eplb def _supports_router_logits_dtype( router_logits_dtype: torch.dtype | None, routing_method: RoutingMethodType, ) -> bool: """ The FlashInfer TRTLLM FP8 kernel expects bfloat16 router_logits by default. Only DeepSeekV3 routing supports float32 router_logits (which is converted internally in the kernel). """ if router_logits_dtype == torch.float32: # Only DeepSeekV3 routing handles float32 logits # https://github.com/flashinfer-ai/flashinfer/issues/2469 return routing_method == RoutingMethodType.DeepSeekV3 return True def is_supported_config_trtllm_fp8( moe_config: FusedMoEConfig, weight_key: QuantKey | None, activation_key: QuantKey | None, activation_format: mk.FusedMoEActivationFormat, ) -> tuple[bool, str | None]: """ This method mirrors mk.FusedMoEPermuteExpertsUnpermute.is_supported_config """ def _make_reason(reason: str) -> str: return f"kernel does not support {reason}" if not _supports_current_device(): return False, _make_reason(f"current device {current_platform.device_name}") elif not (moe_config.is_act_and_mul or _supports_no_act_and_mul()): return False, _make_reason("no act_and_mul MLP layer") elif not _supports_activation(moe_config.activation): return False, _make_reason(f"{moe_config.activation} activation") elif not _supports_quant_scheme(weight_key, activation_key): return False, _make_reason(f"quantization scheme {weight_key}x{activation_key}") elif not _supports_parallel_config(moe_config.moe_parallel_config): return False, _make_reason(f"parallel config {moe_config.moe_parallel_config}") elif not _supports_routing_method( weight_key, activation_key, moe_config.routing_method ): return False, _make_reason(f"routing method {moe_config.routing_method}") elif activation_format != mk.FusedMoEActivationFormat.Standard: return False, _make_reason(f"activation format {activation_format}") elif not _supports_router_logits_dtype( moe_config.router_logits_dtype, moe_config.routing_method ): return False, _make_reason( "float32 router_logits with non-DeepSeekV3 routing " f"{moe_config.router_logits_dtype}x{moe_config.routing_method}" ) return True, None def is_supported_config_trtllm_bf16( moe_config: FusedMoEConfig, activation_format: mk.FusedMoEActivationFormat, ) -> tuple[bool, str | None]: """ This method mirrors mk.FusedMoEPermuteExpertsUnpermute.is_supported_config for BF16 unquantized kernels. """ def _make_reason(reason: str) -> str: return f"kernel does not support {reason}" if not _supports_current_device(): return False, _make_reason(f"current device {current_platform.device_name}") elif not (moe_config.is_act_and_mul or _supports_no_act_and_mul()): return False, _make_reason("no act_and_mul MLP layer") elif not _supports_activation(moe_config.activation): return False, _make_reason(f"{moe_config.activation} activation") elif not _supports_parallel_config(moe_config.moe_parallel_config): return False, _make_reason(f"parallel config {moe_config.moe_parallel_config}") elif not _supports_routing_method_bf16(moe_config.routing_method): return False, _make_reason(f"routing method {moe_config.routing_method}") elif activation_format != mk.FusedMoEActivationFormat.Standard: return False, _make_reason(f"activation format {activation_format}") return True, None def flashinfer_fused_moe_blockscale_fp8( routing_logits: torch.Tensor, routing_bias: torch.Tensor | None, x: torch.Tensor, w13_weight: torch.Tensor, w13_weight_scale_inv: torch.Tensor, w2_weight: torch.Tensor, w2_weight_scale_inv: torch.Tensor, global_num_experts: int, top_k: int, num_expert_group: int | None, topk_group: int | None, intermediate_size: int, expert_offset: int, local_num_experts: int, block_shape: list[int], routing_method_type: int, routed_scaling: float | None = 1.0, ) -> torch.Tensor: from vllm.utils.flashinfer import flashinfer_trtllm_fp8_block_scale_moe num_expert_group = num_expert_group if num_expert_group is not None else 0 topk_group = topk_group if topk_group is not None else 0 assert top_k <= global_num_experts assert top_k <= 10 assert global_num_experts % 4 == 0 assert block_shape == [128, 128] # Routing kernel expects #experts <= #threads 512 assert global_num_experts <= 512 # The DeepSeekV3 routing method requires float32 router logits. if routing_method_type == RoutingMethodType.DeepSeekV3: routing_logits = routing_logits.to(torch.float32) if routing_bias is not None: routing_bias = routing_bias.to(x.dtype) a_q, a_sf = per_token_group_quant_fp8(x, block_shape[1]) # NOTE: scales of hidden states have to be transposed! a_sf_t = a_sf.t().contiguous() return flashinfer_trtllm_fp8_block_scale_moe( routing_logits=routing_logits, routing_bias=routing_bias, hidden_states=a_q, hidden_states_scale=a_sf_t, gemm1_weights=w13_weight, gemm1_weights_scale=w13_weight_scale_inv, gemm2_weights=w2_weight, gemm2_weights_scale=w2_weight_scale_inv, num_experts=global_num_experts, top_k=top_k, n_group=num_expert_group, topk_group=topk_group, intermediate_size=intermediate_size, local_expert_offset=expert_offset, local_num_experts=local_num_experts, routed_scaling_factor=routed_scaling, routing_method_type=routing_method_type, use_shuffled_weight=False, ) def flashinfer_fused_moe_blockscale_fp8_fake( routing_logits: torch.Tensor, routing_bias: torch.Tensor | None, x: torch.Tensor, w13_weight: torch.Tensor, w13_weight_scale_inv: torch.Tensor, w2_weight: torch.Tensor, w2_weight_scale_inv: torch.Tensor, global_num_experts: int, top_k: int, num_expert_group: int, topk_group: int, intermediate_size: int, expert_offset: int, local_num_experts: int, block_shape: list[int], routing_method_type: int, routed_scaling: float = 1.0, ) -> torch.Tensor: return torch.empty_like(x) # TODO(bnell): Does this really need to be a torch.op? direct_register_custom_op( op_name="flashinfer_fused_moe_blockscale_fp8", op_func=flashinfer_fused_moe_blockscale_fp8, fake_impl=flashinfer_fused_moe_blockscale_fp8_fake, tags=(torch.Tag.needs_fixed_stride_order,), ) def fi_trtllm_fp8_per_tensor_moe( routing_logits: torch.Tensor, routing_bias: torch.Tensor | None, hidden_states: torch.Tensor, input_scale: torch.Tensor, gemm1_weights: torch.Tensor, gemm2_weights: torch.Tensor, output1_scales_scalar: torch.Tensor, output1_scales_gate_scalar: torch.Tensor, output2_scales_scalar: torch.Tensor, num_experts: int, top_k: int, num_expert_group: int | None, topk_group: int | None, intermediate_size: int, local_expert_offset: int, local_num_experts: int, use_routing_scales_on_input: bool, routing_method_type: int, activation_type: int, routed_scaling_factor: float = 1.0, ) -> torch.Tensor: num_expert_group = num_expert_group if num_expert_group is not None else 0 topk_group = topk_group if topk_group is not None else 0 quant_hidden_states, _ = moe_kernel_quantize_input( hidden_states, input_scale, quant_dtype=torch.float8_e4m3fn, per_act_token_quant=False, ) from flashinfer.fused_moe.core import ActivationType from vllm.utils.flashinfer import flashinfer_trtllm_fp8_per_tensor_scale_moe # The DeepSeekV3 routing method requires float32 router logits. if routing_method_type == RoutingMethodType.DeepSeekV3: routing_logits = routing_logits.to(torch.float32) return flashinfer_trtllm_fp8_per_tensor_scale_moe( routing_logits=routing_logits, routing_bias=routing_bias, hidden_states=quant_hidden_states, gemm1_weights=gemm1_weights, output1_scales_scalar=output1_scales_scalar, output1_scales_gate_scalar=output1_scales_gate_scalar, gemm2_weights=gemm2_weights, output2_scales_scalar=output2_scales_scalar, num_experts=num_experts, top_k=top_k, n_group=num_expert_group, topk_group=topk_group, intermediate_size=intermediate_size, local_expert_offset=local_expert_offset, local_num_experts=local_num_experts, routed_scaling_factor=routed_scaling_factor, use_routing_scales_on_input=use_routing_scales_on_input, routing_method_type=routing_method_type, # TODO: enum type Required for flashinfer==0.6.3, remove with update # https://github.com/flashinfer-ai/flashinfer/pull/2508 activation_type=ActivationType(activation_type), ) def fi_trtllm_fp8_per_tensor_moe_fake( routing_logits: torch.Tensor, routing_bias: torch.Tensor | None, hidden_states: torch.Tensor, input_scale: torch.Tensor, gemm1_weights: torch.Tensor, gemm2_weights: torch.Tensor, output1_scales_scalar: torch.Tensor, output1_scales_gate_scalar: torch.Tensor, output2_scales_scalar: torch.Tensor, num_experts: int, top_k: int, num_expert_group: int | None, topk_group: int | None, intermediate_size: int, local_expert_offset: int, local_num_experts: int, use_routing_scales_on_input: bool, routing_method_type: int, activation_type: int, routed_scaling_factor: float = 1.0, ) -> torch.Tensor: return torch.empty_like(hidden_states) # TODO(bnell): Does this really need to be a torch.op? direct_register_custom_op( op_name="fi_trtllm_fp8_per_tensor_moe", op_func=fi_trtllm_fp8_per_tensor_moe, mutates_args=["hidden_states"], fake_impl=fi_trtllm_fp8_per_tensor_moe_fake, tags=(torch.Tag.needs_fixed_stride_order,), ) def flashinfer_fused_moe_bf16( routing_logits: torch.Tensor, routing_bias: torch.Tensor | None, hidden_states: torch.Tensor, gemm1_weights: torch.Tensor, gemm2_weights: torch.Tensor, num_experts: int, top_k: int, n_group: int | None, topk_group: int | None, intermediate_size: int, local_expert_offset: int, local_num_experts: int, routing_method_type: int, tune_max_num_tokens: int = 8192, ) -> torch.Tensor: from vllm.utils.flashinfer import flashinfer_trtllm_bf16_moe return flashinfer_trtllm_bf16_moe( routing_logits=routing_logits, routing_bias=routing_bias, hidden_states=hidden_states, gemm1_weights=gemm1_weights, gemm2_weights=gemm2_weights, num_experts=num_experts, top_k=top_k, n_group=n_group, topk_group=topk_group, intermediate_size=intermediate_size, local_expert_offset=local_expert_offset, local_num_experts=local_num_experts, routing_method_type=routing_method_type, tune_max_num_tokens=tune_max_num_tokens, ) def flashinfer_fused_moe_bf16_fake( routing_logits: torch.Tensor, routing_bias: torch.Tensor | None, hidden_states: torch.Tensor, gemm1_weights: torch.Tensor, gemm2_weights: torch.Tensor, num_experts: int, top_k: int, n_group: int | None, topk_group: int | None, intermediate_size: int, local_expert_offset: int, local_num_experts: int, routing_method_type: int = RoutingMethodType.Renormalize, tune_max_num_tokens: int = 8192, ) -> torch.Tensor: return torch.empty_like(hidden_states) direct_register_custom_op( op_name="flashinfer_fused_moe_bf16", op_func=flashinfer_fused_moe_bf16, fake_impl=flashinfer_fused_moe_bf16_fake, tags=(torch.Tag.needs_fixed_stride_order,), )
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py", "license": "Apache License 2.0", "lines": 382, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/kernels/quantization/test_silu_mul_nvfp4_quant.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import pytest import torch from tests.kernels.quantization.nvfp4_utils import ( FLOAT4_E2M1_MAX, FLOAT8_E4M3_MAX, dequantize_nvfp4_to_dtype, ) from vllm._custom_ops import scaled_fp4_quant from vllm.model_executor.layers.activation import SiluAndMul from vllm.platforms import current_platform from vllm.utils.torch_utils import set_random_seed if not current_platform.has_device_capability(100): pytest.skip( reason="Nvfp4 Requires compute capability of 10 or above.", allow_module_level=True, ) FP4_DTYPE = torch.uint8 FP8_DTYPE = current_platform.fp8_dtype() DTYPES = [torch.float16, torch.bfloat16] SHAPES = [(128, 256), (128, 128), (256, 256), (256, 128)] BLOCK_SIZE = 16 @pytest.mark.parametrize("dtype", DTYPES) @pytest.mark.parametrize("shape", SHAPES) @torch.inference_mode() def test_silu_mul_nvfp4_quant( default_vllm_config, dtype: torch.dtype, shape: tuple[int, int], ) -> None: set_random_seed(42) device = "cuda:0" torch.set_default_device(device) x = torch.randn(shape, dtype=dtype) # ref op ref_output = SiluAndMul().forward_native(x) ref_global_scale = (FLOAT8_E4M3_MAX * FLOAT4_E2M1_MAX) / torch.abs( ref_output ).max().to(torch.float32) ref_output_quant, ref_block_scale = scaled_fp4_quant(ref_output, ref_global_scale) # fused op fused_output_quant = torch.empty_like(ref_output_quant) fused_block_scale = torch.empty_like(ref_block_scale) torch.ops._C.silu_and_mul_nvfp4_quant( fused_output_quant, fused_block_scale, x, ref_global_scale ) # check dtype assert ref_output_quant.dtype == FP4_DTYPE assert fused_output_quant.dtype == FP4_DTYPE assert ref_output_quant.shape == fused_output_quant.shape assert ref_block_scale.dtype == FP8_DTYPE assert fused_block_scale.dtype == FP8_DTYPE assert ref_block_scale.shape == fused_block_scale.shape # check dequantized output ref_output_dequant = dequantize_nvfp4_to_dtype( ref_output_quant, ref_block_scale, ref_global_scale, dtype, device ) fused_output_dequant = dequantize_nvfp4_to_dtype( fused_output_quant, fused_block_scale, ref_global_scale, dtype, device ) atol, rtol = 3e-1, 3e-1 torch.testing.assert_close( ref_output_dequant, fused_output_dequant, atol=atol, rtol=rtol )
{ "repo_id": "vllm-project/vllm", "file_path": "tests/kernels/quantization/test_silu_mul_nvfp4_quant.py", "license": "Apache License 2.0", "lines": 65, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/models/multimodal/pooling/test_radio.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import pytest import torch import torch.nn as nn from huggingface_hub import snapshot_download from transformers import AutoConfig, AutoModel, CLIPImageProcessor from vllm.distributed import cleanup_dist_env_and_memory from vllm.model_executor.models.radio import RadioModel from vllm.transformers_utils.configs.radio import RadioConfig from vllm.utils.torch_utils import STR_DTYPE_TO_TORCH_DTYPE from ....conftest import ImageTestAssets # we use snapshot_download to prevent conflicts between # dynamic_module and trust_remote_code for hf_runner DOWNLOAD_PATTERN = ["*.json", "*.py", "*.safetensors", "*.txt", "*.model"] @torch.inference_mode() def run_radio_test( image_assets: ImageTestAssets, model_id: str, *, dtype: str, ): model = snapshot_download(model_id, allow_patterns=DOWNLOAD_PATTERN) torch_dtype = STR_DTYPE_TO_TORCH_DTYPE[dtype] img_processor = CLIPImageProcessor.from_pretrained(model) images = [asset.pil_image for asset in image_assets] # Input resolution must be a multiple of `self.min_resolution_step`. # Using `self.get_nearest_supported_resolution`, for assets 432x642 the # nearest supported resolution is 432x640. pixel_values = [ img_processor(image, return_tensors="pt").pixel_values.to(torch_dtype)[ :, :, :, :640 ] for image in images ] hf_config = AutoConfig.from_pretrained(model_id, trust_remote_code=True) # RADIO model on HF does not properly handle torch_dtype argument # And relies on args["dtype"] which we have to patch manually: hf_config.args["dtype"] = torch_dtype hf_model = AutoModel.from_pretrained( model_id, config=hf_config, dtype=torch_dtype, trust_remote_code=True, ).to("cuda") hf_model.eval() # A HF model has image normalization as a part of model's forward # However in vLLM we don't make normalization a part of the model # forward step since mean/std stored as model's parameters and # subject to precision loss (when using fp16/bf16) which negatively # affects evaluation benchmarks. hf_model.make_preprocessor_external() hf_outputs_per_image = [ hf_model(pixel_value.to("cuda")) for pixel_value in pixel_values ] vllm_config = RadioConfig( model_name=hf_config.args["model"], **hf_config.args, ) vllm_model = RadioModel(vllm_config) vllm_model.load_weights(hf_model.state_dict()) vllm_model = vllm_model.to("cuda", torch_dtype) vllm_outputs_per_image = [ vllm_model(pixel_values=pixel_value.to("cuda")) for pixel_value in pixel_values ] del vllm_model, hf_model cleanup_dist_env_and_memory() cos_similar = nn.CosineSimilarity(dim=-1) for vllm_output, hf_output in zip(vllm_outputs_per_image, hf_outputs_per_image): assert cos_similar(vllm_output[0], hf_output[0]).mean() > 0.99 assert cos_similar(vllm_output[1], hf_output[1]).mean() > 0.99 @pytest.mark.parametrize( "model_id", [ "nvidia/C-RADIOv2-H", ], ) @pytest.mark.parametrize("dtype", ["half", "bfloat16"]) def test_radio( default_vllm_config, dist_init, image_assets, model_id, dtype: str ) -> None: run_radio_test( image_assets, model_id, dtype=dtype, )
{ "repo_id": "vllm-project/vllm", "file_path": "tests/models/multimodal/pooling/test_radio.py", "license": "Apache License 2.0", "lines": 86, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/model_executor/models/radio.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project # Copyright (c) 2023-2024, NVIDIA CORPORATION. All rights reserved. # # NVIDIA CORPORATION and its licensors retain all intellectual property # and proprietary rights in and to this software, related documentation # and any modifications thereto. Any use, reproduction, disclosure or # distribution of this software and related documentation without an express # license agreement from NVIDIA CORPORATION is strictly prohibited. import math from collections.abc import Iterable from itertools import repeat from typing import TypeAlias import torch import torch.nn as nn import torch.nn.functional as F from einops import rearrange from transformers import PretrainedConfig from vllm.model_executor.layers.quantization import QuantizationConfig from vllm.model_executor.model_loader.weight_utils import default_weight_loader from vllm.model_executor.models.intern_vit import ( InternParallelAttention, InternVisionEncoder, InternVisionEncoderLayer, ) input_dim_t: TypeAlias = int | tuple[int, int] norm_t: TypeAlias = tuple[float, float, float] | torch.Tensor def _ntuple(n): def parse(x): if isinstance(x, Iterable) and not isinstance(x, str): return tuple(x) return tuple(repeat(x, n)) return parse to_1tuple = _ntuple(1) to_2tuple = _ntuple(2) to_3tuple = _ntuple(3) to_4tuple = _ntuple(4) to_ntuple = _ntuple def calc_seq_len(size: tuple[int, int], patch_size: int) -> int: h, w = size return (h // patch_size) * (w // patch_size) def calc_seq_lens(sizes: list[tuple[int, int]], patch_size: int) -> list[int]: return [calc_seq_len(size, patch_size) for size in sizes] class ClsToken(nn.Module): def __init__( self, ndim: int, num_tokens: int = 1, enabled: bool = True, register_multiple: int | None = None, num_registers: int | None = None, ): super().__init__() self.ndim = ndim self.enabled = enabled self.num_registers = 0 self.num_tokens = num_tokens if enabled: if num_registers: self.num_registers = num_registers elif register_multiple: self.num_registers = register_multiple - ( num_tokens % register_multiple ) scale = ndim**-0.5 self.token = nn.Parameter( torch.randn(num_tokens + self.num_registers, ndim) * scale ) else: self.token = None self.num_patches = self.num_tokens + self.num_registers def forward(self, x: torch.Tensor): if self.token is None: return x token = self.token.unsqueeze(0).expand(x.shape[0], -1, -1) x = torch.cat( [ token, x, ], dim=1, ) return x class ViTPatchGenerator(nn.Module): def __init__( self, # config: PretrainedConfig, patch_size: int, embed_dim: int, input_dims: input_dim_t, abs_pos: bool = True, normalize_patches: bool = False, cls_token: bool = False, max_input_dims: input_dim_t | None = None, pos_dropout: float = 0.0, return_pos_enc: bool = False, num_cls_tokens: int = 1, register_multiple: int | None = None, num_registers: int | None = None, patch_bias: bool = False, device=None, dtype=None, ): super().__init__() if isinstance(input_dims, int): input_dims = (input_dims, input_dims) if max_input_dims is None: max_input_dims = input_dims if isinstance(max_input_dims, int): max_input_dims = (max_input_dims, max_input_dims) max_input_dims = tuple( int(math.ceil(d / patch_size) * patch_size) for d in max_input_dims ) self.cpe_mode = max_input_dims != input_dims self.pos_dropout = pos_dropout self.return_pos_enc = return_pos_enc factory = dict(device=device, dtype=dtype) self.patch_size = patch_size self.abs_pos = abs_pos self.embed_dim = embed_dim self.num_rows = max_input_dims[0] // patch_size self.num_cols = max_input_dims[1] // patch_size self.input_dims = tuple(d // patch_size for d in input_dims) self.num_patches = self.num_rows * self.num_cols self.max_input_dims = max_input_dims self.im_to_patches = Im2Patches(patch_size) self.embedder = ViTPatchLinear( patch_size, embed_dim, bias=patch_bias, **factory ) if abs_pos: scale = embed_dim**-0.5 self.pos_embed = nn.Parameter( torch.randn(1, self.num_patches, embed_dim, **factory) * scale ) self.cls_token = ClsToken( embed_dim, num_tokens=num_cls_tokens, enabled=cls_token, register_multiple=register_multiple, num_registers=num_registers, ) self.patch_normalizer = ( nn.LayerNorm(embed_dim) if normalize_patches else nn.Identity() ) def forward( self, x: torch.Tensor, imgs_sizes: list[tuple[int, int]] | None = None ) -> torch.Tensor: if imgs_sizes is not None: patches = self.embedder(x) patches, pos_enc = self.apply_pos_enc_dynamic( patches, imgs_sizes=imgs_sizes ) patches = self.cls_token_dynamic(patches, imgs_sizes=imgs_sizes) else: patches = self.embed_patches(x) patches, pos_enc = self.apply_pos_enc(patches, input_size=x.shape[2:]) patches = self.cls_token(patches) patches = self.patch_normalizer(patches) if self.return_pos_enc: return patches, pos_enc return patches def apply_pos_enc_dynamic( self, patches: torch.Tensor, imgs_sizes: list[tuple[int, int]] ) -> tuple[torch.Tensor, torch.Tensor | None]: if not self.abs_pos: return patches, None current_length = 0 pos_enc_list = [] for size in imgs_sizes: seq_length = calc_seq_len(size, self.patch_size) img_patches = patches[:, current_length : current_length + seq_length, :] pos_enc = self.get_pos_enc(patches.shape[0], input_size=size) img_patches_with_pos = img_patches + pos_enc patches = torch.cat( [ patches[:, :current_length, :], img_patches_with_pos, patches[:, current_length + seq_length :, :], ], dim=1, ) pos_enc_list.append(pos_enc) current_length += seq_length full_pos_enc = torch.cat(pos_enc_list, dim=1) if pos_enc_list else None return patches, full_pos_enc def cls_token_dynamic( self, patches: torch.Tensor, imgs_sizes: list[tuple[int, int]] ) -> torch.Tensor: if not self.cls_token.enabled: return patches out = [] current_length = 0 for seq_len in calc_seq_lens(imgs_sizes, self.patch_size): class_token = self.cls_token.token.unsqueeze(0).expand( patches.shape[0], -1, -1 ) out.append(class_token) out.append(patches[:, current_length : current_length + seq_len, :]) current_length += seq_len return torch.cat(out, dim=1) @property def apply_cls_token(self): return self.cls_token.enabled @property def num_cls_tokens(self): return self.cls_token.num_tokens @property def num_cls_patches(self): return self.cls_token.num_patches @property def num_registers(self): return self.cls_token.num_registers @property def num_skip(self): return self.num_cls_tokens + self.num_registers def _load_embed(self, src_embed: torch.Tensor, targ_embed: nn.Parameter): if src_embed.shape != targ_embed.shape: src_size = int(math.sqrt(src_embed.shape[1])) assert src_size**2 == src_embed.shape[1], ( "Unable to interpolate non-square embedding" ) src_embed = rearrange( src_embed, "b (h w) c -> b c h w", h=src_size, w=src_size ) src_embed = F.interpolate( src_embed, size=(self.num_rows, self.num_cols), mode="bicubic", align_corners=True, antialias=False, ) src_embed = rearrange(src_embed, "b c h w -> b (h w) c") targ_embed.data.copy_(src_embed) def _load_projection( self, src_proj_weight: torch.Tensor, targ_proj_weight: torch.Tensor ): if src_proj_weight.shape != targ_proj_weight.shape: src_patch_size = int(math.sqrt(src_proj_weight.shape[1] // 3)) assert (src_patch_size**2) * 3 == src_proj_weight.shape[1], ( "Unable to interpolate non-square patch size" ) src_proj_weight = rearrange( src_proj_weight, "b (c h w) -> b c h w", c=3, h=src_patch_size, w=src_patch_size, ) src_proj_weight = F.interpolate( src_proj_weight, size=(self.patch_size, self.patch_size), mode="bicubic", align_corners=True, antialias=False, ) src_proj_weight = rearrange(src_proj_weight, "b c h w -> b (c h w)") targ_proj_weight.data.copy_(src_proj_weight) def embed_patches(self, x: torch.Tensor) -> torch.Tensor: patches = self.im_to_patches(x) patches = self.embedder(patches) return patches def apply_pos_enc( self, patches: torch.Tensor, patch_idxs: torch.Tensor | None = None, input_size: tuple[int, int] | None = None, ) -> torch.Tensor: if not self.abs_pos: return patches pos_enc = self.get_pos_enc(patches.shape[0], patch_idxs, input_size) if self.training and self.pos_dropout > 0: keeps = ( torch.rand( patches.shape[0], 1, 1, dtype=pos_enc.dtype, device=pos_enc.device ) > self.pos_dropout ) pos_enc_drop = torch.where(keeps, pos_enc, 0) else: pos_enc_drop = pos_enc return patches + pos_enc_drop, pos_enc def get_pos_enc( self, batch_size: int, patch_idxs: torch.Tensor | None = None, input_size: tuple[int, int] | None = None, ) -> torch.Tensor: if input_size is None: input_dims = self.input_dims else: input_dims = tuple(d // self.patch_size for d in input_size) pos_embed = self._get_pos_embeddings(batch_size, input_dims) if patch_idxs is None: return pos_embed exp_patch_idxs = patch_idxs.unsqueeze(-1).expand(-1, -1, pos_embed.shape[-1]) pos_embed = torch.gather( pos_embed.expand(patch_idxs.shape[0], -1, -1), dim=1, index=exp_patch_idxs ) return pos_embed def _get_pos_embeddings(self, batch_size: int, input_dims: tuple[int, int]): if (self.num_rows, self.num_cols) == input_dims: return self.pos_embed pos_embed = self.pos_embed.reshape(1, self.num_rows, self.num_cols, -1).permute( 0, 3, 1, 2 ) def window_select(pos_embed): if input_dims[0] < pos_embed.shape[-2]: pos_embed = pos_embed[..., : input_dims[0], :] if input_dims[1] < pos_embed.shape[-1]: pos_embed = pos_embed[..., :, : input_dims[1]] return pos_embed if self.cpe_mode: if self.training: min_scale = math.sqrt(0.1) scale = ( torch.rand(batch_size, 1, 1, device=pos_embed.device) * (1 - min_scale) + min_scale ) aspect_min = math.log(3 / 4) aspect_max = -aspect_min aspect = torch.exp( torch.rand(batch_size, 1, 1, device=pos_embed.device) * (aspect_max - aspect_min) + aspect_min ) scale_x = scale * aspect scale_y = scale * (1 / aspect) scale_xy = torch.stack([scale_x, scale_y], dim=-1).clamp_(0, 1) pos_xy = torch.rand(batch_size, 1, 1, 2, device=pos_embed.device) * ( 1 - scale_xy ) lin_x = torch.linspace( 0, 1, steps=input_dims[1], device=pos_embed.device )[None, None].expand(batch_size, input_dims[0], -1) lin_y = torch.linspace( 0, 1, steps=input_dims[0], device=pos_embed.device )[None, :, None].expand(batch_size, -1, input_dims[1]) lin_xy = torch.stack([lin_x, lin_y], dim=-1) grid_xy = lin_xy * scale_xy + pos_xy # Convert to [-1, 1] range grid_xy.mul_(2).sub_(1) pos_embed = F.grid_sample( pos_embed.float().expand(batch_size, -1, -1, -1), grid=grid_xy, mode="bilinear", padding_mode="zeros", align_corners=True, ).to(pos_embed.dtype) else: max_dim = max(input_dims) pos_embed = F.interpolate( pos_embed.float(), size=(max_dim, max_dim), align_corners=True, mode="bilinear", ).to(pos_embed.dtype) pos_embed = window_select(pos_embed) else: pos_embed = window_select(pos_embed) if pos_embed.shape[-2:] != input_dims: pos_embed = F.interpolate( pos_embed.float(), size=input_dims, align_corners=True, mode="bilinear" ).to(pos_embed.dtype) pos_embed = pos_embed.flatten(2).permute(0, 2, 1) return pos_embed class Im2Patches(nn.Module): def __init__(self, patch_size: int): super().__init__() self.patch_size = patch_size def forward(self, x: torch.Tensor) -> torch.Tensor: if self.patch_size == 1: patches = x.flatten(2) patches = patches.permute(0, 2, 1) return patches py = x.shape[-2] // self.patch_size px = x.shape[-1] // self.patch_size patches = rearrange( x, "b c (py yy) (px xx) -> b (py px) (c yy xx)", py=py, yy=self.patch_size, px=px, xx=self.patch_size, ) return patches class ViTPatchLinear(nn.Linear): def __init__(self, patch_size: int, embed_dim: int, bias: bool = False, **factory): super().__init__(3 * (patch_size**2), embed_dim, bias=bias, **factory) self.patch_size = patch_size class RadioParallelAttention(InternParallelAttention): def forward( self, x: torch.Tensor, attn_mask: torch.Tensor | None = None ) -> torch.Tensor: if attn_mask is None: return super().forward(x) B, N, _ = x.shape qkv, _ = self.qkv(x) q, k, v = qkv.chunk(3, dim=-1) if self.qk_normalization: q, k = self._apply_qk_norm(q, k) q = q.view(B, N, self.num_heads_per_partition, self.head_dim) k = k.view(B, N, self.num_heads_per_partition, self.head_dim) v = v.view(B, N, self.num_heads_per_partition, self.head_dim) q, k, v = (t.transpose(1, 2) for t in (q, k, v)) out = F.scaled_dot_product_attention( q, k, v, attn_mask=attn_mask, scale=self.scale ) out = out.transpose(1, 2).reshape(B, N, -1) out, _ = self.proj(out) return out class RadioVisionEncoderLayer(InternVisionEncoderLayer): def __init__(self, *args, **kwargs) -> None: super().__init__(*args, attn_cls=RadioParallelAttention, **kwargs) def forward( self, hidden_states: torch.Tensor, attn_mask: torch.Tensor | None = None, ): hidden_states = ( hidden_states + self.attn(self.norm1(hidden_states), attn_mask=attn_mask) * self.ls1 ) hidden_states = hidden_states + self.mlp(self.norm2(hidden_states)) * self.ls2 return hidden_states class RadioVisionEncoder(InternVisionEncoder): def __init__(self, *args, **kwargs) -> None: super().__init__(*args, layer_cls=RadioVisionEncoderLayer, **kwargs) def forward( self, inputs_embeds: torch.Tensor, attn_mask: torch.Tensor | None = None, ): hidden_states = inputs_embeds for encoder_layer in self.layers: hidden_states = encoder_layer(hidden_states, attn_mask=attn_mask) return hidden_states class RadioInternVisionModel(nn.Module): packed_modules_mapping = { "qkv": ["qkv"], } def __init__( self, config: PretrainedConfig = None, quant_config: QuantizationConfig | None = None, *, num_hidden_layers_override: int | None = None, num_dummy_heads: int = 0, prefix: str = "", ) -> None: super().__init__() self.config = config self.img_size, self.grid_size, self.num_patches = self._init_img_size( to_2tuple(config.patch_size), config.image_size ) max_img_size = int( round(config.cpe_max_size / config.patch_size) * config.patch_size ) unique_teachers = set(t["name"] for t in config.teachers) self.patch_generator = ViTPatchGenerator( config.patch_size, config.hidden_size, input_dims=self.img_size, max_input_dims=max_img_size, cls_token=True, num_cls_tokens=len(unique_teachers) if config.cls_token_per_teacher else 1, register_multiple=config.register_multiple, ) self.encoder = RadioVisionEncoder( config=config, quant_config=quant_config, num_hidden_layers_override=num_hidden_layers_override, num_dummy_heads=num_dummy_heads, prefix=f"{prefix}.encoder", ) def _init_img_size(self, patch_size, img_size: int | tuple[int, int]): if img_size is None: return None, None, None img_size = to_2tuple(img_size) grid_size = tuple([s // p for s, p in zip(img_size, patch_size)]) num_patches = grid_size[0] * grid_size[1] return img_size, grid_size, num_patches def get_input_embeddings(self): return self.embeddings def create_inter_image_attention_mask( self, imgs_sizes: list[tuple[int, int]], device: torch.device ) -> torch.Tensor: patch_size = self.patch_generator.patch_size num_skip = self.patch_generator.num_skip seq_lens = calc_seq_lens(imgs_sizes, patch_size) patch_counts = [seq_len + num_skip for seq_len in seq_lens] total_patches = sum(patch_counts) # Create attention mask - default to False (mask out) mask = torch.zeros( total_patches, total_patches, dtype=torch.bool, device=device ) # Each image's patches can only attend to patches from the same image start_idx = 0 for patch_count in patch_counts: end_idx = start_idx + patch_count # Allow attention within this image's patches mask[start_idx:end_idx, start_idx:end_idx] = True start_idx = end_idx return mask def forward( self, x: torch.Tensor, imgs_sizes: torch.Tensor | None = None, ) -> torch.FloatTensor: hidden_states = self.patch_generator(x, imgs_sizes=imgs_sizes) attn_mask = None if imgs_sizes is not None and len(imgs_sizes) > 1: # Dynamic Resolution attn_mask = self.create_inter_image_attention_mask( imgs_sizes, device=x.device ) encoder_outputs = self.encoder(inputs_embeds=hidden_states, attn_mask=attn_mask) return encoder_outputs class RadioModel(nn.Module): packed_modules_mapping = { "qkv": ["qkv"], } def __init__( self, config: PretrainedConfig, quant_config: QuantizationConfig | None = None, *, num_hidden_layers_override: int | None = None, num_dummy_heads: int = 0, prefix: str = "", ) -> None: super().__init__() self.config = config self.model = RadioInternVisionModel( config=config, quant_config=quant_config, num_hidden_layers_override=num_hidden_layers_override, num_dummy_heads=num_dummy_heads, prefix=prefix, ) summary_idxs = None if config.teachers: summary_idxs = torch.tensor( [i for i, t in enumerate(config.teachers) if t.get("use_summary", True)] ) if summary_idxs.numel() > 0: self.register_buffer("summary_idxs", summary_idxs) self.summary_idxs = summary_idxs def forward( self, pixel_values: torch.Tensor | None = None, pixel_embeds: torch.Tensor | None = None, *, imgs_sizes: torch.Tensor | None = None, ) -> tuple[torch.FloatTensor, torch.FloatTensor]: y = self.model(pixel_values, imgs_sizes=imgs_sizes) return self._extract_final(y, imgs_sizes=imgs_sizes) def load_weights(self, weights) -> set[str]: loaded_params: set[str] = set() params_dict = dict(self.named_parameters()) if isinstance(weights, dict): weights_list = list(weights.items()) else: weights_list = list(weights) for name, weight in weights_list: if not name.startswith("radio_model."): # Skip non-radio weights continue sub = name[len("radio_model.") :] # drop "radio_model." prefix # Skip buffers not used in vLLM if sub in {"summary_idxs"}: continue if sub.startswith("input_conditioner."): # we normalize in the input processor, # based on norm and std values from the config continue vllm_key = None if sub.startswith("model.patch_generator."): vllm_key = f"model.patch_generator.{sub.split('.', 2)[-1]}" elif sub.startswith("input_conditioner."): vllm_key = f"input_conditioner.{sub.split('.', 1)[-1]}" elif sub.startswith("model.blocks."): # Encoder blocks: HF 'model.blocks.{i}.' -> # vLLM 'model.encoder.layers.{i}.' parts = sub.split(".") if len(parts) >= 4: layer_idx = parts[2] suffix = ".".join(parts[3:]) # Skip layer-scale entries that vLLM doesn't use if suffix in {"ls1", "ls2"} or suffix.startswith(("ls1.", "ls2.")): continue vllm_key = f"model.encoder.layers.{layer_idx}.{suffix}" if vllm_key and vllm_key in params_dict: param = params_dict[vllm_key] weight_loader = getattr(param, "weight_loader", default_weight_loader) weight_loader(param, weight) loaded_params.add(vllm_key) return loaded_params def _extract_final( self, y: torch.Tensor, imgs_sizes: list[tuple[int, int]] | None = None ) -> tuple[torch.FloatTensor, torch.FloatTensor]: # Remove CLS + REGISTERS tokens num_skip = self.model.patch_generator.num_skip patch_size = self.model.patch_generator.patch_size num_cls_tokens = self.model.patch_generator.num_cls_tokens if imgs_sizes is None: all_summary = y[:, :num_cls_tokens] all_feat = y[:, num_skip:] else: all_patches = [] summaries = [] current_pos = 0 for num_patches in calc_seq_lens(imgs_sizes, patch_size): patches = y[ :, current_pos + num_skip : current_pos + num_skip + num_patches, : ] all_patches.append(patches) summary = y[:, current_pos : current_pos + num_cls_tokens, :] summaries.append(summary) current_pos += num_skip + num_patches all_summary = torch.cat(summaries, dim=1) all_feat = torch.cat(all_patches, dim=1) if self.summary_idxs is not None: bb_summary = all_summary[:, self.summary_idxs] else: bb_summary = all_summary return bb_summary.flatten(1), all_feat
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/model_executor/models/radio.py", "license": "Apache License 2.0", "lines": 627, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/transformers_utils/configs/radio.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """Radio vision model configuration""" from typing import Any from transformers.configuration_utils import PretrainedConfig from transformers.utils import logging logger = logging.get_logger(__name__) VIT_TIMM_DIM_BY_NAME: dict[str, tuple[int, int, int, int]] = { "vit_small_patch16_224": (384, 12, 6, 1536), "vit_base_patch16_224": (768, 12, 12, 3072), "vit_large_patch16_224": (1024, 24, 16, 4096), "vit_huge_patch16_224": (1280, 32, 16, 5120), } OPENAI_CLIP_MEAN = (0.48145466, 0.4578275, 0.40821073) OPENAI_CLIP_STD = (0.26862954, 0.26130258, 0.27577711) class RadioConfig(PretrainedConfig): r""" This is the configuration class to store the configuration of a Radio vision model. It is used to instantiate a Radio model according to the specified arguments, defining the model architecture. Args: model_name: Name of the vision transformer model (e.g., "vit_base_patch16_224"). Used to determine architecture dimensions from `VIT_TIMM_DIM_BY_NAME`. image_size: The size (resolution) of each image. patch_size: The size (resolution) of each patch. qkv_bias: Whether to add a bias to the queries, keys and values. qk_normalization: Whether to apply normalization to queries and keys. norm_type: The normalization type to use. layer_norm_eps: The epsilon used by the layer normalization layers. initializer_factor: A factor for initializing all weight matrices. hidden_act: The non-linear activation function in the encoder. cpe_max_size: Maximum image size for position embeddings. norm_mean: Mean values for image normalization (RGB channels). Defaults to (0.48145466, 0.4578275, 0.40821073)). norm_std: Standard deviation values for image normalization (RGB channels). Defaults to (0.26862954, 0.26130258, 0.27577711)). register_multiple: Number of register tokens to use. teachers: A list of teacher model configurations. Each teacher configuration is a dict with keys like "name" and some may have "use_summary". cls_token_per_teacher: Whether to use a separate CLS token for each teacher. """ model_type = "radio" def __init__( self, model_name: str, image_size: int = 224, patch_size: int = 16, qkv_bias: bool = True, qk_normalization: bool = False, norm_type: str = "layer_norm", layer_norm_eps: float = 1e-6, initializer_factor: float = 1.0, hidden_act: str = "gelu", cpe_max_size: int = 2048, norm_mean: tuple[float, float, float] | list = OPENAI_CLIP_MEAN, norm_std: tuple[float, float, float] | list = OPENAI_CLIP_STD, register_multiple: int | None = None, teachers: list[dict[str, Any]] | None = None, cls_token_per_teacher: bool = False, **kwargs, ): self.model_name = model_name ( self.hidden_size, self.num_hidden_layers, self.num_attention_heads, self.intermediate_size, ) = VIT_TIMM_DIM_BY_NAME[model_name] self.image_size = image_size self.patch_size = patch_size self.qkv_bias = qkv_bias self.qk_normalization = qk_normalization self.norm_type = norm_type self.layer_norm_eps = layer_norm_eps self.initializer_factor = initializer_factor self.hidden_act = hidden_act self.cpe_max_size = cpe_max_size self.norm_mean = ( list(norm_mean) if isinstance(norm_mean, (tuple, list)) else norm_mean ) self.norm_std = ( list(norm_std) if isinstance(norm_std, (tuple, list)) else norm_std ) self.register_multiple = register_multiple self.teachers = teachers if teachers is not None else [] self.cls_token_per_teacher = cls_token_per_teacher super().__init__(**kwargs)
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/transformers_utils/configs/radio.py", "license": "Apache License 2.0", "lines": 88, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/model_executor/models/qwen3_vl.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project # Copyright 2025 The vLLM team. # Copyright 2025 The Qwen Team. # Copyright 2025 The HuggingFace Inc. team. # All rights reserved. # # This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX # and OPT implementations in this library. It has been modified from its # original forms to accommodate minor architectural differences compared # to GPT-NeoX and OPT used by the Meta AI team that trained the model. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """Inference-only Qwen3VL model compatible with HuggingFace weights.""" from collections.abc import Callable, Iterable, Iterator, Mapping, Sequence from functools import lru_cache, partial from itertools import islice from typing import Any import numpy as np import torch import torch.nn as nn import torch.nn.functional as F from transformers import BatchFeature from transformers.models.qwen2_vl import Qwen2VLImageProcessorFast from transformers.models.qwen2_vl.image_processing_qwen2_vl import ( smart_resize as image_smart_resize, ) from transformers.models.qwen3_vl import Qwen3VLProcessor, Qwen3VLVideoProcessor from transformers.models.qwen3_vl.configuration_qwen3_vl import ( Qwen3VLConfig, Qwen3VLVisionConfig, ) from transformers.models.qwen3_vl.video_processing_qwen3_vl import ( smart_resize as video_smart_resize, ) from transformers.video_utils import VideoMetadata from vllm.compilation.decorators import support_torch_compile from vllm.config import VllmConfig from vllm.config.multimodal import BaseDummyOptions, VideoDummyOptions from vllm.distributed import get_pp_group, parallel_state from vllm.logger import init_logger from vllm.model_executor.layers.activation import _ACTIVATION_REGISTRY from vllm.model_executor.layers.attention.mm_encoder_attention import ( MMEncoderAttention, ) from vllm.model_executor.layers.conv import Conv3dLayer from vllm.model_executor.layers.linear import ( ColumnParallelLinear, RowParallelLinear, ) from vllm.model_executor.layers.logits_processor import LogitsProcessor from vllm.model_executor.layers.quantization import QuantizationConfig from vllm.model_executor.layers.rotary_embedding import get_rope from vllm.model_executor.layers.vocab_parallel_embedding import ParallelLMHead from vllm.model_executor.model_loader.weight_utils import default_weight_loader from vllm.model_executor.models.module_mapping import MultiModelKeys from vllm.multimodal import MULTIMODAL_REGISTRY from vllm.multimodal.evs import ( compute_mrope_for_media, compute_retained_tokens_count, compute_retention_mask, recompute_mrope_positions, ) from vllm.multimodal.inputs import ( MultiModalDataDict, MultiModalFeatureSpec, MultiModalFieldConfig, MultiModalKwargsItem, MultiModalKwargsItems, PlaceholderRange, VideoItem, ) from vllm.multimodal.parse import ImageSize, MultiModalDataItems from vllm.multimodal.processing import ( BaseDummyInputsBuilder, BaseMultiModalProcessor, PromptReplacement, PromptUpdate, PromptUpdateDetails, ) from vllm.sequence import IntermediateTensors from vllm.utils.collection_utils import is_list_of from vllm.utils.math_utils import round_up from .interfaces import ( MultiModalEmbeddings, SupportsEagle3, SupportsLoRA, SupportsMRoPE, SupportsMultiModal, SupportsMultiModalPruning, SupportsPP, _require_is_multimodal, ) from .qwen2_5_vl import ( Qwen2_5_VisionAttention, Qwen2_5_VLImageEmbeddingInputs, Qwen2_5_VLImageInputs, Qwen2_5_VLImagePixelInputs, Qwen2_5_VLVideoEmbeddingInputs, Qwen2_5_VLVideoInputs, Qwen2_5_VLVideoPixelInputs, ) from .qwen2_vl import ( Qwen2VLMultiModalDataParser, Qwen2VLProcessingInfo, _create_qwen2vl_field_factory, ) from .qwen3 import Qwen3ForCausalLM, Qwen3Model from .utils import ( AutoWeightsLoader, PPMissingLayer, WeightsMapper, _merge_multimodal_embeddings, maybe_prefix, ) from .vision import ( get_vit_attn_backend, is_vit_use_data_parallel, run_dp_sharded_mrope_vision_model, ) logger = init_logger(__name__) # We use 2048 dummy video frames that would generate vision embeddings # of the maximum size. DUMMY_VIDEO_NUM_FRAMES = 2048 class Qwen3_VisionPatchEmbed(nn.Module): def __init__( self, patch_size: int = 14, temporal_patch_size: int = 2, in_channels: int = 3, hidden_size: int = 1152, ) -> None: super().__init__() self.patch_size = patch_size self.temporal_patch_size = temporal_patch_size self.hidden_size = hidden_size kernel_size = (temporal_patch_size, patch_size, patch_size) self.proj = Conv3dLayer( in_channels, hidden_size, kernel_size=kernel_size, stride=kernel_size, bias=True, ) def forward(self, x: torch.Tensor) -> torch.Tensor: L, C = x.shape x = x.view(L, -1, self.temporal_patch_size, self.patch_size, self.patch_size) x = self.proj(x).view(L, self.hidden_size) return x class Qwen3_VisionMLP(nn.Module): def __init__( self, in_features: int, hidden_features: int, bias: bool = False, act_fn: Callable[[torch.Tensor], torch.Tensor] = F.silu, quant_config: QuantizationConfig | None = None, prefix: str = "", ): super().__init__() use_data_parallel = is_vit_use_data_parallel() self.linear_fc1 = ColumnParallelLinear( in_features, hidden_features, bias=bias, quant_config=quant_config, return_bias=False, prefix=f"{prefix}.linear_fc1", disable_tp=use_data_parallel, ) self.linear_fc2 = RowParallelLinear( hidden_features, in_features, bias=bias, quant_config=quant_config, return_bias=False, prefix=f"{prefix}.linear_fc2", disable_tp=use_data_parallel, ) self.act_fn = act_fn def forward(self, x: torch.Tensor): mlp_output = self.linear_fc2(self.act_fn(self.linear_fc1(x))) return mlp_output class Qwen3_VisionBlock(nn.Module): def __init__( self, dim: int, num_heads: int, mlp_hidden_dim: int, act_fn: Callable[[torch.Tensor], torch.Tensor] = F.silu, norm_layer: Callable[[int], nn.Module] | None = None, quant_config: QuantizationConfig | None = None, prefix: str = "", ) -> None: super().__init__() if norm_layer is None: norm_layer = partial(nn.LayerNorm, eps=1e-6) self.norm1 = norm_layer(dim) self.norm2 = norm_layer(dim) self.attn = Qwen2_5_VisionAttention( embed_dim=dim, num_heads=num_heads, projection_size=dim, quant_config=quant_config, prefix=f"{prefix}.attn", ) self.mlp = Qwen3_VisionMLP( dim, mlp_hidden_dim, act_fn=act_fn, bias=True, quant_config=quant_config, prefix=f"{prefix}.mlp", ) def forward( self, x: torch.Tensor, cu_seqlens: torch.Tensor, rotary_pos_emb_cos: torch.Tensor, rotary_pos_emb_sin: torch.Tensor, max_seqlen: torch.Tensor, # Only used for Flash Attention sequence_lengths: torch.Tensor, # Only used for FlashInfer CuDNN backend ) -> torch.Tensor: x = x + self.attn( self.norm1(x), cu_seqlens=cu_seqlens, rotary_pos_emb_cos=rotary_pos_emb_cos, rotary_pos_emb_sin=rotary_pos_emb_sin, max_seqlen=max_seqlen, sequence_lengths=sequence_lengths, ) x = x + self.mlp(self.norm2(x)) return x class Qwen3_VisionPatchMerger(nn.Module): def __init__( self, d_model: int, context_dim: int, norm_layer: Callable[[int], nn.Module] | None = None, spatial_merge_size: int = 2, use_postshuffle_norm: bool = False, quant_config: QuantizationConfig | None = None, prefix: str = "", ) -> None: super().__init__() use_data_parallel = is_vit_use_data_parallel() self.hidden_size = context_dim * (spatial_merge_size**2) self.use_postshuffle_norm = use_postshuffle_norm if self.use_postshuffle_norm: context_dim = self.hidden_size if norm_layer is None: norm_layer = partial(nn.LayerNorm, eps=1e-6) self.norm = norm_layer(context_dim) self.linear_fc1 = ColumnParallelLinear( self.hidden_size, self.hidden_size, bias=True, quant_config=quant_config, prefix=f"{prefix}.linear_fc1", disable_tp=use_data_parallel, ) self.act_fn = nn.GELU() self.linear_fc2 = RowParallelLinear( self.hidden_size, d_model, bias=True, quant_config=quant_config, prefix=f"{prefix}.linear_fc2", disable_tp=use_data_parallel, ) def forward(self, x: torch.Tensor) -> torch.Tensor: if self.use_postshuffle_norm: x = self.norm(x.view(-1, self.hidden_size)) else: x = self.norm(x).view(-1, self.hidden_size) x_parallel, _ = self.linear_fc1(x) x_parallel = self.act_fn(x_parallel) out, _ = self.linear_fc2(x_parallel) return out class Qwen3_VisionTransformer(nn.Module): def __init__( self, vision_config: Qwen3VLVisionConfig, norm_eps: float = 1e-6, quant_config: QuantizationConfig | None = None, prefix: str = "", ) -> None: super().__init__() self.hidden_size = vision_config.hidden_size self.num_heads = vision_config.num_heads self.num_position_embeddings = vision_config.num_position_embeddings self.patch_size = vision_config.patch_size self.spatial_merge_size = vision_config.spatial_merge_size self.spatial_merge_unit = self.spatial_merge_size**2 self.temporal_patch_size = vision_config.temporal_patch_size self.deepstack_visual_indexes = ( vision_config.deepstack_visual_indexes if hasattr(vision_config, "deepstack_visual_indexes") else [] ) self.num_grid_per_side = int(self.num_position_embeddings**0.5) use_data_parallel = is_vit_use_data_parallel() self.tp_size = ( 1 if use_data_parallel else parallel_state.get_tensor_model_parallel_world_size() ) # NOTE: This is used for creating empty tensor for all_gather for # DP ViT. Here out_hidden_size is enlarged due to deepstack self.out_hidden_size = vision_config.out_hidden_size * ( 1 + len(self.deepstack_visual_indexes) ) self.patch_embed = Qwen3_VisionPatchEmbed( patch_size=self.patch_size, temporal_patch_size=self.temporal_patch_size, in_channels=vision_config.in_channels, hidden_size=self.hidden_size, ) self.pos_embed = nn.Embedding(self.num_position_embeddings, self.hidden_size) norm_layer = partial(nn.LayerNorm, eps=norm_eps) head_dim = self.hidden_size // self.num_heads self.rotary_pos_emb = get_rope( head_size=head_dim, max_position=8192, is_neox_style=True, rope_parameters={"partial_rotary_factor": 0.5}, ) self.merger = Qwen3_VisionPatchMerger( d_model=vision_config.out_hidden_size, context_dim=self.hidden_size, norm_layer=norm_layer, spatial_merge_size=self.spatial_merge_size, quant_config=quant_config, prefix=f"{prefix}.merger", ) self.deepstack_merger_list = nn.ModuleList( [ Qwen3_VisionPatchMerger( d_model=vision_config.out_hidden_size, context_dim=self.hidden_size, spatial_merge_size=self.spatial_merge_size, use_postshuffle_norm=True, norm_layer=norm_layer, quant_config=quant_config, prefix=f"{prefix}.deepstack_merger_list.{layer_idx}", ) for layer_idx in range(len(self.deepstack_visual_indexes)) ] ) self.attn_backend = get_vit_attn_backend( head_size=head_dim, dtype=torch.get_default_dtype(), ) self.blocks = nn.ModuleList( [ Qwen3_VisionBlock( dim=self.hidden_size, num_heads=self.num_heads, mlp_hidden_dim=vision_config.intermediate_size, act_fn=_ACTIVATION_REGISTRY[vision_config.hidden_act], norm_layer=norm_layer, quant_config=quant_config, prefix=f"{prefix}.blocks.{layer_idx}", ) for layer_idx in range(vision_config.depth) ] ) @property def dtype(self) -> torch.dtype: return self.patch_embed.proj.weight.dtype @property def device(self) -> torch.device: return self.patch_embed.proj.weight.device @staticmethod @lru_cache(maxsize=1024) def rot_pos_ids(h: int, w: int, spatial_merge_size: int) -> torch.Tensor: hpos_ids = np.broadcast_to(np.arange(h).reshape(h, 1), (h, w)) h_div = h // spatial_merge_size w_div = w // spatial_merge_size hpos_ids = hpos_ids.reshape( h_div, spatial_merge_size, w_div, spatial_merge_size, ) hpos_ids = hpos_ids.transpose(0, 2, 1, 3) hpos_ids = hpos_ids.flatten() wpos_ids = np.broadcast_to(np.arange(w).reshape(1, w), (h, w)) wpos_ids = wpos_ids.reshape( h_div, spatial_merge_size, w_div, spatial_merge_size, ) wpos_ids = wpos_ids.transpose(0, 2, 1, 3) wpos_ids = wpos_ids.flatten() return torch.from_numpy(np.stack([hpos_ids, wpos_ids], axis=-1)) def rot_pos_emb(self, grid_thw: list[list[int]]): max_grid_size = max(max(h, w) for _, h, w in grid_thw) pos_ids = [ self.rot_pos_ids(h, w, self.spatial_merge_size) if t == 1 else self.rot_pos_ids(h, w, self.spatial_merge_size).repeat(t, 1) for t, h, w in grid_thw ] pos_ids = torch.cat(pos_ids, dim=0).to(self.device, non_blocking=True) # Use pre-computed cos_sin_cache from RotaryEmbedding cos, sin = self.rotary_pos_emb.get_cos_sin(max_grid_size) cos_combined = cos[pos_ids].flatten(1) sin_combined = sin[pos_ids].flatten(1) return cos_combined, sin_combined def fast_pos_embed_interpolate(self, grid_thw: list[list[int]]) -> torch.Tensor: num_grid_per_side = self.num_grid_per_side m_size = self.spatial_merge_size hidden_dim = self.pos_embed.embedding_dim outputs = [] for t, h, w in grid_thw: h_idxs = torch.linspace( 0, num_grid_per_side - 1, h, dtype=torch.float32, device=self.device ) w_idxs = torch.linspace( 0, num_grid_per_side - 1, w, dtype=torch.float32, device=self.device ) h_floor = h_idxs.to(torch.long) w_floor = w_idxs.to(torch.long) h_ceil = torch.clamp(h_floor + 1, max=num_grid_per_side - 1) w_ceil = torch.clamp(w_floor + 1, max=num_grid_per_side - 1) dh = h_idxs - h_floor dw = w_idxs - w_floor # Create meshgrid view for all h, w vars dh_grid, dw_grid = torch.meshgrid(dh, dw, indexing="ij") h_floor_grid, w_floor_grid = torch.meshgrid(h_floor, w_floor, indexing="ij") h_ceil_grid, w_ceil_grid = torch.meshgrid(h_ceil, w_ceil, indexing="ij") # original computation of weights # w00 = (1 - dh_grid) * (1 - dw_grid) # w01 = (1 - dh_grid) * dw_grid # w10 = dh_grid * (1 - dw_grid) # w11 = dh_grid * dw_grid # we reuse w11 here to avoid duplicate # dh_grid * dw_grid computation w11 = dh_grid * dw_grid w10 = dh_grid - w11 w01 = dw_grid - w11 w00 = 1 - dh_grid - w01 h_grid = torch.stack([h_floor_grid, h_floor_grid, h_ceil_grid, h_ceil_grid]) w_grid = torch.stack([w_floor_grid, w_ceil_grid, w_floor_grid, w_ceil_grid]) h_grid_idx = h_grid * num_grid_per_side indices = (h_grid_idx + w_grid).reshape(4, -1) weights = torch.stack([w00, w01, w10, w11], dim=0).reshape(4, -1, 1) weights = weights.to(dtype=self.dtype) embeds = self.pos_embed(indices) embeds *= weights combined = embeds.sum(dim=0) combined = combined.reshape( h // m_size, m_size, w // m_size, m_size, hidden_dim ) combined = combined.permute(0, 2, 1, 3, 4).reshape(1, -1, hidden_dim) repeated = combined.expand(t, -1, -1).reshape(-1, hidden_dim) outputs.append(repeated) return torch.cat(outputs, dim=0) def forward( self, x: torch.Tensor, grid_thw: torch.Tensor | list[list[int]], ) -> torch.Tensor: hidden_states = x.to(device=self.device, dtype=self.dtype, non_blocking=True) hidden_states = self.patch_embed(hidden_states) if isinstance(grid_thw, list): grid_thw_list = grid_thw grid_thw = np.array(grid_thw, dtype=np.int32) else: grid_thw_list = grid_thw.tolist() grid_thw = grid_thw.numpy() pos_embeds = self.fast_pos_embed_interpolate(grid_thw_list) hidden_states = hidden_states + pos_embeds rotary_pos_emb_cos, rotary_pos_emb_sin = self.rot_pos_emb(grid_thw_list) cu_seqlens = np.repeat(grid_thw[:, 1] * grid_thw[:, 2], grid_thw[:, 0]).cumsum( axis=0, dtype=np.int32 ) cu_seqlens = np.concatenate([np.zeros(1, dtype=np.int32), cu_seqlens]) sequence_lengths = MMEncoderAttention.maybe_compute_sequence_lengths( self.attn_backend, cu_seqlens ) if sequence_lengths is not None: sequence_lengths = torch.from_numpy(sequence_lengths).to( self.device, non_blocking=True ) max_seqlen = torch.tensor( MMEncoderAttention.compute_max_seqlen(self.attn_backend, cu_seqlens), dtype=torch.int32, device=self.device, ) cu_seqlens = MMEncoderAttention.maybe_recompute_cu_seqlens( self.attn_backend, cu_seqlens, self.hidden_size, self.tp_size, ) cu_seqlens = torch.from_numpy(cu_seqlens).to(self.device, non_blocking=True) hidden_states = hidden_states.unsqueeze(1) deepstack_feature_lists = [] for layer_num, blk in enumerate(self.blocks): hidden_states = blk( hidden_states, cu_seqlens=cu_seqlens, rotary_pos_emb_cos=rotary_pos_emb_cos, rotary_pos_emb_sin=rotary_pos_emb_sin, max_seqlen=max_seqlen, sequence_lengths=sequence_lengths, ) if layer_num in self.deepstack_visual_indexes: deepstack_merger_idx = self.deepstack_visual_indexes.index(layer_num) deepstack_feature = self.deepstack_merger_list[deepstack_merger_idx]( hidden_states ) deepstack_feature_lists.append(deepstack_feature) hidden_states = self.merger(hidden_states) hidden_states = torch.cat( [hidden_states] + deepstack_feature_lists, dim=1 ) # [seq_len, hidden_size * (1 + depth_of_deepstack)] return hidden_states def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: stacked_params_mapping = [ # (param_name, shard_name, shard_id) ("attn.qkv.", "attn.q.", "q"), ("attn.qkv.", "attn.k.", "k"), ("attn.qkv.", "attn.v.", "v"), ] params_dict = dict(self.named_parameters(remove_duplicate=False)) loaded_params: set[str] = set() for name, loaded_weight in weights: 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: param = params_dict[name] weight_loader = getattr(param, "weight_loader", default_weight_loader) weight_loader(param, loaded_weight) loaded_params.add(name) return loaded_params class Qwen3VLProcessingInfo(Qwen2VLProcessingInfo): def get_hf_config(self): return self.ctx.get_hf_config(Qwen3VLConfig) def get_hf_processor(self, **kwargs: object) -> Qwen3VLProcessor: return self.ctx.get_hf_processor( Qwen3VLProcessor, use_fast=kwargs.pop("use_fast", True), **kwargs, ) def get_image_processor(self, **kwargs: object) -> Qwen2VLImageProcessorFast: return self.get_hf_processor(**kwargs).image_processor def get_video_processor(self, **kwargs: object) -> Qwen3VLVideoProcessor: return self.get_hf_processor(**kwargs).video_processor def get_data_parser(self): return Qwen2VLMultiModalDataParser( self.get_hf_config().vision_config.spatial_merge_size, video_needs_metadata=True, expected_hidden_size=self._get_expected_hidden_size(), ) def _get_vision_info( self, *, image_width: int, image_height: int, num_frames: int = 2, do_resize: bool = True, image_processor: Qwen2VLImageProcessorFast | Qwen3VLVideoProcessor, mm_kwargs: Mapping[str, object], ) -> tuple[ImageSize, int]: is_video = isinstance(image_processor, Qwen3VLVideoProcessor) hf_config = self.get_hf_config() vision_config = hf_config.vision_config patch_size = vision_config.patch_size merge_size = vision_config.spatial_merge_size temporal_patch_size = vision_config.temporal_patch_size mm_kwargs = self.ctx.get_merged_mm_kwargs(mm_kwargs) size = image_processor.size if override_size := mm_kwargs.get("size"): size = size | override_size if (override_min_pixels := mm_kwargs.get("min_pixels")) is not None: size = size | {"shortest_edge": override_min_pixels} if (override_max_pixels := mm_kwargs.get("max_pixels")) is not None: size = size | {"longest_edge": override_max_pixels} if do_resize: if is_video: smart_resize = video_smart_resize extra_kwargs = { "num_frames": num_frames, "temporal_factor": temporal_patch_size, } else: smart_resize = image_smart_resize extra_kwargs = {} resized_height, resized_width = smart_resize( height=image_height, width=image_width, factor=patch_size * merge_size, min_pixels=size["shortest_edge"], max_pixels=size["longest_edge"], **extra_kwargs, ) preprocessed_size = ImageSize(width=resized_width, height=resized_height) else: preprocessed_size = ImageSize(width=image_width, height=image_height) padded_num_frames = round_up(num_frames, temporal_patch_size) grid_t = max(padded_num_frames // temporal_patch_size, 1) grid_h = preprocessed_size.height // patch_size grid_w = preprocessed_size.width // patch_size num_patches = grid_t * grid_h * grid_w num_vision_tokens = num_patches // (merge_size**2) return preprocessed_size, num_vision_tokens def _get_max_video_frames(self, max_tokens: int, start_num_frames: int = 2) -> int: return super()._get_max_video_frames( max_tokens, start_num_frames=start_num_frames ) def get_num_frames_with_most_features( self, seq_len: int, mm_counts: Mapping[str, int], ) -> int: return super().get_num_frames_with_most_features( seq_len, mm_counts, max_frames_per_video=DUMMY_VIDEO_NUM_FRAMES ) def get_max_video_tokens( self, seq_len: int, mm_counts: Mapping[str, int], ) -> int: video_processor = self.get_video_processor() mm_kwargs = self.ctx.get_merged_mm_kwargs({}) video_size = mm_kwargs.get("size", video_processor.size) temporal_patch_size = mm_kwargs.get( "temporal_patch_size", video_processor.temporal_patch_size ) # video_max_pixels contains the temporal compression factor, # so we divide by 2 to get the maximum number of image pixels. video_max_pixels = video_size["longest_edge"] target_width, target_height = self.get_image_size_with_most_features( max_pixels=video_max_pixels // temporal_patch_size ) num_video_soft_tokens = self.get_num_video_tokens( image_width=target_width, image_height=target_height, num_frames=2, image_processor=video_processor, mm_kwargs={}, ) return num_video_soft_tokens def _calculate_timestamps( self, indices: list[int] | torch.Tensor, video_fps: float, merge_size: int ): if not isinstance(indices, list): indices = indices.tolist() if len(indices) % merge_size != 0: # don't update metadata's frames_indices directly indices = indices + [indices[-1]] * (merge_size - len(indices) % merge_size) timestamps = [idx / video_fps for idx in indices] timestamps = [ (timestamps[i] + timestamps[i + merge_size - 1]) / 2 for i in range(0, len(timestamps), merge_size) ] return timestamps def _get_video_second_idx( self, metadata: dict[str, Any], out_item: MultiModalKwargsItem, do_sample_frames: bool | None = None, sampled_fps: float | None = None, ) -> list[int]: video_processor = self.get_video_processor() merge_size = video_processor.merge_size indices = metadata["frames_indices"] # metadata["fps"] refers to the true fps of the input video. video_fps = metadata["fps"] if do_sample_frames is None: do_sample_frames = metadata.get("do_sample_frames", False) # If video frames are sampled in HF processor (instead of vLLM # video loader), we need to re-calculate the indices from original # metadata. if do_sample_frames: # here video_fps is the fps of the sampled video, and # metadata["fps"] refers to the fps of the original video. sampled_fps = sampled_fps if sampled_fps else video_processor.fps total_num_frames = metadata["total_num_frames"] num_frames = int(total_num_frames / metadata["fps"] * sampled_fps) num_frames = min( min( max(num_frames, video_processor.min_frames), video_processor.max_frames, ), total_num_frames, ) indices = ( np.linspace(0, total_num_frames - 1, num_frames) .round() .astype(int) .tolist() ) timestamps = self._calculate_timestamps(indices, video_fps, merge_size) return timestamps class Qwen3VLDummyInputsBuilder(BaseDummyInputsBuilder[Qwen3VLProcessingInfo]): def get_dummy_text(self, mm_counts: Mapping[str, int]) -> str: num_images = mm_counts.get("image", 0) num_videos = mm_counts.get("video", 0) image_token = "<|vision_start|><|image_pad|><|vision_end|>" video_token = "<|vision_start|><|video_pad|><|vision_end|>" return image_token * num_images + video_token * num_videos def get_dummy_mm_data( self, seq_len: int, mm_counts: Mapping[str, int], mm_options: Mapping[str, BaseDummyOptions], ) -> MultiModalDataDict: num_images = mm_counts.get("image", 0) num_videos = mm_counts.get("video", 0) image_overrides = mm_options.get("image") video_overrides = mm_options.get("video") target_image_width, target_image_height = ( self.info.get_image_size_with_most_features() ) # treat videos as special images target_num_frames = 2 if video_overrides: assert isinstance(video_overrides, VideoDummyOptions) num_frames_override = video_overrides.num_frames if num_frames_override: if num_frames_override > target_num_frames: logger.warning( "video.num_frames override (%d) exceeds model's " "maximum number of frames (%d), will be ignored", num_frames_override, target_num_frames, ) if num_frames_override < 2: logger.warning( "video.num_frames override (%d) cannot be less " "than 2, will be ignored", num_frames_override, ) target_num_frames = min(target_num_frames, num_frames_override) target_num_frames = max(target_num_frames, 2) video_processor = self.info.get_video_processor() mm_kwargs = self.info.ctx.get_merged_mm_kwargs({}) video_size = mm_kwargs.get("size", video_processor.size) temporal_patch_size = mm_kwargs.get( "temporal_patch_size", video_processor.temporal_patch_size ) # video_max_pixels contains the temporal compression factor, # so we divide by 2 to get the maximum number of image pixels. video_max_pixels = video_size["longest_edge"] target_video_width, target_video_height = ( self.info.get_image_size_with_most_features( max_pixels=video_max_pixels // temporal_patch_size ) ) target_video_size, _ = self.info._get_vision_info( image_width=target_video_width, image_height=target_video_height, num_frames=target_num_frames, image_processor=video_processor, mm_kwargs={}, ) # NOTE: we need to do this check here since Qwen3-VL resizes video # frames depending on how many frames there are. target_video_width, target_video_height = ( target_video_size.width, target_video_size.height, ) if video_overrides: assert isinstance(video_overrides, VideoDummyOptions) width_override = video_overrides.width if width_override: if width_override > target_video_width: logger.warning( "video.width override (%d) exceeds model's " "maximum width (%d), will be ignored", width_override, target_video_width, ) target_video_width = min(target_video_width, width_override) height_override = video_overrides.height if height_override: if height_override > target_video_height: logger.warning( "video.height override (%d) exceeds model's " "maximum height (%d), will be ignored", height_override, target_video_height, ) target_video_height = min(target_video_height, height_override) return { "image": self._get_dummy_images( width=target_image_width, height=target_image_height, num_images=num_images, overrides=image_overrides, ), "video": self._get_dummy_videos( width=target_video_width, height=target_video_height, num_frames=target_num_frames, num_videos=num_videos, ), } def _get_dummy_videos( self, *, width: int, height: int, num_frames: int, num_videos: int, ) -> list[VideoItem]: video = np.full((num_frames, width, height, 3), 255, dtype=np.uint8) video_items = [] for i in range(num_videos): video_metadata = { "fps": 2.0, "duration": num_frames / 2.0, "total_num_frames": num_frames, "frames_indices": [i for i in range(num_frames)], "video_backend": "opencv", "do_sample_frames": False, } video_item = (video.copy(), video_metadata) video_items.append(video_item) return video_items class Qwen3VLMultiModalProcessor(BaseMultiModalProcessor[Qwen3VLProcessingInfo]): def _call_hf_processor( self, prompt: str, mm_data: Mapping[str, object], mm_kwargs: Mapping[str, object], tok_kwargs: Mapping[str, object], ) -> BatchFeature: mm_data = dict(mm_data) processor = self.info.get_hf_processor(**mm_kwargs) # Separate video processing from image processing. Because the videos # are processed into several image patches if videos := mm_data.pop("videos", []): video_grid_thw_lst = [] pixel_values_videos_lst = [] for item in videos: video_array, metadata = item # NOTE: @JJJYmmm new attr metadata.frames_indices indicates # the sampled frames indices of pre-sampled videos, which is # used to calculate the timestamps. Make sure that # do_sample_frames in mm_kwargs is false for presampled videos. # NOTE: a copy of is created to update do_sample_frames, # otherwise mm_hash for the object will be incorrect. video_mm_kwargs = dict(**mm_kwargs) if "do_sample_frames" not in video_mm_kwargs: # qwen_vl_utils already has "do_sample_frames" in # mm_kwargs, don't overwrite it. video_mm_kwargs["do_sample_frames"] = metadata.get( "do_sample_frames", False ) metadata = VideoMetadata( **{k: metadata[k] for k in metadata if k != "do_sample_frames"} ) video_mm_data = dict() video_mm_data["videos"] = [[video_array]] video_mm_data["video_metadata"] = [[metadata]] video_outputs = super()._call_hf_processor( prompt="<|vision_start|><|video_pad|><|vision_end|>", mm_data=video_mm_data, mm_kwargs=video_mm_kwargs, tok_kwargs=tok_kwargs, ) input_ids = video_outputs.pop("input_ids") video_placeholder = processor.tokenizer.batch_decode(input_ids)[0] prompt = prompt.replace( "<|vision_start|><|video_pad|><|vision_end|>", video_placeholder, 1, ) video_grid_thw_lst.append(video_outputs["video_grid_thw"]) pixel_values_videos_lst.append(video_outputs["pixel_values_videos"]) video_outputs = dict( pixel_values_videos=torch.cat(pixel_values_videos_lst), video_grid_thw=torch.cat(video_grid_thw_lst), ) else: video_outputs = dict() processed_outputs = super()._call_hf_processor( prompt=prompt, mm_data=mm_data, mm_kwargs=mm_kwargs, tok_kwargs=tok_kwargs, ) combined_outputs = dict( processed_outputs, **video_outputs, ) return BatchFeature(combined_outputs) def _get_mm_fields_config( self, hf_inputs: BatchFeature, hf_processor_mm_kwargs: Mapping[str, object], ) -> Mapping[str, MultiModalFieldConfig]: return _create_qwen2vl_field_factory( self.info.get_hf_config().vision_config.spatial_merge_size )(hf_inputs) def _get_prompt_updates( self, mm_items: MultiModalDataItems, hf_processor_mm_kwargs: Mapping[str, Any], out_mm_kwargs: MultiModalKwargsItems, ) -> Sequence[PromptUpdate]: hf_processor = self.info.get_hf_processor(**hf_processor_mm_kwargs) image_processor = self.info.get_image_processor(**hf_processor_mm_kwargs) tokenizer = self.info.get_tokenizer() hf_config = self.info.get_hf_config() video_token_id = hf_config.video_token_id vision_start_token_id = hf_config.vision_start_token_id vision_end_token_id = hf_config.vision_end_token_id merge_length = image_processor.merge_size**2 def get_image_replacement_qwen3vl(item_idx: int): out_item = out_mm_kwargs["image"][item_idx] grid_thw = out_item["image_grid_thw"].data assert isinstance(grid_thw, torch.Tensor) num_tokens = int(grid_thw.prod()) // merge_length return [hf_processor.image_token_id] * num_tokens def get_video_replacement_qwen3vl(item_idx: int): out_item = out_mm_kwargs["video"][item_idx] grid_thw = out_item["video_grid_thw"].data assert isinstance(grid_thw, torch.Tensor) video, metadata = mm_items["video"][item_idx] do_sample_frames = hf_processor_mm_kwargs.get("do_sample_frames") sampled_fps = hf_processor_mm_kwargs.get("fps") if is_list_of(sampled_fps, float): sampled_fps = sampled_fps[item_idx] timestamps = self.info._get_video_second_idx( metadata, out_item, do_sample_frames, sampled_fps ) assert len(timestamps) == grid_thw[0], ( f"The timestamps length({len(timestamps)}) should be equal " f"video length ({grid_thw[0]})." ) frames_idx_token = [ tokenizer.encode(f"<{curr_time:.1f} seconds>", add_special_tokens=False) for curr_time in timestamps ] tokens_per_frame = int(grid_thw[1:].prod()) // merge_length per_frame_token_counts = [tokens_per_frame for _ in frames_idx_token] video_pruning_rate = self.info.ctx.get_mm_config().video_pruning_rate if video_pruning_rate is not None and video_pruning_rate > 0.0: total_retained = compute_retained_tokens_count( tokens_per_frame, len(frames_idx_token), video_pruning_rate, ) if len(frames_idx_token) == 0: per_frame_token_counts = [] elif len(frames_idx_token) == 1: per_frame_token_counts = [tokens_per_frame] else: first_frame_tokens = tokens_per_frame remaining_tokens = max(total_retained - first_frame_tokens, 0) base = remaining_tokens // (len(frames_idx_token) - 1) remainder = remaining_tokens % (len(frames_idx_token) - 1) per_frame_token_counts = [first_frame_tokens] for frame_idx in range(1, len(frames_idx_token)): extra = base + (1 if (frame_idx - 1) < remainder else 0) per_frame_token_counts.append(extra) placeholder = [] for frame_idx, timestamp_tokens in enumerate(frames_idx_token): placeholder.extend(timestamp_tokens) tokens_this_frame = per_frame_token_counts[ frame_idx if frame_idx < len(per_frame_token_counts) else -1 ] placeholder.extend( [vision_start_token_id] + [video_token_id] * tokens_this_frame + [vision_end_token_id] ) return PromptUpdateDetails.select_token_id(placeholder, video_token_id) return [ PromptReplacement( modality="image", target=hf_processor.image_token, replacement=get_image_replacement_qwen3vl, ), # NOTE: We match string on purpose since searching sequence of # token ids takes more time. PromptReplacement( modality="video", target="<|vision_start|><|video_pad|><|vision_end|>", replacement=get_video_replacement_qwen3vl, ), ] @support_torch_compile( dynamic_arg_dims={ "input_ids": 0, # positions is of shape (3, seq_len) if mrope is enabled for qwen2-vl, # otherwise (seq_len, ). "positions": -1, "intermediate_tensors": 0, "inputs_embeds": 0, # the same shape as input_embeds "deepstack_input_embeds": 0, } ) class Qwen3LLMModel(Qwen3Model): def forward( self, input_ids: torch.Tensor | None, positions: torch.Tensor, intermediate_tensors: IntermediateTensors | None = None, inputs_embeds: torch.Tensor | None = None, # args for deepstack deepstack_input_embeds: IntermediateTensors | None = None, ) -> torch.Tensor | IntermediateTensors: if get_pp_group().is_first_rank: if inputs_embeds is not None: hidden_states = inputs_embeds else: hidden_states = self.embed_input_ids(input_ids) residual = None else: assert intermediate_tensors is not None hidden_states = intermediate_tensors["hidden_states"] residual = intermediate_tensors["residual"] aux_hidden_states = [] for layer_idx, layer in islice( enumerate(self.layers), self.start_layer, self.end_layer ): if layer_idx in self.aux_hidden_state_layers: aux_hidden_states.append(hidden_states + residual) hidden_states, residual = layer( positions, hidden_states, residual, ) if deepstack_input_embeds is not None and layer_idx in range( 0, len(deepstack_input_embeds) ): hidden_states = ( hidden_states + deepstack_input_embeds[f"deepstack_input_embeds_{layer_idx}"] ) if not get_pp_group().is_last_rank: return IntermediateTensors( {"hidden_states": hidden_states, "residual": residual} ) hidden_states, _ = self.norm(hidden_states, residual) if len(aux_hidden_states) > 0: return hidden_states, aux_hidden_states return hidden_states class Qwen3LLMForCausalLM(Qwen3ForCausalLM): def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): super(Qwen3ForCausalLM, self).__init__() config = vllm_config.model_config.hf_config quant_config = vllm_config.quant_config self.config = config self.quant_config = quant_config self.model = Qwen3LLMModel( vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model") ) if get_pp_group().is_last_rank: if config.tie_word_embeddings: self.lm_head = self.model.embed_tokens else: self.lm_head = ParallelLMHead( config.vocab_size, config.hidden_size, quant_config=quant_config, prefix="lm_head", ) else: self.lm_head = PPMissingLayer() self.logits_processor = LogitsProcessor(config.vocab_size) self.make_empty_intermediate_tensors = ( self.model.make_empty_intermediate_tensors ) @MULTIMODAL_REGISTRY.register_processor( Qwen3VLMultiModalProcessor, info=Qwen3VLProcessingInfo, dummy_inputs=Qwen3VLDummyInputsBuilder, ) class Qwen3VLForConditionalGeneration( nn.Module, SupportsMultiModal, SupportsLoRA, SupportsPP, SupportsMRoPE, SupportsEagle3, SupportsMultiModalPruning, ): packed_modules_mapping = { "qkv_proj": [ "q_proj", "k_proj", "v_proj", ], "gate_up_proj": [ "gate_proj", "up_proj", ], "qkv": ["qkv"], # For vision tower's already-packed QKV } supports_encoder_tp_data = True # To ensure correct weight loading and mapping. hf_to_vllm_mapper = WeightsMapper( orig_to_new_prefix={ "model.visual.": "visual.", "lm_head.": "language_model.lm_head.", "model.language_model.": "language_model.model.", } ) @classmethod def get_placeholder_str(cls, modality: str, i: int) -> str | None: if modality.startswith("image"): return "<|vision_start|><|image_pad|><|vision_end|>" if modality.startswith("video"): return "<|vision_start|><|video_pad|><|vision_end|>" raise ValueError("Only image or video modality is supported") def __init__(self, *, vllm_config: VllmConfig, prefix: str = "model"): super().__init__() config: Qwen3VLConfig = vllm_config.model_config.hf_config quant_config = vllm_config.quant_config multimodal_config = vllm_config.model_config.multimodal_config self.config = config self.multimodal_config = multimodal_config self.use_data_parallel = multimodal_config.mm_encoder_tp_mode == "data" self.video_pruning_rate = multimodal_config.video_pruning_rate self.is_multimodal_pruning_enabled = ( multimodal_config.is_multimodal_pruning_enabled() ) self.use_deepstack = hasattr(config.vision_config, "deepstack_visual_indexes") self.deepstack_num_level = ( len(config.vision_config.deepstack_visual_indexes) if self.use_deepstack else 0 ) self.visual_dim = config.vision_config.out_hidden_size self.multiscale_dim = self.visual_dim * self.deepstack_num_level with self._mark_tower_model(vllm_config, {"image", "video"}): self.visual = Qwen3_VisionTransformer( config.vision_config, norm_eps=getattr(config, "rms_norm_eps", 1e-6), quant_config=quant_config, prefix=maybe_prefix(prefix, "visual"), ) # register buffer for deepstack if self.use_deepstack: self.deepstack_input_embeds = [ torch.zeros( vllm_config.scheduler_config.max_num_batched_tokens, config.text_config.hidden_size, ) for _ in range(self.deepstack_num_level) ] with self._mark_language_model(vllm_config): self.language_model = Qwen3LLMForCausalLM( vllm_config=vllm_config.with_hf_config(config.text_config), prefix=maybe_prefix(prefix, "language_model"), ) if not get_pp_group().is_first_rank and hasattr( config.vision_config, "deepstack_visual_indexes" ): assert self.language_model.start_layer >= len( config.vision_config.deepstack_visual_indexes ), ( "start_layer should be greater than or equal to " "len(deepstack_visual_indexes)" ) self.make_empty_intermediate_tensors = ( self.language_model.make_empty_intermediate_tensors ) def set_aux_hidden_state_layers(self, layers: tuple[int, ...]) -> None: self.language_model.model.aux_hidden_state_layers = layers def get_eagle3_aux_hidden_state_layers(self) -> tuple[int, ...]: num_layers = len(self.language_model.model.layers) return (2, num_layers // 2, num_layers - 3) def _get_deepstack_input_embeds( self, num_tokens: int, ) -> IntermediateTensors | None: if not getattr(self, "deepstack_input_embeds", None): return None # If vision tower is skipped # get deepstack_input_embeds from buffer, and clear the buffer return IntermediateTensors( { f"deepstack_input_embeds_{idx}": self.deepstack_input_embeds[idx][ :num_tokens ] for idx in range(self.deepstack_num_level) } ) def _set_deepstack_input_embeds(self, deepstack_input_embeds: torch.Tensor) -> None: if not getattr(self, "deepstack_input_embeds", None): return # set deepstack_input_embeds to buffer num_tokens = deepstack_input_embeds.size(1) if num_tokens > self.deepstack_input_embeds[0].size(0): self.deepstack_input_embeds = [ torch.zeros( num_tokens, self.config.text_config.hidden_size, device=self.deepstack_input_embeds[0].device, dtype=self.deepstack_input_embeds[0].dtype, ) for _ in range(self.deepstack_num_level) ] for idx in range(self.deepstack_num_level): self.deepstack_input_embeds[idx][:num_tokens].copy_( deepstack_input_embeds[idx] ) def _clear_deepstack_input_embeds(self, num_tokens: int) -> None: if not getattr(self, "deepstack_input_embeds", None): return # clear deepstack_input_embeds in buffer if num_tokens > 0: for idx in range(self.deepstack_num_level): self.deepstack_input_embeds[idx][:num_tokens].zero_() def _parse_and_validate_image_input( self, **kwargs: object ) -> Qwen2_5_VLImageInputs | None: pixel_values = kwargs.pop("pixel_values", None) image_embeds = kwargs.pop("image_embeds", None) image_grid_thw = kwargs.pop("image_grid_thw", None) if pixel_values is None and image_embeds is None: return None if pixel_values is not None: return Qwen2_5_VLImagePixelInputs( type="pixel_values", pixel_values=pixel_values, image_grid_thw=image_grid_thw, ) if image_embeds is not None: return Qwen2_5_VLImageEmbeddingInputs( type="image_embeds", image_embeds=image_embeds, image_grid_thw=image_grid_thw, ) def _parse_and_validate_video_input( self, **kwargs: object ) -> Qwen2_5_VLVideoInputs | None: pixel_values_videos = kwargs.pop("pixel_values_videos", None) video_embeds = kwargs.pop("video_embeds", None) video_grid_thw = kwargs.pop("video_grid_thw", None) second_per_grid_ts = kwargs.pop("second_per_grid_ts", None) if pixel_values_videos is None and video_embeds is None: return None if pixel_values_videos is not None: return Qwen2_5_VLVideoPixelInputs( type="pixel_values_videos", pixel_values_videos=pixel_values_videos, video_grid_thw=video_grid_thw, second_per_grid_ts=second_per_grid_ts, ) if video_embeds is not None: return Qwen2_5_VLVideoEmbeddingInputs( type="video_embeds", video_embeds=video_embeds, video_grid_thw=video_grid_thw, ) def _process_image_input( self, image_input: Qwen2_5_VLImageInputs ) -> tuple[torch.Tensor, ...]: grid_thw = image_input["image_grid_thw"] assert grid_thw.ndim == 2 if image_input["type"] == "image_embeds": image_embeds = image_input["image_embeds"].type(self.visual.dtype) else: pixel_values = image_input["pixel_values"].type(self.visual.dtype) if self.use_data_parallel: return run_dp_sharded_mrope_vision_model( self.visual, pixel_values, grid_thw.tolist(), rope_type="rope_3d" ) else: image_embeds = self.visual(pixel_values, grid_thw=grid_thw) # Split concatenated embeddings for each image item. merge_size = self.visual.spatial_merge_size sizes = (grid_thw.prod(-1) // merge_size // merge_size).tolist() return image_embeds.split(sizes) def _process_video_input( self, video_input: Qwen2_5_VLVideoInputs ) -> tuple[torch.Tensor, ...]: grid_thw = video_input["video_grid_thw"] assert grid_thw.ndim == 2 if video_input["type"] == "video_embeds": video_embeds = video_input["video_embeds"].type(self.visual.dtype) else: pixel_values_videos = video_input["pixel_values_videos"].type( self.visual.dtype ) if self.use_data_parallel: grid_thw_list = grid_thw.tolist() return run_dp_sharded_mrope_vision_model( self.visual, pixel_values_videos, grid_thw_list, rope_type="rope_3d" ) else: video_embeds = self.visual(pixel_values_videos, grid_thw=grid_thw) # Split concatenated embeddings for each video item. merge_size = self.visual.spatial_merge_size sizes = (grid_thw.prod(-1) // merge_size // merge_size).tolist() return video_embeds.split(sizes) def _postprocess_image_embeds_evs( self, image_embeds_split: tuple[torch.Tensor, ...], image_input: Qwen2_5_VLImageInputs, ) -> tuple[torch.Tensor, ...]: """ Append mrope positions for each for images. This is necessary to recover correct mrope positions after video pruning Args: image_embeds_split: Tuple of image embeddings for each image item. image_input: Image input data. Returns: Tuple of image embeddings for each image item. Resulting embeddings will have extra 4 channels for computed mrope positions. """ merge_size = self.visual.spatial_merge_size grid_thw = image_input["image_grid_thw"] grid_thw_list = grid_thw.tolist() image_embeds_out = [] for emb, size in zip(image_embeds_split, grid_thw_list): positions = compute_mrope_for_media(size, merge_size).to(emb.device) emb = torch.cat([emb, positions], dim=1) image_embeds_out.append(emb) image_embeds_split = image_embeds_out return tuple(image_embeds_split) def _postprocess_video_embeds_evs( self, video_embeds_split: tuple[torch.Tensor, ...], video_input: Qwen2_5_VLVideoInputs, ) -> tuple[torch.Tensor, ...]: """ Prunes video embeddings via Efficient Video Sampling (EVS) and then appends mrope positions for each retained embeddings Args: video_embeds_split: Tuple of video embeddings for each video item. video_input: Video input data. Returns: Tuple of video embeddings for each video item. Resulting embeddings will have extra 4 channels for computed mrope positions. """ grid_thw = video_input["video_grid_thw"] assert grid_thw.ndim == 2 grid_thw_list = grid_thw.tolist() merge_size = self.visual.spatial_merge_size # Cast to long to match the original code # https://github.com/huggingface/transformers/blob/41980ce93e775f6c88500c51c8db7946fc6a2add/src/transformers/models/qwen2_5_vl/modular_qwen2_5_vl.py#L491 # noqa second_per_grid_ts = video_input.get("second_per_grid_ts") if second_per_grid_ts is None: # For Qwen3-VL, second_per_grid_ts might not be available # Use default value of 1.0 for each video second_per_grid_ts = torch.ones(len(grid_thw_list), dtype=torch.long) else: second_per_grid_ts = second_per_grid_ts.long() tokens_per_second = getattr(self.config.vision_config, "tokens_per_second", 1.0) video_embeds_out = [] for emb, size, video_second_per_grid_t in zip( video_embeds_split, grid_thw_list, second_per_grid_ts ): # For each video, we compute retention mask using EVS retention_mask = compute_retention_mask( emb, size, spatial_merge_size=self.visual.spatial_merge_size, q=self.video_pruning_rate, ) # Debug logging for EVS pruning logger.debug( "EVS: Video tokens pruned from %d to %d (T=%d,H=%d,W=%d, " "pruning_rate=%.2f, reduction=%.1f%%)", emb.shape[0], retention_mask.sum().item(), size[0], size[1], size[2], self.video_pruning_rate, (1 - retention_mask.float().mean().item()) * 100, ) positions = compute_mrope_for_media( size, merge_size, tokens_per_second=tokens_per_second, video_second_per_grid=video_second_per_grid_t.item(), ).to(emb.device) emb = emb[retention_mask] positions = positions[retention_mask] emb = torch.cat([emb, positions], dim=1) video_embeds_out.append(emb) return tuple(video_embeds_out) def _parse_and_validate_multimodal_inputs(self, **kwargs: object) -> dict: mm_input_by_modality = {} for input_key in kwargs: if ( input_key in ("pixel_values", "image_embeds") and "image" not in mm_input_by_modality ): mm_input_by_modality["image"] = self._parse_and_validate_image_input( **kwargs ) if ( input_key in ("pixel_values_videos", "video_embeds") and "video" not in mm_input_by_modality ): mm_input_by_modality["video"] = self._parse_and_validate_video_input( **kwargs ) return mm_input_by_modality def iter_mm_grid_hw( self, input_tokens: list[int], mm_features: list[MultiModalFeatureSpec] ) -> Iterator[tuple[int, int, int]]: """ Iterate over multimodal features and yield grid information. For videos with EVS (Efficient Video Sampling) enabled, this function computes the offset based on the pruned token count rather than relying on input_tokens.index(), which would fail when tokens are pruned. Args: input_tokens: List of token IDs in the prompt mm_features: List of multimodal feature specifications Yields: Tuple of (offset, grid_h, grid_w) for each frame/image """ video_token_id = self.config.video_token_id spatial_merge_size = self.config.vision_config.spatial_merge_size for mm_feature in sorted(mm_features, key=lambda f: f.mm_position.offset): offset = mm_feature.mm_position.offset if mm_feature.modality == "image": t, h, w = mm_feature.data["image_grid_thw"].data.tolist() assert t == 1, f"Image must have 1 frame, got {t}" yield offset, h // spatial_merge_size, w // spatial_merge_size elif mm_feature.modality == "video": t, h, w = mm_feature.data["video_grid_thw"].data.tolist() llm_grid_h = h // spatial_merge_size llm_grid_w = w // spatial_merge_size # Check if EVS (Efficient Video Sampling) is enabled is_evs_enabled = ( hasattr(self, "video_pruning_rate") and self.video_pruning_rate is not None and self.video_pruning_rate > 0.0 ) if is_evs_enabled: frame_offsets = self._extract_frame_offsets_from_mask( mm_feature.mm_position, t ) if frame_offsets is not None: for rel_offset in frame_offsets: yield offset + rel_offset, llm_grid_h, llm_grid_w continue # If EVS is enabled but mask is missing, this indicates a bug # in the prompt processing pipeline. The is_embed mask should # always be present when video_pruning_rate > 0. raise RuntimeError( f"EVS is enabled (pruning_rate={self.video_pruning_rate}) " "but is_embed mask is missing from mm_position. " "This indicates a bug in prompt processing." ) else: # Non-EVS mode: Use original logic with input_tokens.index() for _ in range(t): offset = input_tokens.index(video_token_id, offset) yield offset, llm_grid_h, llm_grid_w offset += llm_grid_h * llm_grid_w else: raise ValueError(f"Unsupported modality: {mm_feature.modality}") def _get_evs_mask_segments( self, mm_position: PlaceholderRange, expected_frames: int ) -> list[torch.Tensor] | None: """Extract contiguous segments from EVS is_embed mask. The EVS (Efficient Video Sampling) mask marks which placeholder positions should be filled with video embeddings. This method splits the mask into contiguous segments, where each segment represents one retained frame. This is a pure function - it does not modify any state and always returns the same output for the same input (idempotent). Args: mm_position: MultiModal position containing the is_embed mask expected_frames: Expected number of frame segments Returns: List of tensors, each containing indices for one frame segment, or None if EVS is not enabled or validation fails. """ is_embed_mask = getattr(mm_position, "is_embed", None) if is_embed_mask is None: return None # Find all True positions in the mask mask_tensor = torch.as_tensor(is_embed_mask, dtype=torch.bool).view(-1) true_indices = torch.nonzero(mask_tensor, as_tuple=False).flatten() if true_indices.numel() == 0: return None # Split into contiguous segments (where diff > 1 indicates a gap) if true_indices.numel() == 1: segments = [true_indices] else: diffs = torch.diff(true_indices) split_points = torch.nonzero(diffs != 1, as_tuple=False).flatten() if split_points.numel() == 0: segments = [true_indices] else: segments = torch.tensor_split( true_indices, split_points.add(1).tolist() ) # Validate segment count matches expected frames if len(segments) < expected_frames: logger.debug( "EVS mask segments (%d) do not match expected frames (%d)", len(segments), expected_frames, ) return None return segments[:expected_frames] def _extract_frame_offsets_from_mask( self, mm_position: PlaceholderRange, expected_frames: int ) -> list[int] | None: """Return relative offsets for each EVS-retained frame. The prompt processor stores a boolean mask inside ``mm_position`` that marks which placeholder locations should be populated with video embeddings. By splitting that mask into contiguous runs we can recover the start of every retained frame without probing ``input_tokens``. Args: mm_position: MultiModal position containing the is_embed mask expected_frames: Expected number of frames Returns: List of starting offsets (relative to mm_position) for each frame, or None if EVS is not enabled. """ segments = self._get_evs_mask_segments(mm_position, expected_frames) if segments is None: return None return [int(segment[0].item()) for segment in segments] def _get_actual_frame_token_counts( self, mm_position: PlaceholderRange, expected_frames: int ) -> list[int] | None: """Return actual token count for each EVS-retained frame. This function calculates the actual number of tokens per frame by analyzing the is_embed mask, accounting for EVS pruning. Each frame may have a different token count due to content-aware pruning. Args: mm_position: MultiModal position containing the is_embed mask expected_frames: Expected number of frames Returns: List of token counts for each frame, or None if EVS is not enabled. """ segments = self._get_evs_mask_segments(mm_position, expected_frames) if segments is None: return None return [len(seg) for seg in segments] def recompute_mrope_positions( self, input_ids: list[int], multimodal_embeddings: tuple[torch.Tensor, ...], mrope_positions: torch.LongTensor, num_computed_tokens: int, ) -> tuple[tuple[torch.Tensor, ...], torch.Tensor, int]: """ Update part of input mrope positions (starting with num_computed_tokens index). Original mrope_positions are computed for unpruned sequence and becomes incorrect once pruning occurs, so once we prune media tokens we should reflect this in the mrope_positions before we feed it to LLM. Args: input_ids: (N,) All input tokens of the prompt (Containing entire sequence). multimodal_embeddings: Tuple of multimodal embeddings. mrope_positions: Existing mrope positions (3, N) for entire sequence num_computed_tokens: A number of computed tokens so far. Returns: Tuple of (multimodal_embeddings, mrope_positions, mrope_position_delta). """ image_token_id = self.config.image_token_id video_token_id = self.config.video_token_id vision_start_token_id = self.config.vision_start_token_id # Device device = ( multimodal_embeddings[0].device if len(multimodal_embeddings) else mrope_positions.device ) # Tensors input_ids_t = torch.as_tensor(input_ids, device=device, dtype=torch.long) mm_embeddings_out = [mm[:, :-4] for mm in multimodal_embeddings] mm_embeddings_pos = [ mm[:, -4:].permute(1, 0).long() for mm in multimodal_embeddings ] positions, mrope_positions_delta = recompute_mrope_positions( input_ids_t, mm_embeddings_pos, mrope_positions, num_computed_tokens, vision_start_token_id, image_token_id, video_token_id, ) return tuple(mm_embeddings_out), positions, mrope_positions_delta def get_mrope_input_positions( self, input_tokens: list[int], mm_features: list[MultiModalFeatureSpec], ) -> tuple[torch.Tensor, int]: # Pre-collect actual frame token counts for EVS mode frame_token_counts_map = {} for mm_feature in mm_features: if mm_feature.modality == "video": is_evs_enabled = ( hasattr(self, "video_pruning_rate") and self.video_pruning_rate is not None and self.video_pruning_rate > 0.0 ) if is_evs_enabled: t = mm_feature.data["video_grid_thw"].data.tolist()[0] token_counts = self._get_actual_frame_token_counts( mm_feature.mm_position, t ) assert token_counts is not None, ( "EVS enabled but failed to extract frame token counts " "from is_embed mask" ) frame_token_counts_map[mm_feature.mm_position.offset] = token_counts llm_pos_ids_list = [] st = 0 frame_counts_idx = {} for offset, llm_grid_h, llm_grid_w in self.iter_mm_grid_hw( input_tokens, mm_features ): text_len = offset - st st_idx = llm_pos_ids_list[-1].max() + 1 if len(llm_pos_ids_list) > 0 else 0 # Determine actual token count for this frame base_offset = None for feat_offset in frame_token_counts_map: if offset >= feat_offset: base_offset = feat_offset if base_offset is not None: # EVS mode: use actual token count from is_embed mask assert base_offset in frame_token_counts_map, ( f"Found base_offset {base_offset} but not in frame_token_counts_map" ) if base_offset not in frame_counts_idx: frame_counts_idx[base_offset] = 0 counts = frame_token_counts_map[base_offset] idx = frame_counts_idx[base_offset] assert idx < len(counts), ( f"EVS frame index {idx} out of range (total frames: {len(counts)})" ) actual_frame_tokens = counts[idx] frame_counts_idx[base_offset] += 1 else: # Non-EVS mode (or image): use theoretical grid size actual_frame_tokens = llm_grid_h * llm_grid_w # Add text segment text_positions = ( np.broadcast_to(np.arange(text_len), (3, text_len)) + st_idx ) llm_pos_ids_list.append(text_positions) st_idx += text_len # Add frame segment with actual token count (not theoretical) grid_indices = np.indices((1, llm_grid_h, llm_grid_w)).reshape(3, -1) # Only take the first actual_frame_tokens positions frame_positions = grid_indices[:, :actual_frame_tokens] + st_idx llm_pos_ids_list.append(frame_positions) # Update st using actual token count st = offset + actual_frame_tokens # Handle final text segment if st < len(input_tokens): st_idx = llm_pos_ids_list[-1].max() + 1 if len(llm_pos_ids_list) > 0 else 0 text_len = len(input_tokens) - st final_text_positions = ( np.broadcast_to(np.arange(text_len), (3, text_len)) + st_idx ) llm_pos_ids_list.append(final_text_positions) llm_positions = np.concatenate(llm_pos_ids_list, axis=1).reshape(3, -1) mrope_position_delta = (llm_positions.max() + 1 - len(input_tokens)).item() return torch.from_numpy(llm_positions), mrope_position_delta def embed_multimodal(self, **kwargs: object) -> MultiModalEmbeddings | None: mm_input_by_modality = self._parse_and_validate_multimodal_inputs(**kwargs) if not mm_input_by_modality: return None # The result multimodal_embeddings is tuple of tensors, with each # tensor correspoending to a multimodal data item (image or video). multimodal_embeddings: tuple[torch.Tensor, ...] = () # NOTE: It is important to iterate over the keys in this dictionary # to preserve the order of the modalities. for modality in mm_input_by_modality: multimodal_input = mm_input_by_modality[modality] if modality == "image": image_embeddings = self._process_image_input(multimodal_input) if self.is_multimodal_pruning_enabled: image_embeddings = self._postprocess_image_embeds_evs( image_embeddings, multimodal_input ) multimodal_embeddings += tuple(image_embeddings) if modality == "video": video_embeddings = self._process_video_input(multimodal_input) if self.is_multimodal_pruning_enabled: video_embeddings = self._postprocess_video_embeds_evs( video_embeddings, multimodal_input ) multimodal_embeddings += tuple(video_embeddings) return multimodal_embeddings def _compute_deepstack_embeds( self, inputs_embeds: torch.Tensor, multimodal_embeddings: MultiModalEmbeddings, is_multimodal: torch.Tensor, ) -> tuple[torch.Tensor, MultiModalEmbeddings]: visual_lens = [len(x) for x in multimodal_embeddings] multimodal_embeddings_cat = torch.cat(multimodal_embeddings, dim=0) ( multimodal_embeddings_main, multimodal_embeddings_multiscale, ) = torch.split( multimodal_embeddings_cat, [self.visual_dim, self.multiscale_dim], dim=-1, ) multimodal_embeddings = torch.split( multimodal_embeddings_main, visual_lens, dim=0 ) multimodal_embeddings_multiscale = torch.split( multimodal_embeddings_multiscale, visual_lens, dim=0 ) deepstack_input_embeds = inputs_embeds.new_zeros( inputs_embeds.size(0), self.deepstack_num_level * inputs_embeds.size(1) ) deepstack_input_embeds = _merge_multimodal_embeddings( inputs_embeds=deepstack_input_embeds, multimodal_embeddings=multimodal_embeddings_multiscale, is_multimodal=is_multimodal, ) deepstack_input_embeds = deepstack_input_embeds.view( inputs_embeds.shape[0], self.deepstack_num_level, self.visual_dim ) deepstack_input_embeds = deepstack_input_embeds.permute(1, 0, 2) return deepstack_input_embeds, multimodal_embeddings def embed_input_ids( self, input_ids: torch.Tensor, multimodal_embeddings: MultiModalEmbeddings | None = None, *, is_multimodal: torch.Tensor | None = None, handle_oov_mm_token: bool = False, ) -> torch.Tensor: inputs_embeds = self._embed_text_input_ids( input_ids, self.language_model.embed_input_ids, is_multimodal=is_multimodal, handle_oov_mm_token=handle_oov_mm_token, ) if multimodal_embeddings is None or len(multimodal_embeddings) == 0: return inputs_embeds is_multimodal = _require_is_multimodal(is_multimodal) if self.use_deepstack: ( deepstack_input_embeds, multimodal_embeddings, ) = self._compute_deepstack_embeds( inputs_embeds=inputs_embeds, multimodal_embeddings=multimodal_embeddings, is_multimodal=is_multimodal, ) else: deepstack_input_embeds = None inputs_embeds = _merge_multimodal_embeddings( inputs_embeds=inputs_embeds, multimodal_embeddings=multimodal_embeddings, is_multimodal=is_multimodal, ) if deepstack_input_embeds is not None: self._set_deepstack_input_embeds(deepstack_input_embeds) return inputs_embeds def forward( self, input_ids: torch.Tensor | None, positions: torch.Tensor, intermediate_tensors: IntermediateTensors | None = None, inputs_embeds: torch.Tensor | None = None, **kwargs: object, ) -> torch.Tensor | IntermediateTensors: """Run forward pass for Qwen3VL. Args: input_ids: Flattened (concatenated) input_ids corresponding to a batch. positions: Flattened (concatenated) position ids corresponding to a batch. **NOTE**: If mrope is enabled (default setting for Qwen3VL opensource models), the shape will be `(3, seq_len)`, otherwise it will be `(seq_len,). intermediate_tensors: Intermediate tensors from previous pipeline stages. inputs_embeds: Pre-computed input embeddings. **kwargs: Additional keyword arguments including: - pixel_values: Pixel values to be fed to a model. `None` if no images are passed. - image_grid_thw: Tensor `(n_images, 3)` of image 3D grid in LLM. `None` if no images are passed. - pixel_values_videos: Pixel values of videos to be fed to a model. `None` if no videos are passed. - video_grid_thw: Tensor `(n_videos, 3)` of video 3D grid in LLM. `None` if no videos are passed. """ if intermediate_tensors is not None: inputs_embeds = None if inputs_embeds is not None and get_pp_group().is_first_rank: deepstack_input_embeds = self._get_deepstack_input_embeds( inputs_embeds.size(0) ) else: deepstack_input_embeds = None hidden_states = self.language_model.model( input_ids=input_ids, positions=positions, intermediate_tensors=intermediate_tensors, inputs_embeds=inputs_embeds, # args for deepstack deepstack_input_embeds=deepstack_input_embeds, ) if inputs_embeds is not None and get_pp_group().is_first_rank: self._clear_deepstack_input_embeds(inputs_embeds.size(0)) return hidden_states def compute_logits( self, hidden_states: torch.Tensor, ) -> torch.Tensor | None: return self.language_model.compute_logits(hidden_states) def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: loader = AutoWeightsLoader(self) return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper) def get_mm_mapping(self) -> MultiModelKeys: """ Get the module prefix in multimodal models """ return MultiModelKeys.from_string_field( language_model="language_model", connector=["visual.merger", "visual.deepstack_merger_list"], tower_model="visual.", ) def get_num_mm_encoder_tokens( self, num_image_tokens: int, ) -> int: hf_config = self.config vision_config = hf_config.vision_config merge_size = vision_config.spatial_merge_size return num_image_tokens * merge_size**2 def get_num_mm_connector_tokens( self, num_vision_tokens: int, ) -> int: hf_config = self.config vision_config = hf_config.vision_config merge_size = vision_config.spatial_merge_size return num_vision_tokens // merge_size**2
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/model_executor/models/qwen3_vl.py", "license": "Apache License 2.0", "lines": 1867, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/model_executor/models/qwen3_vl_moe.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project # Copyright 2025 The vLLM team. # Copyright 2025 The Qwen Team. # Copyright 2025 The HuggingFace Inc. team. # All rights reserved. # # This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX # and OPT implementations in this library. It has been modified from its # original forms to accommodate minor architectural differences compared # to GPT-NeoX and OPT used by the Meta AI team that trained the model. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """Inference-only Qwen3-VL-MoE model compatible with HuggingFace weights.""" import typing from collections.abc import Callable, Iterable from itertools import islice import torch from transformers.models.qwen3_vl_moe.configuration_qwen3_vl_moe import ( Qwen3VLMoeConfig, ) from vllm.compilation.decorators import support_torch_compile from vllm.config import VllmConfig from vllm.distributed import get_pp_group from vllm.logger import init_logger from vllm.model_executor.layers.logits_processor import LogitsProcessor from vllm.model_executor.layers.vocab_parallel_embedding import ParallelLMHead from vllm.model_executor.model_loader.weight_utils import ( default_weight_loader, maybe_remap_kv_scale_name, ) from vllm.multimodal import MULTIMODAL_REGISTRY from vllm.sequence import IntermediateTensors from .interfaces import MixtureOfExperts from .qwen3_moe import ( Qwen3MoeForCausalLM, Qwen3MoeModel, Qwen3MoeSparseMoeBlock, ) from .qwen3_vl import ( Qwen3_VisionTransformer, Qwen3VLDummyInputsBuilder, Qwen3VLForConditionalGeneration, Qwen3VLMultiModalProcessor, Qwen3VLProcessingInfo, ) from .utils import is_pp_missing_parameter, maybe_prefix logger = init_logger(__name__) class Qwen3VLMoeProcessingInfo(Qwen3VLProcessingInfo): def get_hf_config(self): return self.ctx.get_hf_config(Qwen3VLMoeConfig) @support_torch_compile( dynamic_arg_dims={ "input_ids": 0, # positions is of shape (3, seq_len) if mrope is enabled for qwen2-vl, # otherwise (seq_len, ). "positions": -1, "intermediate_tensors": 0, "inputs_embeds": 0, # the same shape as input_embeds "deepstack_input_embeds": 0, } ) class Qwen3MoeLLMModel(Qwen3MoeModel): def forward( self, input_ids: torch.Tensor | None, positions: torch.Tensor, intermediate_tensors: IntermediateTensors | None = None, inputs_embeds: torch.Tensor | None = None, deepstack_input_embeds: IntermediateTensors | None = None, ) -> torch.Tensor | IntermediateTensors: if get_pp_group().is_first_rank: if inputs_embeds is not None: hidden_states = inputs_embeds else: hidden_states = self.embed_input_ids(input_ids) residual = None else: assert intermediate_tensors is not None hidden_states = intermediate_tensors["hidden_states"] residual = intermediate_tensors["residual"] aux_hidden_states = [] for layer_idx, layer in islice( enumerate(self.layers), self.start_layer, self.end_layer ): if layer_idx in self.aux_hidden_state_layers: aux_hidden_states.append(hidden_states + residual) hidden_states, residual = layer( positions, hidden_states, residual, ) if deepstack_input_embeds is not None and layer_idx in range( 0, len(deepstack_input_embeds) ): hidden_states = ( hidden_states + deepstack_input_embeds[f"deepstack_input_embeds_{layer_idx}"] ) if not get_pp_group().is_last_rank: return IntermediateTensors( {"hidden_states": hidden_states, "residual": residual} ) hidden_states, _ = self.norm(hidden_states, residual) if len(aux_hidden_states) > 0: return hidden_states, aux_hidden_states return hidden_states def load_fused_expert_weights( self, name: str, params_dict: dict, loaded_weight: torch.Tensor, shard_id: str, num_experts: int, ) -> bool: param = params_dict[name] weight_loader = typing.cast(Callable[..., bool], param.weight_loader) loaded_local_expert = False for expert_id in range(num_experts): curr_expert_weight = loaded_weight[expert_id] success = weight_loader( param, curr_expert_weight, name, shard_id, expert_id, return_success=True, ) if success: loaded_local_expert = True return loaded_local_expert def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: stacked_params_mapping = [ # (param_name, shard_name, shard_id) ("qkv_proj", "q_proj", "q"), ("qkv_proj", "k_proj", "k"), ("qkv_proj", "v_proj", "v"), ("gate_up_proj", "gate_proj", 0), ("gate_up_proj", "up_proj", 1), ] # Skip loading extra parameters for GPTQ/modelopt models. ignore_suffixes = ( ".bias", "_bias", ".k_scale", "_k_scale", ".v_scale", "_v_scale", ".weight_scale", "_weight_scale", ".input_scale", "_input_scale", ) params_dict = dict(self.named_parameters()) loaded_params: set[str] = set() expert_params_mapping = self.get_expert_mapping() is_fused_expert = False fused_expert_params_mapping = [ ("experts.w13_weight", "experts.gate_up_proj", 0, "w1"), ("experts.w2_weight", "experts.down_proj", 0, "w2"), ] num_experts = self.config.num_experts for name, loaded_weight in weights: for param_name, weight_name, shard_id in stacked_params_mapping: if "experts.gate_up_proj" in name or "experts.down_proj" in name: is_fused_expert = True expert_params_mapping = fused_expert_params_mapping # Skip non-stacked layers and experts (experts handled below). if weight_name not in name: continue # We have mlp.experts[0].gate_proj in the checkpoint. # Since we handle the experts below in expert_params_mapping, # we need to skip here BEFORE we update the name, otherwise # name will be updated to mlp.experts[0].gate_up_proj, which # will then be updated below in expert_params_mapping # for mlp.experts[0].gate_gate_up_proj, which breaks load. if "mlp.experts" in name: continue name = name.replace(weight_name, param_name) # Skip loading extra parameters for GPTQ/modelopt models. if name.endswith(ignore_suffixes) and name not in params_dict: continue # Skip layers on other devices. if is_pp_missing_parameter(name, self): continue if name.endswith("scale"): # Remapping the name of FP8 kv-scale. name = maybe_remap_kv_scale_name(name, params_dict) if name is None: continue if name not in params_dict: continue param = params_dict[name] weight_loader = getattr(param, "weight_loader", default_weight_loader) if weight_loader == default_weight_loader: weight_loader(param, loaded_weight) else: weight_loader(param, loaded_weight, shard_id) break else: is_expert_weight = False for mapping in expert_params_mapping: param_name, weight_name, expert_id, shard_id = mapping if weight_name not in name: continue # Anyway, this is an expert weight and should not be # attempted to load as other weights later is_expert_weight = True name_mapped = name.replace(weight_name, param_name) if is_pp_missing_parameter(name_mapped, self): continue if is_fused_expert: loaded_weight = loaded_weight.transpose(-1, -2) # no bias if "experts.gate_up_proj" in name: loaded_weight = loaded_weight.chunk(2, dim=-2) success_w1 = self.load_fused_expert_weights( name_mapped, params_dict, loaded_weight[0], "w1", num_experts, ) success_w3 = self.load_fused_expert_weights( name_mapped, params_dict, loaded_weight[1], "w3", num_experts, ) success = success_w1 and success_w3 else: # down_proj success = self.load_fused_expert_weights( name_mapped, params_dict, loaded_weight, shard_id, num_experts, ) else: # Skip loading extra parameters for GPTQ/modelopt models if ( name_mapped.endswith(ignore_suffixes) and name_mapped not in params_dict ): continue param = params_dict[name_mapped] # We should ask the weight loader to return success or # not here since otherwise we may skip experts with # other available replicas. weight_loader = typing.cast( Callable[..., bool], param.weight_loader ) success = weight_loader( param, loaded_weight, name_mapped, shard_id=shard_id, expert_id=expert_id, return_success=True, ) if success: name = name_mapped break else: if is_expert_weight: # We've checked that this is an expert weight # However it's not mapped locally to this rank # So we simply skip it continue # Skip loading extra parameters for GPTQ/modelopt models. if name.endswith(ignore_suffixes) and name not in params_dict: continue # Skip layers on other devices. if is_pp_missing_parameter(name, self): continue # Remapping the name of FP8 kv-scale. if name.endswith("kv_scale"): remapped_kv_scale_name = name.replace( ".kv_scale", ".attn.kv_scale" ) if remapped_kv_scale_name not in params_dict: logger.warning_once( "Found kv scale in the checkpoint (e.g. %s), but not found the expected name in the model (e.g. %s). kv-scale is not loaded.", # noqa: E501 name, remapped_kv_scale_name, ) continue else: name = remapped_kv_scale_name param = params_dict[name] weight_loader = getattr( param, "weight_loader", default_weight_loader ) weight_loader(param, loaded_weight) loaded_params.add(name) return loaded_params class Qwen3MoeLLMForCausalLM(Qwen3MoeForCausalLM): def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): super(Qwen3MoeForCausalLM, self).__init__() self.config = vllm_config.model_config.hf_config self.quant_config = vllm_config.quant_config self.model = Qwen3MoeLLMModel( vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model") ) self.lm_head = ParallelLMHead( self.config.vocab_size, self.config.hidden_size, quant_config=self.quant_config, prefix=maybe_prefix(prefix, "lm_head"), ) if self.config.tie_word_embeddings: self.lm_head.weight = self.model.embed_tokens.weight self.logits_processor = LogitsProcessor(self.config.vocab_size) self.make_empty_intermediate_tensors = ( self.model.make_empty_intermediate_tensors ) class Qwen3VLMoeMixtureOfExperts(MixtureOfExperts): def update_physical_experts_metadata( self, num_physical_experts: int, num_local_physical_experts: int, ) -> None: assert self.num_local_physical_experts == num_local_physical_experts self.num_physical_experts = num_physical_experts self.num_local_physical_experts = num_local_physical_experts self.num_redundant_experts = num_physical_experts - self.num_logical_experts for layer in self.language_model.model.layers: if isinstance(layer.mlp, Qwen3MoeSparseMoeBlock): moe = layer.mlp moe.n_local_physical_experts = num_local_physical_experts moe.n_physical_experts = num_physical_experts moe.n_redundant_experts = self.num_redundant_experts moe.experts.update_expert_map() def set_moe_parameters(self): self.expert_weights = [] self.moe_layers = [] example_moe = None for layer in self.language_model.model.layers: if hasattr(layer, "mlp") and isinstance(layer.mlp, Qwen3MoeSparseMoeBlock): example_moe = layer.mlp self.moe_layers.append(layer.mlp.experts) if example_moe is None: raise RuntimeError("No Qwen3Moe layer found in the language_model.") # Set MoE hyperparameters self.num_moe_layers = len(self.moe_layers) self.num_expert_groups = 1 self.num_shared_experts = 0 self.num_logical_experts = example_moe.n_logical_experts self.num_physical_experts = example_moe.n_physical_experts self.num_local_physical_experts = example_moe.n_local_physical_experts self.num_routed_experts = example_moe.n_routed_experts self.num_redundant_experts = example_moe.n_redundant_experts @MULTIMODAL_REGISTRY.register_processor( Qwen3VLMultiModalProcessor, info=Qwen3VLMoeProcessingInfo, dummy_inputs=Qwen3VLDummyInputsBuilder, ) class Qwen3VLMoeForConditionalGeneration( Qwen3VLForConditionalGeneration, Qwen3VLMoeMixtureOfExperts ): is_3d_moe_weight: bool = True packed_modules_mapping = { "qkv_proj": [ "q_proj", "k_proj", "v_proj", ], } def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): super(Qwen3VLForConditionalGeneration, self).__init__() config: Qwen3VLMoeConfig = vllm_config.model_config.hf_config quant_config = vllm_config.quant_config multimodal_config = vllm_config.model_config.multimodal_config self.config = config self.multimodal_config = multimodal_config self.use_data_parallel = multimodal_config.mm_encoder_tp_mode == "data" self.video_pruning_rate = multimodal_config.video_pruning_rate self.is_multimodal_pruning_enabled = ( multimodal_config.is_multimodal_pruning_enabled() ) self.use_deepstack = hasattr(config.vision_config, "deepstack_visual_indexes") self.deepstack_num_level = ( len(config.vision_config.deepstack_visual_indexes) if self.use_deepstack else 0 ) self.visual_dim = config.vision_config.out_hidden_size self.multiscale_dim = self.visual_dim * self.deepstack_num_level with self._mark_tower_model(vllm_config, {"image", "video"}): self.visual = Qwen3_VisionTransformer( config.vision_config, norm_eps=getattr(config, "rms_norm_eps", 1e-6), quant_config=quant_config, prefix=maybe_prefix(prefix, "visual"), ) # register buffer for deepstack if self.use_deepstack: self.deepstack_input_embeds = [ torch.zeros( vllm_config.scheduler_config.max_num_batched_tokens, config.text_config.hidden_size, ) for _ in range(self.deepstack_num_level) ] with self._mark_language_model(vllm_config): self.language_model = Qwen3MoeLLMForCausalLM( vllm_config=vllm_config.with_hf_config(config.text_config), prefix=maybe_prefix(prefix, "language_model"), ) if not get_pp_group().is_first_rank and hasattr( config.vision_config, "deepstack_visual_indexes" ): assert self.language_model.start_layer >= len( config.vision_config.deepstack_visual_indexes ), ( "start_layer should be greater than or equal to " "len(deepstack_visual_indexes)" ) # Whether to include the gate_up_proj mapping is determined by # the language model. self.packed_modules_mapping = ( self.packed_modules_mapping | self.language_model.packed_modules_mapping ) self.make_empty_intermediate_tensors = ( self.language_model.make_empty_intermediate_tensors ) # Set MoE hyperparameters self.set_moe_parameters()
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/model_executor/models/qwen3_vl_moe.py", "license": "Apache License 2.0", "lines": 440, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/kernels/quantization/test_fp8_quant_group.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """Tests for QuantFP8 Group Quantization implementation.""" import pytest import torch from vllm.model_executor.layers.quantization.input_quant_fp8 import QuantFP8 from vllm.model_executor.layers.quantization.utils.quant_utils import GroupShape from vllm.utils.torch_utils import set_random_seed @pytest.mark.parametrize( "batch_size,hidden_dim,group_size", [ (16, 256, 32), # Small (64, 1024, 64), # Medium (128, 2048, 128), # Large (8, 513, 64), # Non-divisible (native only) ], ) @pytest.mark.parametrize("seed", [42]) @pytest.mark.parametrize("use_ue8m0", [True, False]) @torch.inference_mode() def test_quantfp8_group_functionality( default_vllm_config, batch_size: int, hidden_dim: int, group_size: int, seed: int, use_ue8m0: bool, ) -> None: """Test QuantFP8 group quantization with various configurations. Tests both CUDA and native implementations, column-major scales, and verifies consistency between implementations. """ set_random_seed(seed) x = torch.randn((batch_size, hidden_dim), dtype=torch.bfloat16, device="cuda") * 8 expected_num_groups = (hidden_dim + group_size - 1) // group_size is_divisible = hidden_dim % group_size == 0 group_shape = GroupShape(1, group_size) quant_op = QuantFP8( static=False, group_shape=group_shape, column_major_scales=False, use_ue8m0=use_ue8m0, ) # 1. Test native implementation (always available) x_quant_native, scales_native = quant_op.forward_native(x.clone()) assert x_quant_native.shape == x.shape assert scales_native.shape == (batch_size, expected_num_groups) # 2. Test column-major scales configuration quant_op_col = QuantFP8( static=False, group_shape=group_shape, column_major_scales=True, use_ue8m0=use_ue8m0, ) _, scales_col = quant_op_col.forward_native(x.clone()) assert scales_col.shape == (batch_size, expected_num_groups) assert scales_col.stride(0) == 1 assert scales_col.stride(1) == batch_size # Test column-major scales consistency torch.testing.assert_close(scales_col, scales_native, rtol=1e-9, atol=1e-8) # 3. Test CUDA implementation (only for divisible dimensions) if is_divisible: x_quant_cuda, scales_cuda = quant_op.forward_cuda(x.clone()) assert x_quant_cuda.shape == x.shape assert scales_cuda.shape == (batch_size, expected_num_groups) # Verify CUDA/native consistency torch.testing.assert_close(scales_cuda, scales_native, rtol=2e-7, atol=2e-8) # Quantized values should mostly match diff_count = (x_quant_cuda != x_quant_native).sum().item() diff_ratio = diff_count / x_quant_cuda.numel() assert diff_ratio < 0.002, f"Too many differences: {diff_ratio:.4%}" @pytest.mark.parametrize("seed", [42]) @pytest.mark.parametrize("use_ue8m0", [True, False]) @torch.inference_mode() def test_quantfp8_group_multidimensional( default_vllm_config, seed: int, use_ue8m0: bool ) -> None: set_random_seed(seed) group_size = 64 # Test with 3D input batch1, batch2, hidden_dim = 4, 8, 1024 x_3d = ( torch.randn((batch1, batch2, hidden_dim), dtype=torch.bfloat16, device="cuda") * 8 ) group_shape = GroupShape(1, group_size) quant_op = QuantFP8( static=False, group_shape=group_shape, column_major_scales=False, use_ue8m0=use_ue8m0, ) x_quant, scales = quant_op.forward_native(x_3d.clone()) assert x_quant.shape == x_3d.shape assert scales.shape == (batch1, batch2, hidden_dim // group_size) # Test column_major_scales with multi-dim quant_op_col = QuantFP8( static=False, group_shape=group_shape, column_major_scales=True, use_ue8m0=use_ue8m0, ) _, scales_col = quant_op_col.forward_native(x_3d.clone()) assert scales_col.shape == (batch1, batch2, hidden_dim // group_size) # Test with 4D input batch1, batch2, batch3, hidden_dim = 2, 3, 4, 256 x_4d = ( torch.randn( (batch1, batch2, batch3, hidden_dim), dtype=torch.bfloat16, device="cuda" ) * 8 ) x_quant_4d, scales_4d = quant_op.forward_native(x_4d.clone()) assert x_quant_4d.shape == x_4d.shape assert scales_4d.shape == (batch1, batch2, batch3, hidden_dim // group_size) _, scales_4d_col = quant_op_col.forward_native(x_4d.clone()) assert scales_4d_col.shape == (batch1, batch2, hidden_dim // group_size, batch3) @pytest.mark.parametrize("seed", [42]) @torch.inference_mode() def test_quantfp8_group_edge_cases(default_vllm_config, seed: int) -> None: set_random_seed(seed) batch_size = 16 group_size = 64 # Test with single group (group_size >= hidden_dim) x_small = torch.randn((batch_size, 32), dtype=torch.bfloat16, device="cuda") * 8 group_shape = GroupShape(1, group_size) quant_op = QuantFP8( static=False, group_shape=group_shape, column_major_scales=False ) x_quant_small, scales_small = quant_op.forward_native(x_small.clone()) assert x_quant_small.shape == x_small.shape assert scales_small.shape == (batch_size, 1) # Test with zero inputs x_zero = torch.zeros((batch_size, 256), dtype=torch.bfloat16, device="cuda") x_quant_zero, scales_zero = quant_op.forward_native(x_zero.clone()) assert x_quant_zero.shape == x_zero.shape assert (scales_zero > 0).all(), "Scales should be clamped to minimum" # Test very large values x_large = torch.full((batch_size, 256), 1000.0, dtype=torch.bfloat16, device="cuda") x_quant_large, scales_large = quant_op.forward_native(x_large.clone()) assert x_quant_large.shape == x_large.shape # FP8 max is typically 448 or 224, so scales should be > 1 assert (scales_large > 1.0).all(), "Large values should have scales > 1"
{ "repo_id": "vllm-project/vllm", "file_path": "tests/kernels/quantization/test_fp8_quant_group.py", "license": "Apache License 2.0", "lines": 143, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/evals/gpt_oss/test_gpqa_correctness.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """ GPQA evaluation using vLLM server and GPT-OSS evaluation package. Usage: pytest -s -v tests/evals/gpt_oss/test_gpqa_correctness.py \ --config-list-file=configs/models-h200.txt """ import os import shlex import subprocess import sys import urllib.request from pathlib import Path import regex as re import yaml from tests.utils import RemoteOpenAIServer TOL = 0.05 # Absolute tolerance for accuracy comparison # Path to tiktoken encoding files TIKTOKEN_DATA_DIR = Path(__file__).parent / "data" # Tiktoken encoding files to download TIKTOKEN_FILES = { "cl100k_base.tiktoken": "https://openaipublic.blob.core.windows.net/encodings/cl100k_base.tiktoken", "o200k_base.tiktoken": "https://openaipublic.blob.core.windows.net/encodings/o200k_base.tiktoken", } def ensure_tiktoken_files(): """Download tiktoken encoding files if they don't exist.""" TIKTOKEN_DATA_DIR.mkdir(parents=True, exist_ok=True) for filename, url in TIKTOKEN_FILES.items(): filepath = TIKTOKEN_DATA_DIR / filename if not filepath.exists(): print(f"Downloading {filename} from {url}...") urllib.request.urlretrieve(url, filepath) print(f" Downloaded to {filepath}") else: print(f" {filename} already exists.") def run_gpqa_eval(model_name: str, base_url: str, reasoning_effort: str) -> float: """Run GPQA evaluation using the gpt-oss evaluation package.""" # Build the command to run the evaluation cmd = [ sys.executable, "-m", "gpt_oss.evals", "--eval", "gpqa", "--model", model_name, "--reasoning-effort", reasoning_effort, "--base-url", base_url, "--n-threads", "200", ] try: # Set up environment for the evaluation subprocess # Inherit current environment and add required variables eval_env = os.environ.copy() eval_env["OPENAI_API_KEY"] = "dummy" # Run the evaluation result = subprocess.run( cmd, text=True, capture_output=True, timeout=1800, # 30 minute timeout env=eval_env, ) print("Evaluation process stdout:\n", result.stdout) print("Evaluation process stderr:\n", result.stderr) print(f"Evaluation process return code: {result.returncode}") if result.returncode != 0: raise RuntimeError( f"Evaluation failed with exit code {result.returncode}:\n" f"stdout: {result.stdout}\nstderr: {result.stderr}" ) # Parse the output to extract the score match = re.search(r"'metric':\s*([\d.]+)", result.stdout) if match: return float(match.group(1)) # If we still can't find it, raise an error raise ValueError( f"Could not parse score from evaluation output:\n{result.stdout}" ) except subprocess.TimeoutExpired as e: raise RuntimeError("Evaluation timed out") from e def test_gpqa_correctness(config_filename): """Test GPQA correctness for a given model configuration.""" # Ensure tiktoken files are downloaded ensure_tiktoken_files() # Verify tiktoken files exist for filename in TIKTOKEN_FILES: filepath = TIKTOKEN_DATA_DIR / filename assert filepath.exists(), f"Tiktoken file not found: {filepath}" eval_config = yaml.safe_load(config_filename.read_text(encoding="utf-8")) # Parse server arguments from config (use shlex to handle quoted strings) server_args_str = eval_config.get("server_args", "") server_args = shlex.split(server_args_str) if server_args_str else [] # Add standard server arguments server_args.extend( [ "--trust-remote-code", "--enforce-eager", "--disable-uvicorn-access-log", ] ) # Build server environment with tiktoken path and any config-specified vars server_env = {"TIKTOKEN_ENCODINGS_BASE": str(TIKTOKEN_DATA_DIR)} if eval_config.get("env"): server_env.update(eval_config["env"]) reasoning_effort = eval_config.get("reasoning_effort", "low") print(f"Starting GPQA evaluation for model: {eval_config['model_name']}") print(f"Expected metric threshold: {eval_config['metric_threshold']}") print(f"Reasoning effort: {reasoning_effort}") print(f"Server args: {' '.join(server_args)}") print(f"Server environment variables: {server_env}") # Launch server and run evaluation with RemoteOpenAIServer( eval_config["model_name"], server_args, env_dict=server_env, max_wait_seconds=eval_config.get("startup_max_wait_seconds", 1800), ) as remote_server: base_url = remote_server.url_for("v1") print(f"Server started at: {base_url}") measured_metric = run_gpqa_eval( eval_config["model_name"], base_url, reasoning_effort ) expected_metric = eval_config["metric_threshold"] print(f"GPQA Results for {eval_config['model_name']}:") print(f" Measured metric: {measured_metric:.4f}") print(f" Expected metric: {expected_metric:.4f}") print(f" Tolerance: {TOL:.4f}") # Verify metric is within tolerance assert measured_metric >= expected_metric - TOL, ( f"GPQA metric too low: {measured_metric:.4f} < " f"{expected_metric:.4f} - {TOL:.4f} = {expected_metric - TOL:.4f}" ) print(f"GPQA test passed for {eval_config['model_name']}")
{ "repo_id": "vllm-project/vllm", "file_path": "tests/evals/gpt_oss/test_gpqa_correctness.py", "license": "Apache License 2.0", "lines": 138, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/v1/worker/gpu_ubatch_wrapper.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import threading from collections.abc import Callable from dataclasses import dataclass from typing import Any import torch import vllm.envs as envs from vllm.compilation.cuda_graph import CUDAGraphWrapper from vllm.config import CUDAGraphMode, VllmConfig from vllm.distributed import get_ep_group from vllm.distributed.device_communicators.pynccl_allocator import set_graph_pool_id from vllm.forward_context import ( DPMetadata, create_forward_context, get_forward_context, override_forward_context, ) from vllm.logger import init_logger from vllm.model_executor.offloader.base import get_offloader from vllm.platforms import current_platform from vllm.sequence import IntermediateTensors from vllm.utils.import_utils import has_deep_gemm from vllm.utils.platform_utils import num_compute_units from vllm.v1.worker.ubatching import UBatchContext, make_ubatch_contexts logger = init_logger(__name__) @dataclass class UbatchMetadata: context: UBatchContext input_ids: torch.Tensor positions: torch.Tensor inputs_embeds: torch.Tensor | None intermediate_tensors: IntermediateTensors | None num_tokens: int @dataclass class CUDAGraphMetaData: cudagraph: torch.cuda.CUDAGraph ubatch_metadata: UbatchMetadata outputs: Any | None = None class SMControlContextManager: def __init__( self, comm_sms: int, set_comm_sms: Callable[[int], None], set_compute_sms: Callable[[int], None], ): """ Context manager for controlling SM (Streaming Multiprocessor) allocation. Upon entering the context, it sets the number of SMs allocated for communication and computation to comm_sms and total_sms - comm_sms respectively. Upon exiting, it restores the allocation to use all available SMs (i.e. total_sms). Args: comm_sms (int): The number of SMs to allocate for communication. (The remainder will be used for computation.) set_comm_sms (Callable[[int], None]): A function that sets the number of SMs for communication. set_compute_sms (Callable[[int], None]): A function that sets the number of SMs for computation. """ assert current_platform.is_cuda(), ( "SM control is currently only supported on CUDA" ) total_sms = num_compute_units(torch.cuda.current_device()) assert comm_sms < total_sms self.total_sms = total_sms self.compute_sms = total_sms - comm_sms self.comm_sms = comm_sms self.set_comm_sms = set_comm_sms self.set_compute_sms = set_compute_sms def __enter__(self): self.set_comm_sms(self.comm_sms) self.set_compute_sms(self.compute_sms) def __exit__(self, exc_type, exc_value, traceback): self.set_comm_sms(self.total_sms) self.set_compute_sms(self.total_sms) class UBatchWrapper: def __init__( self, runnable: Callable, vllm_config: VllmConfig, runtime_mode: CUDAGraphMode, device: torch.cuda.device, ): self.runnable = runnable self.vllm_config = vllm_config self.compilation_config = vllm_config.compilation_config self.comm_stream = torch.cuda.Stream(device=device) # Ubatch threads plus the main thread self.ready_barrier = threading.Barrier( self.vllm_config.parallel_config.num_ubatches + 1 ) self.cudagraphs: dict[int, CUDAGraphMetaData] = {} self.cudagraph_wrapper = None self.graph_pool = None if runtime_mode is not CUDAGraphMode.NONE: self.cudagraph_wrapper = CUDAGraphWrapper( runnable, vllm_config, runtime_mode=runtime_mode ) self.graph_pool = current_platform.get_global_graph_pool() self.sm_control = self._create_sm_control_context(vllm_config) self.device = device @staticmethod def _create_sm_control_context(vllm_config: VllmConfig): comm_sms: int = envs.VLLM_DBO_COMM_SMS set_comm_sms = lambda sms: None if vllm_config.parallel_config.enable_expert_parallel: # Currently only DeepEP highthroughput supports SM control so this # only affects that case. ep_group = get_ep_group() device_communicator = ep_group.device_communicator all2all_manager = None if device_communicator is not None: all2all_manager = device_communicator.all2all_manager if all2all_manager is not None: max_sms_used = all2all_manager.max_sms_used() if max_sms_used is not None: comm_sms = min(comm_sms, max_sms_used) if comm_sms > 0 and all2all_manager is not None: set_comm_sms = lambda sms: all2all_manager.set_num_sms(sms) # TODO(lucas): support other kernels besides DeepGEMM set_compute_sms = lambda sms: None if has_deep_gemm() and comm_sms > 0: import deep_gemm as dg set_compute_sms = lambda sms: dg.set_num_sms(sms) return SMControlContextManager( comm_sms=comm_sms, set_comm_sms=set_comm_sms, set_compute_sms=set_compute_sms, ) def __getattr__(self, key: str): # allow accessing the attributes of the runnable. if hasattr(self.runnable, key): return getattr(self.runnable, key) raise AttributeError( f"Attribute {key} not exists in the runnable of " f"cudagraph wrapper: {self.runnable}" ) def unwrap(self) -> Callable: # in case we need to access the original runnable. return self.runnable def _capture_ubatches(self, ubatch_metadata, model) -> torch.Tensor: """ Capture a cudagraph for a microbatched run. The logic here is somewhat complicated because we need to make sure that each of the ubatch threads initialize the cuda context before we start the graph capture. The flow is as follows: 1. The main thread starts up each ubatch thread. Each thread will initialize its cuda context (torch.cuda.current_blas_handle()) before going to sleep upon entering the ubatch_context. 2. The main thread starts the graph capture and wakes up the first ubatch thread. 3. Each ubatch thread runs the model to completion and returns the completed output tensors back to the main thread. 4. The main thread stores the captured cudagraph along with its metadata and returns """ @torch.inference_mode() def _capture_ubatch_thread(results, ubatch_metadata): torch.cuda.set_device(self.device) ubatch_context = ubatch_metadata.context with torch.cuda.stream(ubatch_context.compute_stream): _ = torch.cuda.current_blas_handle() with torch.cuda.stream(ubatch_context.comm_stream): _ = torch.cuda.current_blas_handle() with ubatch_context: model_output = model( input_ids=ubatch_metadata.input_ids, positions=ubatch_metadata.positions, intermediate_tensors=ubatch_metadata.intermediate_tensors, inputs_embeds=ubatch_metadata.inputs_embeds, ) results.append((ubatch_metadata.context.id, model_output)) results: list[tuple[int, torch.Tensor]] = [] compute_stream = ubatch_metadata[0].context.compute_stream num_tokens = ubatch_metadata[0].num_tokens + ubatch_metadata[1].num_tokens # Ubatches will manually manage the forward context, so we override # it to None here so we can have it restored correctly later with override_forward_context(None): ubatch_threads = [] for metadata in ubatch_metadata: thread = threading.Thread( target=_capture_ubatch_thread, args=( results, metadata, ), ) ubatch_threads.append(thread) thread.start() self.ready_barrier.wait() # Wait for both threads to be ready # Capture the cudagraph cudagraph_metadata = CUDAGraphMetaData( cudagraph=torch.cuda.CUDAGraph(), ubatch_metadata=ubatch_metadata, ) if self.graph_pool is not None: set_graph_pool_id(self.graph_pool) else: set_graph_pool_id(current_platform.graph_pool_handle()) # Sync offloader's copy stream before capture. # Ensure any pre-capture prefetches from offloader are complete. get_offloader().sync_prev_onload() with torch.cuda.graph( cudagraph_metadata.cudagraph, stream=compute_stream, pool=self.graph_pool, ): ubatch_metadata[0].context.cpu_wait_event.set() for thread in ubatch_threads: thread.join() sorted_results = [value for position, value in sorted(results)] result = torch.cat(sorted_results, dim=0) cudagraph_metadata.outputs = result # Join offloader's copy stream after forward to avoid unjoined # stream error. The last layer's start_prefetch forks copy_stream, # but wait_prefetch only happens in the next forward pass. get_offloader().join_after_forward() self.cudagraphs[num_tokens] = cudagraph_metadata return cudagraph_metadata.outputs def _run_ubatches(self, ubatch_metadata, model) -> torch.Tensor: @torch.inference_mode() def _ubatch_thread(results, model, ubatch_metadata): with ubatch_metadata.context: model_output = model( input_ids=ubatch_metadata.input_ids, positions=ubatch_metadata.positions, intermediate_tensors=ubatch_metadata.intermediate_tensors, inputs_embeds=ubatch_metadata.inputs_embeds, ) results.append((ubatch_metadata.context.id, model_output)) results: list[tuple[int, torch.Tensor]] = [] # Ubatch threads will manually manage the forward context, so we # override it to None here so we can have it restored correctly # after both threads have finished with override_forward_context(None): ubatch_threads = [] for metadata in ubatch_metadata: thread = threading.Thread( target=_ubatch_thread, args=( results, model, metadata, ), ) ubatch_threads.append(thread) thread.start() self.ready_barrier.wait() # Wait for both threads to be ready ubatch_metadata[0].context.cpu_wait_event.set() for thread in ubatch_threads: thread.join() sorted_results = [value for position, value in sorted(results)] result = torch.cat(sorted_results, dim=0) return result def _make_ubatch_metadata( self, ubatch_slices, attn_metadata, slot_mapping, input_ids, positions, inputs_embeds, intermediate_tensors, compute_stream, dp_metadata, batch_descriptor, cudagraph_runtime_mode, ) -> list[UbatchMetadata]: # Create one forward context per ubatch forward_contexts = [] # slot_mapping can be None, an empty dict (from create_forward_context # converting None to {}), or a list of dicts (one per ubatch) has_slot_mapping = slot_mapping and isinstance(slot_mapping, list) for i, ubatch_slice in enumerate(ubatch_slices): forward_contexts.append( create_forward_context( attn_metadata[i] if attn_metadata is not None else None, self.vllm_config, dp_metadata=dp_metadata[i], batch_descriptor=batch_descriptor, cudagraph_runtime_mode=cudagraph_runtime_mode, slot_mapping=slot_mapping[i] if has_slot_mapping else None, ) ) ubatch_ctxs = make_ubatch_contexts( num_micro_batches=len(ubatch_slices), comm_stream=self.comm_stream, compute_stream=compute_stream, forward_contexts=forward_contexts, ready_barrier=self.ready_barrier, ) ubatch_metadata: list[UbatchMetadata] = [] for i, ubatch_slice in enumerate(ubatch_slices): ( sliced_input_ids, sliced_positions, sliced_inputs_embeds, sliced_intermediate_tensors, ) = self._slice_model_inputs( ubatch_slice.token_slice, input_ids, positions, inputs_embeds, intermediate_tensors, ) ubatch_metadata.append( UbatchMetadata( context=ubatch_ctxs[i], input_ids=sliced_input_ids, positions=sliced_positions, inputs_embeds=sliced_inputs_embeds, intermediate_tensors=sliced_intermediate_tensors, num_tokens=ubatch_slice.token_slice.stop - ubatch_slice.token_slice.start, ) ) return ubatch_metadata def _slice_model_inputs( self, tokens_slice: slice, input_ids, positions, inputs_embeds, intermediate_tensors, ): sliced_input_ids = input_ids[tokens_slice] # if we are using mrope. Mrope adds an additional dimension to the # positions tensor if positions.ndim == 2: sliced_positions = positions[:, tokens_slice] else: sliced_positions = positions[tokens_slice] sliced_inputs_embeds = inputs_embeds[tokens_slice] if inputs_embeds else None sliced_intermediate_tensors = ( intermediate_tensors[tokens_slice] if intermediate_tensors else None ) return ( sliced_input_ids, sliced_positions, sliced_inputs_embeds, sliced_intermediate_tensors, ) def __call__(self, *args, **kwargs): forward_context = get_forward_context() batch_descriptor = forward_context.batch_descriptor ubatch_slices = forward_context.ubatch_slices cudagraph_runtime_mode = forward_context.cudagraph_runtime_mode # If there's no ubatching, just run the runnable object if ubatch_slices is None: # This is to account for the case where ubatching was aborted. # When we capture full graphs we only capture one graph per shape, # meaning that if we have a ubatched cudagraph for the current # num_tokens, we don't have a non-ubatched one. Without this # check, the cudagraph wrapper will try to capture a cudagraph # for this shape during a normal run. if cudagraph_runtime_mode is CUDAGraphMode.FULL: assert batch_descriptor is not None if batch_descriptor.num_tokens in self.cudagraphs: cudagraph_runtime_mode = CUDAGraphMode.NONE if cudagraph_runtime_mode in (CUDAGraphMode.NONE, CUDAGraphMode.PIECEWISE): return self.runnable(*args, **kwargs) else: assert self.cudagraph_wrapper is not None return self.cudagraph_wrapper(*args, **kwargs) attn_metadata = forward_context.attn_metadata slot_mapping = forward_context.slot_mapping num_tokens = sum(ubatch_slice.num_tokens for ubatch_slice in ubatch_slices) input_ids = kwargs["input_ids"] positions = kwargs["positions"] intermediate_tensors = kwargs["intermediate_tensors"] inputs_embeds = kwargs["inputs_embeds"] compute_stream = torch.cuda.current_stream() dp_metadata = forward_context.dp_metadata # We shouldn't be here unless we are running with multiple DP ranks assert dp_metadata is not None ubatch_dp_metadata = [] for ubatch_slice in ubatch_slices: dp_size = self.vllm_config.parallel_config.data_parallel_size ubatch_num_tokens_across_dp = torch.tensor( [ubatch_slice.num_tokens] * dp_size, device="cpu", dtype=torch.int32 ) ubatch_dp_metadata.append( DPMetadata.make( self.vllm_config.parallel_config, ubatch_slice.num_tokens, ubatch_num_tokens_across_dp, ) ) if ( num_tokens not in self.cudagraphs and cudagraph_runtime_mode is CUDAGraphMode.FULL ): ubatch_metadata = self._make_ubatch_metadata( ubatch_slices=ubatch_slices, attn_metadata=attn_metadata, slot_mapping=slot_mapping, input_ids=input_ids, positions=positions, intermediate_tensors=intermediate_tensors, inputs_embeds=inputs_embeds, compute_stream=compute_stream, dp_metadata=ubatch_dp_metadata, batch_descriptor=batch_descriptor, cudagraph_runtime_mode=CUDAGraphMode.NONE, ) with self.sm_control: return self._capture_ubatches(ubatch_metadata, self.model) elif ( num_tokens in self.cudagraphs and cudagraph_runtime_mode is CUDAGraphMode.FULL ): cudagraph_metadata = self.cudagraphs[num_tokens] # Sync offloader before replay - ensures any external dependencies # from pre-capture prefetches are satisfied. get_offloader().sync_prev_onload() cudagraph_metadata.cudagraph.replay() return cudagraph_metadata.outputs else: ubatch_metadata = self._make_ubatch_metadata( ubatch_slices=ubatch_slices, attn_metadata=attn_metadata, slot_mapping=slot_mapping, input_ids=input_ids, positions=positions, intermediate_tensors=intermediate_tensors, inputs_embeds=inputs_embeds, compute_stream=compute_stream, dp_metadata=ubatch_dp_metadata, batch_descriptor=batch_descriptor, cudagraph_runtime_mode=CUDAGraphMode.NONE, ) with self.sm_control: return self._run_ubatches(ubatch_metadata, self.model)
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/worker/gpu_ubatch_wrapper.py", "license": "Apache License 2.0", "lines": 435, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/v1/worker/ubatch_utils.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from dataclasses import dataclass from typing import TypeAlias import numpy as np import torch from vllm.config import ParallelConfig from vllm.v1.attention.backend import CommonAttentionMetadata @dataclass class UBatchSlice: request_slice: slice token_slice: slice def is_empty(self) -> bool: return ( self.request_slice.start == self.request_slice.stop or self.token_slice.start == self.token_slice.stop ) @property def num_tokens(self) -> int: return self.token_slice.stop - self.token_slice.start UBatchSlices: TypeAlias = list[UBatchSlice] def is_last_ubatch_empty( orig_num_tokens: int, padded_num_tokens: int, num_ubatches: int ) -> bool: return (padded_num_tokens // num_ubatches) * (num_ubatches - 1) >= orig_num_tokens def check_ubatch_thresholds( config: ParallelConfig, num_tokens: int, uniform_decode: bool ) -> bool: if not config.use_ubatching: return False if uniform_decode: return num_tokens >= config.dbo_decode_token_threshold else: return num_tokens >= config.dbo_prefill_token_threshold # This pads the last ubatch slice out to the total number of tokens # (num_tokens + padding) since we do `create_ubatch_slices` before applying DP padding. def _pad_out_ubatch_slices( ubatch_slices: UBatchSlices, num_total_tokens: int, num_reqs_padded: int ) -> UBatchSlices: last_slice = ubatch_slices[-1] padded_last_request_slice = slice(last_slice.request_slice.start, num_reqs_padded) padded_last_token_slice = slice(last_slice.token_slice.start, num_total_tokens) return ubatch_slices[:-1] + [ UBatchSlice(padded_last_request_slice, padded_last_token_slice) ] def maybe_create_ubatch_slices( should_ubatch: bool, num_scheduled_tokens: np.ndarray, num_tokens_padded: int, num_reqs_padded: int, num_ubatches: int, split_point: list[int] | int | None = None, ) -> tuple[UBatchSlices | None, UBatchSlices | None]: if not should_ubatch: return None, None if split_point is None: split_point = int(num_tokens_padded) // num_ubatches token_split_points = [split_point * i for i in range(1, num_ubatches)] # TODO(lucas): Refactor the gpu_model_runner.py so we can pass # in cu_num_tokens directly (i.e. query_start_loc) cu_num_tokens = np.zeros(len(num_scheduled_tokens) + 1, dtype=np.int32) np.cumsum(num_scheduled_tokens, dtype=np.int32, out=cu_num_tokens[1:]) ubatch_slices = [] start_token = 0 # Add the end point to the split points to make iteration easier all_points = token_split_points + [cu_num_tokens[-1]] for end_token in all_points: token_slice = slice(start_token, end_token) # Determine request slices using exclusive stop semantics # Ubatch includes requests whose tokens overlap [start_token, end_token) # Start at the request that contains the start_token # or the request starting exactly at start_token (if on boundary) req_start = int(np.searchsorted(cu_num_tokens, start_token, side="right") - 1) # Stop at the request that starts at or after end_token req_stop = int(np.searchsorted(cu_num_tokens, end_token, side="left")) req_slice = slice(req_start, req_stop) ubatch_slices.append(UBatchSlice(req_slice, token_slice)) start_token = end_token ubatch_slices_padded = _pad_out_ubatch_slices( ubatch_slices, num_tokens_padded, num_reqs_padded ) assert sum(s.num_tokens for s in ubatch_slices_padded) == num_tokens_padded return ubatch_slices, ubatch_slices_padded def slice_query_start_locs( query_start_loc: torch.Tensor, request_slice: slice, ) -> torch.Tensor: """ Creates a new query_start_loc that corresponds to the requests in request_slice. Note: This function creates a new tensor to hold the new query_start_locs. This will break cudagraph compatibility. """ return ( query_start_loc[request_slice.start : request_slice.stop + 1] - query_start_loc[request_slice.start] ) def _make_metadata_with_slice( ubatch_slice: UBatchSlice, attn_metadata: CommonAttentionMetadata ) -> CommonAttentionMetadata: """ This function creates a new CommonAttentionMetadata that corresponds to the requests included in ubatch_slice """ assert not ubatch_slice.is_empty(), f"Ubatch slice {ubatch_slice} is empty" request_slice = ubatch_slice.request_slice token_slice = ubatch_slice.token_slice start_locs = attn_metadata.query_start_loc_cpu first_req = request_slice.start first_tok = token_slice.start last_req = request_slice.stop - 1 last_tok = token_slice.stop - 1 assert start_locs[first_req] <= first_tok < start_locs[first_req + 1], ( "Token slice start outside of first request" ) # NOTE: last token can be outside of the last request if we have CG padding. # If the request is split across ubatches, we have to adjust the metadata. # splits_first_request: The first request in this slice is the continuation of # a request that started in a previous slice. # splits_last_request: The last request in this slice continues into the # next slice. splits_first_request = first_tok > start_locs[first_req] splits_last_request = last_tok < start_locs[last_req + 1] - 1 query_start_loc_cpu = slice_query_start_locs(start_locs, request_slice) query_start_loc = slice_query_start_locs( attn_metadata.query_start_loc, request_slice ) assert len(query_start_loc) >= 2, ( f"query_start_loc must have at least 2 elements, got {len(query_start_loc)}" ) if splits_first_request: tokens_skipped = first_tok - start_locs[first_req] query_start_loc[1:] -= tokens_skipped query_start_loc_cpu[1:] -= tokens_skipped seq_lens = attn_metadata.seq_lens[request_slice] seq_lens_cpu = attn_metadata.seq_lens_cpu[request_slice] if splits_last_request: # NOTE: We use start_locs (the original query_start_loc_cpu) to calculate # the tokens skipped because query_start_loc_cpu might have been modified # if splits_first_request is True. tokens_skipped = start_locs[last_req + 1] - token_slice.stop query_start_loc[-1] -= tokens_skipped query_start_loc_cpu[-1] -= tokens_skipped # Make sure we don't modify the seq_lens tensors # (not cudagraph compatible) seq_lens = seq_lens.clone() seq_lens_cpu = seq_lens_cpu.clone() seq_lens[-1] -= tokens_skipped seq_lens_cpu[-1] -= tokens_skipped max_seq_len = int(seq_lens_cpu.max()) num_computed_tokens_cpu = attn_metadata.num_computed_tokens_cpu[request_slice] num_requests = request_slice.stop - request_slice.start num_actual_tokens = token_slice.stop - token_slice.start max_query_len = int( torch.max(torch.abs(query_start_loc_cpu[1:] - query_start_loc_cpu[:-1])).item() ) # This is to account for the case where we are in a dummy # run and query_start_loc_cpu is full of 0s if max_query_len == 0: max_query_len = attn_metadata.max_query_len block_table_tensor = attn_metadata.block_table_tensor[request_slice] slot_mapping = attn_metadata.slot_mapping[token_slice] return CommonAttentionMetadata( query_start_loc=query_start_loc, query_start_loc_cpu=query_start_loc_cpu, seq_lens=seq_lens, num_reqs=num_requests, num_actual_tokens=num_actual_tokens, max_query_len=max_query_len, max_seq_len=max_seq_len, block_table_tensor=block_table_tensor, slot_mapping=slot_mapping, _seq_lens_cpu=seq_lens_cpu, _num_computed_tokens_cpu=num_computed_tokens_cpu, ) def split_attn_metadata( ubatch_slices: list[UBatchSlice], common_attn_metadata: CommonAttentionMetadata, ) -> list[CommonAttentionMetadata]: """ Creates a new CommonAttentionMetadata instance that corresponds to the requests for each UBatchSlice in ubatch_slices. Note: This function does not modify common_attn_metadata """ results = [] for ubatch_slice in ubatch_slices: results.append(_make_metadata_with_slice(ubatch_slice, common_attn_metadata)) return results
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/worker/ubatch_utils.py", "license": "Apache License 2.0", "lines": 188, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/v1/worker/ubatching.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import threading import torch from vllm import forward_context from vllm.forward_context import ForwardContext from vllm.logger import init_logger from vllm.utils.torch_utils import current_stream logger = init_logger(__name__) _THREAD_ID_TO_CONTEXT: dict = {} # Here we hardcode the number of microbatches to 2 for default. _NUM_UBATCHES: int = 2 _CURRENT_CONTEXTS: list["UBatchContext | None"] = [] class UBatchContext: """ Context manager for micro-batching synchronization using threading events. """ def __init__( self, id: int, comm_stream: torch.cuda.Stream, compute_stream: torch.cuda.Stream, forward_context: ForwardContext, ready_barrier: threading.Barrier, cpu_wait_event: threading.Event, cpu_signal_event: threading.Event, gpu_comm_done_event: torch.Event, gpu_compute_done_event: torch.Event, schedule: str = "default", ): self.id = id self.comm_stream = comm_stream self.compute_stream = compute_stream self.forward_context = forward_context self.ready_barrier = ready_barrier self.cpu_wait_event = cpu_wait_event self.cpu_signal_event = cpu_signal_event self.current_stream = compute_stream self.gpu_comm_done_event = gpu_comm_done_event self.gpu_compute_done_event = gpu_compute_done_event self.schedule = schedule self.recv_hook = None def __enter__(self): global _CURRENT_CONTEXTS, _THREAD_ID_TO_CONTEXT _THREAD_ID_TO_CONTEXT[threading.get_ident()] = self.id _CURRENT_CONTEXTS[self.id] = self # _NUM_UBATCHES is set in make_ubatch_contexts self.ready_barrier.wait() self.cpu_wait_event.wait() self.cpu_wait_event.clear() self._restore_context() # Assume we want to start on the compute stream self.update_stream(self.compute_stream) return self def __exit__(self, exc_type, exc_val, exc_tb): global _CURRENT_CONTEXTS, _THREAD_ID_TO_CONTEXT _CURRENT_CONTEXTS[self.id] = None del _THREAD_ID_TO_CONTEXT[threading.get_ident()] self.maybe_run_recv_hook() self.cpu_signal_event.set() self.cpu_wait_event.clear() return False def _restore_context(self): forward_context._forward_context = self.forward_context def update_stream(self, stream): self.current_stream = stream if current_stream() != self.current_stream: torch.cuda.set_stream(self.current_stream) def _signal_comm_done(self): self.gpu_comm_done_event.record(self.comm_stream) def _signal_compute_done(self): self.gpu_compute_done_event.record(self.compute_stream) def _wait_compute_done(self): self.comm_stream.wait_event(self.gpu_compute_done_event) def _wait_comm_done(self): self.compute_stream.wait_event(self.gpu_comm_done_event) def _cpu_yield(self): # It is critical for correctness that only one thread is running # at a time. These asserts just make sure that this is the only # thread running before waking the other one up and going to sleep assert forward_context._forward_context == self.forward_context assert current_stream() == self.current_stream assert not self.cpu_wait_event.is_set() self.cpu_signal_event.set() self.cpu_wait_event.wait() self.cpu_wait_event.clear() self._restore_context() def switch_to_comm(self): self.update_stream(self.comm_stream) def switch_to_compute(self): self.update_stream(self.compute_stream) def switch_to_comm_sync(self): self._signal_compute_done() self.update_stream(self.comm_stream) self._wait_compute_done() def switch_to_compute_sync(self): self._signal_comm_done() self.update_stream(self.compute_stream) self._wait_comm_done() def maybe_run_recv_hook(self): if self.recv_hook is not None: self.recv_hook() self.recv_hook = None def yield_(self): self.current_stream = current_stream() self._cpu_yield() self.update_stream(self.current_stream) def yield_and_switch_from_compute_to_comm(self): assert current_stream() == self.compute_stream self._signal_compute_done() self._cpu_yield() assert self.current_stream == self.compute_stream self.update_stream(self.comm_stream) self._wait_compute_done() def yield_and_switch_from_comm_to_compute(self): assert current_stream() == self.comm_stream self._signal_comm_done() self._cpu_yield() assert self.current_stream == self.comm_stream self.update_stream(self.compute_stream) self._wait_comm_done() def dbo_enabled() -> bool: return len(_THREAD_ID_TO_CONTEXT) > 0 def dbo_current_ubatch_id() -> int: if len(_THREAD_ID_TO_CONTEXT) == 0: return 0 return _THREAD_ID_TO_CONTEXT[threading.get_ident()] def _register_ubatch_function(func): def wrapper(*args, **kwargs): if len(_THREAD_ID_TO_CONTEXT) > 0: ctx_idx = _THREAD_ID_TO_CONTEXT[threading.get_ident()] ctx = _CURRENT_CONTEXTS[ctx_idx] func(ctx, *args, **kwargs) return wrapper dbo_maybe_run_recv_hook = _register_ubatch_function(UBatchContext.maybe_run_recv_hook) dbo_yield = _register_ubatch_function(UBatchContext.yield_) dbo_yield_and_switch_from_compute_to_comm = _register_ubatch_function( UBatchContext.yield_and_switch_from_compute_to_comm ) dbo_yield_and_switch_from_comm_to_compute = _register_ubatch_function( UBatchContext.yield_and_switch_from_comm_to_compute ) dbo_switch_to_comm = _register_ubatch_function(UBatchContext.switch_to_comm) dbo_switch_to_compute = _register_ubatch_function(UBatchContext.switch_to_compute) dbo_switch_to_comm_sync = _register_ubatch_function(UBatchContext.switch_to_comm_sync) dbo_switch_to_compute_sync = _register_ubatch_function( UBatchContext.switch_to_compute_sync ) def dbo_register_recv_hook(recv_hook): if len(_THREAD_ID_TO_CONTEXT) > 0: ctx_idx = _THREAD_ID_TO_CONTEXT[threading.get_ident()] next_ctx = _CURRENT_CONTEXTS[(ctx_idx + 1) % _NUM_UBATCHES] next_ctx.recv_hook = recv_hook def dbo_get_previous_event(func, *args, **kwargs): if len(_THREAD_ID_TO_CONTEXT) > 0: ctx_idx = _THREAD_ID_TO_CONTEXT[threading.get_ident()] ctx = _CURRENT_CONTEXTS[ctx_idx] # execute callable on the ubatch compute stream to record/wait events there with torch.cuda.stream(ctx.compute_stream): return func(*args, **kwargs) def make_ubatch_contexts( num_micro_batches: int, compute_stream: torch.cuda.Stream, comm_stream: torch.cuda.Stream, forward_contexts: list[ForwardContext], ready_barrier: threading.Barrier, schedule: str = "default", ) -> list[UBatchContext]: global _NUM_UBATCHES, _CURRENT_CONTEXTS assert num_micro_batches > 1, "num_micro_batches must be greater than 1" _NUM_UBATCHES = num_micro_batches # Ensure the global context list is large enough if len(_CURRENT_CONTEXTS) < num_micro_batches: _CURRENT_CONTEXTS.extend([None] * (num_micro_batches - len(_CURRENT_CONTEXTS))) """ Create a context manager for micro-batching synchronization. """ cpu_events = [threading.Event() for _ in range(num_micro_batches)] gpu_comm_done_events = [torch.Event() for _ in range(num_micro_batches)] gpu_compute_done_events = [torch.Event() for _ in range(num_micro_batches)] ctxs = [] for i in range(num_micro_batches): ctx = UBatchContext( id=i, compute_stream=compute_stream, comm_stream=comm_stream, forward_context=forward_contexts[i], ready_barrier=ready_barrier, cpu_wait_event=cpu_events[i], cpu_signal_event=cpu_events[(i + 1) % num_micro_batches], gpu_comm_done_event=gpu_comm_done_events[i], gpu_compute_done_event=gpu_compute_done_events[i], schedule=schedule, ) ctxs.append(ctx) return ctxs
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/worker/ubatching.py", "license": "Apache License 2.0", "lines": 196, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/config/speculative.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import ast from typing import TYPE_CHECKING, Any, Literal, get_args from pydantic import Field, SkipValidation, model_validator from typing_extensions import Self from vllm.config import LoadConfig from vllm.config.model import ModelConfig from vllm.config.parallel import ParallelConfig from vllm.config.utils import config from vllm.logger import init_logger from vllm.transformers_utils.config import get_hf_text_config from vllm.utils.hashing import safe_hash from vllm.utils.import_utils import LazyLoader, has_arctic_inference if TYPE_CHECKING: from transformers import PretrainedConfig import vllm.model_executor.layers.quantization as me_quant else: PretrainedConfig = Any me_quant = LazyLoader( "model_executor", globals(), "vllm.model_executor.layers.quantization" ) logger = init_logger(__name__) MTPModelTypes = Literal[ "deepseek_mtp", "mimo_mtp", "glm4_moe_mtp", "glm4_moe_lite_mtp", "glm_ocr_mtp", "ernie_mtp", "nemotron_h_mtp", "exaone_moe_mtp", "qwen3_next_mtp", "qwen3_5_mtp", "longcat_flash_mtp", "mtp", "pangu_ultra_moe_mtp", "step3p5_mtp", ] EagleModelTypes = Literal["eagle", "eagle3", MTPModelTypes] SpeculativeMethod = Literal[ "ngram", "medusa", "mlp_speculator", "draft_model", "suffix", EagleModelTypes, ] @config class SpeculativeConfig: """Configuration for speculative decoding.""" enforce_eager: bool | None = None """Override the default enforce_eager from model_config""" # General speculative decoding control num_speculative_tokens: int = Field(default=None, gt=0) """The number of speculative tokens, if provided. It will default to the number in the draft model config if present, otherwise, it is required.""" model: str | None = None """The name of the draft model, eagle head, or additional weights, if provided.""" method: SpeculativeMethod | None = None """The name of the speculative method to use. If users provide and set the `model` param, the speculative method type will be detected automatically if possible, if `model` param is not provided, the method name must be provided. If using `ngram` method, the related configuration `prompt_lookup_max` and `prompt_lookup_min` should be considered.""" draft_tensor_parallel_size: int | None = Field(default=None, ge=1) """The degree of the tensor parallelism for the draft model. Can only be 1 or the same as the target model's tensor parallel size.""" tensor_parallel_size: int | None = None """Users should pass "draft_tensor_parallel_size". This parameter's purpose is to warn users when they mistakenly provide the wrong argument.""" # Draft model configuration quantization: me_quant.QuantizationMethods | None = None """Quantization method that was used to quantize the draft model weights. If `None`, we assume the model weights are not quantized. Note that it only takes effect when using the draft model-based speculative method.""" max_model_len: int | None = Field(default=None, ge=1) """The maximum model length of the draft model. Used when testing the ability to skip speculation for some sequences.""" revision: str | None = None """The specific model version to use for the draft model. It can be a branch name, a tag name, or a commit id. If unspecified, will use the default version.""" code_revision: str | None = None """The specific revision to use for the draft model code on Hugging Face Hub. It can be a branch name, a tag name, or a commit id. If unspecified, will use the default version.""" # Advanced control disable_padded_drafter_batch: bool = False """Disable input padding for speculative decoding. If set to True, speculative input batches can contain sequences of different lengths, which may only be supported by certain attention backends. This currently only affects the EAGLE method of speculation.""" use_local_argmax_reduction: bool = False """Use vocab-parallel local argmax instead of all-gathering full logits for draft token generation. Reduces communication from O(vocab_size) to O(2 * tp_size) per token. Only applies to greedy draft selection in non-tree speculation.""" # Ngram proposer configuration prompt_lookup_max: int | None = Field(default=None, ge=1) """Maximum size of ngram token window when using Ngram proposer, required when method is set to ngram.""" prompt_lookup_min: int | None = Field(default=None, ge=1) """Minimum size of ngram token window when using Ngram proposer, if provided. Defaults to 1.""" # Alternative drafting strategies speculative_token_tree: str | None = None """Specifies the tree structure for speculative token generation. """ parallel_drafting: bool = False """Enable parallel drafting, where all speculative tokens are generated in parallel rather than sequentially. This can improve performance but requires the speculative model be trained to support parallel drafting. Only compatible with EAGLE and draft model methods.""" # required configuration params passed from engine target_model_config: SkipValidation[ModelConfig] = None # type: ignore """The configuration of the target model.""" target_parallel_config: SkipValidation[ParallelConfig] = None # type: ignore """The parallel configuration for the target model.""" # params generated in the post-init stage draft_model_config: SkipValidation[ModelConfig] = None # type: ignore """The configuration of the draft model initialized internal.""" draft_parallel_config: SkipValidation[ParallelConfig] = None # type: ignore """The parallel configuration for the draft model initialized internal.""" # Suffix decoding configuration suffix_decoding_max_tree_depth: int = 24 """The maximum depth of the suffix decoding global and prompt trees. The tree depth limits the sum of the prefix match and speculation lengths.""" suffix_decoding_max_cached_requests: int = 10000 """The maximum number of requests to cache in the global suffix tree. If exceeded, will trigger eviction in FIFO order. If set to 0, the global suffix tree is disabled and past responses are not cached (prompt trees are still used).""" suffix_decoding_max_spec_factor: float = 1.0 """The maximum spec factor for suffix decoding. The spec factor controls speculation lengths based on the prefix match length: max_spec_tokens = max_spec_factor * prefix_match_length.""" suffix_decoding_min_token_prob: float = 0.1 """The minimum token probability for suffix decoding. Will only speculate tokens with estimated probability (based on frequency counts) greater than or equal to this value.""" draft_load_config: LoadConfig | None = None """Load config for the draft model. If not specified, will use the load config from the target model.""" def compute_hash(self) -> str: """ WARNING: Whenever a new field is added to this config, ensure that it is included in the factors list if it affects the computation graph. Provide a hash that uniquely identifies all the configs that affect the structure of the computation graph from input ids/embeddings to the final hidden states, excluding anything before input ids/embeddings and after the final hidden states. """ factors: list[Any] = [] # Eagle3 affects the computation graph because it returns intermediate # hidden states in addition to the final hidden state. factors.append(self.method == "eagle3") hash_str = safe_hash(str(factors).encode(), usedforsecurity=False).hexdigest() return hash_str @staticmethod def hf_config_override(hf_config: PretrainedConfig) -> PretrainedConfig: initial_architecture = hf_config.architectures[0] if hf_config.model_type in ("deepseek_v3", "deepseek_v32", "glm_moe_dsa"): hf_config.model_type = "deepseek_mtp" if hf_config.model_type == "deepseek_mtp": n_predict = getattr(hf_config, "num_nextn_predict_layers", None) hf_config.update( {"n_predict": n_predict, "architectures": ["DeepSeekMTPModel"]} ) if hf_config.model_type in ("pangu_ultra_moe"): hf_config.model_type = "pangu_ultra_moe_mtp" if hf_config.model_type == "pangu_ultra_moe_mtp": n_predict = getattr(hf_config, "num_nextn_predict_layers", None) hf_config.update( {"n_predict": n_predict, "architectures": ["OpenPanguMTPModel"]} ) if hf_config.architectures[0] == "MiMoForCausalLM": hf_config.model_type = "mimo_mtp" n_predict = getattr(hf_config, "num_nextn_predict_layers", None) hf_config.update( { "num_hidden_layers": 0, "n_predict": n_predict, "architectures": ["MiMoMTPModel"], } ) if hf_config.architectures[0] == "Glm4MoeForCausalLM": hf_config.model_type = "glm4_moe_mtp" n_predict = getattr(hf_config, "num_nextn_predict_layers", None) hf_config.update( { "n_predict": n_predict, "architectures": ["Glm4MoeMTPModel"], } ) if hf_config.architectures[0] == "Glm4MoeLiteForCausalLM": hf_config.model_type = "glm4_moe_lite_mtp" n_predict = getattr(hf_config, "num_nextn_predict_layers", None) hf_config.update( { "num_hidden_layers": 0, "n_predict": n_predict, "architectures": ["Glm4MoeLiteMTPModel"], } ) if hf_config.architectures[0] == "GlmOcrForConditionalGeneration": hf_config.model_type = "glm_ocr_mtp" n_predict = getattr(hf_config, "num_nextn_predict_layers", None) hf_config.update( { "num_hidden_layers": 0, "n_predict": n_predict, "architectures": ["GlmOcrMTPModel"], } ) if hf_config.model_type == "ernie4_5_moe": hf_config.model_type = "ernie_mtp" if hf_config.model_type == "ernie_mtp": n_predict = getattr(hf_config, "num_nextn_predict_layers", None) hf_config.update( {"n_predict": n_predict, "architectures": ["ErnieMTPModel"]} ) if ( hf_config.model_type == "nemotron_h" and hasattr(hf_config, "num_nextn_predict_layers") and hf_config.num_nextn_predict_layers > 0 ): # Check if this is an MTP variant hf_config.model_type = "nemotron_h_mtp" if hf_config.model_type == "nemotron_h_mtp": n_predict = getattr(hf_config, "num_nextn_predict_layers", 1) hf_config.update( {"n_predict": n_predict, "architectures": ["NemotronHMTPModel"]} ) if hf_config.model_type == "qwen3_next": hf_config.model_type = "qwen3_next_mtp" if hf_config.model_type == "qwen3_next_mtp": n_predict = getattr(hf_config, "num_nextn_predict_layers", None) hf_config.update( {"n_predict": n_predict, "architectures": ["Qwen3NextMTP"]} ) if hf_config.model_type == "exaone_moe": hf_config.model_type = "exaone_moe_mtp" if hf_config.model_type == "exaone_moe_mtp": n_predict = getattr(hf_config, "num_nextn_predict_layers", None) hf_config.update( {"n_predict": n_predict, "architectures": ["ExaoneMoeMTP"]} ) if hf_config.model_type in ("qwen3_5", "qwen3_5_moe"): is_moe = hf_config.model_type == "qwen3_5_moe" hf_config.model_type = "qwen3_5_mtp" n_predict = getattr(hf_config, "mtp_num_hidden_layers", None) hf_config.update( { "n_predict": n_predict, "architectures": ["Qwen3_5MoeMTP" if is_moe else "Qwen3_5MTP"], } ) if hf_config.model_type == "longcat_flash": hf_config.model_type = "longcat_flash_mtp" n_predict = getattr(hf_config, "num_nextn_predict_layers", 1) hf_config.update( {"n_predict": n_predict, "architectures": ["LongCatFlashMTPModel"]} ) if hf_config.model_type == "step3p5": hf_config.model_type = "step3p5_mtp" n_predict = getattr(hf_config, "num_nextn_predict_layers", 1) hf_config.update({"n_predict": n_predict, "architectures": ["Step3p5MTP"]}) if initial_architecture == "MistralLarge3ForCausalLM": hf_config.update({"architectures": ["EagleMistralLarge3ForCausalLM"]}) return hf_config def __post_init__(self): # Note: "method" is a new parameter that helps to extend the # configuration of non-model-based proposers, and the "model" parameter # will be used to set the draft model, eagle head, or additional weight # when needed. If users do not specify "method", the speculative method # will be detected automatically if possible. If the speculative method # can not be detected, it will be considered as the "draft_model" by # default. # infer method from user args if self.method is None: if self.model in ("ngram", "[ngram]"): self.method = "ngram" else: self.method = "draft_model" if self.method in get_args(MTPModelTypes) and self.method != "mtp": logger.warning( "method `%s` is deprecated and replaced with mtp.", self.method ) self.method = "mtp" if self.model is None and self.num_speculative_tokens is not None: if self.method == "mtp": if self.target_model_config is None: raise ValueError("target_model_config must be present for mtp") if self.target_model_config.hf_text_config.model_type == "deepseek_v32": # FIXME(luccafong): cudagraph with v32 MTP is not supported, # remove this when the issue is fixed. self.enforce_eager = True # use the draft model from the same model: self.model = self.target_model_config.model # Align the quantization of draft model for cases such as # --quantization fp8 with a bf16 checkpoint. if not self.quantization: self.quantization = self.target_model_config.quantization elif self.method in ("ngram", "[ngram]"): self.model = "ngram" elif self.method == "suffix": self.model = "suffix" else: raise ValueError( "num_speculative_tokens was provided but without speculative model." ) if self.method in ("ngram", "[ngram]"): # Unified to "ngram" internally self.method = "ngram" # Set default values if not provided if self.prompt_lookup_min is None and self.prompt_lookup_max is None: # TODO(woosuk): Tune these values. They are arbitrarily chosen. self.prompt_lookup_min = 5 self.prompt_lookup_max = 5 elif self.prompt_lookup_min is None: if self.prompt_lookup_max is None: raise ValueError( "Either prompt_lookup_max or prompt_lookup_min must be " "provided when using the ngram method." ) self.prompt_lookup_min = self.prompt_lookup_max elif self.prompt_lookup_max is None: if self.prompt_lookup_min is None: raise ValueError( "Either prompt_lookup_max or prompt_lookup_min must be " "provided when using the ngram method." ) self.prompt_lookup_max = self.prompt_lookup_min # Validate values if self.prompt_lookup_min > self.prompt_lookup_max: raise ValueError( f"prompt_lookup_min={self.prompt_lookup_min} must " f"be <= prompt_lookup_max={self.prompt_lookup_max}" ) # TODO: current we still need extract vocab_size from target model # config, in future, we may try refactor it out, and set # draft related config as None here. self.draft_model_config = self.target_model_config self.draft_parallel_config = self.target_parallel_config elif self.method == "suffix": self._validate_suffix_decoding() else: self.prompt_lookup_max = 0 self.prompt_lookup_min = 0 if self.model is not None: self.draft_model_config = ModelConfig( model=self.model, runner="draft", tokenizer=self.target_model_config.tokenizer, tokenizer_mode=self.target_model_config.tokenizer_mode, trust_remote_code=self.target_model_config.trust_remote_code, allowed_local_media_path=self.target_model_config.allowed_local_media_path, allowed_media_domains=self.target_model_config.allowed_media_domains, dtype=self.target_model_config.dtype, seed=self.target_model_config.seed, revision=self.revision, code_revision=self.code_revision, tokenizer_revision=self.target_model_config.tokenizer_revision, spec_target_max_model_len=self.target_model_config.max_model_len, quantization=self.quantization, enforce_eager=self.target_model_config.enforce_eager, max_logprobs=self.target_model_config.max_logprobs, hf_overrides=SpeculativeConfig.hf_config_override, config_format=self.target_model_config.config_format, ) # Automatically detect the method if self.method in ("eagle", "eagle3"): pass # examples: # yuhuili/EAGLE-LLaMA3-Instruct-8B # yuhuili/EAGLE3-LLaMA3.1-Instruct-8B # AngelSlim/Qwen3-8B_eagle3 elif "eagle-" in self.draft_model_config.model.lower(): self.method = "eagle" elif "eagle3" in self.draft_model_config.model.lower(): self.method = "eagle3" elif self.draft_model_config.hf_config.model_type == "medusa": self.method = "medusa" elif self.draft_model_config.hf_config.model_type == "mlp_speculator": self.method = "mlp_speculator" elif self.draft_model_config.hf_config.model_type in get_args( MTPModelTypes ): self.method = "mtp" if self.num_speculative_tokens > 1: logger.warning( "Enabling num_speculative_tokens > 1 will run " "multiple times of forward on same MTP layer" ",which may result in lower acceptance rate" ) elif self.draft_model_config.hf_config.model_type in ( "longcat_flash_mtp" ): self.method = "longcat_flash_mtp" if self.num_speculative_tokens > 1: logger.warning( "LongCat MTP models only have " "one layer. Might need some code changes " "to support multiple layers." ) elif self.method == "draft_model": pass else: raise NotImplementedError( f"Unsupported speculative method: '{self.method}'" ) # Replace hf_config for EAGLE draft_model if self.method in ("eagle", "eagle3"): from vllm.transformers_utils.configs import SpeculatorsConfig from vllm.transformers_utils.configs.eagle import EAGLEConfig if isinstance( self.draft_model_config.hf_config, (EAGLEConfig, SpeculatorsConfig), ): pass else: eagle_config = EAGLEConfig( self.draft_model_config.hf_config, method=self.method, model_type="eagle", ) # EAGLEConfig primarily updates architectures, so update # all architectures-related fields in draft_model_config self.draft_model_config.hf_config = eagle_config self.draft_model_config.hf_text_config = get_hf_text_config( self.draft_model_config.hf_config ) self.draft_model_config.model_arch_config = ( self.draft_model_config.get_model_arch_config() ) model_info, arch = ( self.draft_model_config.registry.inspect_model_cls( self.draft_model_config.architectures, self.draft_model_config, ) ) self.draft_model_config._model_info = model_info self.draft_model_config._architecture = arch if self.num_speculative_tokens is not None and hasattr( self.draft_model_config.hf_config, "num_lookahead_tokens" ): self.draft_model_config.hf_config.num_lookahead_tokens = ( self.num_speculative_tokens ) n_predict = getattr( self.draft_model_config.hf_config, "n_predict", None ) if n_predict is not None: if self.num_speculative_tokens is None: # Default to max value defined in draft model config. self.num_speculative_tokens = n_predict elif ( self.num_speculative_tokens > n_predict and self.num_speculative_tokens % n_predict != 0 ): # Ensure divisibility for MTP module reuse. raise ValueError( f"num_speculative_tokens:{self.num_speculative_tokens}" f" must be divisible by {n_predict=}" ) if self.speculative_token_tree is None: if self.num_speculative_tokens is None: raise ValueError( "A speculative model was provided, but neither " "`speculative_token_tree` nor `num_speculative_tokens` " "was provided" ) # Generate chain of tokens. self.speculative_token_tree = str( [(i + 1) * (0,) for i in range(self.num_speculative_tokens)] ) else: # Sort the token tree breadth-first. tree_choices = ast.literal_eval(self.speculative_token_tree) self.speculative_token_tree = str( sorted(tree_choices, key=lambda t: (len(t), t)) ) self.draft_tensor_parallel_size = ( SpeculativeConfig._verify_and_get_draft_tp( self.target_parallel_config, self.draft_tensor_parallel_size, self.draft_model_config.hf_config, ) ) self.draft_model_config.max_model_len = ( SpeculativeConfig._maybe_override_draft_max_model_len( self.max_model_len, self.draft_model_config.max_model_len, self.target_model_config.max_model_len, ) ) self.draft_parallel_config = ( SpeculativeConfig.create_draft_parallel_config( self.target_parallel_config, self.draft_tensor_parallel_size ) ) return self def _validate_suffix_decoding(self): if not has_arctic_inference(): raise ImportError( "Arctic Inference is required for suffix decoding. " "Install via `pip install arctic-inference==0.1.1`." ) if self.num_speculative_tokens is None: # Suffix decoding decides the actual number of speculative tokens # dynamically and treats num_speculative_tokens as a maximum limit. self.num_speculative_tokens = self.suffix_decoding_max_tree_depth logger.warning( "Defaulted num_speculative_tokens to %s for suffix decoding.", self.num_speculative_tokens, ) # Validate values if self.suffix_decoding_max_tree_depth < 1: raise ValueError( f"suffix_decoding_max_tree_depth=" f"{self.suffix_decoding_max_tree_depth} must be >= 1" ) if self.suffix_decoding_max_cached_requests < 0: raise ValueError( f"suffix_decoding_max_cached_requests=" f"{self.suffix_decoding_max_cached_requests} must be >= 0" ) if self.suffix_decoding_max_spec_factor < 0: raise ValueError( f"suffix_decoding_max_spec_factor=" f"{self.suffix_decoding_max_spec_factor} must be >= 0" ) if not 0 <= self.suffix_decoding_min_token_prob <= 1: raise ValueError( f"suffix_decoding_min_token_prob=" f"{self.suffix_decoding_min_token_prob} must be in [0, 1]" ) @staticmethod def _maybe_override_draft_max_model_len( speculative_max_model_len: int | None, draft_max_model_len: int, target_max_model_len: int, ) -> int: """Determine the max sequence len for the draft model. This is usually the draft_max_model_len, but may be the target_max_model_len if it is less than the draft_max_model_len, or may be speculative_max_model_len if it is specified. This is necessary so that sequences do not exceed the capacity of the draft model or the target model. speculative_max_model_len is mainly used for testing that sequences can skip speculation. """ if speculative_max_model_len is not None: if speculative_max_model_len > draft_max_model_len: raise ValueError( f"{speculative_max_model_len=} cannot be " f"larger than {draft_max_model_len=}" ) if speculative_max_model_len > target_max_model_len: raise ValueError( f"{speculative_max_model_len=} cannot be " f"larger than {target_max_model_len=}" ) return speculative_max_model_len return min( draft_max_model_len, target_max_model_len, ) @staticmethod def _verify_and_get_draft_tp( target_parallel_config: ParallelConfig, speculative_draft_tensor_parallel_size: int | None, draft_hf_config: PretrainedConfig, ) -> int: """ Verifies and adjusts the tensor parallel size for a draft model specified using speculative_draft_tensor_parallel_size. """ # If speculative_draft_tensor_parallel_size is unset then set it # appropriately else verify that it is set correctly. if speculative_draft_tensor_parallel_size is None: if draft_hf_config.model_type == "mlp_speculator": speculative_draft_tensor_parallel_size = 1 if target_parallel_config.tensor_parallel_size > 1: logger.warning( "%s cannot currently be run with tp>1; " "setting speculative_draft_tensor_parallel_size=1", draft_hf_config.model_type, ) else: speculative_draft_tensor_parallel_size = ( target_parallel_config.tensor_parallel_size ) elif speculative_draft_tensor_parallel_size not in ( 1, target_parallel_config.tensor_parallel_size, ): raise ValueError( f"{speculative_draft_tensor_parallel_size=} cannot be " f"other value than 1 or target model tensor_parallel_size" ) return speculative_draft_tensor_parallel_size @staticmethod def create_draft_parallel_config( target_parallel_config: ParallelConfig, speculative_draft_tensor_parallel_size: int, ) -> ParallelConfig: """Create a parallel config for use by the draft worker. This is mostly a copy of the target parallel config, except the tp_size. """ draft_parallel_config = ParallelConfig( pipeline_parallel_size=target_parallel_config.pipeline_parallel_size, tensor_parallel_size=speculative_draft_tensor_parallel_size, distributed_executor_backend=target_parallel_config.distributed_executor_backend, max_parallel_loading_workers=target_parallel_config.max_parallel_loading_workers, disable_custom_all_reduce=target_parallel_config.disable_custom_all_reduce, ray_workers_use_nsight=target_parallel_config.ray_workers_use_nsight, placement_group=target_parallel_config.placement_group, ) return draft_parallel_config @model_validator(mode="after") def _verify_args(self) -> Self: if self.tensor_parallel_size is not None: raise ValueError( "'tensor_parallel_size' is not a valid argument in the " "speculative_config. Please pass 'draft_tensor_parallel_size' instead." ) if self.num_speculative_tokens is None: raise ValueError( "num_speculative_tokens must be provided with " "speculative model unless the draft model config contains an " "n_predict parameter." ) if self.num_speculative_tokens <= 0: raise ValueError( "Expected num_speculative_tokens to be greater " f"than zero ({self.num_speculative_tokens})." ) if self.draft_model_config: self.draft_model_config.verify_with_parallel_config( self.draft_parallel_config ) eagle3_target_supported = [ "llama", "qwen", "minicpm", "gpt_oss", "hunyuan_vl", "hunyuan_v1_dense", "afmoe", "nemotron_h", ] if ( self.method == "eagle3" and self.target_model_config and not any( supported_model in self.target_model_config.hf_text_config.model_type for supported_model in eagle3_target_supported ) ): raise ValueError( f"Eagle3 is only supported for {eagle3_target_supported} models. " # noqa: E501 f"Got {self.target_model_config.hf_text_config.model_type=}" ) self.verify_equal_vocab_size_if_draft_model() return self def verify_equal_vocab_size_if_draft_model(self): if ( self.method == "draft_model" and self.target_model_config is not None and self.draft_model_config is not None ): target_vocab_size = self.target_model_config.get_vocab_size() draft_vocab_size = self.draft_model_config.get_vocab_size() if target_vocab_size != draft_vocab_size: raise ValueError( f"Target and draft model should have the same vocabulary size. " f"Target model vocab_size={target_vocab_size}. " f"Draft model vocab_size={draft_vocab_size}. " f"Using models with different tokenizers can cause out-of-bounds " f"errors during speculative decoding." ) @property def max_num_new_slots_for_drafting(self) -> int: """ Calculate the maximum number of new slots that might be added to the batch when drafting. """ slots_per_req = 0 # for serial non-draft-model methods, no change needed if self.parallel_drafting: # For parallel drafting, we need one new slot per 'masked' token slots_per_req = self.num_speculative_tokens - 1 if self.uses_draft_model(): # For draft model-based speculation, we need one new slot per request # Since we do not slice the draft tokens slots_per_req += 1 return slots_per_req def use_eagle(self) -> bool: return self.method in ("eagle", "eagle3", "mtp") def uses_draft_model(self) -> bool: return self.method == "draft_model" def __repr__(self) -> str: method = self.method model = None if method in ("ngram", "suffix") else self.draft_model_config.model num_spec_tokens = self.num_speculative_tokens return f"SpeculativeConfig({method=}, {model=}, {num_spec_tokens=})"
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/config/speculative.py", "license": "Apache License 2.0", "lines": 711, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/distributed/test_expert_placement.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import pytest from vllm.model_executor.layers.fused_moe.layer import determine_expert_map def verify_round_robin_pattern(expert_map, ep_rank, ep_size, global_num_experts): """Verify that the expert map follows the round_robin pattern.""" # Calculate expected local experts (supporting non-divisible cases) base_experts = global_num_experts // ep_size remainder = global_num_experts % ep_size local_num_experts = base_experts + 1 if ep_rank < remainder else base_experts # Expected expert IDs for this rank in round_robin pattern # For non-divisible cases, ranks with extra experts start earlier expected_expert_ids = [] for expert_idx in range(local_num_experts): global_expert_id = ep_rank + expert_idx * ep_size expected_expert_ids.append(global_expert_id) # Check that only expected experts are mapped to this rank for global_expert_id in range(global_num_experts): if global_expert_id in expected_expert_ids: local_expert_id = expert_map[global_expert_id] expected_local_id = expected_expert_ids.index(global_expert_id) assert local_expert_id == expected_local_id, ( f"Global expert {global_expert_id} should map to local expert " f"{expected_local_id}, got {local_expert_id}" ) else: assert expert_map[global_expert_id] == -1, ( f"Global expert {global_expert_id} should not be mapped to this rank" ) # Verify that all local expert IDs are consecutive starting from 0 local_expert_ids = [expert_map[global_id] for global_id in expected_expert_ids] expected_local_ids = list(range(local_num_experts)) assert local_expert_ids == expected_local_ids, ( f"Expected local expert IDs {expected_local_ids}, got {local_expert_ids}" ) @pytest.mark.parametrize("expert_placement_strategy", ["round_robin"]) @pytest.mark.parametrize("world_size", [2, 4]) def test_expert_placement_various_sizes(expert_placement_strategy, world_size): """Test round_robin expert placement with various expert counts.""" # Test with different global_num_experts values # Include both divisible and non-divisible cases if world_size == 2: test_cases = [ (4, 2), # 4 experts (divisible) (8, 2), # 8 experts (divisible) (9, 2), # 9 experts (non-divisible) (16, 2), # 16 experts (divisible) (17, 2), # 17 experts (non-divisible) ] elif world_size == 4: test_cases = [ (8, 4), # 8 experts (divisible) (16, 4), # 16 experts (divisible) (18, 4), # 18 experts (non-divisible) (32, 4), # 32 experts (divisible) (33, 4), # 33 experts (non-divisible) ] else: test_cases = [] for test_global_experts, test_ep_size in test_cases: # Ensure ep_size matches world_size assert test_ep_size == world_size, ( f"ep_size {test_ep_size} must equal world_size {world_size}" ) # Test each rank for ep_rank in range(world_size): # Calculate expected local experts base_experts = test_global_experts // test_ep_size remainder = test_global_experts % test_ep_size if ep_rank < remainder: expected_test_local = base_experts + 1 else: expected_test_local = base_experts test_local_experts, test_expert_map, _ = determine_expert_map( ep_size=test_ep_size, ep_rank=ep_rank, global_num_experts=test_global_experts, expert_placement_strategy=expert_placement_strategy, ) assert test_local_experts == expected_test_local, ( f"For {test_global_experts} experts on {test_ep_size} ranks, " f"rank {ep_rank}: expected {expected_test_local} local" f"experts, got {test_local_experts}" ) if test_expert_map is not None: assert test_expert_map.shape == (test_global_experts,), ( f"Expected expert map shape ({test_global_experts},), " f"got {test_expert_map.shape}" ) # Verify round_robin pattern for this test case verify_round_robin_pattern( test_expert_map, ep_rank, test_ep_size, test_global_experts ) @pytest.mark.parametrize("expert_placement_strategy", ["round_robin"]) @pytest.mark.parametrize("world_size", [2, 4]) def test_expert_placement_edge_cases(expert_placement_strategy, world_size): """Test edge cases for round_robin expert placement.""" # Test case 1: ep_size = 1 (should return None for expert_map) local_num_experts, expert_map, _ = determine_expert_map( ep_size=1, ep_rank=0, global_num_experts=8, expert_placement_strategy=expert_placement_strategy, ) assert local_num_experts == 8, "For ep_size=1, should get all experts" assert expert_map is None, "For ep_size=1, expert_map should be None" # Test case 2: ep_size = 0 (should raise assertion) with pytest.raises(AssertionError): determine_expert_map( ep_size=0, ep_rank=0, global_num_experts=8, expert_placement_strategy=expert_placement_strategy, ) def test_determine_expert_map_comprehensive(): """Test of determine_expert_map function with various configurations.""" # Test cases: (ep_size, ep_rank, global_num_experts, # expert_placement_strategy, expected_local, expected_map_pattern) test_cases = [ # Round robin placement tests ( 2, 0, 8, "round_robin", 4, [0, -1, 1, -1, 2, -1, 3, -1], ), # rank 0 gets even experts ( 2, 1, 8, "round_robin", 4, [-1, 0, -1, 1, -1, 2, -1, 3], ), # rank 1 gets odd experts ( 2, 0, 9, "round_robin", 5, [0, -1, 1, -1, 2, -1, 3, -1, 4], ), # rank 0 gets 5 experts (even + last) ( 2, 1, 9, "round_robin", 4, [-1, 0, -1, 1, -1, 2, -1, 3, -1], ), # rank 1 gets 4 experts (odd) # 4-rank tests ( 4, 0, 8, "round_robin", 2, [0, -1, -1, -1, 1, -1, -1, -1], ), # rank 0 gets experts 0, 4 ( 4, 1, 8, "round_robin", 2, [-1, 0, -1, -1, -1, 1, -1, -1], ), # rank 1 gets experts 1, 5 ( 4, 2, 8, "round_robin", 2, [-1, -1, 0, -1, -1, -1, 1, -1], ), # rank 2 gets experts 2, 6 ( 4, 3, 8, "round_robin", 2, [-1, -1, -1, 0, -1, -1, -1, 1], ), # rank 3 gets experts 3, 7 ] for ( ep_size, ep_rank, global_num_experts, expert_placement_strategy, expected_local, expected_map_pattern, ) in test_cases: local_num_experts, expert_map, _ = determine_expert_map( ep_size=ep_size, ep_rank=ep_rank, global_num_experts=global_num_experts, expert_placement_strategy=expert_placement_strategy, ) assert local_num_experts == expected_local, ( f"ep_size={ep_size}, ep_rank={ep_rank}, " f"global_num_experts={global_num_experts}, " f"expert_placement_strategy={expert_placement_strategy}: " f"expected {expected_local} local experts, got {local_num_experts}" ) if expected_map_pattern is None: assert expert_map is None, "Expected expert_map to be None" else: assert expert_map is not None, "Expected expert_map to not be None" actual_map = expert_map.tolist() assert actual_map == expected_map_pattern, ( f"ep_size={ep_size}, ep_rank={ep_rank}, " f"global_num_experts={global_num_experts}, " f"expert_placement_strategy={expert_placement_strategy}: " f"expected map {expected_map_pattern}, got {actual_map}" )
{ "repo_id": "vllm-project/vllm", "file_path": "tests/distributed/test_expert_placement.py", "license": "Apache License 2.0", "lines": 217, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/kernels/quantization/test_hadacore.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import math import pytest import torch from compressed_tensors.transform import deterministic_hadamard_matrix from vllm import _custom_ops as ops from vllm.platforms import current_platform if current_platform.is_rocm(): pytest.skip( "These tests require hadacore_transform, not supported on ROCm.", allow_module_level=True, ) @pytest.mark.parametrize("batch_size", [1, 32]) @pytest.mark.parametrize("hidden_dim", [2**n for n in range(10)]) def test_hadacore(batch_size, hidden_dim, dtype=torch.bfloat16, device="cuda"): x = torch.eye(hidden_dim, dtype=dtype, device=device) hadamard = deterministic_hadamard_matrix( hidden_dim, dtype=torch.float64, device="cuda" ) / math.sqrt(hidden_dim) y = ops.hadacore_transform(x.clone()) y_true = (x.to(hadamard.dtype) @ hadamard.T).to(y.dtype) assert torch.allclose(y, y_true) y = ops.hadacore_transform(y) assert torch.allclose(y, x)
{ "repo_id": "vllm-project/vllm", "file_path": "tests/kernels/quantization/test_hadacore.py", "license": "Apache License 2.0", "lines": 25, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/config/multimodal.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from collections.abc import Mapping from typing import Any, Literal, TypeAlias, TypedDict, final from pydantic import ConfigDict, Field, field_validator, model_validator from pydantic.dataclasses import dataclass from vllm.config.utils import config from vllm.utils.hashing import safe_hash from vllm.v1.attention.backends.registry import AttentionBackendEnum @dataclass class BaseDummyOptions: """Base options for generating dummy data during profiling.""" count: int = Field(999, ge=0) @dataclass(config=ConfigDict(extra="forbid")) class VideoDummyOptions(BaseDummyOptions): """Options for generating dummy video data during profiling.""" num_frames: int | None = Field(None, gt=0) width: int | None = Field(None, gt=0) height: int | None = Field(None, gt=0) @dataclass(config=ConfigDict(extra="forbid")) class ImageDummyOptions(BaseDummyOptions): """Options for generating dummy image data during profiling.""" width: int | None = Field(None, gt=0) height: int | None = Field(None, gt=0) @dataclass(config=ConfigDict(extra="forbid")) class AudioDummyOptions(BaseDummyOptions): """Options for generating dummy audio data during profiling.""" length: int | None = Field(None, gt=0) @final class MultiModalDummyOptionsBuiltins(TypedDict, total=False): """Type annotations for modality types predefined by vLLM.""" image: ImageDummyOptions """Options for dummy images.""" video: VideoDummyOptions """Options for dummy videos.""" audio: AudioDummyOptions """Options for dummy audios.""" MMEncoderTPMode = Literal["weights", "data"] MMCacheType = Literal["shm", "lru"] MMDummyOptions: TypeAlias = dict[str, BaseDummyOptions] """ A dictionary containing an entry for each modality type of dummy data. The built-in modalities are defined by [`MultiModalDummyOptionsBuiltins`][vllm.config.multimodal.MultiModalDummyOptionsBuiltins]. """ @config class MultiModalConfig: """Controls the behavior of multimodal models.""" language_model_only: bool = False """If True, disables all multimodal inputs by setting all modality limits to 0. Equivalent to setting `--limit-mm-per-prompt` to 0 for every modality.""" limit_per_prompt: MMDummyOptions = Field(default_factory=dict) """The maximum number of input items and options allowed per prompt for each modality. Defaults to 999 for each modality. Legacy format (count only): {"image": 16, "video": 2} Configurable format (with options): {"video": {"count": 1, "num_frames": 32, "width": 512, "height": 512}, "image": {"count": 5, "width": 512, "height": 512}} Mixed format (combining both): {"image": 16, "video": {"count": 1, "num_frames": 32, "width": 512, "height": 512}} """ enable_mm_embeds: bool = False """If `True`, enables passing multimodal embeddings: for `LLM` class, this refers to tensor inputs under `multi_modal_data`; for the OpenAI-compatible server, this refers to chat messages with content `"type": "*_embeds"`. When enabled with `--limit-mm-per-prompt` set to 0 for a modality, precomputed embeddings skip count validation for that modality, saving memory by not loading encoder modules while still enabling embeddings as an input. Limits greater than 0 still apply to embeddings. WARNING: The vLLM engine may crash if incorrect shape of embeddings is passed. Only enable this flag for trusted users!""" media_io_kwargs: dict[str, dict[str, Any]] = Field(default_factory=dict) """Additional args passed to process media inputs, keyed by modalities. For example, to set num_frames for video, set `--media-io-kwargs '{"video": {"num_frames": 40} }'`""" mm_processor_kwargs: dict[str, object] | None = None """Arguments to be forwarded to the model's processor for multi-modal data, e.g., image processor. Overrides for the multi-modal processor obtained from `transformers.AutoProcessor.from_pretrained`. The available overrides depend on the model that is being run. For example, for Phi-3-Vision: `{"num_crops": 4}`.""" mm_processor_cache_gb: float = Field(default=4, ge=0) """The size (in GiB) of the multi-modal processor cache, which is used to avoid re-processing past multi-modal inputs. This cache is duplicated for each API process and engine core process, resulting in a total memory usage of `mm_processor_cache_gb * (api_server_count + data_parallel_size)`. Set to `0` to disable this cache completely (not recommended).""" mm_processor_cache_type: MMCacheType = "lru" """Type of cache to use for the multi-modal preprocessor/mapper. If `shm`, use shared memory FIFO cache. If `lru`, use mirrored LRU cache.""" mm_shm_cache_max_object_size_mb: int = Field(default=128, ge=0) """Size limit (in MiB) for each object stored in the multi-modal processor shared memory cache. Only effective when `mm_processor_cache_type` is `"shm"`.""" mm_encoder_only: bool = False """ When enabled, skips the language component of the model. This is usually only valid in disaggregated Encoder process. """ mm_encoder_tp_mode: MMEncoderTPMode = "weights" """Indicates how to optimize multi-modal encoder inference using tensor parallelism (TP). - `"weights"`: Within the same vLLM engine, split the weights of each layer across TP ranks. (default TP behavior)\n - `"data"`: Within the same vLLM engine, split the batched input data across TP ranks to process the data in parallel, while hosting the full weights on each TP rank. This batch-level DP is not to be confused with API request-level DP (which is controlled by `--data-parallel-size`). This is only supported on a per-model basis and falls back to `"weights"` if the encoder does not support DP.""" mm_encoder_attn_backend: AttentionBackendEnum | None = None """Optional override for the multi-modal encoder attention backend when using vision transformers. Accepts any value from `vllm.v1.attention.backends.registry.AttentionBackendEnum` (e.g. `FLASH_ATTN`).""" interleave_mm_strings: bool = False """Enable fully interleaved support for multimodal prompts, while using --chat-template-content-format=string.""" skip_mm_profiling: bool = False """When enabled, skips multimodal memory profiling and only profiles with language backbone model during engine initialization. This reduces engine startup time but shifts the responsibility to users for estimating the peak memory usage of the activation of multimodal encoder and embedding cache.""" video_pruning_rate: float | None = Field(default=None, ge=0.0, lt=1.0) """Sets pruning rate for video pruning via Efficient Video Sampling. Value sits in range [0;1) and determines fraction of media tokens from each video to be pruned. """ @field_validator("limit_per_prompt", mode="before") @classmethod def _validate_limit_per_prompt( cls, value: dict[str, int | dict[str, int]], ) -> MMDummyOptions: out: MMDummyOptions = {} for k, v in value.items(): # Handle legacy format where only count is specified if isinstance(v, int): v = {"count": v} # Convert to the appropriate DummyOptions subclass if k == "video": out[k] = VideoDummyOptions(**v) elif k == "image": out[k] = ImageDummyOptions(**v) elif k == "audio": out[k] = AudioDummyOptions(**v) else: out[k] = BaseDummyOptions(**v) return out @field_validator("mm_encoder_attn_backend", mode="before") @classmethod def _validate_mm_encoder_attn_backend( cls, value: str | AttentionBackendEnum | None ) -> AttentionBackendEnum | None: if isinstance(value, str) and value.upper() == "XFORMERS": raise ValueError( "Attention backend 'XFORMERS' has been removed (See PR #29262 for " "details). Please select a supported attention backend." ) if value is None or isinstance(value, AttentionBackendEnum): return value assert isinstance(value, str), ( "mm_encoder_attn_backend must be a string or an AttentionBackendEnum." ) return AttentionBackendEnum[value.upper()] @model_validator(mode="after") def _validate_multimodal_config(self): if self.mm_processor_cache_type != "shm" and ( self.mm_shm_cache_max_object_size_mb != MultiModalConfig.mm_shm_cache_max_object_size_mb ): raise ValueError( "'mm_shm_cache_max_object_size_mb' should only be set when " "'mm_processor_cache_type' is 'shm'." ) return self def compute_hash(self) -> str: """ WARNING: Whenever a new field is added to this config, ensure that it is included in the factors list if it affects the computation graph. Provide a hash that uniquely identifies all the configs that affect the structure of the computation graph from input ids/embeddings to the final hidden states, excluding anything before input ids/embeddings and after the final hidden states. """ factors: list[Any] = [ self.mm_encoder_attn_backend.name if self.mm_encoder_attn_backend is not None else None, self.mm_encoder_tp_mode, ] hash_str = safe_hash(str(factors).encode(), usedforsecurity=False).hexdigest() return hash_str def get_limit_per_prompt(self, modality: str) -> int: """ Get the maximum number of input items allowed per prompt for the given modality (backward compatible). """ if self.language_model_only: return 0 limit_data = self.limit_per_prompt.get(modality) if limit_data is None: # Unspecified modality is set to 999 by default return 999 return limit_data.count def merge_mm_processor_kwargs( self, inference_kwargs: Mapping[str, object], ) -> dict[str, object]: """ Get the keyword arguments to pass to the multi-modal processor according to the extra arguments passed during inference. """ kwargs = self.mm_processor_kwargs or {} return kwargs | dict(inference_kwargs) def is_multimodal_pruning_enabled(self): return self.video_pruning_rate is not None and self.video_pruning_rate > 0
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/config/multimodal.py", "license": "Apache License 2.0", "lines": 226, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/transformers_utils/configs/olmo3.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from transformers.configuration_utils import PretrainedConfig class Olmo3Config(PretrainedConfig): model_type = "olmo3" keys_to_ignore_at_inference = ["past_key_values"] def __init__( self, vocab_size=50304, hidden_size=4096, intermediate_size=11008, num_hidden_layers=32, num_attention_heads=32, num_key_value_heads=None, hidden_act="silu", max_position_embeddings=2048, initializer_range=0.02, use_cache=True, pad_token_id=1, bos_token_id=None, eos_token_id=50279, tie_word_embeddings=False, rope_parameters=None, attention_bias=False, attention_dropout=0.0, rms_norm_eps=1e-5, sliding_window=4096, layer_types=None, **kwargs, ): # This model uses Olmo3ForCausalLM in transformers but Olmo2ForCausalLM # in vLLM. if "architectures" not in kwargs: kwargs["architectures"] = ["Olmo2ForCausalLM"] elif "Olmo3ForCausalLM" in kwargs["architectures"]: kwargs["architectures"].remove("Olmo3ForCausalLM") kwargs["architectures"].append("Olmo2ForCausalLM") super().__init__( pad_token_id=pad_token_id, bos_token_id=bos_token_id, eos_token_id=eos_token_id, tie_word_embeddings=tie_word_embeddings, **kwargs, ) self.vocab_size = vocab_size self.max_position_embeddings = max_position_embeddings self.hidden_size = hidden_size self.intermediate_size = intermediate_size self.num_hidden_layers = num_hidden_layers self.num_attention_heads = num_attention_heads # for backward compatibility if num_key_value_heads is None: num_key_value_heads = num_attention_heads self.num_key_value_heads = num_key_value_heads self.hidden_act = hidden_act self.initializer_range = initializer_range self.use_cache = use_cache # Try to set `rope_scaling` if available, otherwise use `rope_parameters` rope_scaling = kwargs.pop("rope_scaling", None) rope_parameters = rope_scaling or rope_parameters or {"rope_type": "default"} rope_theta = kwargs.pop("rope_theta", 10000.0) if "rope_theta" not in rope_parameters: rope_parameters["rope_theta"] = rope_theta self.rope_parameters = rope_parameters self.attention_bias = attention_bias self.attention_dropout = attention_dropout self.rms_norm_eps = rms_norm_eps self.sliding_window = sliding_window self.layer_types = layer_types if self.layer_types is None: self.layer_types = [ "sliding_attention" if (i + 1) % 4 != 0 else "full_attention" for i in range(self.num_hidden_layers) ]
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/transformers_utils/configs/olmo3.py", "license": "Apache License 2.0", "lines": 74, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/distributed/test_shm_buffer.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import traceback import unittest import numpy as np from vllm.distributed.device_communicators.shm_object_storage import ( SingleWriterShmRingBuffer, ) class TestSingleWriterShmRingBuffer(unittest.TestCase): """Test suite for the ring buffer implementation""" def setUp(self): """Set up test fixtures""" self.buffer_size = 4096 self.ring_buffer = None def tearDown(self): """Clean up after tests""" if self.ring_buffer: self.ring_buffer.close() def test_buffer_opening(self): """Test opening an existing buffer""" # First create a buffer self.ring_buffer = SingleWriterShmRingBuffer( data_buffer_size=self.buffer_size, create=True ) # Then open it with another instance reader_buffer = SingleWriterShmRingBuffer(*self.ring_buffer.handle()) self.assertFalse(reader_buffer.is_writer) self.assertEqual( reader_buffer.shared_memory.name, self.ring_buffer.shared_memory.name ) def test_buffer_access(self): """Test accessing allocated buffers""" self.ring_buffer = SingleWriterShmRingBuffer( data_buffer_size=self.buffer_size, create=True ) size = 100 address, monotonic_id = self.ring_buffer.allocate_buf(size) # Write some test data test_data = b"Hello, World!" * 7 # 91 bytes with self.ring_buffer.access_buf(address) as (data_buf, metadata): data_buf[0 : len(test_data)] = test_data # Read it back with self.ring_buffer.access_buf(address) as (data_buf2, metadata2): read_data = bytes(data_buf2[0 : len(test_data)]) read_id = metadata2[0] self.assertEqual(read_data, test_data) self.assertEqual(read_id, monotonic_id) def test_memory_error_on_full_buffer(self): """Test that MemoryError is raised when buffer is full""" small_buffer_size = 200 self.ring_buffer = SingleWriterShmRingBuffer( data_buffer_size=small_buffer_size, create=True ) # Fill up the buffer self.ring_buffer.allocate_buf(100) self.ring_buffer.allocate_buf(80) # Total: 196 bytes used # This should fail with self.assertRaises(MemoryError): self.ring_buffer.allocate_buf(1) # Would exceed buffer capacity def test_allocation_and_free(self): """Test allocation and freeing of buffers""" small_buffer_size = 200 self.ring_buffer = SingleWriterShmRingBuffer( data_buffer_size=small_buffer_size, create=True ) size = 80 # Write some data test_data = b"Repeated test data" for i in range(5): address, monotonic_id = self.ring_buffer.allocate_buf(size) with self.ring_buffer.access_buf(address) as (data_buf, metadata): data_buf[0:4] = (0).to_bytes(4, "little") # 0 for not in-use data_buf[4 : len(test_data) + 4] = test_data print(self.ring_buffer.metadata) freed_ids = self.ring_buffer.free_buf(lambda *args: True) print(f" Freed IDs: {freed_ids}") self.assertEqual(freed_ids[0], i) def test_clear_buffer(self): """Test clearing the buffer""" self.ring_buffer = SingleWriterShmRingBuffer( data_buffer_size=self.buffer_size, create=True ) # Allocate some buffers for _ in range(3): self.ring_buffer.allocate_buf(100) # Clear the buffer self.ring_buffer.clear() # Check that metadata is empty and IDs reset self.assertEqual(len(self.ring_buffer.metadata), 0) self.assertEqual(self.ring_buffer.monotonic_id_start, 0) self.assertEqual(self.ring_buffer.monotonic_id_end, 0) self.assertEqual(self.ring_buffer.data_buffer_start, 0) self.assertEqual(self.ring_buffer.data_buffer_end, 0) def test_allocation_cycles(self): buffer_size = 100 ring = SingleWriterShmRingBuffer(data_buffer_size=buffer_size, create=True) # tracking allocations for assertions allocated_bitmap = np.zeros( (buffer_size,), dtype=np.bool_ ) # addr -> is_allocated allocation_map = dict() # monotonic_id -> (addr, size) def count_allocated(bitmap) -> int: return np.sum(bitmap).item() def is_free_fn(a, b) -> bool: return True def mark_allocated_with_assertion(id, addr, size): addr = addr % buffer_size self.assertEqual(count_allocated(allocated_bitmap[addr : addr + size]), 0) allocated_bitmap[addr : addr + size] = True allocation_map[id] = (addr, size) def mark_freed_with_assertion(id): self.assertTrue(id in allocation_map) addr, size = allocation_map.pop(id) addr = addr % buffer_size self.assertEqual( count_allocated(allocated_bitmap[addr : addr + size]), size ) allocated_bitmap[addr : addr + size] = False def ring_free(free_size=None): freed_ids = ring.free_buf(is_free_fn, free_size) for freed_id in freed_ids: mark_freed_with_assertion(freed_id) def ring_allocate(allocate_size): allocate_size_with_md = allocate_size + ring.MD_SIZE try: addr, monotonic_id = ring.allocate_buf(allocate_size) mark_allocated_with_assertion(monotonic_id, addr, allocate_size_with_md) except MemoryError: # free 2x size for enough space if wrapping happened ring_free(allocate_size_with_md * 2) # retry allocating addr, monotonic_id = ring.allocate_buf(allocate_size) mark_allocated_with_assertion(monotonic_id, addr, allocate_size_with_md) # 1. allocation & free cycles for _ in range(33): # will consume 2 + 8 = 10 bytes per allocation ring_allocate(2) # 2. free all allocations ring_free() # 3. try allocate the largest possible buffer ring_allocate(buffer_size - ring.MD_SIZE) def main(): """Main function demonstrating usage and running tests""" print("=== SingleWriterShmRingBuffer Test Suite ===\n") # Run unit tests print("Running unit tests...") unittest.main(argv=[""], exit=False, verbosity=2) print("\n" + "=" * 50) print("=== Manual Demo ===\n") # Manual demonstration try: print("Creating ring buffer...") writer_buffer = SingleWriterShmRingBuffer(data_buffer_size=2048, create=True) reader_buffer = SingleWriterShmRingBuffer(*writer_buffer.handle()) print(f"Buffer created with name: {writer_buffer.shared_memory.name}") # Allocate some buffers print("\nAllocating buffers...") address_array = [] for i in range(3): size = 100 + i * 50 try: writer_buffer.free_buf(lambda *args: True) address, monotonic_id = writer_buffer.allocate_buf(size) address_array.append((address, size, monotonic_id)) # Write some test data with writer_buffer.access_buf(address) as (data_buf, metadata): test_message = f"Test message {i}".encode() data_buf[0 : len(test_message)] = test_message except MemoryError as e: print(f" Failed to allocate {size} bytes: {e}") print("\nBuffer state:") print(f" Data buffer start: {writer_buffer.data_buffer_start}") print(f" Data buffer end: {writer_buffer.data_buffer_end}") print(f" Monotonic ID start: {writer_buffer.monotonic_id_start}") print(f" Monotonic ID end: {writer_buffer.monotonic_id_end}") print(f" Metadata entries: {len(writer_buffer.metadata)}") # Try to read back the data print("\nReading back data...") for address, size, monotonic_id in address_array: with reader_buffer.access_buf(address) as (data_buf, metadata): # Find null terminator or read first 50 chars data_bytes = bytes(data_buf[0:size]) message = data_bytes.decode() print(f" ID {monotonic_id}: '{message}'") except Exception as e: print(f"Demo error: {e}") traceback.print_exc() print("\n=== Demo Complete ===") if __name__ == "__main__": main()
{ "repo_id": "vllm-project/vllm", "file_path": "tests/distributed/test_shm_buffer.py", "license": "Apache License 2.0", "lines": 190, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/distributed/test_shm_storage.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import multiprocessing import random import time import traceback import unittest from multiprocessing import Lock import torch # Assuming these are imported from your module from vllm.distributed.device_communicators.shm_object_storage import ( MsgpackSerde, SingleWriterShmObjectStorage, SingleWriterShmRingBuffer, ) from vllm.multimodal.inputs import ( MultiModalFieldElem, MultiModalKwargsItem, MultiModalSharedField, ) def _dummy_elem(size: int): return MultiModalFieldElem( data=torch.empty((size,), dtype=torch.int8), field=MultiModalSharedField(batch_size=1), ) def _dummy_item(size_by_key: dict[str, int]): return MultiModalKwargsItem( {key: _dummy_elem(size) for key, size in size_by_key.items()} ) class TestSingleWriterShmObjectStorage(unittest.TestCase): def setUp(self): """Set up test fixtures before each test method.""" ring_buffer = SingleWriterShmRingBuffer( data_buffer_size=1024 * 100, create=True, # 10 MB buffer ) self.storage = SingleWriterShmObjectStorage( max_object_size=1024 * 10, # 10KB max object n_readers=2, ring_buffer=ring_buffer, serde_class=MsgpackSerde, reader_lock=Lock(), ) def tearDown(self): """Clean up after each test.""" if self.storage: self.storage.close() def test_minimal_put_get_cycle(self): """Test basic put and get operations.""" key = "test_key" value = _dummy_item({"field1": 10, "field2": 20}) # Put operation address, monotonic_id = self.storage.put(key, value) # Verify key is in index self.assertIn(key, self.storage.key_index) self.assertEqual(self.storage.key_index[key], (address, monotonic_id)) self.assertEqual(self.storage.id_index[monotonic_id], key) # Get operation result = self.storage.get(address, monotonic_id) # Verify result self.assertEqual(result, value) def test_put_same_key_twice(self): """Test behavior when putting the same key multiple times.""" key = "duplicate_key" value1 = "first value" value2 = "second value" # First put address1, id1 = self.storage.put(key, value1) retrieved1 = self.storage.get(address1, id1) self.assertEqual(retrieved1, value1) # should raise an error on second put with self.assertRaises(ValueError) as context: self.storage.put(key, value2) self.assertIn("already exists in the storage", str(context.exception)) def test_large_object_rejection(self): """Test that objects exceeding max_object_size are rejected.""" # Create an object larger than max_object_size large_data = "x" * (self.storage.max_object_size + 100) with self.assertRaises(ValueError) as context: self.storage.put("large_key", large_data) self.assertIn("exceeds max object size", str(context.exception)) def test_buffer_overflow_and_cleanup(self): """Test behavior when buffer fills up and needs cleanup.""" # Fill up the buffer with many small objects stored_items = [] try: for i in range(1000): # Try to store many items key = f"item_{i}" value = f"data_{i}" * 100 # Make it reasonably sized address, monotonic_id = self.storage.put(key, value) stored_items.append((key, value, address, monotonic_id)) except MemoryError: print(f"Buffer filled after {len(stored_items)} items") # Verify that some items are still accessible accessible_count = 0 for key, original_value, address, monotonic_id in stored_items: for i in range(self.storage.n_readers): retrieved = self.storage.get(address, monotonic_id) if retrieved == original_value: accessible_count += 1 self.assertEqual(accessible_count, len(stored_items)) try: for i in range(len(stored_items), 1000): # Try to store many items key = f"item_{i}" value = f"data_{i}" * 100 # Make it reasonably sized address, monotonic_id = self.storage.put(key, value) stored_items.append((key, value, address, monotonic_id)) except MemoryError: print(f"Buffer filled after {len(stored_items)} items") # Verify that some items are still accessibles for key, original_value, address, monotonic_id in stored_items: try: for i in range(self.storage.n_readers): retrieved = self.storage.get(address, monotonic_id) if retrieved == original_value: accessible_count += 1 except ValueError as e: print(f"Error retrieving {key}: {e}") # some items from the first batch may still be accessible self.assertGreaterEqual(accessible_count, len(stored_items)) def test_blocking_unread_object(self): """Test behavior when buffer fills up and needs cleanup.""" # Fill up the buffer with many small objects stored_items = [] try: for i in range(1000): # Try to store many items key = f"item_{i}" value = f"data_{i}" * 100 # Make it reasonably sized address, monotonic_id = self.storage.put(key, value) stored_items.append((key, value, address, monotonic_id)) except MemoryError: print(f"Buffer filled after {len(stored_items)} items") # read all items except the first one # to simulate a blocking situation accessible_count = 0 for key, original_value, address, monotonic_id in stored_items[1:]: for i in range(self.storage.n_readers): retrieved = self.storage.get(address, monotonic_id) if retrieved == original_value: accessible_count += 1 self.assertEqual(accessible_count, len(stored_items) - 1) try: key = f"item_{len(stored_items)}" value = f"data_{len(stored_items)}" * 100 address, monotonic_id = self.storage.put(key, value) except MemoryError: print(f"Buffer filled after {len(stored_items)} items") # read the first item for i in range(self.storage.n_readers): key, original_value, address, monotonic_id = stored_items[0] retrieved = self.storage.get(address, monotonic_id) self.assertEqual(retrieved, original_value) try: for i in range(len(stored_items), 1000): # Try to store many items key = f"item_{i}" value = f"data_{i}" * 100 # Make it reasonably sized address, monotonic_id = self.storage.put(key, value) stored_items.append((key, value, address, monotonic_id)) except MemoryError: print(f"Buffer filled after {len(stored_items)} items") # some items from the first batch may still be accessible self.assertGreaterEqual(len(stored_items), accessible_count + 10) def test_invalid_get_operations(self): """Test various invalid get operations.""" # Test with non-existent address with self.assertRaises(ValueError): # Could be various exceptions self.storage.get(99999, 1) # Store something first address, monotonic_id = self.storage.put("test", "value") # Test with wrong monotonic_id with self.assertRaises(ValueError) as context: self.storage.get(address, monotonic_id + 100) self.assertIn("has been modified or is invalid", str(context.exception)) def test_clear_storage(self): """Test clearing the storage.""" # Store some items for i in range(5): self.storage.put(f"item_{i}", f"value_{i}") # Clear the storage self.storage.clear() # Verify that all indices are empty self.assertEqual(len(self.storage.key_index), 0) self.assertEqual(len(self.storage.id_index), 0) self.assertEqual(len(self.storage.ring_buffer.metadata), 0) # Verify that new items can be added after clearing address, monotonic_id = self.storage.put("new_item", "new_value") self.assertIn("new_item", self.storage.key_index) self.assertEqual((address, monotonic_id), (0, 0)) # Reader process function def reader_process(process_id, storage_handle, items_to_read): """Reader process that connects to existing shared memory and reads data.""" reader_storage = SingleWriterShmObjectStorage.create_from_handle(storage_handle) print(f"Reader {process_id} started") errors = [] for key, original_value, address, monotonic_id in items_to_read: time.sleep(random.random() / 100) try: # Read data from shared memory retrieved_value = reader_storage.get(address, monotonic_id) # Verify data integrity assert retrieved_value == original_value print(f"Reader {process_id} retrieved {key}: {retrieved_value}") except Exception as e: errors.append((key, str(e), type(e).__name__)) def run_multiprocess_example(): """Run a minimal working example with real shared memory.""" print("=== Minimal Object Storage Example ===") try: # Create storage instance ring_buffer = SingleWriterShmRingBuffer( data_buffer_size=1024 * 100, create=True, # 10 MB buffer ) storage = SingleWriterShmObjectStorage( max_object_size=1024, n_readers=3, ring_buffer=ring_buffer, serde_class=MsgpackSerde, reader_lock=Lock(), ) print(f"Created storage (writer: {storage.is_writer})") # Test basic data types test_data = [ ("user_data", {"name": "Alice", "age": 30, "scores": [95, 87, 92]}), ("simple_string", "Hello, World!"), ("number", 42), ("list_data", [1, 2, 3, "four", 5.0]), ] stored_items = [] # Store all data for key, value in test_data: print(f"Storing {key}: {value}") address, monotonic_id = storage.put(key, value) stored_items.append((key, value, address, monotonic_id)) print(f" -> Stored at address {address}, ID {monotonic_id}") print("\n--- Retrieving Data ---") processes = [] handle = storage.handle() # initialize lock for reader processes handle.reader_lock = Lock() for i in range(storage.n_readers): p = multiprocessing.Process( target=reader_process, args=(i, handle, stored_items) ) processes.append(p) p.start() for p in processes: p.join(timeout=10) if p.is_alive(): p.terminate() p.join() except Exception as e: print(f"Error in minimal example: {e}") traceback.print_exc() if __name__ == "__main__": # Run the minimal example first run_multiprocess_example() print("\n" + "=" * 50 + "\n") # Run the test suite print("Running comprehensive test suite...") unittest.main(verbosity=2, exit=False)
{ "repo_id": "vllm-project/vllm", "file_path": "tests/distributed/test_shm_storage.py", "license": "Apache License 2.0", "lines": 261, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/distributed/device_communicators/shm_object_storage.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import pickle from abc import ABC, abstractmethod from collections.abc import Callable, Iterable from contextlib import contextmanager, suppress from dataclasses import dataclass from itertools import chain from multiprocessing import shared_memory from multiprocessing.synchronize import Lock as LockType from typing import Any from unittest.mock import patch import torch from vllm.logger import init_logger logger = init_logger(__name__) class SingleWriterShmRingBuffer: """ A single-writer, multiple-reader ring buffer implementation using shared memory. This class provides a thread-safe ring buffer where one process can write data while multiple processes/threads can read from it. Architecture: - Uses shared memory for cross-process communication - Maintains metadata for each allocated buffer chunk in the writer process - Supports custom "is_free_fn" functions to determine when buffers can be reused - Each buffer chunk contains: `[4-byte id][4-byte size][actual_data]` Key Concepts: - monotonic_id_start/end: Track the range of active buffer IDs - data_buffer_start/end: Track the physical memory range in use - Automatic wraparound when reaching buffer end - Lazy garbage collection based on is_free_fn checks Example Usage Scenarios: Scenario 1: Simple Linear Allocation ``` Buffer size: 100 bytes Initial state: [................................................. ] ^start=end(0) After allocating 20 bytes (id=0): [id:0|size:20|data........][...................................] ^start(0) ^end(28) After allocating 30 bytes (id=1): [id:0|size:20|data........][id:1|size:30|data..............][..] ^start(0) ^end(66) ``` Scenario 2: Memory Reclamation ``` Before freeing (both buffers still in use): [id:0|size:20|data........][id:1|size:30|data..............][..] ^start(0) ^end(66) After id:0 is marked free by readers: [FREED.................... ][id:1|size:30|data..............][..] ^start(28) ^end(66) After both are freed: [FREED..............................................][..] ^start=end(66) ``` Scenario 3: Wraparound Allocation (continuing from Scenario 2) ``` Starting from after memory reclamation in Scenario 2: [FREED..............................................][..] ^start=end(66) Allocate 40 bytes (id=2) - only 34 bytes available at end, so wraparound: [id:2|size:40|data........................][FREED.............][..] ^end(148) ^start(66) ``` Scenario 4: Error Handling - Out of Space ``` Starting from after wraparound allocation in Scenario 3: [id:2|size:40|data........................][FREED.............][..] ^end(148) ^start(66) Trying to allocate 20 more bytes: occupied_size_new = end + size - start = 148 + 28 - 66 > buffer_size(100) -> Raises MemoryError: "Not enough space in the data buffer" ``` Thread Safety: - Single writer: Only one process/thread should write (allocate_buf) - Multiple readers: Multiple processes/threads can read (access_buf) - Reader synchronization handled by is_free_fn callback - Writer handles garbage collection (free_buf) based on reader feedback Memory Layout per Buffer Chunk: `[4-byte monotonic_id][4-byte chunk_size][actual_data...]` ^metadata_start ^data_start The monotonic_id ensures data integrity - readers can verify they're accessing the correct data even after buffer wraparound or reuse. """ def __init__( self, data_buffer_size: int, name: str | None = None, create: bool = False, ): self.data_buffer_size = data_buffer_size self.is_writer = create self.ID_NBYTES = 4 self.ID_MAX = 2**31 # exclusive, so 2**31 - 1 is the max value self.SIZE_NBYTES = 4 # 4 bytes for id, 4 bytes for buffer size self.MD_SIZE = self.ID_NBYTES + self.SIZE_NBYTES self.monotonic_id_end = 0 self.monotonic_id_start = 0 self.data_buffer_start = 0 self.data_buffer_end = 0 if create: logger.debug("Creating new shared memory buffer: %s", name) # we are creating a buffer self.metadata: dict[int, int] = {} # monotonic_id -> start address self.shared_memory = shared_memory.SharedMemory( create=True, size=self.data_buffer_size, name=name ) else: # we are opening an existing buffer # fix to https://stackoverflow.com/q/62748654/9191338 # Python incorrectly tracks shared memory even if it is not # created by the process. The following patch is a workaround. with patch( "multiprocessing.resource_tracker.register", lambda *args, **kwargs: None, ): self.shared_memory = shared_memory.SharedMemory(name=name) # See https://docs.python.org/3/library/multiprocessing.shared_memory.html # noqa # Some platforms allocate memory based on page size, # so the shared memory block size may be larger or equal # to the requested size. The size parameter is ignored # when attaching to an existing block. assert self.shared_memory.size >= self.data_buffer_size logger.debug( "Shared memory created/opened with name: %s, size: %d", self.shared_memory.name, self.data_buffer_size, ) def handle(self): return ( self.data_buffer_size, self.shared_memory.name, ) def clear(self) -> None: """Clear the ring buffer.""" assert self.is_writer, "Only the writer can clear the buffer." self.metadata.clear() self.monotonic_id_end = 0 self.monotonic_id_start = 0 self.data_buffer_start = 0 self.data_buffer_end = 0 def close(self) -> None: """Close the shared memory.""" if hasattr(self, "shared_memory"): self.shared_memory.close() if self.is_writer: with suppress(FileNotFoundError): self.shared_memory.unlink() def __del__(self): self.close() def int2byte(self, integer: int) -> bytes: """Convert an integer to bytes.""" return integer.to_bytes(self.ID_NBYTES, "little", signed=True) def byte2int(self, byte_data: bytes) -> int: """Convert bytes back to an integer.""" return int.from_bytes(byte_data, "little", signed=True) def allocate_buf(self, size: int) -> tuple[int, int]: """ Allocate a buffer `MD_SIZE` + `size` bytes in the shared memory. Memory layout: `[4-byte monotonic_id][4-byte size][buffer data...]` """ assert self.is_writer, "Only the writer can allocate buffers." assert size > 0, "Size must be greater than 0" size += self.MD_SIZE # add metadata size to the buffer size # reset to beginning if the buffer does have enough contiguous space buffer_end_reset = self.data_buffer_end % self.data_buffer_size if buffer_end_reset + size > self.data_buffer_size: buffer_end_reset = ( self.data_buffer_end // self.data_buffer_size + 1 ) * self.data_buffer_size else: # no reset needed buffer_end_reset = self.data_buffer_end # check if we have enough space in the data buffer # i.e. if the new end (self.data_buffer_end + size) # exceeds the start of the data buffer occupied_size_new = buffer_end_reset + size - self.data_buffer_start if occupied_size_new > self.data_buffer_size: raise MemoryError( "Not enough space in the data buffer, " "try calling free_buf() to free up space" ) self.data_buffer_end = buffer_end_reset # first 4 bytes as the monotonic id buf_idx = self.data_buffer_end % self.data_buffer_size self.shared_memory.buf[buf_idx : buf_idx + self.ID_NBYTES] = self.int2byte( self.monotonic_id_end ) # next 4 bytes as the size of the data buffer self.shared_memory.buf[buf_idx + self.ID_NBYTES : buf_idx + self.MD_SIZE] = ( self.int2byte(size) ) # record metadata self.metadata[self.monotonic_id_end % self.ID_MAX] = self.data_buffer_end # update buffer and monotonic id indices current_buffer_end = self.data_buffer_end current_id_end = self.monotonic_id_end self.data_buffer_end += size self.monotonic_id_end = (self.monotonic_id_end + 1) % self.ID_MAX return current_buffer_end, current_id_end @contextmanager def access_buf(self, address: int): buf_idx = address % self.data_buffer_size # read metadata metadata_buff = self.shared_memory.buf[buf_idx : buf_idx + self.MD_SIZE] id = self.byte2int(metadata_buff[: self.ID_NBYTES]) size = self.byte2int(metadata_buff[self.ID_NBYTES : self.MD_SIZE]) # yield the data buffer and metadata data_buff = self.shared_memory.buf[buf_idx + self.MD_SIZE : buf_idx + size] with ( memoryview(data_buff) as data_view, ): yield data_view, (id, size) def free_buf( self, is_free_fn: Callable[[int, memoryview], bool], nbytes: int | None = None, ) -> Iterable[int]: """ Free a buffer of the given size. This is a no-op in shared memory, but we need to keep track of the metadata. If freed memory spreads across the end and start of the ring buffer, the actual freed memory will be in two segments. In this case there still might not be a contiguous space of `nbytes` available. Args: nbytes (int, optional): The size of the buffer to free. If None, frees the maximum size of the ring buffer. """ assert self.is_writer, "Only the writer can free buffers." logger.debug( "Freeing up space in the ring buffer, " "monotonic_id_start: %d, monotonic_id_end: %d", self.monotonic_id_start, self.monotonic_id_end, ) monotonic_id_before = self.monotonic_id_start # if nbytes is None, free up the maximum size of the ring buffer if nbytes is None: nbytes = self.data_buffer_size freed_bytes = 0 while self.monotonic_id_start in self.metadata and freed_bytes < nbytes: address = self.metadata[self.monotonic_id_start] with self.access_buf(address) as (data_buff, metadata): if is_free_fn(self.monotonic_id_start, data_buff): # check passed, we can free the buffer del self.metadata[self.monotonic_id_start] self.monotonic_id_start = ( self.monotonic_id_start + 1 ) % self.ID_MAX if self.monotonic_id_start in self.metadata: # pointing to the start addr of next allocation self.data_buffer_start += ( self.metadata[self.monotonic_id_start] - self.data_buffer_start ) % self.data_buffer_size else: # no remaining allocation, reset to zero self.data_buffer_start = self.data_buffer_end = 0 freed_bytes += metadata[1] else: # there are still readers, we cannot free the buffer break logger.debug( "Freed %d bytes from the ring buffer, " "monotonic_id_start: %d, monotonic_id_end: %d", freed_bytes, self.monotonic_id_start, self.monotonic_id_end, ) # buffer wrap around if self.data_buffer_start >= self.data_buffer_size: self.data_buffer_start -= self.data_buffer_size self.data_buffer_end -= self.data_buffer_size monotonic_id_after = self.monotonic_id_start # id wrap around if monotonic_id_after >= monotonic_id_before: return range(monotonic_id_before, monotonic_id_after) else: return chain( range(monotonic_id_before, self.ID_MAX), range(0, monotonic_id_after) ) class ObjectSerde(ABC): @abstractmethod def serialize(self, value: Any) -> tuple[Any, int, bytes, int]: """Serialize an object to bytes.""" raise NotImplementedError @abstractmethod def deserialize(self, data: memoryview) -> Any: """Deserialize bytes back to an object.""" raise NotImplementedError class MsgpackSerde(ObjectSerde): def __init__(self): # Delayed import to avoid circular dependency from vllm.multimodal.inputs import MultiModalKwargsItem from vllm.v1.serial_utils import MsgpackDecoder, MsgpackEncoder self.encoder = MsgpackEncoder() self.tensor_decoder = MsgpackDecoder(torch.Tensor, share_mem=False) self.mm_decoder = MsgpackDecoder(MultiModalKwargsItem, share_mem=False) self._mm_kwargs_item_cls = MultiModalKwargsItem def serialize(self, value: Any) -> tuple[bytes | list[bytes], int, bytes, int]: len_arr = None if isinstance(value, (torch.Tensor, self._mm_kwargs_item_cls)): type_name = type(value).__name__ value = self.encoder.encode(value) len_arr = [len(s) for s in value] nbytes = sum(len_arr) else: value = pickle.dumps(value, protocol=pickle.HIGHEST_PROTOCOL) type_name = type(value).__name__ nbytes = len(value) object_metadata = (type_name, nbytes, len_arr) serialized_metadata = pickle.dumps( object_metadata, protocol=pickle.HIGHEST_PROTOCOL ) return value, nbytes, serialized_metadata, len(serialized_metadata) def deserialize(self, data_view: memoryview) -> Any: # pickle.loads do not read past the end of a pickled object # within a large buffer, so we can skip storing the metadata size type_name, nbytes, len_arr = pickle.loads(data_view) serialized_data = data_view[-nbytes:] if type_name == torch.Tensor.__name__: obj = [] start_idx = 0 for length in len_arr: item_bytes = serialized_data[start_idx : start_idx + length] obj.append(item_bytes) start_idx += length obj = self.tensor_decoder.decode(obj) elif type_name == self._mm_kwargs_item_cls.__name__: obj = [] start_idx = 0 for length in len_arr: item_bytes = serialized_data[start_idx : start_idx + length] obj.append(item_bytes) start_idx += length obj = self.mm_decoder.decode(obj) elif type_name == bytes.__name__: obj = pickle.loads(serialized_data) else: raise ValueError(f"Unsupported object type '{type_name}' in metadata") return obj @dataclass class ShmObjectStorageHandle: max_object_size: int n_readers: int ring_buffer_handle: tuple[int, str] serde_class: type[ObjectSerde] reader_lock: LockType | None class SingleWriterShmObjectStorage: """ A single-writer, multiple-reader object storage system built on top of a shared memory ring buffer. Provides key-value storage with automatic memory management and cross-process serialization support. This storage system follows a FIFO (First-In-First-Out) eviction policy where the oldest objects are automatically freed when memory runs low. Memory is reclaimed based on reader reference counting - objects are only freed when all readers have finished accessing them. Architecture: - Single writer process can put(key, value) objects - Multiple reader processes can get(address, monotonic_id) objects - Built on SingleWriterShmRingBuffer for efficient shared memory management - Thread-safe operations with reader synchronization via locks Key Features: - FIFO Eviction: Oldest objects are evicted first when memory is full - Reference Counting: Objects are only freed when no readers are accessing them - Duplicate Key Handling: Existing keys are not overwritten, just re-referenced - Customized Serialization: By default uses Msgpack for efficient serialization of Python objects, but can be extended for custom types - Cross-Process Safety: Uses shared memory with proper synchronization - Automatic Cleanup: Garbage collection happens transparently during allocation Memory Layout per Object: `[4-byte reference_count][metadata_size][serialized_object_data]` Thread Safety: - Writer operations (put, clear) are single-threaded by design - Reader operations (get) are thread-safe with lock-based reference counting - Memory reclamation is handled exclusively by the writer process """ def __init__( self, max_object_size: int, n_readers: int, ring_buffer: SingleWriterShmRingBuffer, serde_class: type[ObjectSerde] = MsgpackSerde, reader_lock: LockType | None = None, ): """ Initialize the object storage. Args: max_object_size: Maximum size for a single object in bytes. n_readers: Number of reader processes that can access the storage. ring_buffer: The shared memory ring buffer for storing objects. serde_class: Serializer/deserializer for objects. reader_lock: Optional lock for synchronizing reader access. Raises: ValueError: If reader_lock is None for readers. """ self.max_object_size = max_object_size self.n_readers = n_readers self.serde_class = serde_class self.ser_de = serde_class() self.ring_buffer = ring_buffer self.is_writer = self.ring_buffer.is_writer self.flag_bytes = 4 # for in-use flag if self.is_writer: # Key-value mapping: key -> (address, monotonic_id) self.key_index: dict[str, tuple[int, int]] = {} # Reverse mapping: monotonic_id -> key self.id_index: dict[int, str] = {} # Writer flag to track in-use status: monotonic_id -> count self.writer_flag: dict[int, int] = {} else: if reader_lock is None: raise ValueError("Lock must be provided for readers.") self._reader_lock = reader_lock def clear(self) -> None: """Clear the object storage.""" if self.is_writer: self.ring_buffer.clear() self.key_index.clear() self.id_index.clear() self.writer_flag.clear() logger.debug("Object storage cleared and reinitialized.") def copy_to_buffer( self, data: bytes | list[bytes], data_bytes: int, metadata: bytes, md_bytes: int, data_view: memoryview, ) -> None: data_view[self.flag_bytes : self.flag_bytes + md_bytes] = metadata if isinstance(data, bytes): data_view[-data_bytes:] = data elif isinstance(data, list): start_idx = self.flag_bytes + md_bytes for item_bytes in data: item_size = len(item_bytes) data_view[start_idx : start_idx + item_size] = item_bytes start_idx += item_size else: raise ValueError(f"Unsupported data type for serialization: {type(data)}") def increment_writer_flag(self, id: int) -> None: """Set the in-use flag for the writer.""" self.writer_flag[id] = self.writer_flag.get(id, 0) + 1 def increment_reader_flag(self, data_view: memoryview) -> None: """Set the in-use flag for the reader.""" # >0 for in-use flag reader_count = self.ring_buffer.byte2int(data_view) data_view[:] = self.ring_buffer.int2byte(reader_count + 1) def free_unused(self) -> None: """Free unused buffers in the ring buffer.""" # try to free up 2*max_object_size bytes of space in the ring buffer, # since the buffer might be fragmented freed_ids = self.ring_buffer.free_buf( self.default_is_free_check, 2 * self.max_object_size ) # update the metadata after freeing up space for freed_id in freed_ids: key_to_free = self.id_index[freed_id] del self.key_index[key_to_free] del self.id_index[freed_id] del self.writer_flag[freed_id] def is_cached(self, key: str) -> bool: """ Check if the object with the given key is cached. """ return key in self.key_index def get_cached(self, key: str) -> tuple[int, int]: """ Get the cached object by key if it exists. """ address, monotonic_id = self.key_index[key] self.increment_writer_flag(monotonic_id) return address, monotonic_id def put(self, key: str, value: Any) -> tuple[int, int]: """ Store a key-value pair in the object storage. Attempts to free max_object_size bytes using FIFO order when the ring buffer runs out of space during a put() operation. Args: key: String key to identify the object value: Any serializable Python object Raises: MemoryError: If there's not enough space in the buffer ValueError: If the serialized object is too large ValueError: If the key already exists in the storage """ if key in self.key_index: raise ValueError(f"Key '{key}' already exists in the storage.") object_data, data_bytes, object_metadata, md_bytes = self.ser_de.serialize( value ) buffer_size = self.flag_bytes + data_bytes + md_bytes # Sanity checks if buffer_size > self.max_object_size: raise ValueError( f"Serialized object size ({buffer_size} bytes) exceeds " f"max object size ({self.max_object_size} bytes)" ) # Allocate new buffer try: address, monotonic_id = self.ring_buffer.allocate_buf(buffer_size) except MemoryError: self.free_unused() # try again after freeing up space address, monotonic_id = self.ring_buffer.allocate_buf(buffer_size) # Write data to buffer with self.ring_buffer.access_buf(address) as (data_view, metadata): data_view[: self.flag_bytes] = self.ring_buffer.int2byte(0) self.copy_to_buffer( object_data, data_bytes, object_metadata, md_bytes, data_view ) self.increment_writer_flag(monotonic_id) # Update key index self.key_index[key] = (address, monotonic_id) self.id_index[monotonic_id] = key return address, monotonic_id def get(self, address: int, monotonic_id: int) -> Any: # Read data from buffer with self.ring_buffer.access_buf(address) as (data_view, buf_metadata): # check id from metadata if buf_metadata[0] != monotonic_id: raise ValueError( f"Data for address:id '{address}:{monotonic_id}'" " has been modified or is invalid." ) obj = self.ser_de.deserialize(data_view[self.flag_bytes :]) # decrease the in-use flag for reader reads if self._reader_lock is not None: with self._reader_lock: self.increment_reader_flag(data_view[: self.flag_bytes]) else: # if self._reader_lock is None, it means we are the writer # in this case, we do not need to decrease the reader count assert self.is_writer return obj def touch( self, key: str, address: int = 0, monotonic_id: int = 0, ) -> None: """ Touch an existing cached item to update its eviction status. For writers (ShmObjectStoreSenderCache): Increment writer_flag For readers (ShmObjectStoreReceiverCache): Increment reader_count Args: key: String key of the object to touch address: Address of the object (only for readers) monotonic_id: Monotonic ID of the object (only for readers) """ if self._reader_lock is None: if key not in self.key_index: return None address, monotonic_id = self.key_index[key] # Writer side: increment writer_flag to raise eviction threshold self.increment_writer_flag(monotonic_id) else: with ( self._reader_lock, self.ring_buffer.access_buf(address) as (data_view, _), ): reader_count = self.ring_buffer.byte2int(data_view[: self.flag_bytes]) # NOTE(Long): # Avoid increasing flag on newly added item (sync with sender) # Since when a new item is added # pre-touch has no effect on writer side if reader_count >= self.n_readers: self.increment_reader_flag(data_view[: self.flag_bytes]) def close(self) -> None: """Close the shared memory.""" self.ring_buffer.close() def handle(self): """Get handle for sharing across processes.""" return ShmObjectStorageHandle( max_object_size=self.max_object_size, n_readers=self.n_readers, ring_buffer_handle=self.ring_buffer.handle(), serde_class=self.serde_class, reader_lock=self._reader_lock, ) @staticmethod def create_from_handle( handle: ShmObjectStorageHandle, ) -> "SingleWriterShmObjectStorage": logger.debug("Creating storage from handle: %s", handle) ring_buffer = SingleWriterShmRingBuffer(*handle.ring_buffer_handle) return SingleWriterShmObjectStorage( max_object_size=handle.max_object_size, n_readers=handle.n_readers, ring_buffer=ring_buffer, serde_class=handle.serde_class, reader_lock=handle.reader_lock, ) def default_is_free_check(self, id: int, buf: memoryview) -> bool: """ Default is_free function that checks if the first 4 bytes are zero. This indicates that the buffer is free. """ reader_count = int.from_bytes(buf[0:4], "little", signed=True) writer_count = self.writer_flag[id] return reader_count >= writer_count * self.n_readers
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/distributed/device_communicators/shm_object_storage.py", "license": "Apache License 2.0", "lines": 608, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/ci_envs.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """ These envs only work for a small part of the tests, fix what you need! """ import os from collections.abc import Callable from typing import TYPE_CHECKING, Any from vllm.envs import maybe_convert_bool if TYPE_CHECKING: VLLM_CI_NO_SKIP: bool = False VLLM_CI_DTYPE: str | None = None VLLM_CI_HEAD_DTYPE: str | None = None VLLM_CI_HF_DTYPE: str | None = None environment_variables: dict[str, Callable[[], Any]] = { # A model family has many models with the same architecture. # By default, a model family tests only one model. # Through this flag, all models can be tested. "VLLM_CI_NO_SKIP": lambda: bool(int(os.getenv("VLLM_CI_NO_SKIP", "0"))), # Allow changing the dtype used by vllm in tests "VLLM_CI_DTYPE": lambda: os.getenv("VLLM_CI_DTYPE", None), # Allow changing the head dtype used by vllm in tests "VLLM_CI_HEAD_DTYPE": lambda: os.getenv("VLLM_CI_HEAD_DTYPE", None), # Allow changing the head dtype used by transformers in tests "VLLM_CI_HF_DTYPE": lambda: os.getenv("VLLM_CI_HF_DTYPE", None), # Allow control over whether tests use enforce_eager "VLLM_CI_ENFORCE_EAGER": lambda: maybe_convert_bool( os.getenv("VLLM_CI_ENFORCE_EAGER", None) ), } def __getattr__(name: str): # lazy evaluation of environment variables if name in environment_variables: return environment_variables[name]() raise AttributeError(f"module {__name__!r} has no attribute {name!r}") def __dir__(): return list(environment_variables.keys()) def is_set(name: str): """Check if an environment variable is explicitly set.""" if name in environment_variables: return name in os.environ raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
{ "repo_id": "vllm-project/vllm", "file_path": "tests/ci_envs.py", "license": "Apache License 2.0", "lines": 42, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/models/language/pooling/test_mm_classifier_conversion.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from vllm.config.pooler import PoolerConfig def test_idefics_multimodal( vllm_runner, ) -> None: prompts = [ "Hello, my name is", "The president of the United States is", "The capital of France is", "The future of AI is", ] with vllm_runner( model_name="HuggingFaceM4/Idefics3-8B-Llama3", runner="pooling", convert="classify", load_format="dummy", max_model_len=512, enforce_eager=True, tensor_parallel_size=1, disable_log_stats=True, dtype="bfloat16", ) as vllm_model: llm = vllm_model.get_llm() outputs = llm.classify(prompts) for output in outputs: assert len(output.outputs.probs) == 2 def update_config(config): config.text_config.update( { "architectures": ["Gemma3ForSequenceClassification"], "classifier_from_token": ["A", "B", "C", "D", "E"], "method": "no_post_processing", "id2label": { "A": "Chair", "B": "Couch", "C": "Table", "D": "Bed", "E": "Cupboard", }, } ) return config def test_gemma_multimodal( vllm_runner, ) -> None: messages = [ { "role": "system", "content": """ You are a helpful assistant. You will be given a product description which may also include an image. Classify the following product into one of the categories: A = chair B = couch C = table D = bed E = cupboard You'll answer with exactly one letter (A, B, C, D, or E).""", }, { "role": "user", "content": [ { "type": "image_url", "image_url": { "url": "https://vllm-public-assets.s3.us-west-2.amazonaws.com/multimodal_asset/red_chair.jpg" }, }, {"type": "text", "text": "A fine 19th century piece of furniture."}, ], }, ] with vllm_runner( model_name="google/gemma-3-4b-it", runner="pooling", convert="classify", load_format="auto", hf_overrides=update_config, pooler_config=PoolerConfig(seq_pooling_type="LAST"), max_model_len=512, enforce_eager=True, tensor_parallel_size=1, disable_log_stats=True, dtype="bfloat16", ) as vllm_model: llm = vllm_model.get_llm() prompts = llm._preprocess_chat([messages]) result = llm.classify(prompts) assert result[0].outputs.probs[0] > 0.95 assert all(c < 0.05 for c in result[0].outputs.probs[1:])
{ "repo_id": "vllm-project/vllm", "file_path": "tests/models/language/pooling/test_mm_classifier_conversion.py", "license": "Apache License 2.0", "lines": 91, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/v1/tracing/test_tracing.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project # ruff: noqa # type: ignore import pytest import time from opentelemetry.sdk.environment_variables import OTEL_EXPORTER_OTLP_TRACES_INSECURE from vllm import LLM, SamplingParams from vllm.tracing import SpanAttributes # Import shared fixtures from the tracing conftest from tests.tracing.conftest import ( # noqa: F401 FAKE_TRACE_SERVER_ADDRESS, FakeTraceService, trace_service, ) def test_traces( monkeypatch: pytest.MonkeyPatch, trace_service: FakeTraceService, ): with monkeypatch.context() as m: m.setenv(OTEL_EXPORTER_OTLP_TRACES_INSECURE, "true") sampling_params = SamplingParams( temperature=0.01, top_p=0.1, max_tokens=256, ) model = "facebook/opt-125m" llm = LLM( model=model, otlp_traces_endpoint=FAKE_TRACE_SERVER_ADDRESS, gpu_memory_utilization=0.3, disable_log_stats=False, ) prompts = ["This is a short prompt"] outputs = llm.generate(prompts, sampling_params=sampling_params) print(f"test_traces outputs is : {outputs}") # Wait for the "llm_request" span to be exported. # The BatchSpanProcessor batches spans and exports them periodically, # so we need to wait specifically for the llm_request span to appear. timeout = 15 deadline = time.time() + timeout llm_request_spans = [] while time.time() < deadline: all_spans = trace_service.get_all_spans() llm_request_spans = [s for s in all_spans if s["name"] == "llm_request"] if llm_request_spans: break time.sleep(0.5) assert len(llm_request_spans) == 1, ( f"Expected exactly 1 'llm_request' span, but got {len(llm_request_spans)}. " f"All span names: {[s['name'] for s in all_spans]}" ) attributes = llm_request_spans[0]["attributes"] # assert attributes.get(SpanAttributes.GEN_AI_RESPONSE_MODEL) == model assert attributes.get(SpanAttributes.GEN_AI_REQUEST_ID) == outputs[0].request_id assert ( attributes.get(SpanAttributes.GEN_AI_REQUEST_TEMPERATURE) == sampling_params.temperature ) assert ( attributes.get(SpanAttributes.GEN_AI_REQUEST_TOP_P) == sampling_params.top_p ) assert ( attributes.get(SpanAttributes.GEN_AI_REQUEST_MAX_TOKENS) == sampling_params.max_tokens ) assert attributes.get(SpanAttributes.GEN_AI_REQUEST_N) == sampling_params.n assert attributes.get(SpanAttributes.GEN_AI_USAGE_PROMPT_TOKENS) == len( outputs[0].prompt_token_ids ) completion_tokens = sum(len(o.token_ids) for o in outputs[0].outputs) assert ( attributes.get(SpanAttributes.GEN_AI_USAGE_COMPLETION_TOKENS) == completion_tokens ) assert attributes.get(SpanAttributes.GEN_AI_LATENCY_TIME_IN_QUEUE) > 0 assert attributes.get(SpanAttributes.GEN_AI_LATENCY_TIME_TO_FIRST_TOKEN) > 0 assert attributes.get(SpanAttributes.GEN_AI_LATENCY_E2E) > 0
{ "repo_id": "vllm-project/vllm", "file_path": "tests/v1/tracing/test_tracing.py", "license": "Apache License 2.0", "lines": 78, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:benchmarks/kernels/benchmark_device_communicators.py
#!/usr/bin/env python3 # SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """ Benchmark script for device communicators: CustomAllreduce (oneshot, twoshot), PyNcclCommunicator, and SymmMemCommunicator (multimem, two-shot). for NCCL symmetric memory you need to set the environment variables NCCL_NVLS_ENABLE=1 NCCL_CUMEM_ENABLE=1 VLLM_USE_NCCL_SYMM_MEM=1, otherwise NCCL does not use fast NVLS implementation for all reduce. Usage: torchrun --nproc_per_node=<N> benchmark_device_communicators.py [options] Example: torchrun --nproc_per_node=2 benchmark_device_communicators.py --sequence-lengths 512 1024 2048 --num-warmup 10 --num-trials 100 """ import json import os import time from collections.abc import Callable from contextlib import nullcontext import torch import torch.distributed as dist from torch.distributed import ProcessGroup from vllm.distributed.device_communicators.custom_all_reduce import CustomAllreduce from vllm.distributed.device_communicators.flashinfer_all_reduce import ( FlashInferAllReduce, ) from vllm.distributed.device_communicators.pynccl import ( PyNcclCommunicator, register_nccl_symmetric_ops, ) from vllm.distributed.device_communicators.pynccl_allocator import ( set_graph_pool_id, ) from vllm.distributed.device_communicators.symm_mem import SymmMemCommunicator from vllm.logger import init_logger from vllm.utils.argparse_utils import FlexibleArgumentParser logger = init_logger(__name__) # Default sequence lengths to benchmark DEFAULT_SEQUENCE_LENGTHS = [16, 64, 128, 512, 1024, 2048, 4096, 8192] # Fixed hidden size and dtype for all benchmarks HIDDEN_SIZE = 8192 BENCHMARK_DTYPE = torch.bfloat16 # CUDA graph settings CUDA_GRAPH_CAPTURE_CYCLES = 10 class CommunicatorBenchmark: """Benchmark class for testing device communicators.""" def __init__( self, rank: int, world_size: int, device: torch.device, cpu_group: ProcessGroup, sequence_lengths: list[int], ): self.rank = rank self.world_size = world_size self.device = device self.cpu_group = cpu_group # Calculate max_size_override based on largest sequence length max_seq_len = max(sequence_lengths) max_tensor_elements = max_seq_len * HIDDEN_SIZE self.max_size_override = max_tensor_elements * BENCHMARK_DTYPE.itemsize + 1 # Initialize communicators self.custom_allreduce = None self.pynccl_comm = None self.symm_mem_comm = None self.symm_mem_comm_multimem = None self.symm_mem_comm_two_shot = None self.fi_ar_comm = None self._init_communicators() def _init_communicators(self): """Initialize all available communicators.""" try: self.custom_allreduce = CustomAllreduce( group=self.cpu_group, device=self.device, max_size=self.max_size_override, ) if not self.custom_allreduce.disabled: logger.info("Rank %s: CustomAllreduce initialized", self.rank) else: logger.info("Rank %s: CustomAllreduce disabled", self.rank) except Exception as e: logger.warning( "Rank %s: Failed to initialize CustomAllreduce: %s", self.rank, e ) self.custom_allreduce = None try: self.pynccl_comm = PyNcclCommunicator( group=self.cpu_group, device=self.device ) if not self.pynccl_comm.disabled: logger.info("Rank %s: PyNcclCommunicator initialized", self.rank) register_nccl_symmetric_ops(self.pynccl_comm) else: logger.info("Rank %s: PyNcclCommunicator disabled", self.rank) self.pynccl_comm = None except Exception as e: logger.warning( "Rank %s: Failed to initialize PyNcclCommunicator: %s", self.rank, e ) self.pynccl_comm = None # Initialize variants for SymmMemCommunicator try: self.symm_mem_comm_multimem = SymmMemCommunicator( group=self.cpu_group, device=self.device, force_multimem=True, max_size_override=self.max_size_override, ) if not self.symm_mem_comm_multimem.disabled: logger.info( "Rank %s: SymmMemCommunicator (multimem) initialized", self.rank ) else: self.symm_mem_comm_multimem = None except Exception as e: logger.warning( "Rank %s: Failed to initialize SymmMemCommunicator (multimem): %s", self.rank, e, ) self.symm_mem_comm_multimem = None try: self.symm_mem_comm_two_shot = SymmMemCommunicator( group=self.cpu_group, device=self.device, force_multimem=False, max_size_override=self.max_size_override, ) if not self.symm_mem_comm_two_shot.disabled: logger.info( "Rank %s: SymmMemCommunicator (two_shot) initialized", self.rank ) else: self.symm_mem_comm_two_shot = None except Exception as e: logger.warning( "Rank %s: Failed to initialize SymmMemCommunicator (two_shot): %s", self.rank, e, ) self.symm_mem_comm_two_shot = None try: self.fi_ar_comm = FlashInferAllReduce( group=self.cpu_group, device=self.device, ) if not self.fi_ar_comm.disabled: logger.info("Rank %s: FlashInferAllReduce initialized", self.rank) else: logger.info("Rank %s: FlashInferAllReduce disabled", self.rank) self.fi_ar_comm = None except Exception as e: logger.warning( "Rank %s: Failed to initialize FlashInferAllReduce: %s", self.rank, e ) self.fi_ar_comm = None def benchmark_allreduce( self, sequence_length: int, num_warmup: int, num_trials: int ) -> dict[str, float]: """Benchmark allreduce operations for all available communicators.""" results = {} # Define communicators with their benchmark functions communicators = [] if self.custom_allreduce is not None: comm = self.custom_allreduce # CustomAllreduce one-shot communicators.append( ( "ca_1stage", lambda t, c=comm: c.custom_all_reduce(t), lambda t, c=comm: c.should_custom_ar(t), comm.capture(), {"VLLM_CUSTOM_ALLREDUCE_ALGO": "1stage"}, None, # no destroy function ) ) # CustomAllreduce two-shot communicators.append( ( "ca_2stage", lambda t, c=comm: c.custom_all_reduce(t), lambda t, c=comm: c.should_custom_ar(t), comm.capture(), {"VLLM_CUSTOM_ALLREDUCE_ALGO": "2stage"}, None, # no destroy function ) ) if self.pynccl_comm is not None: comm = self.pynccl_comm communicators.append( ( "pynccl", lambda t, c=comm: c.all_reduce(t), lambda t: True, # Always available if initialized nullcontext(), {}, # no env variable needed None, # no destroy function ) ) communicators.append( ( "pynccl-symm", lambda t: torch.ops.vllm.all_reduce_symmetric_with_copy(t), lambda t: True, # Always available if initialized nullcontext(), {}, # no env variable needed None, # no destroy function ) ) if self.symm_mem_comm_multimem is not None: comm = self.symm_mem_comm_multimem communicators.append( ( "symm_mem_multimem", lambda t, c=comm: c.all_reduce(t), lambda t, c=comm: c.should_use_symm_mem(t), nullcontext(), {}, # no env variable needed None, # no destroy function ) ) if self.symm_mem_comm_two_shot is not None: comm = self.symm_mem_comm_two_shot communicators.append( ( "symm_mem_two_shot", lambda t, c=comm: c.all_reduce(t), lambda t, c=comm: c.should_use_symm_mem(t), nullcontext(), {}, # no env variable needed None, # no destroy function needed ) ) if self.fi_ar_comm is not None: comm = self.fi_ar_comm communicators.append( ( "flashinfer_trtllm", lambda t, c=comm: c.all_reduce(t), lambda t, c=comm: c.should_use_fi_ar(t), nullcontext(), {"VLLM_FLASHINFER_ALLREDUCE_BACKEND": "trtllm"}, lambda c=comm: c.destroy(), ) ) communicators.append( ( "flashinfer_mnnvl", lambda t, c=comm: c.all_reduce(t), lambda t, c=comm: c.should_use_fi_ar(t), nullcontext(), {"VLLM_FLASHINFER_ALLREDUCE_BACKEND": "mnnvl"}, lambda c=comm: c.destroy(), ) ) # Benchmark each communicator for ( name, allreduce_fn, should_use_fn, context, env_dict, destroy_fn, ) in communicators: # Save original values and apply new environment variables saved_env = {key: os.environ.get(key) for key in env_dict} for key, value in env_dict.items(): os.environ[key] = value try: latency = self.benchmark_allreduce_single( sequence_length, allreduce_fn, should_use_fn, context, num_warmup, num_trials, ) if latency is not None: results[name] = latency finally: if destroy_fn is not None: destroy_fn() # Restore environment variables to their original state for key, original_value in saved_env.items(): if original_value is None: os.environ.pop(key, None) else: os.environ[key] = original_value return results def benchmark_allreduce_single( self, sequence_length: int, allreduce_fn: Callable[[torch.Tensor], torch.Tensor | None], should_use_fn: Callable[[torch.Tensor], bool], context, num_warmup: int, num_trials: int, ) -> float | None: """Benchmark method with CUDA graph optimization.""" try: # Create test tensor (2D: sequence_length x hidden_size) tensor = torch.randn( sequence_length, HIDDEN_SIZE, dtype=BENCHMARK_DTYPE, device=self.device ) if not should_use_fn(tensor): return None torch.cuda.synchronize() stream = torch.cuda.Stream() with torch.cuda.stream(stream): graph_input = tensor.clone() # Warmup before capture for _ in range(3): allreduce_fn(graph_input) # Capture the graph using context manager with context: graph = torch.cuda.CUDAGraph() graph_pool = torch.cuda.graph_pool_handle() set_graph_pool_id(graph_pool) with torch.cuda.graph(graph, pool=graph_pool, stream=stream): for _ in range(CUDA_GRAPH_CAPTURE_CYCLES): allreduce_fn(graph_input) torch.cuda.synchronize() for _ in range(num_warmup): graph.replay() torch.cuda.synchronize() torch.cuda.synchronize() start_time = time.perf_counter() for _ in range(num_trials): graph.replay() torch.cuda.synchronize() end_time = time.perf_counter() # Convert to ms and divide by CUDA_GRAPH_CAPTURE_CYCLES return ( (end_time - start_time) / num_trials / CUDA_GRAPH_CAPTURE_CYCLES * 1000 ) except Exception as e: logger.error("CUDA graph benchmark failed: %s", e) raise RuntimeError( f"CUDA graph benchmark failed for communicator: {e}" ) from e def _calculate_speedup_info(comm_results: dict[str, float]) -> str: """Calculate speedup information for a single tensor size.""" if not comm_results: return "N/A" # Find the fastest communicator fastest_comm = min(comm_results.keys(), key=lambda k: comm_results[k]) fastest_time = comm_results[fastest_comm] # Calculate speedup vs PyNccl if available if "pynccl" in comm_results: pynccl_time = comm_results["pynccl"] speedup = pynccl_time / fastest_time return f"{fastest_comm} ({speedup:.2f}x)" else: return f"{fastest_comm} (N/A)" def print_results( results: dict[str, dict[str, float]], sequence_lengths: list[int], world_size: int ): """Print benchmark results in a formatted table.""" print(f"\n{'=' * 130}") print("Device Communicator Benchmark Results") print( f"World Size: {world_size}, Data Type: {BENCHMARK_DTYPE}, " f"Hidden Size: {HIDDEN_SIZE}" ) print(f"{'=' * 130}") # Get all communicator names all_comms = set() for size_results in results.values(): all_comms.update(size_results.keys()) all_comms = sorted(list(all_comms)) # Print header header = f"{'Tensor Shape':<20}{'Tensor Size':<15}" for comm in all_comms: header += f"{comm:<20}" header += f"{'Best (Speedup vs PyNccl)':<30}" print(header) print("-" * len(header)) # Print results for each sequence length for seq_len in sequence_lengths: if seq_len in results: # Calculate tensor size in elements and bytes tensor_elements = seq_len * HIDDEN_SIZE tensor_bytes = tensor_elements * BENCHMARK_DTYPE.itemsize # Format tensor size (MB) tensor_size_mb = tensor_bytes / (1024 * 1024) tensor_size_str = f"{tensor_size_mb:.2f} MB" # Format tensor shape tensor_shape = f"({seq_len}, {HIDDEN_SIZE})" row = f"{tensor_shape:<20}{tensor_size_str:<15}" for comm in all_comms: if comm in results[seq_len]: row += f"{results[seq_len][comm]:<20.3f}" else: row += f"{'N/A':<20}" # Calculate speedup information speedup_info = _calculate_speedup_info(results[seq_len]) row += f"{speedup_info:<30}" print(row) print(f"{'=' * 130}") print("All times are in milliseconds (ms) per allreduce operation") print("Speedup column shows: fastest_algorithm (speedup_vs_pynccl)") def main(): parser = FlexibleArgumentParser(description="Benchmark device communicators") parser.add_argument( "--sequence-lengths", type=int, nargs="+", default=DEFAULT_SEQUENCE_LENGTHS, help="Sequence lengths to benchmark (tensor shape: seq_len x hidden_size)", ) parser.add_argument( "--num-warmup", type=int, default=5, help="Number of warmup iterations" ) parser.add_argument( "--num-trials", type=int, default=50, help="Number of benchmark trials" ) parser.add_argument("--output-json", type=str, help="Output results to JSON file") args = parser.parse_args() # Initialize distributed if not dist.is_initialized(): dist.init_process_group(backend="gloo") rank = dist.get_rank() world_size = dist.get_world_size() # Set device device = torch.device(f"cuda:{rank}") torch.cuda.set_device(device) # Get CPU process group cpu_group = dist.new_group(backend="gloo") # Disable USE_SYMM_MEM to avoid affecting the max_sizes # in symm_mem and custom_all_reduce for benchmark os.environ["VLLM_ALLREDUCE_USE_SYMM_MEM"] = "0" # Initialize benchmark benchmark = CommunicatorBenchmark( rank, world_size, device, cpu_group, args.sequence_lengths ) # Run benchmarks all_results = {} for seq_len in args.sequence_lengths: if rank == 0: logger.info( "Benchmarking sequence length: %s (tensor shape: %s x %s)", seq_len, seq_len, HIDDEN_SIZE, ) results = benchmark.benchmark_allreduce( sequence_length=seq_len, num_warmup=args.num_warmup, num_trials=args.num_trials, ) all_results[seq_len] = results # Synchronize between ranks dist.barrier() # Print results (only rank 0) if rank == 0: print_results(all_results, args.sequence_lengths, world_size) # Save to JSON if requested if args.output_json: # Add speedup information to results enhanced_results = {} for seq_len, comm_results in all_results.items(): enhanced_results[seq_len] = { "timings": comm_results, "speedup_info": _calculate_speedup_info(comm_results), } output_data = { "world_size": world_size, "dtype": str(BENCHMARK_DTYPE), "hidden_size": HIDDEN_SIZE, "sequence_lengths": args.sequence_lengths, "num_warmup": args.num_warmup, "num_trials": args.num_trials, "cuda_graph_capture_cycles": CUDA_GRAPH_CAPTURE_CYCLES, "results": enhanced_results, } with open(args.output_json, "w") as f: json.dump(output_data, f, indent=2) logger.info("Results saved to %s", args.output_json) # Cleanup if cpu_group != dist.group.WORLD: dist.destroy_process_group(cpu_group) if __name__ == "__main__": main()
{ "repo_id": "vllm-project/vllm", "file_path": "benchmarks/kernels/benchmark_device_communicators.py", "license": "Apache License 2.0", "lines": 487, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/config/lora.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from typing import TYPE_CHECKING, Any, Literal import torch from pydantic import ConfigDict, Field, model_validator from typing_extensions import Self from vllm.config.utils import config from vllm.logger import init_logger from vllm.utils.hashing import safe_hash if TYPE_CHECKING: from vllm.config import ModelConfig from vllm.config.cache import CacheConfig else: ModelConfig = Any CacheConfig = Any logger = init_logger(__name__) LoRADType = Literal["auto", "float16", "bfloat16"] MaxLoRARanks = Literal[1, 8, 16, 32, 64, 128, 256, 320, 512] LoRAExtraVocabSize = Literal[256, 512] @config(config=ConfigDict(arbitrary_types_allowed=True)) class LoRAConfig: """Configuration for LoRA.""" max_lora_rank: MaxLoRARanks = 16 """Max LoRA rank.""" max_loras: int = Field(default=1, ge=1) """Max number of LoRAs in a single batch.""" fully_sharded_loras: bool = False """By default, only half of the LoRA computation is sharded with tensor parallelism. Enabling this will use the fully sharded layers. At high sequence length, max rank or tensor parallel size, this is likely faster. """ max_cpu_loras: int | None = None """Maximum number of LoRAs to store in CPU memory. Must be >= than `max_loras`.""" lora_dtype: torch.dtype | LoRADType = "auto" """Data type for LoRA. If auto, will default to base model dtype.""" default_mm_loras: dict[str, str] | None = None """Dictionary mapping specific modalities to LoRA model paths; this field is only applicable to multimodal models and should be leveraged when a model always expects a LoRA to be active when a given modality is present. Note that currently, if a request provides multiple additional modalities, each of which have their own LoRA, we do NOT apply default_mm_loras because we currently only support one lora adapter per prompt. When run in offline mode, the lora IDs for n modalities will be automatically assigned to 1-n with the names of the modalities in alphabetic order.""" enable_tower_connector_lora: bool = False """If `True`, LoRA support for the tower (vision encoder) and connector of multimodal models will be enabled. This is an experimental feature and currently only supports some MM models such as the Qwen VL series. The default is False.""" specialize_active_lora: bool = False """Whether to construct lora kernel grid by the number of active LoRA adapters. When set to True, separate cuda graphs will be captured for different counts of active LoRAs (powers of 2 up to max_loras), which can improve performance for variable LoRA usage patterns at the cost of increased startup time and memory usage. Only takes effect when cudagraph_specialize_lora is True. """ def compute_hash(self) -> str: """ WARNING: Whenever a new field is added to this config, ensure that it is included in the factors list if it affects the computation graph. Provide a hash that uniquely identifies all the configs that affect the structure of the computation graph from input ids/embeddings to the final hidden states, excluding anything before input ids/embeddings and after the final hidden states. """ factors: list[Any] = [] factors.append(self.max_lora_rank) factors.append(self.max_loras) factors.append(self.fully_sharded_loras) factors.append(self.lora_dtype) factors.append(self.enable_tower_connector_lora) hash_str = safe_hash(str(factors).encode(), usedforsecurity=False).hexdigest() return hash_str @model_validator(mode="after") def _validate_lora_config(self) -> Self: if self.max_cpu_loras is None: self.max_cpu_loras = self.max_loras elif self.max_cpu_loras < self.max_loras: raise ValueError( f"max_cpu_loras ({self.max_cpu_loras}) must be >= " f"max_loras ({self.max_loras})." ) return self def verify_with_model_config(self, model_config: ModelConfig): if self.lora_dtype in (None, "auto"): self.lora_dtype = model_config.dtype elif isinstance(self.lora_dtype, str): self.lora_dtype = getattr(torch, self.lora_dtype)
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/config/lora.py", "license": "Apache License 2.0", "lines": 92, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/model_executor/models/qwen3_next.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """Inference-only Qwen3Next model.""" from collections.abc import Iterable from itertools import islice import torch from einops import rearrange from torch import nn from transformers.activations import ACT2FN from vllm.compilation.decorators import support_torch_compile from vllm.config import ( CacheConfig, ModelConfig, SpeculativeConfig, VllmConfig, get_current_vllm_config, ) from vllm.distributed import ( divide, get_ep_group, get_pp_group, get_tensor_model_parallel_rank, get_tensor_model_parallel_world_size, tensor_model_parallel_all_gather, ) from vllm.forward_context import ForwardContext, get_forward_context from vllm.logger import init_logger from vllm.model_executor.custom_op import CustomOp from vllm.model_executor.layers.attention import Attention from vllm.model_executor.layers.fla.ops import ( chunk_gated_delta_rule as fla_chunk_gated_delta_rule, ) from vllm.model_executor.layers.fla.ops import ( fused_recurrent_gated_delta_rule, ) from vllm.model_executor.layers.fla.ops.chunk import l2norm_fwd from vllm.model_executor.layers.fused_moe import SharedFusedMoE from vllm.model_executor.layers.layernorm import ( GemmaRMSNorm as Qwen3NextRMSNorm, ) from vllm.model_executor.layers.layernorm import RMSNormGated from vllm.model_executor.layers.linear import ( ColumnParallelLinear, MergedColumnParallelLinear, QKVParallelLinear, ReplicatedLinear, RowParallelLinear, ) from vllm.model_executor.layers.logits_processor import LogitsProcessor from vllm.model_executor.layers.mamba.abstract import MambaBase from vllm.model_executor.layers.mamba.mamba_mixer2 import mamba_v2_sharded_weight_loader from vllm.model_executor.layers.mamba.mamba_utils import ( MambaStateCopyFunc, MambaStateCopyFuncCalculator, MambaStateDtypeCalculator, MambaStateShapeCalculator, ) from vllm.model_executor.layers.mamba.ops.causal_conv1d import ( causal_conv1d_fn, causal_conv1d_update, ) from vllm.model_executor.layers.quantization import QuantizationConfig 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.model_loader.weight_utils import ( default_weight_loader, maybe_remap_kv_scale_name, sharded_weight_loader, ) from vllm.model_executor.models.qwen2_moe import Qwen2MoeMLP as Qwen3NextMLP from vllm.model_executor.models.utils import sequence_parallel_chunk from vllm.model_executor.utils import set_weight_attrs from vllm.platforms import current_platform from vllm.sequence import IntermediateTensors from vllm.transformers_utils.configs import Qwen3NextConfig from vllm.triton_utils import tl, triton from vllm.utils.torch_utils import direct_register_custom_op from vllm.v1.attention.backend import AttentionMetadata from vllm.v1.attention.backends.gdn_attn import GDNAttentionMetadata from .interfaces import ( HasInnerState, IsHybrid, MixtureOfExperts, SupportsLoRA, SupportsPP, ) from .utils import ( AutoWeightsLoader, PPMissingLayer, extract_layer_index, is_pp_missing_parameter, make_empty_intermediate_tensors_factory, make_layers, maybe_prefix, ) logger = init_logger(__name__) KVCache = tuple[torch.Tensor, torch.Tensor] def fi_chunk_gated_delta_rule( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, beta: torch.Tensor, initial_state: torch.Tensor, output_final_state: bool, cu_seqlens: torch.LongTensor | None = None, use_qk_l2norm_in_kernel: bool = True, ): from flashinfer.gdn_prefill import ( chunk_gated_delta_rule as chunk_gated_delta_rule_fi, ) if use_qk_l2norm_in_kernel: q = l2norm_fwd(q) k = l2norm_fwd(k) # use flashinfer implementation q = q.squeeze(0).contiguous() k = k.squeeze(0).contiguous() v = v.squeeze(0).contiguous() g = g.squeeze(0).contiguous() beta = beta.squeeze(0).contiguous() fi_state = initial_state.to(torch.float32) fi_g = g.to(torch.float32) fi_beta = beta.to(torch.float32) output, final_state = chunk_gated_delta_rule_fi( q=q, k=k, v=v, g=torch.exp(fi_g), beta=fi_beta, initial_state=fi_state, output_final_state=output_final_state, cu_seqlens=cu_seqlens, ) # Unsqueeze back to 4D (1, L, H, D) to match fla output format return output.unsqueeze(0), final_state @CustomOp.register("chunk_gated_delta_rule") class ChunkGatedDeltaRule(CustomOp): def __init__(self) -> None: super().__init__() if current_platform.is_cuda() and current_platform.is_device_capability(90): logger.info_once( "Using FlashInfer GDN prefill kernel on CUDA compute capability 90" ) self._forward_method = self.forward_cuda else: self._forward_method = self.forward_native def forward_cuda( self, q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, beta: torch.Tensor, initial_state: torch.Tensor, output_final_state: bool, cu_seqlens: torch.LongTensor | None = None, use_qk_l2norm_in_kernel: bool = True, ): return fi_chunk_gated_delta_rule( q=q, k=k, v=v, g=g, beta=beta, initial_state=initial_state, output_final_state=output_final_state, cu_seqlens=cu_seqlens, use_qk_l2norm_in_kernel=use_qk_l2norm_in_kernel, ) def forward_native( self, q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, beta: torch.Tensor, initial_state: torch.Tensor, output_final_state: bool, cu_seqlens: torch.LongTensor | None = None, use_qk_l2norm_in_kernel: bool = True, ): return fla_chunk_gated_delta_rule( q=q, k=k, v=v, g=g, beta=beta, initial_state=initial_state, output_final_state=output_final_state, cu_seqlens=cu_seqlens, use_qk_l2norm_in_kernel=use_qk_l2norm_in_kernel, ) class Qwen3NextSparseMoeBlock(nn.Module): def __init__(self, vllm_config: VllmConfig, prefix: str = ""): super().__init__() config = vllm_config.model_config.hf_text_config parallel_config = vllm_config.parallel_config quant_config = vllm_config.quant_config self.tp_size = get_tensor_model_parallel_world_size() self.ep_group = get_ep_group().device_group self.ep_rank = get_ep_group().rank_in_group self.ep_size = self.ep_group.size() self.n_routed_experts = config.num_experts self.is_sequence_parallel = parallel_config.use_sequence_parallel_moe if self.tp_size > config.num_experts: raise ValueError( f"Tensor parallel size {self.tp_size} is greater than " f"the number of experts {config.num_experts}." ) # Load balancing settings. vllm_config = get_current_vllm_config() eplb_config = vllm_config.parallel_config.eplb_config self.enable_eplb = parallel_config.enable_eplb self.n_logical_experts = self.n_routed_experts self.n_redundant_experts = eplb_config.num_redundant_experts self.n_physical_experts = self.n_logical_experts + self.n_redundant_experts self.n_local_physical_experts = self.n_physical_experts // self.ep_size self.physical_expert_start = self.ep_rank * self.n_local_physical_experts self.physical_expert_end = ( self.physical_expert_start + self.n_local_physical_experts ) self.gate = ReplicatedLinear( config.hidden_size, config.num_experts, bias=False, quant_config=None, prefix=f"{prefix}.gate", ) self.shared_expert_gate = ReplicatedLinear( config.hidden_size, 1, bias=False, quant_config=None, prefix=f"{prefix}.shared_expert_gate", ) if config.shared_expert_intermediate_size > 0: self.shared_expert = Qwen3NextMLP( hidden_size=config.hidden_size, intermediate_size=config.shared_expert_intermediate_size, hidden_act=config.hidden_act, quant_config=quant_config, reduce_results=False, expert_gate=self.shared_expert_gate, prefix=f"{prefix}.shared_expert", ) else: self.shared_expert = None self.experts = SharedFusedMoE( shared_experts=self.shared_expert, gate=self.gate, num_experts=self.n_routed_experts, top_k=config.num_experts_per_tok, hidden_size=config.hidden_size, intermediate_size=config.moe_intermediate_size, reduce_results=False, renormalize=getattr(config, "norm_topk_prob", True), quant_config=quant_config, prefix=f"{prefix}.experts", enable_eplb=self.enable_eplb, num_redundant_experts=self.n_redundant_experts, is_sequence_parallel=self.is_sequence_parallel, ) def forward(self, hidden_states: torch.Tensor) -> torch.Tensor: # NOTE: hidden_states can have either 1D or 2D shape. orig_shape = hidden_states.shape num_tokens, hidden_dim = hidden_states.shape hidden_states = hidden_states.view(-1, hidden_dim) if self.is_sequence_parallel: hidden_states = sequence_parallel_chunk(hidden_states) if self.experts.is_internal_router: # In this case, the gate/router runs inside the FusedMoE class final_hidden_states = self.experts( hidden_states=hidden_states, router_logits=hidden_states ) else: # router_logits: (num_tokens, n_experts) router_logits, _ = self.gate(hidden_states) final_hidden_states = self.experts( hidden_states=hidden_states, router_logits=router_logits ) if self.shared_expert is not None: final_hidden_states = final_hidden_states[0] + final_hidden_states[1] if self.is_sequence_parallel: final_hidden_states = tensor_model_parallel_all_gather( final_hidden_states, 0 ) final_hidden_states = final_hidden_states[:num_tokens] elif self.tp_size > 1: final_hidden_states = self.experts.maybe_all_reduce_tensor_model_parallel( # noqa E501 final_hidden_states ) return final_hidden_states.view(orig_shape) class Qwen3NextGatedDeltaNet(nn.Module, MambaBase): @property def mamba_type(self) -> str: return "gdn_attention" def get_state_dtype(self) -> tuple[torch.dtype, torch.dtype]: return MambaStateDtypeCalculator.gated_delta_net_state_dtype( self.model_config.dtype, self.cache_config.mamba_cache_dtype, self.cache_config.mamba_ssm_cache_dtype, ) def get_state_shape(self) -> tuple[tuple[int, ...], tuple[int, ...]]: return MambaStateShapeCalculator.gated_delta_net_state_shape( self.tp_size, self.num_k_heads, self.num_v_heads, self.head_k_dim, self.head_v_dim, self.conv_kernel_size, self.num_spec, ) def __init__( self, config: Qwen3NextConfig, model_config: ModelConfig | None = None, cache_config: CacheConfig | None = None, quant_config: QuantizationConfig | None = None, speculative_config: SpeculativeConfig | None = None, prefix: str = "", ) -> None: super().__init__() self.tp_size = get_tensor_model_parallel_world_size() self.tp_rank = get_tensor_model_parallel_rank() self.hidden_size = config.hidden_size self.num_v_heads = config.linear_num_value_heads self.num_k_heads = config.linear_num_key_heads self.head_k_dim = config.linear_key_head_dim self.head_v_dim = config.linear_value_head_dim self.key_dim = self.head_k_dim * self.num_k_heads self.value_dim = self.head_v_dim * self.num_v_heads self.conv_kernel_size = config.linear_conv_kernel_dim self.layer_idx = extract_layer_index(prefix) self.activation = config.hidden_act self.act = ACT2FN[config.hidden_act] self.layer_norm_epsilon = config.rms_norm_eps self.prefix = prefix self.config = config self.model_config = model_config self.cache_config = cache_config self.quant_config = quant_config self.speculative_config = speculative_config self.num_spec = ( self.speculative_config.num_speculative_tokens if self.speculative_config else 0 ) # QKV self.conv_dim = self.key_dim * 2 + self.value_dim self.conv1d = ColumnParallelLinear( input_size=self.conv_kernel_size, output_size=self.conv_dim, bias=False, prefix=f"{prefix}.conv1d", ) self.conv1d.weight.data = self.conv1d.weight.data.unsqueeze(1) # projection of the input hidden states # Qwen3-Next and Qwen3.5 has a different qkv_proj layout, # we need to create qkvz_proj adaptively here. self.in_proj_qkvz = self.create_qkvz_proj( hidden_size=self.hidden_size, key_dim=self.key_dim, value_dim=self.value_dim, quant_config=quant_config, prefix=f"{prefix}.in_proj_qkvz", ) # ba_proj doesn't support blockwise fp8 quantization. # # in_proj_ba is defined as MergedColumnParallelLinear for # compatibility with Qwen3_5. self.in_proj_ba = MergedColumnParallelLinear( input_size=self.hidden_size, output_sizes=[self.num_v_heads] * 2, bias=False, quant_config=quant_config, prefix=f"{prefix}.in_proj_ba", ) query_key_settings = (self.key_dim, 0, False) value_settings = (self.value_dim, 0, False) delattr(self.conv1d.weight, "weight_loader") set_weight_attrs( self.conv1d.weight, { "weight_loader": mamba_v2_sharded_weight_loader( [ query_key_settings, query_key_settings, value_settings, ], self.tp_size, self.tp_rank, ) }, ) # selective projection used to make dt, B and C input dependant # time step projection (discretization) # instantiate once and copy inv_dt in init_weights of PretrainedModel self.dt_bias = nn.Parameter( torch.ones(self.num_v_heads // self.tp_size), ) self.A_log = nn.Parameter( torch.empty( divide(self.num_v_heads, self.tp_size), ) ) set_weight_attrs(self.A_log, {"weight_loader": sharded_weight_loader(0)}) set_weight_attrs(self.dt_bias, {"weight_loader": sharded_weight_loader(0)}) self.norm = RMSNormGated( self.head_v_dim, eps=self.layer_norm_epsilon, group_size=None, norm_before_gate=True, device=current_platform.current_device(), ) self.out_proj = RowParallelLinear( self.value_dim, self.hidden_size, bias=False, input_is_parallel=True, quant_config=quant_config, prefix=f"{prefix}.out_proj", ) self.chunk_gated_delta_rule = ChunkGatedDeltaRule() compilation_config = get_current_vllm_config().compilation_config if prefix in compilation_config.static_forward_context: raise ValueError(f"Duplicate layer name: {prefix}") compilation_config.static_forward_context[prefix] = self def create_qkvz_proj( self, hidden_size: int, key_dim: int, value_dim: int, quant_config: QuantizationConfig | None, prefix: str, ) -> MergedColumnParallelLinear: return MergedColumnParallelLinear( input_size=hidden_size, output_sizes=[sum((key_dim, key_dim, value_dim, value_dim))], bias=False, quant_config=quant_config, prefix=prefix, ) def fix_query_key_value_ordering( self, mixed_qkvz: torch.Tensor, mixed_ba: torch.Tensor, ): """ Derives `query`, `key` and `value` tensors from `mixed_qkvzba`. """ new_tensor_shape_qkvz = mixed_qkvz.size()[:-1] + ( self.num_k_heads // self.tp_size, ( self.head_k_dim + self.head_k_dim + (self.head_v_dim + self.head_v_dim) * self.num_v_heads // self.num_k_heads ), ) new_tensor_shape_ba = mixed_qkvz.size()[:-1] + ( self.num_k_heads // self.tp_size, 2 * self.num_v_heads // self.num_k_heads, ) mixed_qkvz = mixed_qkvz.view(*new_tensor_shape_qkvz) mixed_ba = mixed_ba.view(*new_tensor_shape_ba) split_arg_list_qkvz = [ self.head_k_dim, self.head_k_dim, (self.num_v_heads // self.num_k_heads * self.head_v_dim), (self.num_v_heads // self.num_k_heads * self.head_v_dim), ] split_arg_list_ba = [ self.num_v_heads // self.num_k_heads, self.num_v_heads // self.num_k_heads, ] # [b, sq, ng, (hn + hn + np/ng * hn + np/ng + np/ng)] # --> [b, sq, ng, hn], [b, sq, ng, hn], [b, sq, ng, np/ng * hn], # [b, sq, ng, np/ng * hn], [b, sq, ng, np/ng], [b, sq, ng, np/ng] (query, key, value, z) = torch.split(mixed_qkvz, split_arg_list_qkvz, dim=2) (b, a) = torch.split(mixed_ba, split_arg_list_ba, dim=2) # [b, sq, ng, np/ng * hn] -> [b, sq, np, hn] value = value.reshape(value.size(0), -1, self.head_v_dim) z = z.reshape(z.size(0), -1, self.head_v_dim) b = b.reshape(b.size(0), self.num_v_heads // self.tp_size) a = a.reshape(a.size(0), self.num_v_heads // self.tp_size) return query, key, value, z, b, a def rearrange_mixed_qkv(self, mixed_qkv): if mixed_qkv is None: return None, None, None query, key, value = torch.split( mixed_qkv, [ self.key_dim // self.tp_size, self.key_dim // self.tp_size, self.value_dim // self.tp_size, ], dim=-1, ) query, key = map( lambda x: rearrange(x, "l (h d) -> 1 l h d", d=self.head_k_dim), (query, key), ) value = rearrange(value, "l (h d) -> 1 l h d", d=self.head_v_dim) return query.contiguous(), key.contiguous(), value.contiguous() def forward( self, hidden_states: torch.Tensor, output: torch.Tensor, ): """ Forward pass with three parts: 1. Input projection 2. Core attention (custom op) 3. Output projection """ num_tokens = hidden_states.size(0) # ============================================================ # Part 1: Input Projection # ============================================================ projected_states_qkvz, _ = self.in_proj_qkvz(hidden_states) projected_states_ba, _ = self.in_proj_ba(hidden_states) query, key, value, z, b, a = self.fix_query_key_value_ordering( projected_states_qkvz, projected_states_ba ) query, key, value = map( lambda x: rearrange(x, "l p d -> l (p d)"), (query, key, value) ) mixed_qkv = torch.cat((query, key, value), dim=-1) # ============================================================ # Part 2: Core Attention (Custom Op) # ============================================================ # Note: we should not use torch.empty here like other attention backends, # see discussions in https://github.com/vllm-project/vllm/pull/28182 core_attn_out = torch.zeros( (num_tokens, self.num_v_heads // self.tp_size, self.head_v_dim), dtype=hidden_states.dtype, device=hidden_states.device, ) torch.ops.vllm.gdn_attention_core( mixed_qkv, b, a, core_attn_out, self.prefix, ) # ============================================================ # Part 3: Output Projection # ============================================================ z_shape_og = z.shape # Reshape input data into 2D tensor core_attn_out = core_attn_out.reshape(-1, core_attn_out.shape[-1]) z = z.reshape(-1, z.shape[-1]) core_attn_out = self.norm(core_attn_out, z) core_attn_out = core_attn_out.reshape(z_shape_og) core_attn_out = rearrange(core_attn_out, "... h d -> ... (h d)") output[:num_tokens], _ = self.out_proj(core_attn_out) def _forward_core( self, mixed_qkv: torch.Tensor, b: torch.Tensor, a: torch.Tensor, core_attn_out: torch.Tensor, ): """ Core attention computation (called by custom op). """ forward_context = get_forward_context() attn_metadata: AttentionMetadata = forward_context.attn_metadata if attn_metadata is None: # V1 profile run return assert isinstance(attn_metadata, dict) attn_metadata = attn_metadata[self.prefix] assert isinstance(attn_metadata, GDNAttentionMetadata) has_initial_state = attn_metadata.has_initial_state spec_query_start_loc = attn_metadata.spec_query_start_loc non_spec_query_start_loc = attn_metadata.non_spec_query_start_loc spec_sequence_masks = attn_metadata.spec_sequence_masks spec_token_indx = attn_metadata.spec_token_indx non_spec_token_indx = attn_metadata.non_spec_token_indx spec_state_indices_tensor = attn_metadata.spec_state_indices_tensor # noqa: E501 non_spec_state_indices_tensor = attn_metadata.non_spec_state_indices_tensor # noqa: E501 self_kv_cache = self.kv_cache[forward_context.virtual_engine] conv_state = self_kv_cache[0].transpose(-1, -2) ssm_state = self_kv_cache[1] num_actual_tokens = attn_metadata.num_actual_tokens num_accepted_tokens = attn_metadata.num_accepted_tokens mixed_qkv = mixed_qkv[:num_actual_tokens] b = b[:num_actual_tokens] a = a[:num_actual_tokens] # 1. Convolution sequence transformation conv_weights = self.conv1d.weight.view( self.conv1d.weight.size(0), self.conv1d.weight.size(2) ) if spec_sequence_masks is not None: if attn_metadata.num_prefills == 0 and attn_metadata.num_decodes == 0: mixed_qkv_spec = mixed_qkv mixed_qkv_non_spec = None else: mixed_qkv_spec = mixed_qkv.index_select(0, spec_token_indx) mixed_qkv_non_spec = mixed_qkv.index_select(0, non_spec_token_indx) else: mixed_qkv_spec = None mixed_qkv_non_spec = mixed_qkv # 1.1: Process the multi-query part if spec_sequence_masks is not None: mixed_qkv_spec = causal_conv1d_update( mixed_qkv_spec, conv_state, conv_weights, self.conv1d.bias, self.activation, conv_state_indices=spec_state_indices_tensor[:, 0][ : attn_metadata.num_spec_decodes ], num_accepted_tokens=num_accepted_tokens, query_start_loc=spec_query_start_loc, max_query_len=spec_state_indices_tensor.size(-1), validate_data=False, ) # 1.2: Process the remaining part if attn_metadata.num_prefills > 0: mixed_qkv_non_spec_T = mixed_qkv_non_spec.transpose(0, 1) # - "cache_indices" updates the conv_state cache in positions # pointed to by "state_indices_tensor" mixed_qkv_non_spec = causal_conv1d_fn( mixed_qkv_non_spec_T, conv_weights, self.conv1d.bias, activation=self.activation, conv_states=conv_state, has_initial_state=has_initial_state, cache_indices=non_spec_state_indices_tensor, query_start_loc=non_spec_query_start_loc, metadata=attn_metadata, ).transpose(0, 1) elif attn_metadata.num_decodes > 0: mixed_qkv_non_spec = causal_conv1d_update( mixed_qkv_non_spec, conv_state, conv_weights, self.conv1d.bias, self.activation, conv_state_indices=non_spec_state_indices_tensor[ : attn_metadata.num_actual_tokens ], validate_data=True, ) else: mixed_qkv_non_spec = None query_spec, key_spec, value_spec = self.rearrange_mixed_qkv(mixed_qkv_spec) query_non_spec, key_non_spec, value_non_spec = self.rearrange_mixed_qkv( mixed_qkv_non_spec ) g, beta = fused_gdn_gating(self.A_log, a, b, self.dt_bias) if spec_sequence_masks is not None: if attn_metadata.num_prefills == 0 and attn_metadata.num_decodes == 0: g_spec = g beta_spec = beta g_non_spec = None beta_non_spec = None else: g_spec = g.index_select(1, spec_token_indx) beta_spec = beta.index_select(1, spec_token_indx) g_non_spec = g.index_select(1, non_spec_token_indx) beta_non_spec = beta.index_select(1, non_spec_token_indx) else: g_spec = None beta_spec = None g_non_spec = g beta_non_spec = beta # 2. Recurrent attention # 2.1: Process the multi-query part if spec_sequence_masks is not None: core_attn_out_spec, last_recurrent_state = fused_recurrent_gated_delta_rule( q=query_spec, k=key_spec, v=value_spec, g=g_spec, beta=beta_spec, initial_state=ssm_state, inplace_final_state=True, cu_seqlens=spec_query_start_loc[: attn_metadata.num_spec_decodes + 1], ssm_state_indices=spec_state_indices_tensor, num_accepted_tokens=num_accepted_tokens, use_qk_l2norm_in_kernel=True, ) else: core_attn_out_spec, last_recurrent_state = None, None # 2.2: Process the remaining part if attn_metadata.num_prefills > 0: initial_state = ssm_state[non_spec_state_indices_tensor].contiguous() initial_state[~has_initial_state, ...] = 0 ( core_attn_out_non_spec, last_recurrent_state, ) = self.chunk_gated_delta_rule( q=query_non_spec, k=key_non_spec, v=value_non_spec, g=g_non_spec, beta=beta_non_spec, initial_state=initial_state, output_final_state=True, cu_seqlens=non_spec_query_start_loc, use_qk_l2norm_in_kernel=True, ) # Init cache ssm_state[non_spec_state_indices_tensor] = last_recurrent_state.to( ssm_state.dtype ) elif attn_metadata.num_decodes > 0: core_attn_out_non_spec, last_recurrent_state = ( fused_recurrent_gated_delta_rule( q=query_non_spec, k=key_non_spec, v=value_non_spec, g=g_non_spec, beta=beta_non_spec, initial_state=ssm_state, inplace_final_state=True, cu_seqlens=non_spec_query_start_loc[ : attn_metadata.num_decodes + 1 ], ssm_state_indices=non_spec_state_indices_tensor, use_qk_l2norm_in_kernel=True, ) ) else: core_attn_out_non_spec, last_recurrent_state = None, None # 3. Merge core attention output if spec_sequence_masks is not None and core_attn_out_non_spec is not None: merged_out = torch.empty( (1, num_actual_tokens, *core_attn_out_spec.shape[2:]), dtype=core_attn_out_non_spec.dtype, device=core_attn_out_non_spec.device, ) merged_out.index_copy_(1, spec_token_indx, core_attn_out_spec) merged_out.index_copy_(1, non_spec_token_indx, core_attn_out_non_spec) core_attn_out[:num_actual_tokens] = merged_out.squeeze(0) elif spec_sequence_masks is not None: core_attn_out[:num_actual_tokens] = core_attn_out_spec.squeeze(0) else: core_attn_out[:num_actual_tokens] = core_attn_out_non_spec.squeeze(0) class Qwen3NextAttention(nn.Module): def __init__( self, config: Qwen3NextConfig, model_config: ModelConfig | None = None, cache_config: CacheConfig | None = None, quant_config: QuantizationConfig | None = None, prefix: str = "", ) -> None: super().__init__() self.config = config self.hidden_size = config.hidden_size tp_size = get_tensor_model_parallel_world_size() self.total_num_heads = config.num_attention_heads assert self.total_num_heads % tp_size == 0 self.num_heads = self.total_num_heads // tp_size self.total_num_kv_heads = config.num_key_value_heads if self.total_num_kv_heads >= tp_size: # Number of KV heads is greater than TP size, so we partition # the KV heads across multiple tensor parallel GPUs. assert self.total_num_kv_heads % tp_size == 0 else: # Number of KV heads is less than TP size, so we replicate # the KV heads across multiple tensor parallel GPUs. assert tp_size % self.total_num_kv_heads == 0 self.num_kv_heads = max(1, self.total_num_kv_heads // tp_size) self.head_dim = config.head_dim or (self.hidden_size // self.num_heads) self.q_size = self.num_heads * self.head_dim self.kv_size = self.num_kv_heads * self.head_dim self.scaling = self.head_dim**-0.5 self.dual_chunk_attention_config = getattr( config, "dual_chunk_attention_config", None ) self.attn_output_gate = getattr(config, "attn_output_gate", True) self.qkv_proj = QKVParallelLinear( config.hidden_size, self.head_dim, self.total_num_heads * (1 + self.attn_output_gate), self.total_num_kv_heads, bias=getattr(config, "qkv_bias", False), quant_config=quant_config, prefix=f"{prefix}.qkv_proj", ) self.o_proj = RowParallelLinear( self.total_num_heads * self.head_dim, config.hidden_size, bias=False, quant_config=quant_config, prefix=f"{prefix}.o_proj", ) self.rotary_emb = get_rope( head_size=self.head_dim, max_position=config.max_position_embeddings, rope_parameters=config.rope_parameters, dual_chunk_attention_config=self.dual_chunk_attention_config, ) self.attn = Attention( self.num_heads, self.head_dim, self.scaling, num_kv_heads=self.num_kv_heads, cache_config=cache_config, quant_config=quant_config, prefix=f"{prefix}.attn", **{ "layer_idx": extract_layer_index(prefix), "dual_chunk_attention_config": self.dual_chunk_attention_config, } if self.dual_chunk_attention_config else {}, ) self.q_norm = Qwen3NextRMSNorm(self.head_dim, eps=config.rms_norm_eps) self.k_norm = Qwen3NextRMSNorm(self.head_dim, eps=config.rms_norm_eps) def forward( self, positions: torch.Tensor, output: torch.Tensor, hidden_states: torch.Tensor, ): qkv, _ = self.qkv_proj(hidden_states) if self.attn_output_gate: q_gate, k, v = qkv.split( [self.q_size * 2, self.kv_size, self.kv_size], dim=-1 ) orig_shape = q_gate.shape[:-1] q_gate = q_gate.view(*orig_shape, self.num_heads, -1) q, gate = torch.chunk(q_gate, 2, dim=-1) q = q.reshape(*orig_shape, -1) gate = gate.reshape(*orig_shape, -1) else: q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1) q = self.q_norm(q.view(-1, self.num_heads, self.head_dim)).view( -1, self.num_heads * self.head_dim ) k = self.k_norm(k.view(-1, self.num_kv_heads, self.head_dim)).view( -1, self.num_kv_heads * self.head_dim ) q, k = self.rotary_emb(positions, q, k) attn_output = self.attn(q, k, v) if self.attn_output_gate: gate = torch.sigmoid(gate) attn_output = attn_output * gate output[:], _ = self.o_proj(attn_output) class Qwen3NextDecoderLayer(nn.Module): def __init__( self, vllm_config: VllmConfig, layer_type: str, prefix: str = "", ) -> None: super().__init__() config = vllm_config.model_config.hf_config model_config = vllm_config.model_config cache_config = vllm_config.cache_config quant_config = vllm_config.quant_config speculative_config = vllm_config.speculative_config self.layer_type = layer_type self.layer_idx = extract_layer_index(prefix) if self.layer_type == "linear_attention": self.linear_attn = Qwen3NextGatedDeltaNet( config, model_config=model_config, cache_config=cache_config, quant_config=quant_config, speculative_config=speculative_config, prefix=f"{prefix}.linear_attn", ) elif self.layer_type == "full_attention": self.self_attn = Qwen3NextAttention( config, model_config=model_config, cache_config=cache_config, quant_config=quant_config, prefix=f"{prefix}.self_attn", ) else: raise ValueError(f"Invalid layer_type {self.layer_type}") mlp_only_layers = ( [] if not hasattr(config, "mlp_only_layers") else config.mlp_only_layers ) if (self.layer_idx not in mlp_only_layers) and ( config.num_experts > 0 and (self.layer_idx + 1) % config.decoder_sparse_step == 0 ): self.mlp = Qwen3NextSparseMoeBlock( vllm_config=vllm_config, prefix=f"{prefix}.mlp", ) else: self.mlp = Qwen3NextMLP( hidden_size=config.hidden_size, intermediate_size=config.intermediate_size, hidden_act=config.hidden_act, quant_config=quant_config, prefix=f"{prefix}.mlp", ) self.input_layernorm = Qwen3NextRMSNorm( config.hidden_size, eps=config.rms_norm_eps ) self.post_attention_layernorm = Qwen3NextRMSNorm( config.hidden_size, eps=config.rms_norm_eps ) self.layer_scale = getattr(config, "layer_scale", False) if self.layer_scale: self.attn_layer_scale = torch.nn.Parameter( torch.zeros( 1, 1, config.hidden_size, ), ) self.ffn_layer_scale = torch.nn.Parameter( torch.zeros( 1, 1, config.hidden_size, ), ) def forward( self, hidden_states: torch.Tensor, residual: torch.Tensor | None, positions: torch.Tensor = None, **kwargs: object, ): if residual is None: residual = hidden_states hidden_states = self.input_layernorm(hidden_states) else: hidden_states, residual = self.input_layernorm(hidden_states, residual) self_attention_output = torch.empty_like(hidden_states) if self.layer_type == "linear_attention": self.linear_attn( hidden_states=hidden_states, output=self_attention_output, ) elif self.layer_type == "full_attention": self.self_attn( hidden_states=hidden_states, output=self_attention_output, positions=positions, ) else: raise ValueError("Invalid layer_type") hidden_states = self_attention_output if self.layer_scale: if len(hidden_states.shape) == 2: hidden_states = hidden_states * ( self.attn_layer_scale.to(hidden_states.dtype)[0] + 1 ) else: hidden_states = hidden_states * ( self.attn_layer_scale.to(hidden_states.dtype) + 1 ) # Fully Connected hidden_states, residual = self.post_attention_layernorm(hidden_states, residual) hidden_states = self.mlp(hidden_states) if self.layer_scale: if len(hidden_states.shape) == 2: hidden_states = hidden_states * ( self.ffn_layer_scale.to(hidden_states.dtype)[0] + 1 ) else: assert len(hidden_states.shape) == len(self.ffn_layer_scale.shape), ( f"shape must be the same {len(hidden_states.shape)}, " f"{len(self.ffn_layer_scale.shape)}" ) hidden_states = hidden_states * ( self.ffn_layer_scale.to(hidden_states.dtype) + 1 ) return hidden_states, residual @support_torch_compile class Qwen3NextModel(nn.Module): def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): super().__init__() config: Qwen3NextConfig = vllm_config.model_config.hf_text_config parallel_config = vllm_config.parallel_config eplb_config = parallel_config.eplb_config self.num_redundant_experts = eplb_config.num_redundant_experts self.config = config self.vocab_size = config.vocab_size self.embed_tokens = VocabParallelEmbedding( self.vocab_size, config.hidden_size, ) def get_layer(prefix: str): return Qwen3NextDecoderLayer( vllm_config, layer_type=config.layer_types[extract_layer_index(prefix)], prefix=prefix, ) self.start_layer, self.end_layer, self.layers = make_layers( config.num_hidden_layers, get_layer, prefix=f"{prefix}.layers" ) self.make_empty_intermediate_tensors = make_empty_intermediate_tensors_factory( ["hidden_states", "residual"], config.hidden_size ) if get_pp_group().is_last_rank: self.norm = Qwen3NextRMSNorm(config.hidden_size, eps=config.rms_norm_eps) else: self.norm = PPMissingLayer() def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor: return self.embed_tokens(input_ids) def forward( self, input_ids: torch.Tensor | None, positions: torch.Tensor, intermediate_tensors: IntermediateTensors | None = None, inputs_embeds: torch.Tensor | None = None, ) -> torch.Tensor: if get_pp_group().is_first_rank: if inputs_embeds is not None: hidden_states = inputs_embeds else: hidden_states = self.embed_input_ids(input_ids) residual = None else: assert intermediate_tensors is not None hidden_states = intermediate_tensors["hidden_states"] residual = intermediate_tensors["residual"] for layer in islice(self.layers, self.start_layer, self.end_layer): hidden_states, residual = layer( positions=positions, hidden_states=hidden_states, residual=residual, ) if not get_pp_group().is_last_rank: return IntermediateTensors( {"hidden_states": hidden_states, "residual": residual} ) hidden_states, _ = self.norm(hidden_states, residual) return hidden_states def get_expert_mapping(self) -> list[tuple[str, str, int, str]]: # Params for weights, fp8 weight scales, fp8 activation scales # (param_name, weight_name, expert_id, shard_id) return SharedFusedMoE.make_expert_params_mapping( self, ckpt_gate_proj_name="gate_proj", ckpt_down_proj_name="down_proj", ckpt_up_proj_name="up_proj", num_experts=getattr(self.config, "num_experts", 0), num_redundant_experts=self.num_redundant_experts, ) def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: stacked_params_mapping = [ # (param_name, shard_name, shard_id) ("qkv_proj", "q_proj", "q"), ("qkv_proj", "k_proj", "k"), ("qkv_proj", "v_proj", "v"), ("gate_up_proj", "gate_proj", 0), ("gate_up_proj", "up_proj", 1), ] params_dict = dict(self.named_parameters()) loaded_params: set[str] = set() expert_params_mapping = self.get_expert_mapping() for name, loaded_weight in weights: if "rotary_emb.inv_freq" in name: continue if name.startswith("mtp."): continue # Remapping the name of FP8 kv-scale. if name.endswith("scale"): name = maybe_remap_kv_scale_name(name, params_dict) if name is None: continue for param_name, weight_name, shard_id in stacked_params_mapping: if weight_name not in name: continue if "mlp.experts" 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 # Skip layers on other devices. if is_pp_missing_parameter(name, self): continue # name = apply_attn_prefix(name, params_dict) if name not in params_dict: continue param = params_dict[name] weight_loader = param.weight_loader weight_loader(param, loaded_weight, shard_id) break else: for mapping in expert_params_mapping: param_name, weight_name, expert_id, shard_id = mapping if weight_name not in name: continue name = name.replace(weight_name, param_name) # Skip layers on other devices. if is_pp_missing_parameter(name, self): continue # Skip loading extra bias for GPTQ models. if ( name.endswith(".bias") or name.endswith("_bias") ) and name not in params_dict: continue if name not in params_dict: continue param = params_dict[name] weight_loader = param.weight_loader weight_loader( param, loaded_weight, name, shard_id=shard_id, expert_id=expert_id, ) break else: # Skip loading extra bias for GPTQ models. if name.endswith(".bias") and name not in params_dict: continue if is_pp_missing_parameter(name, self): continue if name not in params_dict: logger.warning_once( f"Parameter {name} not found in params_dict, skip loading" ) continue param = params_dict[name] weight_loader = getattr( param, "weight_loader", default_weight_loader ) weight_loader(param, loaded_weight) loaded_params.add(name) return loaded_params class QwenNextMixtureOfExperts(MixtureOfExperts): def update_physical_experts_metadata( self, num_physical_experts: int, num_local_physical_experts: int, ) -> None: assert self.num_local_physical_experts == num_local_physical_experts self.num_physical_experts = num_physical_experts self.num_local_physical_experts = num_local_physical_experts self.num_redundant_experts = num_physical_experts - self.num_logical_experts for layer in self.model.layers: if isinstance(layer.mlp, Qwen3NextSparseMoeBlock): moe = layer.mlp moe.n_local_physical_experts = num_local_physical_experts moe.n_physical_experts = num_physical_experts moe.n_redundant_experts = self.num_redundant_experts moe.experts.update_expert_map() def set_moe_parameters(self): self.expert_weights = [] self.moe_layers = [] example_moe = None for layer in self.model.layers: if isinstance(layer, Qwen3NextDecoderLayer) and isinstance( layer.mlp, Qwen3NextSparseMoeBlock ): example_moe = layer.mlp self.moe_layers.append(layer.mlp.experts) if example_moe is None: raise RuntimeError("No Qwen3Next layer found in the model.layers.") # Set MoE hyperparameters self.num_moe_layers = len(self.moe_layers) self.num_expert_groups = 1 self.num_shared_experts = 0 self.num_logical_experts = example_moe.n_logical_experts self.num_physical_experts = example_moe.n_physical_experts self.num_local_physical_experts = example_moe.n_local_physical_experts self.num_routed_experts = example_moe.n_routed_experts self.num_redundant_experts = example_moe.n_redundant_experts class Qwen3NextForCausalLM( nn.Module, HasInnerState, SupportsLoRA, SupportsPP, QwenNextMixtureOfExperts, IsHybrid, ): packed_modules_mapping = { "qkv_proj": [ "q_proj", "k_proj", "v_proj", ], "gate_up_proj": ["gate_proj", "up_proj"], "in_proj_qkvz": ["in_proj_qkvz"], "in_proj_ba": ["in_proj_ba"], } def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): config = vllm_config.model_config.hf_text_config self.vllm_config = vllm_config self.model_config = vllm_config.model_config cache_config = vllm_config.cache_config scheduler_config = vllm_config.scheduler_config if cache_config.mamba_cache_mode == "all": raise NotImplementedError( "Qwen3Next currently does not support 'all' prefix caching, " "please use '--mamba-cache-mode=align' instead" ) self.quant_config = vllm_config.quant_config super().__init__() self.config = config self.scheduler_config = scheduler_config self.model = Qwen3NextModel( vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model") ) self.lm_head = ParallelLMHead( config.vocab_size, config.hidden_size, prefix=maybe_prefix(prefix, "lm_head"), ) self.logits_processor = LogitsProcessor(config.vocab_size) self.make_empty_intermediate_tensors = ( self.model.make_empty_intermediate_tensors ) # Set MoE hyperparameters self.set_moe_parameters() def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor: return self.model.embed_input_ids(input_ids) def forward( self, input_ids: torch.Tensor | None, positions: torch.Tensor, intermediate_tensors: IntermediateTensors | None = None, inputs_embeds: torch.Tensor | None = None, **kwargs: object, ): hidden_states = self.model( input_ids, positions, intermediate_tensors, inputs_embeds ) return hidden_states @classmethod def get_mamba_state_dtype_from_config( cls, vllm_config: "VllmConfig", ) -> tuple[torch.dtype, torch.dtype]: return MambaStateDtypeCalculator.gated_delta_net_state_dtype( vllm_config.model_config.dtype, vllm_config.cache_config.mamba_cache_dtype, vllm_config.cache_config.mamba_ssm_cache_dtype, ) @classmethod def get_mamba_state_shape_from_config( cls, vllm_config: "VllmConfig" ) -> tuple[tuple[int, int], tuple[int, int]]: parallel_config = vllm_config.parallel_config hf_config = vllm_config.model_config.hf_text_config tp_size = parallel_config.tensor_parallel_size num_spec = ( vllm_config.speculative_config.num_speculative_tokens if vllm_config.speculative_config else 0 ) return MambaStateShapeCalculator.gated_delta_net_state_shape( tp_size, hf_config.linear_num_key_heads, hf_config.linear_num_value_heads, hf_config.linear_key_head_dim, hf_config.linear_value_head_dim, hf_config.linear_conv_kernel_dim, num_spec, ) @classmethod def get_mamba_state_copy_func(cls) -> tuple[MambaStateCopyFunc, MambaStateCopyFunc]: return MambaStateCopyFuncCalculator.gated_delta_net_state_copy_func() def compute_logits( self, hidden_states: torch.Tensor, ) -> torch.Tensor | None: return self.logits_processor(self.lm_head, hidden_states) def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: loader = AutoWeightsLoader( self, skip_prefixes=["mtp."], ) return loader.load_weights(weights) def get_expert_mapping(self) -> list[tuple[str, str, int, str]]: return self.model.get_expert_mapping() def gdn_attention_core( mixed_qkv: torch.Tensor, b: torch.Tensor, a: torch.Tensor, core_attn_out: torch.Tensor, layer_name: str, ) -> None: """ Custom op for the core attention computation. Only handles the convolution + recurrent attention part. Input/output projections are handled outside this op. """ forward_context: ForwardContext = get_forward_context() self = forward_context.no_compile_layers[layer_name] self._forward_core( mixed_qkv=mixed_qkv, b=b, a=a, core_attn_out=core_attn_out, ) def gdn_attention_core_fake( mixed_qkv: torch.Tensor, b: torch.Tensor, a: torch.Tensor, core_attn_out: torch.Tensor, layer_name: str, ) -> None: """Fake implementation for torch.compile.""" return direct_register_custom_op( op_name="gdn_attention_core", op_func=gdn_attention_core, mutates_args=["core_attn_out"], fake_impl=gdn_attention_core_fake, ) @triton.jit def fused_gdn_gating_kernel( g, beta_output, A_log, a, b, dt_bias, seq_len, NUM_HEADS: tl.constexpr, beta: tl.constexpr, threshold: tl.constexpr, BLK_HEADS: tl.constexpr, ): i_b, i_s, i_d = tl.program_id(0), tl.program_id(1), tl.program_id(2) head_off = i_d * BLK_HEADS + tl.arange(0, BLK_HEADS) off = i_b * seq_len * NUM_HEADS + i_s * NUM_HEADS + head_off mask = head_off < NUM_HEADS blk_A_log = tl.load(A_log + head_off, mask=mask) blk_a = tl.load(a + off, mask=mask) blk_b = tl.load(b + off, mask=mask) blk_bias = tl.load(dt_bias + head_off, mask=mask) # If the model is loaded in fp16, without the .float() here, A might be -inf x = blk_a.to(tl.float32) + blk_bias.to(tl.float32) softplus_x = tl.where( beta * x <= threshold, (1 / beta) * tl.log(1 + tl.exp(beta * x)), x ) blk_g = -tl.exp(blk_A_log.to(tl.float32)) * softplus_x tl.store(g + off, blk_g.to(g.dtype.element_ty), mask=mask) # compute beta_output = sigmoid(b) blk_beta_output = tl.sigmoid(blk_b.to(tl.float32)) tl.store( beta_output + off, blk_beta_output.to(beta_output.dtype.element_ty), mask=mask ) def fused_gdn_gating( A_log: torch.Tensor, a: torch.Tensor, b: torch.Tensor, dt_bias: torch.Tensor, beta: float = 1.0, threshold: float = 20.0, ) -> tuple[torch.Tensor, torch.Tensor]: """ Fused computation of g and beta for Gated Delta Net. g = -self.A_log.float().exp() * F.softplus(a.float() + self.dt_bias) beta_output = b.sigmoid() TODO maybe use torch.compile to replace this triton kernel """ batch, num_heads = a.shape seq_len = 1 grid = (batch, seq_len, triton.cdiv(num_heads, 8)) g = torch.empty(1, batch, num_heads, dtype=torch.float32, device=a.device) beta_output = torch.empty(1, batch, num_heads, dtype=b.dtype, device=b.device) fused_gdn_gating_kernel[grid]( g, beta_output, A_log, a, b, dt_bias, seq_len, num_heads, beta, threshold, 8, num_warps=1, ) return g, beta_output
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/model_executor/models/qwen3_next.py", "license": "Apache License 2.0", "lines": 1384, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/model_executor/models/qwen3_next_mtp.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """Inference-only Qwen3Next MTP model.""" from collections.abc import Iterable import torch from torch import nn from vllm.compilation.decorators import support_torch_compile from vllm.config import VllmConfig from vllm.distributed.parallel_state import get_pp_group from vllm.logger import init_logger from vllm.model_executor.layers.fused_moe import FusedMoE from vllm.model_executor.layers.linear import ColumnParallelLinear from vllm.model_executor.layers.logits_processor import LogitsProcessor from vllm.model_executor.layers.vocab_parallel_embedding import ( ParallelLMHead, VocabParallelEmbedding, ) from vllm.model_executor.model_loader.weight_utils import default_weight_loader from vllm.model_executor.models.qwen3_next import ( Qwen3NextDecoderLayer, Qwen3NextRMSNorm, QwenNextMixtureOfExperts, ) from vllm.sequence import IntermediateTensors from vllm.transformers_utils.configs import Qwen3NextConfig from .utils import ( AutoWeightsLoader, is_pp_missing_parameter, make_empty_intermediate_tensors_factory, maybe_prefix, ) logger = init_logger(__name__) KVCache = tuple[torch.Tensor, torch.Tensor] @support_torch_compile class Qwen3NextMultiTokenPredictor(nn.Module): def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): super().__init__() model_config = vllm_config.model_config quant_config = vllm_config.quant_config config: Qwen3NextConfig = model_config.hf_config self.config = config self.vocab_size = config.vocab_size self.mtp_start_layer_idx = config.num_hidden_layers self.num_mtp_layers = getattr(config, "num_nextn_predict_layers", 1) self.embed_tokens = VocabParallelEmbedding( self.vocab_size, config.hidden_size, ) self.fc = ColumnParallelLinear( self.config.hidden_size * 2, self.config.hidden_size, gather_output=True, bias=False, return_bias=False, quant_config=quant_config, prefix=f"{prefix}.fc", ) self.layers = torch.nn.ModuleList( Qwen3NextDecoderLayer( vllm_config, layer_type="full_attention", prefix=f"{prefix}.layers.{idx}", ) for idx in range(self.num_mtp_layers) ) self.make_empty_intermediate_tensors = make_empty_intermediate_tensors_factory( ["hidden_states", "residual"], config.hidden_size ) self.norm = Qwen3NextRMSNorm(config.hidden_size, eps=config.rms_norm_eps) self.pre_fc_norm_hidden = Qwen3NextRMSNorm( config.hidden_size, eps=config.rms_norm_eps ) self.pre_fc_norm_embedding = Qwen3NextRMSNorm( config.hidden_size, eps=config.rms_norm_eps ) def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor: return self.embed_tokens(input_ids) def forward( self, input_ids: torch.Tensor, positions: torch.Tensor, hidden_states: torch.Tensor, intermediate_tensors: IntermediateTensors | None = None, inputs_embeds: torch.Tensor | None = None, spec_step_idx: int = 0, ) -> torch.Tensor: if get_pp_group().is_first_rank: if inputs_embeds is None: inputs_embeds = self.embed_input_ids(input_ids) assert hidden_states.shape[-1] == inputs_embeds.shape[-1] inputs_embeds = self.pre_fc_norm_embedding(inputs_embeds) hidden_states = self.pre_fc_norm_hidden(hidden_states) hidden_states = torch.cat([inputs_embeds, hidden_states], dim=-1) hidden_states = self.fc(hidden_states) residual = None else: assert intermediate_tensors is not None hidden_states = intermediate_tensors["hidden_states"] residual = intermediate_tensors["residual"] current_step_idx = spec_step_idx % self.num_mtp_layers hidden_states, residual = self.layers[current_step_idx]( positions=positions, hidden_states=hidden_states, residual=residual, ) if not get_pp_group().is_last_rank: return IntermediateTensors( {"hidden_states": hidden_states, "residual": residual} ) hidden_states, _ = self.norm(hidden_states, residual) return hidden_states def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: stacked_params_mapping = [ # (param_name, shard_name, shard_id) ("qkv_proj", "q_proj", "q"), ("qkv_proj", "k_proj", "k"), ("qkv_proj", "v_proj", "v"), ("gate_up_proj", "gate_proj", 0), ("gate_up_proj", "up_proj", 1), ] # Params for weights, fp8 weight scales, fp8 activation scales # (param_name, weight_name, expert_id, shard_id) expert_params_mapping = FusedMoE.make_expert_params_mapping( self, ckpt_gate_proj_name="gate_proj", ckpt_down_proj_name="down_proj", ckpt_up_proj_name="up_proj", num_experts=self.config.num_experts, ) params_dict = dict(self.named_parameters()) loaded_params: set[str] = set() for name, loaded_weight in weights: 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 if "mlp.experts" 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 # Skip layers on other devices. if is_pp_missing_parameter(name, self): continue if name not in params_dict: continue param = params_dict[name] weight_loader = param.weight_loader weight_loader(param, loaded_weight, shard_id) break else: for mapping in expert_params_mapping: param_name, weight_name, expert_id, shard_id = mapping if weight_name not in name: continue name = name.replace(weight_name, param_name) # Skip layers on other devices. if is_pp_missing_parameter(name, self): continue # Skip loading extra bias for GPTQ models. if ( name.endswith(".bias") or name.endswith("_bias") ) and name not in params_dict: continue param = params_dict[name] weight_loader = param.weight_loader weight_loader( param, loaded_weight, name, shard_id=shard_id, expert_id=expert_id, ) break else: # Skip loading extra bias for GPTQ models. if name.endswith(".bias") and name not in params_dict: continue if is_pp_missing_parameter(name, self): continue param = params_dict[name] weight_loader = getattr( param, "weight_loader", default_weight_loader ) weight_loader(param, loaded_weight) loaded_params.add(name) return loaded_params @support_torch_compile class Qwen3NextMTP(nn.Module, QwenNextMixtureOfExperts): packed_modules_mapping = { "qkv_proj": [ "q_proj", "k_proj", "v_proj", ], "gate_up_proj": ["up_proj", "down_proj"], } def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): config = vllm_config.model_config.hf_config self.vllm_config = vllm_config cache_config = vllm_config.cache_config if cache_config.mamba_cache_mode == "all": raise NotImplementedError( "Qwen3NextMTP currently does not support 'all' prefix caching, " "please use '--mamba-cache-mode=align' instead" ) self.quant_config = vllm_config.quant_config super().__init__() self.config = config self.model = Qwen3NextMultiTokenPredictor( vllm_config=vllm_config, prefix=maybe_prefix(prefix, "mtp") ) self.lm_head = ParallelLMHead( config.vocab_size, config.hidden_size, prefix=maybe_prefix(prefix, "lm_head"), ) self.logits_processor = LogitsProcessor(config.vocab_size) self.set_moe_parameters() def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor: return self.model.embed_input_ids(input_ids) def forward( self, input_ids: torch.Tensor | None, positions: torch.Tensor, hidden_states: torch.Tensor, intermediate_tensors: IntermediateTensors | None = None, inputs_embeds: torch.Tensor | None = None, **kwargs: object, ): hidden_states = self.model( input_ids, positions, hidden_states, intermediate_tensors, inputs_embeds ) return hidden_states def compute_logits( self, hidden_states: torch.Tensor, spec_step_idx: int = 0, ) -> torch.Tensor | None: return self.logits_processor(self.lm_head, hidden_states) def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: shared_weight_names = ["embed_tokens", "lm_head"] def remap_weight_names(weights): for name, weight in weights: if name.startswith("mtp."): name = name.replace("mtp.", "model.") elif not any(key in name for key in shared_weight_names): continue yield name, weight loader = AutoWeightsLoader(self) return loader.load_weights(remap_weight_names(weights))
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/model_executor/models/qwen3_next_mtp.py", "license": "Apache License 2.0", "lines": 253, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/transformers_utils/configs/qwen3_next.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project # Copyright 2025 The Qwen team, Alibaba Group and the HuggingFace Inc. team. # All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """Qwen3-Next model configuration""" from transformers.configuration_utils import PretrainedConfig, layer_type_validation from transformers.utils import logging logger = logging.get_logger(__name__) class Qwen3NextConfig(PretrainedConfig): r""" This is the configuration class to store the configuration of a [`Qwen3NextModel`]. It is used to instantiate a Qwen3-Next model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of Qwen3-Next-80B-A3B-Instruct [Qwen/Qwen3-Next-80B-A3B-Instruct](https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct). Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the documentation from [`PretrainedConfig`] for more information. Args: vocab_size (`int`, *optional*, defaults to 151936): Vocabulary size of the model. Defines the number of different tokens that can be represented by the `inputs_ids`. hidden_size (`int`, *optional*, defaults to 2048): Dimension of the hidden representations. intermediate_size (`int`, *optional*, defaults to 5632): Dimension of the MLP representations. num_hidden_layers (`int`, *optional*, defaults to 48): Number of hidden layers in the Transformer encoder. num_attention_heads (`int`, *optional*, defaults to 16): Number of attention heads for each attention layer in the Transformer encoder. num_key_value_heads (`int`, *optional*, defaults to 2): This is the number of key_value heads that should be used to implement Grouped Query Attention. If `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed by meanpooling all the original heads within that group. For more details checkout [this paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to `32`. hidden_act (`str`, *optional*, defaults to `"silu"`): The non-linear activation function in the decoder. max_position_embeddings (`int`, *optional*, defaults to 32768): The maximum sequence length that this model might ever be used with. initializer_range (`float`, *optional*, defaults to 0.02): The standard deviation of the truncated_normal_initializer for initializing all weight matrices. rms_norm_eps (`float`, *optional*, defaults to 1e-06): The epsilon used by the rms normalization layers. use_cache (`bool`, *optional*, defaults to `True`): Whether or not the model should return the last key/values attentions (not used by all models). Only relevant if `config.is_decoder=True`. tie_word_embeddings (`bool`, *optional*, defaults to `False`): Whether the model's input and output word embeddings should be tied. rope_parameters (`dict`, *optional*): Dictionary containing the scaling configuration for the RoPE embeddings. NOTE: if you apply new rope type and you expect the model to work on longer `max_position_embeddings`, we recommend you to update this value accordingly. Expected contents: `rope_theta` (`float`): The base period of the RoPE embeddings. `rope_type` (`str`): The sub-variant of RoPE to use. Can be one of ['default', 'linear', 'dynamic', 'yarn', 'longrope', 'llama3'], with 'default' being the original RoPE implementation. `factor` (`float`, *optional*): Used with all rope types except 'default'. The scaling factor to apply to the RoPE embeddings. In most scaling types, a `factor` of x will enable the model to handle sequences of length x * original maximum pre-trained length. `original_max_position_embeddings` (`int`, *optional*): Used with 'dynamic', 'longrope' and 'llama3'. The original max position embeddings used during pretraining. `attention_factor` (`float`, *optional*): Used with 'yarn' and 'longrope'. The scaling factor to be applied on the attention computation. If unspecified, it defaults to value recommended by the implementation, using the `factor` field to infer the suggested value. `beta_fast` (`float`, *optional*): Only used with 'yarn'. Parameter to set the boundary for extrapolation (only) in the linear ramp function. If unspecified, it defaults to 32. `beta_slow` (`float`, *optional*): Only used with 'yarn'. Parameter to set the boundary for interpolation (only) in the linear ramp function. If unspecified, it defaults to 1. `short_factor` (`List[float]`, *optional*): Only used with 'longrope'. The scaling factor to be applied to short contexts (< `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden size divided by the number of attention heads divided by 2 `long_factor` (`List[float]`, *optional*): Only used with 'longrope'. The scaling factor to be applied to long contexts (< `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden size divided by the number of attention heads divided by 2 `low_freq_factor` (`float`, *optional*): Only used with 'llama3'. Scaling factor applied to low frequency components of the RoPE `high_freq_factor` (`float`, *optional*): Only used with 'llama3'. Scaling factor applied to high frequency components of the RoPE `partial_rotary_factor` (`float`, *optional*, defaults to 0.25): Percentage of the query and keys which will have rotary embedding. attention_bias (`bool`, *optional*, defaults to `False`): Whether to use a bias in the query, key, value and output projection layers during self-attention. attention_dropout (`float`, *optional*, defaults to 0.0): The dropout ratio for the attention probabilities. head_dim (`int`, *optional*, defaults to 256): Projection weights dimension in multi-head attention. linear_conv_kernel_dim (`int`, *optional*, defaults to 4): Kernel size of the convolution used in linear attention layers. linear_key_head_dim (`int`, *optional*, defaults to 128): Dimension of each key head in linear attention. linear_value_head_dim (`int`, *optional*, defaults to 128): Dimension of each value head in linear attention. linear_num_key_heads (`int`, *optional*, defaults to 16): Number of key heads used in linear attention layers. linear_num_value_heads (`int`, *optional*, defaults to 32): Number of value heads used in linear attention layers. decoder_sparse_step (`int`, *optional*, defaults to 1): The frequency of the MoE layer. moe_intermediate_size (`int`, *optional*, defaults to 512): Intermediate size of the routed expert. shared_expert_intermediate_size (`int`, *optional*, defaults to 512): Intermediate size of the shared expert. num_experts_per_tok (`int`, *optional*, defaults to 10): Number of selected experts. num_experts (`int`, *optional*, defaults to 512): Number of routed experts. norm_topk_prob (`bool`, *optional*, defaults to `True`): Whether to normalize the topk probabilities. output_router_logits (`bool`, *optional*, defaults to `False`): Whether or not the router logits should be returned by the model. Enabling this will also allow the model to output the auxiliary loss, including load balancing loss and router z-loss. router_aux_loss_coef (`float`, *optional*, defaults to 0.001): The aux loss factor for the total loss. mlp_only_layers (`list[int]`, *optional*, defaults to `[]`): Indicate which layers use Qwen3NextMLP rather than Qwen3NextSparseMoeBlock The list contains layer index, from 0 to num_layers-1 if we have num_layers layers If `mlp_only_layers` is empty, `decoder_sparse_step` is used to determine the sparsity. layer_types (`list[str]`, *optional*): Types of each layer (attention or linear). ```python >>> from transformers import Qwen3NextModel, Qwen3NextConfig >>> # Initializing a Qwen3Next style configuration >>> configuration = Qwen3NextConfig() >>> # Initializing a model from the Qwen3-Next-80B-A3B style configuration >>> model = Qwen3NextModel(configuration) >>> # Accessing the model configuration >>> configuration = model.config ``` """ # noqa: E501 model_type = "qwen3_next" keys_to_ignore_at_inference = ["past_key_values"] base_model_tp_plan = { "layers.*.self_attn.q_proj": "colwise", "layers.*.self_attn.k_proj": "colwise", "layers.*.self_attn.v_proj": "colwise", "layers.*.self_attn.o_proj": "rowwise", "layers.*.mlp.experts.*.gate_proj": "colwise", "layers.*.mlp.experts.*.up_proj": "colwise", "layers.*.mlp.experts.*.down_proj": "rowwise", "layers.*.mlp.shared_experts.gate_proj": "colwise", "layers.*.mlp.shared_experts.up_proj": "colwise", "layers.*.mlp.shared_experts.down_proj": "rowwise", "layers.*.mlp.gate_proj": "colwise", "layers.*.mlp.up_proj": "colwise", "layers.*.mlp.down_proj": "rowwise", } base_model_pp_plan = { "embed_tokens": (["input_ids"], ["inputs_embeds"]), "layers": (["hidden_states", "attention_mask"], ["hidden_states"]), "norm": (["hidden_states"], ["hidden_states"]), } def __init__( self, vocab_size=151936, hidden_size=2048, intermediate_size=5632, num_hidden_layers=48, num_attention_heads=16, num_key_value_heads=2, hidden_act="silu", max_position_embeddings=32768, initializer_range=0.02, rms_norm_eps=1e-6, use_cache=True, tie_word_embeddings=False, rope_parameters=None, attention_bias=False, attention_dropout=0.0, head_dim=256, linear_conv_kernel_dim=4, linear_key_head_dim=128, linear_value_head_dim=128, linear_num_key_heads=16, linear_num_value_heads=32, decoder_sparse_step=1, moe_intermediate_size=512, shared_expert_intermediate_size=512, num_experts_per_tok=10, num_experts=512, norm_topk_prob=True, output_router_logits=False, router_aux_loss_coef=0.001, mlp_only_layers=None, layer_types=None, **kwargs, ): if mlp_only_layers is None: mlp_only_layers = [] super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs) self.vocab_size = vocab_size self.max_position_embeddings = max_position_embeddings self.hidden_size = hidden_size self.intermediate_size = intermediate_size self.num_hidden_layers = num_hidden_layers self.num_attention_heads = num_attention_heads self.num_key_value_heads = num_key_value_heads self.hidden_act = hidden_act self.initializer_range = initializer_range self.rms_norm_eps = rms_norm_eps self.use_cache = use_cache # Try to set `rope_scaling` if available, otherwise use `rope_parameters` rope_scaling = kwargs.pop("rope_scaling", None) rope_parameters = rope_scaling or rope_parameters or {"rope_type": "default"} rope_theta = kwargs.pop("rope_theta", 10000.0) if "rope_theta" not in rope_parameters: rope_parameters["rope_theta"] = rope_theta partial_rotary_factor = kwargs.pop("partial_rotary_factor", 0.25) if "partial_rotary_factor" not in rope_parameters: rope_parameters["partial_rotary_factor"] = partial_rotary_factor self.rope_parameters = rope_parameters self.partial_rotary_factor = partial_rotary_factor self.attention_bias = attention_bias self.attention_dropout = attention_dropout self.head_dim = head_dim self.layer_types = layer_types if self.layer_types is None: self.layer_types = [ "linear_attention" if bool((i + 1) % 4) else "full_attention" for i in range(self.num_hidden_layers) ] layer_type_validation(self.layer_types) # linear attention part self.linear_conv_kernel_dim = linear_conv_kernel_dim self.linear_key_head_dim = linear_key_head_dim self.linear_value_head_dim = linear_value_head_dim self.linear_num_key_heads = linear_num_key_heads self.linear_num_value_heads = linear_num_value_heads # MoE arguments self.decoder_sparse_step = decoder_sparse_step self.moe_intermediate_size = moe_intermediate_size self.shared_expert_intermediate_size = shared_expert_intermediate_size self.num_experts_per_tok = num_experts_per_tok self.num_experts = num_experts self.norm_topk_prob = norm_topk_prob self.output_router_logits = output_router_logits self.router_aux_loss_coef = router_aux_loss_coef self.mlp_only_layers = mlp_only_layers __all__ = ["Qwen3NextConfig"]
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/transformers_utils/configs/qwen3_next.py", "license": "Apache License 2.0", "lines": 258, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/v1/attention/backends/gdn_attn.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """Backend for GatedDeltaNet attention.""" from dataclasses import dataclass import torch from vllm.config import VllmConfig from vllm.v1.attention.backend import ( AttentionBackend, AttentionCGSupport, AttentionMetadataBuilder, CommonAttentionMetadata, ) from vllm.v1.attention.backends.utils import ( PAD_SLOT_ID, compute_causal_conv1d_metadata, mamba_get_block_table_tensor, split_decodes_and_prefills, ) from vllm.v1.kv_cache_interface import AttentionSpec, MambaSpec class GDNAttentionBackend(AttentionBackend): @staticmethod def get_name() -> str: return "GDN_ATTN" @staticmethod def get_builder_cls() -> type["GDNAttentionMetadataBuilder"]: return GDNAttentionMetadataBuilder @dataclass class GDNAttentionMetadata: num_prefills: int num_prefill_tokens: int num_decodes: int num_decode_tokens: int num_spec_decodes: int num_spec_decode_tokens: int num_actual_tokens: int has_initial_state: torch.Tensor | None = None spec_query_start_loc: torch.Tensor | None = None # shape: [num_spec_decodes + 1,] non_spec_query_start_loc: torch.Tensor | None = ( None # shape: [batch - num_spec_decodes + 1,] ) spec_state_indices_tensor: torch.Tensor | None = None # shape: [batch, num_spec] non_spec_state_indices_tensor: torch.Tensor | None = ( None # shape: [batch - num_spec_decodes,] ) spec_sequence_masks: torch.Tensor | None = None # shape: [batch,] spec_token_indx: torch.Tensor | None = None non_spec_token_indx: torch.Tensor | None = None num_accepted_tokens: torch.Tensor | None = None # shape: [batch,] # The following attributes are for triton implementation of causal_conv1d nums_dict: dict | None = None batch_ptr: torch.Tensor | None = None token_chunk_offset_ptr: torch.Tensor | None = None class GDNAttentionMetadataBuilder(AttentionMetadataBuilder[GDNAttentionMetadata]): _cudagraph_support = AttentionCGSupport.UNIFORM_BATCH reorder_batch_threshold: int = 1 def __init__( self, kv_cache_spec: AttentionSpec, layer_names: list[str], vllm_config: VllmConfig, device: torch.device, ): assert isinstance(kv_cache_spec, MambaSpec) self.vllm_config = vllm_config self.compilation_config = vllm_config.compilation_config self.speculative_config = vllm_config.speculative_config self.kv_cache_spec = kv_cache_spec if self.speculative_config: assert self.speculative_config.num_speculative_tokens is not None self.num_spec: int = self.speculative_config.num_speculative_tokens else: self.num_spec = 0 self.use_spec_decode = self.num_spec > 0 self._init_reorder_batch_threshold(1, self.use_spec_decode) self.use_full_cuda_graph = ( self.compilation_config.cudagraph_mode.has_full_cudagraphs() ) self.decode_cudagraph_max_bs = ( self.vllm_config.scheduler_config.max_num_seqs * (self.num_spec + 1) ) if self.compilation_config.max_cudagraph_capture_size is not None: self.decode_cudagraph_max_bs = min( self.decode_cudagraph_max_bs, self.compilation_config.max_cudagraph_capture_size, ) self.spec_state_indices_tensor = torch.empty( (self.decode_cudagraph_max_bs, self.num_spec + 1), dtype=torch.int32, device=device, ) self.non_spec_state_indices_tensor = torch.empty( (self.decode_cudagraph_max_bs,), dtype=torch.int32, device=device, ) self.spec_sequence_masks = torch.empty( (self.decode_cudagraph_max_bs,), dtype=torch.bool, device=device, ) self.spec_token_indx = torch.empty( (self.decode_cudagraph_max_bs * (self.num_spec + 1),), dtype=torch.int32, device=device, ) self.non_spec_token_indx = torch.empty( (self.decode_cudagraph_max_bs * (self.num_spec + 1),), dtype=torch.int32, device=device, ) self.spec_query_start_loc = torch.empty( (self.decode_cudagraph_max_bs + 1,), dtype=torch.int32, device=device, ) self.non_spec_query_start_loc = torch.empty( (self.decode_cudagraph_max_bs + 1,), dtype=torch.int32, device=device, ) self.num_accepted_tokens = torch.empty( (self.decode_cudagraph_max_bs,), dtype=torch.int32, device=device, ) def build( # type: ignore[override] self, common_prefix_len: int, common_attn_metadata: CommonAttentionMetadata, num_accepted_tokens: torch.Tensor | None = None, num_decode_draft_tokens_cpu: torch.Tensor | None = None, fast_build: bool = False, ) -> GDNAttentionMetadata: m = common_attn_metadata query_start_loc = m.query_start_loc query_start_loc_cpu = m.query_start_loc_cpu context_lens_tensor = m.compute_num_computed_tokens() nums_dict, batch_ptr, token_chunk_offset_ptr = None, None, None block_table_tensor = mamba_get_block_table_tensor( m.block_table_tensor, m.seq_lens, self.kv_cache_spec, self.vllm_config.cache_config.mamba_cache_mode, ) spec_sequence_masks_cpu: torch.Tensor | None = None if ( not self.use_spec_decode or num_decode_draft_tokens_cpu is None or num_decode_draft_tokens_cpu[num_decode_draft_tokens_cpu >= 0] .sum() .item() == 0 ): spec_sequence_masks = None num_spec_decodes = 0 else: spec_sequence_masks_cpu = num_decode_draft_tokens_cpu >= 0 num_spec_decodes = spec_sequence_masks_cpu.sum().item() if num_spec_decodes == 0: spec_sequence_masks = None spec_sequence_masks_cpu = None else: spec_sequence_masks = spec_sequence_masks_cpu.to( query_start_loc.device, non_blocking=True ) if spec_sequence_masks is None: num_decodes, num_prefills, num_decode_tokens, num_prefill_tokens = ( split_decodes_and_prefills(m, decode_threshold=1) ) num_spec_decode_tokens = 0 spec_token_indx = None non_spec_token_indx = None spec_state_indices_tensor = None non_spec_state_indices_tensor = block_table_tensor[:, 0] spec_query_start_loc = None non_spec_query_start_loc = query_start_loc non_spec_query_start_loc_cpu = query_start_loc_cpu num_accepted_tokens = None else: query_lens = query_start_loc[1:] - query_start_loc[:-1] assert spec_sequence_masks_cpu is not None query_lens_cpu = query_start_loc_cpu[1:] - query_start_loc_cpu[:-1] # Use CPU tensors to avoid CPU-GPU sync non_spec_query_lens_cpu = query_lens_cpu[~spec_sequence_masks_cpu] num_decodes = (non_spec_query_lens_cpu == 1).sum().item() # Exclude zero-length padded sequences from prefill count. num_zero_len = (non_spec_query_lens_cpu == 0).sum().item() num_prefills = non_spec_query_lens_cpu.size(0) - num_decodes - num_zero_len num_decode_tokens = num_decodes num_prefill_tokens = ( non_spec_query_lens_cpu.sum().item() - num_decode_tokens ) num_spec_decode_tokens = ( query_lens_cpu.sum().item() - num_prefill_tokens - num_decode_tokens ) if num_prefills == 0 and num_decodes == 0: spec_token_size = min( num_spec_decodes * (self.num_spec + 1), query_start_loc_cpu[-1].item(), ) spec_token_indx = torch.arange( spec_token_size, dtype=torch.int32, device=query_start_loc.device, ) non_spec_token_indx = torch.empty( 0, dtype=torch.int32, device=query_start_loc.device ) # Filter by spec_sequence_masks to exclude padded sequences spec_state_indices_tensor = block_table_tensor[ spec_sequence_masks, : self.num_spec + 1 ] non_spec_state_indices_tensor = None # Padded sequences are always at the back, so the first # num_spec_decodes + 1 entries of query_start_loc already # contain the correct cumulative token counts. spec_query_start_loc = query_start_loc[: num_spec_decodes + 1] non_spec_query_start_loc = None non_spec_query_start_loc_cpu = None else: spec_token_masks = torch.repeat_interleave( spec_sequence_masks, query_lens ) index = torch.argsort(spec_token_masks, stable=True) num_non_spec_tokens = num_prefill_tokens + num_decode_tokens non_spec_token_indx = index[:num_non_spec_tokens] spec_token_indx = index[num_non_spec_tokens:] spec_state_indices_tensor = block_table_tensor[ spec_sequence_masks, : self.num_spec + 1 ] non_spec_state_indices_tensor = block_table_tensor[ ~spec_sequence_masks, 0 ] spec_query_start_loc = torch.zeros( num_spec_decodes + 1, dtype=torch.int32, device=query_start_loc.device, ) torch.cumsum( query_lens[spec_sequence_masks], dim=0, out=spec_query_start_loc[1:] ) non_spec_query_start_loc = torch.zeros( query_lens.size(0) - num_spec_decodes + 1, dtype=torch.int32, device=query_start_loc.device, ) torch.cumsum( query_lens[~spec_sequence_masks], dim=0, out=non_spec_query_start_loc[1:], ) non_spec_query_start_loc_cpu = torch.zeros( query_lens_cpu.size(0) - num_spec_decodes + 1, dtype=torch.int32, ) torch.cumsum( query_lens_cpu[~spec_sequence_masks_cpu], dim=0, out=non_spec_query_start_loc_cpu[1:], ) assert num_accepted_tokens is not None num_accepted_tokens = num_accepted_tokens[spec_sequence_masks] if num_prefills > 0: has_initial_state = context_lens_tensor > 0 if spec_sequence_masks is not None: has_initial_state = has_initial_state[~spec_sequence_masks] assert non_spec_query_start_loc_cpu is not None nums_dict, batch_ptr, token_chunk_offset_ptr = ( compute_causal_conv1d_metadata( non_spec_query_start_loc_cpu, device=query_start_loc.device, ) ) else: has_initial_state = None # Function code counted on either presency non-spec decode or spec decode, # but not both. assert not (num_decodes > 0 and num_spec_decodes > 0), ( f"num_decodes: {num_decodes}, num_spec_decodes: {num_spec_decodes}" ) # Prepare tensors for cudagraph # Note: m.num_actual_tokens is already padded by the model runner for CUDAGraph batch_size = m.num_actual_tokens if ( self.use_full_cuda_graph and num_prefills == 0 and num_decodes == 0 and num_spec_decodes <= self.decode_cudagraph_max_bs and num_spec_decode_tokens <= self.decode_cudagraph_max_bs ): self.spec_state_indices_tensor[:num_spec_decodes].copy_( spec_state_indices_tensor, non_blocking=True ) spec_state_indices_tensor = self.spec_state_indices_tensor[:batch_size] spec_state_indices_tensor[num_spec_decodes:].fill_(PAD_SLOT_ID) self.spec_sequence_masks[:num_spec_decodes].copy_( spec_sequence_masks[:num_spec_decodes], non_blocking=True ) spec_sequence_masks = self.spec_sequence_masks[:batch_size] spec_sequence_masks[num_spec_decodes:].fill_(False) assert non_spec_token_indx is not None and spec_token_indx is not None self.non_spec_token_indx[: non_spec_token_indx.size(0)].copy_( non_spec_token_indx, non_blocking=True ) non_spec_token_indx = self.non_spec_token_indx[ : non_spec_token_indx.size(0) ] self.spec_token_indx[: spec_token_indx.size(0)].copy_( spec_token_indx, non_blocking=True ) spec_token_indx = self.spec_token_indx[: spec_token_indx.size(0)] self.spec_query_start_loc[: num_spec_decodes + 1].copy_( spec_query_start_loc, non_blocking=True ) spec_num_query_tokens = spec_query_start_loc[-1] # type: ignore[index] spec_query_start_loc = self.spec_query_start_loc[: batch_size + 1] spec_query_start_loc[num_spec_decodes + 1 :].fill_(spec_num_query_tokens) self.num_accepted_tokens[:num_spec_decodes].copy_( num_accepted_tokens, non_blocking=True ) num_accepted_tokens = self.num_accepted_tokens[:batch_size] num_accepted_tokens[num_spec_decodes:].fill_(1) if ( self.use_full_cuda_graph and num_prefills == 0 and num_spec_decodes == 0 and num_decodes <= self.decode_cudagraph_max_bs ): self.non_spec_state_indices_tensor[:num_decodes].copy_( non_spec_state_indices_tensor, non_blocking=True ) non_spec_state_indices_tensor = self.non_spec_state_indices_tensor[ :batch_size ] non_spec_state_indices_tensor[num_decodes:].fill_(PAD_SLOT_ID) self.non_spec_query_start_loc[: num_decodes + 1].copy_( non_spec_query_start_loc, non_blocking=True ) non_spec_num_query_tokens = non_spec_query_start_loc[-1] # type: ignore[index] non_spec_query_start_loc = self.non_spec_query_start_loc[: batch_size + 1] non_spec_query_start_loc[num_decodes + 1 :].fill_(non_spec_num_query_tokens) attn_metadata = GDNAttentionMetadata( num_prefills=num_prefills, num_prefill_tokens=num_prefill_tokens, num_decodes=num_decodes, num_decode_tokens=num_decode_tokens, num_spec_decodes=num_spec_decodes, num_spec_decode_tokens=num_spec_decode_tokens, num_actual_tokens=m.num_actual_tokens, has_initial_state=has_initial_state, spec_query_start_loc=spec_query_start_loc, non_spec_query_start_loc=non_spec_query_start_loc, spec_state_indices_tensor=spec_state_indices_tensor, non_spec_state_indices_tensor=non_spec_state_indices_tensor, spec_sequence_masks=spec_sequence_masks, spec_token_indx=spec_token_indx, non_spec_token_indx=non_spec_token_indx, num_accepted_tokens=num_accepted_tokens, nums_dict=nums_dict, batch_ptr=batch_ptr, token_chunk_offset_ptr=token_chunk_offset_ptr, ) return attn_metadata def build_for_cudagraph_capture( self, common_attn_metadata: CommonAttentionMetadata ): """ This method builds the metadata for full cudagraph capture. Currently, only decode is supported for full cudagraphs with Mamba. """ m = common_attn_metadata assert ( m.num_reqs <= self.decode_cudagraph_max_bs and m.num_actual_tokens <= self.decode_cudagraph_max_bs ), ( f"GDN only supports decode-only full CUDAGraph capture. " f"Make sure batch size ({m.num_reqs}) <= " f"cudagraph capture sizes ({self.decode_cudagraph_max_bs}), " f"and number of tokens ({m.num_actual_tokens}) <= " f"cudagraph capture sizes ({self.decode_cudagraph_max_bs})." ) num_accepted_tokens = torch.diff(m.query_start_loc) num_decode_draft_tokens_cpu = (num_accepted_tokens - 1).cpu() return self.build(0, m, num_accepted_tokens, num_decode_draft_tokens_cpu)
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/attention/backends/gdn_attn.py", "license": "Apache License 2.0", "lines": 384, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/kernels/attention/test_flashinfer_mla_decode.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import pytest import torch import torch.nn.functional as F from torch import Tensor from vllm.platforms import current_platform FLASHINFER_WORKSPACE_BUFFER_SIZE = 128 * 1024 * 1024 if not current_platform.has_device_capability(100): pytest.skip( reason="FlashInfer MLA Requires compute capability of 10 or above.", allow_module_level=True, ) else: from flashinfer.decode import trtllm_batch_decode_with_kv_cache_mla def ref_mla( out: Tensor, # (bs, num_heads, v_head_dim) query: Tensor, # (bs, num_heads, head_dim) kv_cache: Tensor, # (num_blocks, block_size, head_dim) scale: float, block_tables: Tensor, # (bs, max_num_blocks) seq_lens: Tensor, # (bs,) ): bs, num_heads, v_head_dim = out.shape head_dim = query.shape[2] for i in range(bs): # gather and flatten KV-cache kv = kv_cache[block_tables[i]] # (max_num_blocks, block_size, head_dim) kv = kv.view(1, -1, head_dim)[:, : seq_lens[i]] # (1, seq_len, head_dim) v = kv[:, :, :v_head_dim] q = query[i].view(num_heads, 1, head_dim) o = F.scaled_dot_product_attention(q, kv, v, scale=scale, enable_gqa=True) out[i] = o.view(num_heads, v_head_dim) return out @pytest.mark.parametrize("dtype", [torch.bfloat16]) @pytest.mark.parametrize("bs", [1, 2, 4, 16]) @pytest.mark.parametrize("block_size", [32, 64]) def test_flashinfer_mla_decode(dtype: torch.dtype, bs: int, block_size: int): torch.set_default_device("cuda") torch.manual_seed(42) # Deepseek R1 config num_heads = 128 kv_lora_rank = 512 qk_nope_head_dim = 128 qk_rope_head_dim = 64 qk_head_dim = kv_lora_rank + qk_rope_head_dim scale = (qk_nope_head_dim + qk_rope_head_dim) ** -0.5 MAX_SEQ_LEN = 1024 seq_lens = [torch.randint(2, MAX_SEQ_LEN, (1,)).item() for _ in range(bs)] seq_lens[-1] = MAX_SEQ_LEN max_seq_len = max(seq_lens) seq_lens_tensor = torch.tensor(seq_lens, dtype=torch.int32) # Generate block tables with random but unique block IDs # From https://github.com/flashinfer-ai/flashinfer/pull/1222 blocks_per_seq = (seq_lens_tensor + block_size - 1) // block_size max_num_blocks_per_seq = max(blocks_per_seq.max().item(), 4) total_blocks_needed = sum(blocks_per_seq) # Get random unique IDs for all blocks all_block_ids = torch.randperm(total_blocks_needed) block_id = 0 block_tables = torch.zeros( (bs, max_num_blocks_per_seq), dtype=torch.int32, ) # Populate block tables and track block assignments block_id = 0 for i in range(bs): num_blocks_needed = blocks_per_seq[i] block_tables[i, :num_blocks_needed] = all_block_ids[ block_id : block_id + num_blocks_needed ] block_id += num_blocks_needed kv_cache = torch.randn(block_tables.numel(), block_size, qk_head_dim).to(dtype) q = torch.randn(bs, num_heads, qk_head_dim).to(dtype) out_ref = q.new_zeros(bs, num_heads, kv_lora_rank) ref_mla(out_ref, q, kv_cache, scale, block_tables, seq_lens_tensor) workspace_buffer = torch.zeros( FLASHINFER_WORKSPACE_BUFFER_SIZE, dtype=torch.uint8, device=q.device, ) # Flashinfer MLA expects the query to be of shape # (bs, q_len_per_request, num_heads, qk_head_dim), # where q_len_per_request is the MTP query length (=1 without MTP) q = q.unsqueeze(1) out_ans = trtllm_batch_decode_with_kv_cache_mla( query=q, kv_cache=kv_cache.unsqueeze(1), workspace_buffer=workspace_buffer, qk_nope_head_dim=qk_nope_head_dim, kv_lora_rank=kv_lora_rank, qk_rope_head_dim=qk_rope_head_dim, block_tables=block_tables, seq_lens=seq_lens_tensor, max_seq_len=max_seq_len, bmm1_scale=scale, ) out_ans = out_ans.squeeze(1) torch.testing.assert_close(out_ans, out_ref, atol=1e-2, rtol=1e-2)
{ "repo_id": "vllm-project/vllm", "file_path": "tests/kernels/attention/test_flashinfer_mla_decode.py", "license": "Apache License 2.0", "lines": 99, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/v1/attention/backends/mla/flashinfer_mla.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from typing import ClassVar import torch from flashinfer.decode import trtllm_batch_decode_with_kv_cache_mla from vllm.config.cache import CacheDType from vllm.logger import init_logger from vllm.model_executor.layers.attention.mla_attention import ( MLACommonBackend, MLACommonImpl, MLACommonMetadata, MLACommonMetadataBuilder, QueryLenSupport, ) from vllm.platforms.interface import DeviceCapability from vllm.v1.attention.backend import ( AttentionCGSupport, AttentionLayer, AttentionType, MultipleOf, ) from vllm.v1.attention.backends.utils import KVCacheLayoutType logger = init_logger(__name__) FLASHINFER_MLA_WORKSPACE_BUFFER_SIZE = 128 * 1024 * 1024 class FlashInferMLAMetadataBuilder(MLACommonMetadataBuilder[MLACommonMetadata]): _cudagraph_support: ClassVar[AttentionCGSupport] = AttentionCGSupport.UNIFORM_BATCH query_len_support: ClassVar[QueryLenSupport] = QueryLenSupport.UNIFORM class FlashInferMLABackend(MLACommonBackend): supported_dtypes: ClassVar[list[torch.dtype]] = [torch.float16, torch.bfloat16] supported_kv_cache_dtypes: ClassVar[list[CacheDType]] = [ "auto", "bfloat16", "fp8", "fp8_e4m3", ] @staticmethod def get_supported_kernel_block_sizes() -> list[int | MultipleOf]: return [32, 64] @staticmethod def get_name() -> str: return "FLASHINFER_MLA" @staticmethod def get_impl_cls() -> type["FlashInferMLAImpl"]: return FlashInferMLAImpl @staticmethod def get_builder_cls() -> type["FlashInferMLAMetadataBuilder"]: return FlashInferMLAMetadataBuilder @classmethod def supports_compute_capability(cls, capability: DeviceCapability) -> bool: return capability.major == 10 @classmethod def supports_combination( cls, head_size: int, dtype: torch.dtype, kv_cache_dtype: CacheDType | None, block_size: int, use_mla: bool, has_sink: bool, use_sparse: bool, device_capability: DeviceCapability, ) -> str | None: # FlashInfer MLA kernel requires qk_nope_head_dim == 128 from vllm.config import get_current_vllm_config vllm_config = get_current_vllm_config() if vllm_config.model_config is not None: hf_text_config = vllm_config.model_config.hf_text_config qk_nope_head_dim = getattr(hf_text_config, "qk_nope_head_dim", 1) if qk_nope_head_dim != 128: return ( f"FlashInfer MLA kernel requires qk_nope_head_dim == 128, " f"but got {qk_nope_head_dim}" ) return None @classmethod def get_required_kv_cache_layout(cls) -> "KVCacheLayoutType | None": return "HND" g_fi_workspace = torch.zeros( FLASHINFER_MLA_WORKSPACE_BUFFER_SIZE, dtype=torch.uint8, device="cuda", ) class FlashInferMLAImpl(MLACommonImpl[MLACommonMetadata]): def __init__( self, num_heads: int, head_size: int, scale: float, num_kv_heads: int, alibi_slopes: list[float] | None, sliding_window: int | None, kv_cache_dtype: str, logits_soft_cap: float | None, attn_type: str, kv_sharing_target_layer_name: str | None, # MLA Specific Arguments **mla_args, ) -> None: super().__init__( num_heads, head_size, scale, num_kv_heads, alibi_slopes, sliding_window, kv_cache_dtype, logits_soft_cap, attn_type, kv_sharing_target_layer_name, **mla_args, ) unsupported_features = [alibi_slopes, sliding_window, logits_soft_cap] if any(unsupported_features): raise NotImplementedError( "FlashInferMLAImpl does not support one of the following: " "alibi_slopes, sliding_window, logits_soft_cap" ) if attn_type != AttentionType.DECODER: raise NotImplementedError( "Encoder self-attention and " "encoder/decoder cross-attention " "are not implemented for " "FlashInferMLAImpl" ) self._workspace_buffer = g_fi_workspace self.bmm1_scale: float | None = None self.bmm2_scale: float | None = None def forward_mqa( self, q: torch.Tensor | tuple[torch.Tensor, torch.Tensor], kv_c_and_k_pe_cache: torch.Tensor, attn_metadata: MLACommonMetadata, layer: AttentionLayer, ) -> tuple[torch.Tensor, torch.Tensor | None]: assert kv_c_and_k_pe_cache.numel() > 0 assert attn_metadata.decode is not None if isinstance(q, tuple): q_nope, q_pe = q q = torch.cat([q_nope, q_pe], dim=-1) # trtllm API requires extra dimension q_len_per_request for MTP if attn_metadata.num_decode_tokens % attn_metadata.num_decodes != 0: logger.warning_once( """FlashInferMLAImpl got a query of uneven length. This usually indicates an issue in batch reordering or incorrect setup in dummy_run.""" ) q = q.unsqueeze(1) else: q = q.view(attn_metadata.num_decodes, -1, q.shape[-2], q.shape[-1]) if self.bmm1_scale is None: self.bmm1_scale = layer._q_scale_float * layer._k_scale_float * self.scale if self.bmm2_scale is None: self.bmm2_scale = layer._v_scale_float o = trtllm_batch_decode_with_kv_cache_mla( query=q, kv_cache=kv_c_and_k_pe_cache.unsqueeze(1), workspace_buffer=self._workspace_buffer, qk_nope_head_dim=self.qk_nope_head_dim, kv_lora_rank=self.kv_lora_rank, qk_rope_head_dim=self.qk_rope_head_dim, block_tables=attn_metadata.decode.block_table, seq_lens=attn_metadata.decode.seq_lens, max_seq_len=attn_metadata.max_seq_len, bmm1_scale=self.bmm1_scale, bmm2_scale=self.bmm2_scale, ) # Flatten the output for consistent shape o = o.view(-1, o.shape[-2], o.shape[-1]) # TODO: Return LSE pending support from Flashinfer API: # https://github.com/flashinfer-ai/flashinfer/pull/1566 return o, None
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/v1/attention/backends/mla/flashinfer_mla.py", "license": "Apache License 2.0", "lines": 171, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/transformers_utils/test_config_parser_registry.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from pathlib import Path import pytest from transformers import PretrainedConfig from vllm.transformers_utils.config import get_config_parser, register_config_parser from vllm.transformers_utils.config_parser_base import ConfigParserBase @register_config_parser("custom_config_parser") class CustomConfigParser(ConfigParserBase): def parse( self, model: str | Path, trust_remote_code: bool, revision: str | None = None, code_revision: str | None = None, **kwargs, ) -> tuple[dict, PretrainedConfig]: raise NotImplementedError def test_register_config_parser(): assert isinstance(get_config_parser("custom_config_parser"), CustomConfigParser) def test_invalid_config_parser(): with pytest.raises(ValueError): @register_config_parser("invalid_config_parser") class InvalidConfigParser: pass
{ "repo_id": "vllm-project/vllm", "file_path": "tests/transformers_utils/test_config_parser_registry.py", "license": "Apache License 2.0", "lines": 25, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/transformers_utils/config_parser_base.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from abc import ABC, abstractmethod from pathlib import Path from transformers import PretrainedConfig class ConfigParserBase(ABC): @abstractmethod def parse( self, model: str | Path, trust_remote_code: bool, revision: str | None = None, code_revision: str | None = None, **kwargs, ) -> tuple[dict, PretrainedConfig]: raise NotImplementedError
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/transformers_utils/config_parser_base.py", "license": "Apache License 2.0", "lines": 16, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/lora/layers/base.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from typing import TYPE_CHECKING import torch import torch.nn as nn from transformers import PretrainedConfig from vllm.config.lora import LoRAConfig if TYPE_CHECKING: from vllm.lora.punica_wrapper import PunicaWrapperBase class BaseLayerWithLoRA(nn.Module): def slice_lora_a( self, lora_a: torch.Tensor | list[torch.Tensor | None] ) -> torch.Tensor | list[torch.Tensor | None]: """Slice lora a if splitting for tensor parallelism.""" ... def slice_lora_b( self, lora_b: torch.Tensor | list[torch.Tensor | None] ) -> torch.Tensor | list[torch.Tensor | None]: """Slice lora b if splitting with tensor parallelism.""" ... def create_lora_weights( self, max_loras: int, lora_config: LoRAConfig, model_config: PretrainedConfig | None = None, ) -> None: """Initializes lora matrices.""" ... def reset_lora(self, index: int): """Resets the lora weights at index back to 0.""" ... def set_lora( self, index: int, lora_a: torch.Tensor | list[torch.Tensor], lora_b: torch.Tensor | list[torch.Tensor], ): """Overwrites lora tensors at index.""" ... def set_mapping( self, punica_wrapper, ): self.punica_wrapper: PunicaWrapperBase = punica_wrapper @classmethod def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: """Returns True if the layer can be replaced by this LoRA layer.""" raise NotImplementedError
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/lora/layers/base.py", "license": "Apache License 2.0", "lines": 54, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/lora/layers/base_linear.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import torch from transformers import PretrainedConfig from vllm.config.lora import LoRAConfig from vllm.distributed.utils import divide from vllm.model_executor.layers.linear import ( ColumnParallelLinear, LinearBase, ReplicatedLinear, RowParallelLinear, ) from vllm.platforms import current_platform from .base import BaseLayerWithLoRA from .utils import _get_lora_device class BaseLinearLayerWithLoRA(BaseLayerWithLoRA): def __init__(self, base_layer: LinearBase): super().__init__() self.base_layer = base_layer self.input_size = self.base_layer.input_size # Ensure tp_size and tp_rank consistency with the base_layer. self.tp_size = self.base_layer.tp_size self.tp_rank = self.base_layer.tp_rank self.device = _get_lora_device(self.base_layer) self.output_slices: tuple[int, ...] self.output_size: int self.n_slices: int def create_lora_weights( self, max_loras: int, lora_config: LoRAConfig, model_config: PretrainedConfig | None = None, ) -> None: self.lora_config = lora_config # if isinstance(self.base_layer, ReplicatedLinear): lora_a_out_size = lora_config.max_lora_rank lora_b_out_size = self.output_size elif isinstance(self.base_layer, ColumnParallelLinear): lora_a_out_size = ( lora_config.max_lora_rank if not lora_config.fully_sharded_loras else divide(lora_config.max_lora_rank, self.tp_size) ) lora_b_out_size = self.output_size elif isinstance(self.base_layer, RowParallelLinear): lora_a_out_size = lora_config.max_lora_rank lora_b_out_size = ( self.output_size if not lora_config.fully_sharded_loras else divide(self.output_size, self.tp_size) ) else: raise NotImplementedError self.lora_a_stacked = tuple( torch.zeros( max_loras, 1, lora_a_out_size, self.input_size, dtype=lora_config.lora_dtype, device=self.device, ) for _ in range(self.n_slices) ) self.lora_b_stacked = tuple( torch.zeros( max_loras, 1, lora_b_out_size, lora_config.max_lora_rank, dtype=lora_config.lora_dtype, device=self.device, ) for _ in range(self.n_slices) ) self.output_slices = (self.lora_b_stacked[0].shape[2],) def reset_lora(self, index: int): for s_index in range(self.n_slices): self.lora_a_stacked[s_index][index] = 0 self.lora_b_stacked[s_index][index] = 0 def set_lora( self, index: int, lora_a: torch.Tensor | list[torch.Tensor], lora_b: torch.Tensor | list[torch.Tensor], ): # Except for QKVParallelLinearWithLoRA and # MergedColumnParallelLinearWithLoRA, all other linear LoRA layers # store weights in a tuple of size 1. These two layers will # override this function. assert isinstance(lora_a, torch.Tensor) assert isinstance(lora_b, torch.Tensor) assert ( len(self.lora_a_stacked) == len(self.lora_b_stacked) == self.n_slices == 1 ) self.reset_lora(index) if self.tp_size > 1: lora_a = self.slice_lora_a(lora_a) lora_b = self.slice_lora_b(lora_b) self.lora_a_stacked[0][index, 0, : lora_a.shape[0], : lora_a.shape[1]].copy_( lora_a, non_blocking=True ) self.lora_b_stacked[0][index, 0, : lora_b.shape[0], : lora_b.shape[1]].copy_( lora_b, non_blocking=True ) def apply(self, x: torch.Tensor, bias: torch.Tensor | None = None) -> torch.Tensor: output = self.base_layer.quant_method.apply(self.base_layer, x, bias) original_shape = output.shape if output.ndim == 3 else None # In transformers backend, x and output have extra batch dimension like # (1, seq_len, hidden_dim), while punica expects (seq_len, hidden_dim), # therefore we need to flatten the batch dimensions. if x.ndim == 3 and output.ndim == 3: output = output.flatten(0, 1) x = x.flatten(0, 1) lora_output: torch.Tensor | None = self.punica_wrapper.add_lora_linear( output, x, self.lora_a_stacked, self.lora_b_stacked, 1.0, self.output_slices ) if not current_platform.can_update_inplace(): output = lora_output # Reshape the flattened output back to its original shape, # as some MM encoders cannot handle flattened inputs. if original_shape is not None: output = output.reshape(original_shape) return output @property def weight(self) -> torch.Tensor: # unquantizedLinear if hasattr(self.base_layer, "weight"): return self.base_layer.weight # Compressed Tensor elif hasattr(self.base_layer, "weight_packed"): return self.base_layer.weight_packed # GPTQ/AWQ elif hasattr(self.base_layer, "qweight"): return self.base_layer.qweight # marlin elif hasattr(self.base_layer, "B"): return self.base_layer.B else: raise ValueError(f"Unsupported base layer: {self.base_layer}") @property def bias(self) -> torch.Tensor | None: if hasattr(self.base_layer, "bias"): return self.base_layer.bias else: return None
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/lora/layers/base_linear.py", "license": "Apache License 2.0", "lines": 147, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/lora/layers/column_parallel_linear.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import torch import torch.nn as nn from transformers import PretrainedConfig from vllm.config.lora import LoRAConfig from vllm.distributed import tensor_model_parallel_all_gather from vllm.distributed.utils import divide from vllm.model_executor.layers.linear import ( ColumnParallelLinear, MergedColumnParallelLinear, QKVParallelLinear, ) from vllm.platforms import current_platform from .base_linear import BaseLinearLayerWithLoRA from .utils import _fully_sharded_can_replace, _not_fully_sharded_can_replace def _mcp_apply(x, bias, layer: "ColumnParallelLinearWithLoRA"): """ For `ColumnParallelLinearWithLoRA` or classes that inherit from `ColumnParallelLinearWithLoRA`, they share the same `apply` logic. """ assert ( layer.n_slices == len(layer.lora_a_stacked) == len(layer.lora_b_stacked) == len(layer.output_slices) ) output = layer.base_layer.quant_method.apply(layer.base_layer, x, bias) x = x.view(-1, x.shape[-1]) output, out_orig_shape = output.view(-1, output.shape[-1]), output.shape # Since communication is needed, the buffer is directly initialized as a # tensor rather than a tuple of tensor. buffers = torch.zeros( (layer.n_slices, x.shape[0], layer.lora_a_stacked[0].shape[2]), dtype=torch.float32, device=x.device, ) shrunk_buffers: torch.Tensor | None = layer.punica_wrapper.add_shrink( buffers, x, layer.lora_a_stacked, 1.0 ) if not current_platform.can_update_inplace(): buffers = shrunk_buffers buffers = tensor_model_parallel_all_gather(buffers) lora_output: torch.Tensor | None = layer.punica_wrapper.add_expand( output, buffers, layer.lora_b_stacked, layer.output_slices, offset_start=0, add_input=True, ) if not current_platform.can_update_inplace(): output = lora_output output = output.view(*out_orig_shape) # now have column partitioned and packed output return output class ColumnParallelLinearWithLoRA(BaseLinearLayerWithLoRA): """ LoRA on top of ColumnParallelLinear layer. LoRA B is sliced for tensor parallelism. There are two types for the `base_layer`: 1. ColumnParallelLinear, e.g.`dense_h_to_4h` in `FalconForCausalLM`. 2. MergedColumnParallelLinear, e.g.`gate_up_proj` in `Phi3ForCausalLM`. """ def __init__(self, base_layer: ColumnParallelLinear) -> None: super().__init__(base_layer) # The base_layer type is ColumnParallelLinear or # MergedColumnParallelLinear, their weight sharding logic is # inconsistent when TP is greater than 1. self.is_merged_col_linear = type(base_layer) is MergedColumnParallelLinear self.output_size = self.base_layer.output_size_per_partition # There is only one LoRA layer self.n_slices = 1 def slice_lora_a(self, lora_a: torch.Tensor) -> torch.Tensor: return lora_a def slice_lora_b(self, lora_b: torch.Tensor) -> torch.Tensor: # Applicable to cases where the base_layer is # MergedColumnParallelLinear. if self.is_merged_col_linear: shard_size = self.output_size // 2 offset = lora_b.shape[0] // 2 left_weight = lora_b[ self.tp_rank * shard_size : (self.tp_rank + 1) * shard_size, : ] right_weight = lora_b[ offset + self.tp_rank * shard_size : offset + (self.tp_rank + 1) * shard_size, :, ] lora_b = torch.cat([left_weight, right_weight], dim=0) # Applicable to cases where the base_layer is # ColumnParallelLinear. else: shard_size = self.output_size start_idx = self.tp_rank * shard_size end_idx = (self.tp_rank + 1) * shard_size lora_b = lora_b[start_idx:end_idx, :] return lora_b def forward( self, input_: torch.Tensor ) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor | None]: """Forward of ColumnParallelLinear Args: input_: Tensor whose last dimension is `input_size`. Returns: - output - bias """ bias = self.base_layer.bias if not self.base_layer.skip_bias_add else None # Matrix multiply. output_parallel = self.apply(input_, bias) if self.base_layer.gather_output and self.tp_size > 1: # All-gather across the partitions. output = tensor_model_parallel_all_gather(output_parallel) else: output = output_parallel if not self.base_layer.return_bias: return output output_bias = self.base_layer.bias if self.base_layer.skip_bias_add else None return output, output_bias @classmethod @_not_fully_sharded_can_replace def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: if type(source_layer) is ColumnParallelLinear: return True if type(source_layer) is MergedColumnParallelLinear: if len(packed_modules_list) != 1: return False # Exclude layers with 3+ output sizes - those are handled by # MergedColumnParallelLinearVariableSliceWithLoRA since this # class's slice_lora_b assumes exactly 2 slices. return not ( hasattr(source_layer, "output_sizes") and len(source_layer.output_sizes) >= 3 ) return False class MergedColumnParallelLinearWithLoRA(ColumnParallelLinearWithLoRA): """ColumnParallelLinear layer that is composed of 2 sublayers (slices) packed together (e.g. gate_proj + up_proj -> gate_up_proj). This means we have 2 LoRAs, each applied to one half of the layer. Both slices must have the same size. """ def __init__( self, base_layer: MergedColumnParallelLinear | QKVParallelLinear ) -> None: super().__init__(base_layer) # There are two LoRA layers # the output_sizes in MergedColumnParallelLinear is not sharded by tp # we need to divide it by the tp_size to get correct slices size output_sizes = self.base_layer.output_sizes self.output_slices = tuple( divide(output_size, self.tp_size) for output_size in output_sizes ) self.n_slices = len(self.output_slices) self.output_ids = (self.tp_rank,) * self.n_slices def create_lora_weights( self, max_loras: int, lora_config: LoRAConfig, model_config: PretrainedConfig | None = None, ) -> None: """ The main reason for overriding this function is to enhance code maintainability. """ self.lora_config = lora_config lora_a_output_size_per_partition = ( lora_config.max_lora_rank if not lora_config.fully_sharded_loras else divide(lora_config.max_lora_rank, self.tp_size) ) self.lora_a_stacked = tuple( torch.zeros( max_loras, 1, lora_a_output_size_per_partition, self.input_size, dtype=lora_config.lora_dtype, device=self.device, ) for _ in range(self.n_slices) ) self.lora_b_stacked = tuple( torch.zeros( max_loras, 1, output_size, lora_config.max_lora_rank, dtype=lora_config.lora_dtype, device=self.device, ) for output_size in self.output_slices ) def slice_lora_a( self, lora_a: list[torch.Tensor | None] ) -> list[torch.Tensor | None]: return lora_a def slice_lora_b( self, lora_b: list[torch.Tensor | None] ) -> list[torch.Tensor | None]: sliced_lora_b = [None] * self.n_slices for i, (shard_id, shard_size) in enumerate( zip(self.output_ids, self.output_slices) ): if (lora_b_i := lora_b[i]) is not None: sliced_lora_b[i] = lora_b_i[ shard_size * shard_id : shard_size * (shard_id + 1), : ] return sliced_lora_b def set_lora( self, index: int, lora_a: torch.Tensor | list[torch.Tensor], lora_b: torch.Tensor | list[torch.Tensor], ): self.reset_lora(index) if self.tp_size > 1: lora_a = self.slice_lora_a(lora_a) lora_b = self.slice_lora_b(lora_b) for i in range(self.n_slices): if (lora_a_i := lora_a[i]) is not None: self.lora_a_stacked[i][ index, 0, : lora_a_i.shape[0], : lora_a_i.shape[1] ].copy_(lora_a_i, non_blocking=True) if (lora_b_i := lora_b[i]) is not None: self.lora_b_stacked[i][ index, 0, : lora_b_i.shape[0], : lora_b_i.shape[1] ].copy_(lora_b_i, non_blocking=True) @classmethod @_not_fully_sharded_can_replace def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: return ( type(source_layer) is MergedColumnParallelLinear and len(packed_modules_list) == 2 ) class QKVParallelLinearWithLoRA(ColumnParallelLinearWithLoRA): """ ColumnParallelLinear layer that is specifically designed for qkv_proj. Certain models, such as chatglm3 and baichuan-7b, only contains a single LoRA within their qkv_proj layer. During inference with Tensor Parallel, the weights of lora_b must be accurately partitioned according to the respective ranks. Q slice may have different shape than K and V slices (which both have the same shape). """ def __init__(self, base_layer: QKVParallelLinear) -> None: super().__init__(base_layer) self.q_proj_total_size = ( self.base_layer.total_num_heads * self.base_layer.head_size ) self.q_proj_shard_size = self.base_layer.num_heads * self.base_layer.head_size self.kv_proj_shard_size = ( self.base_layer.num_kv_heads * self.base_layer.head_size ) self.kv_proj_total_size = ( self.base_layer.total_num_kv_heads * self.base_layer.head_size ) # There is only one LoRA layer self.n_slices = 1 def slice_lora_b(self, lora_b: torch.Tensor) -> torch.Tensor: self.q_shard_id = self.tp_rank self.kv_shard_id = self.tp_rank // self.base_layer.num_kv_head_replicas lora_b_q = lora_b[ self.q_proj_shard_size * self.q_shard_id : self.q_proj_shard_size * (self.q_shard_id + 1), :, ] k_offset = self.q_proj_total_size lora_b_k = lora_b[ k_offset + self.kv_proj_shard_size * self.kv_shard_id : k_offset + self.kv_proj_shard_size * (self.kv_shard_id + 1), :, ] v_offset = k_offset + self.kv_proj_total_size lora_b_v = lora_b[ v_offset + self.kv_proj_shard_size * self.kv_shard_id : v_offset + self.kv_proj_shard_size * (self.kv_shard_id + 1), :, ] lora_b = torch.cat([lora_b_q, lora_b_k, lora_b_v], dim=0) return lora_b @classmethod @_not_fully_sharded_can_replace def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: return type(source_layer) is QKVParallelLinear and len(packed_modules_list) == 1 class MergedQKVParallelLinearWithLoRA(MergedColumnParallelLinearWithLoRA): """MergedColumnParallelLinear layer that is composed of 3 sublayers (slices) packed together in qkv proj fashion (q_proj + k_proj + v_proj -> qkv_proj). This means we have 3 LoRAs, each applied to one slice of the layer. Q slice may have different shape than K and V slices (which both have the same shape). """ def __init__(self, base_layer: QKVParallelLinear) -> None: super().__init__(base_layer) # There are three LoRA layer. self.n_slices = len(self.base_layer.output_sizes) self.q_proj_shard_size = self.base_layer.num_heads * self.base_layer.head_size self.kv_proj_shard_size = ( self.base_layer.num_kv_heads * self.base_layer.head_size ) self.q_shard_id = self.tp_rank self.kv_shard_id = self.tp_rank // self.base_layer.num_kv_head_replicas self.output_slices = ( self.q_proj_shard_size, self.kv_proj_shard_size, self.kv_proj_shard_size, ) self.output_ids = ( self.q_shard_id, self.kv_shard_id, self.kv_shard_id, ) def create_lora_weights( self, max_loras: int, lora_config: LoRAConfig, model_config: PretrainedConfig | None = None, ) -> None: """ The main reason for overloading this function is to handle inconsistent weight dimensions in qkv lora. """ super().create_lora_weights(max_loras, lora_config, model_config) @classmethod @_not_fully_sharded_can_replace def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: return type(source_layer) is QKVParallelLinear and len(packed_modules_list) == 3 # These following layers are based on the tensor parallelism strategy given in # Y. Sheng et al., S-LoRA: Serving Thousands of Concurrent LoRA Adapters. 2023, # https://arxiv.org/abs/2311.03285. class ColumnParallelLinearWithShardedLoRA(ColumnParallelLinearWithLoRA): """ Differs from ColumnParallelLinearWithLoRA by slicing LoRA A also. Based on S-LoRA, slicing happens along the rank dim. """ # For all LoRA layers where the `base_layer` is `ColumnParallelLinear`, # their `lora_a` and `lora_b` have different sharding patterns. After # completing the `lora_a` GEMM , a gather operation is performed. # Therefore, the sharding of `lora_a` only needs to correspond with the # gather operation. def slice_lora_a(self, lora_a: torch.Tensor) -> torch.Tensor: shard_size = self.lora_a_stacked[0].shape[2] start_idx = self.tp_rank * shard_size lora_a = lora_a[start_idx : start_idx + shard_size, :] return lora_a def apply(self, x: torch.Tensor, bias: torch.Tensor | None = None) -> torch.Tensor: return _mcp_apply(x, bias, self) @classmethod @_fully_sharded_can_replace def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: # specifying kwargs so they can be easily accessed in decorator return super().can_replace_layer( source_layer=source_layer, lora_config=lora_config, packed_modules_list=packed_modules_list, model_config=model_config, decorate=False, ) class MergedColumnParallelLinearWithShardedLoRA(MergedColumnParallelLinearWithLoRA): """ Differs from MergedColumnParallelLinearWithLoRA by slicing the LoRA A's also. Based on S-LoRA, slicing happens along the rank dim. """ def slice_lora_a( self, lora_a: list[torch.Tensor | None] ) -> list[torch.Tensor | None]: # NOTE: lora_a contains 2 subloras, and each sublora could be None. output_shard_size = self.lora_a_stacked[0].shape[2] output_start_idx = self.tp_rank * output_shard_size lora_a = [ lora_a[0][output_start_idx : output_start_idx + output_shard_size, :] if lora_a[0] is not None else None, lora_a[1][output_start_idx : output_start_idx + output_shard_size, :] if lora_a[1] is not None else None, ] return lora_a def apply(self, x: torch.Tensor, bias: torch.Tensor | None = None) -> torch.Tensor: return _mcp_apply(x, bias, self) @classmethod @_fully_sharded_can_replace def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: # specifying kwargs so they can be easily accessed in decorator return super().can_replace_layer( source_layer=source_layer, lora_config=lora_config, packed_modules_list=packed_modules_list, model_config=model_config, decorate=False, ) class QKVParallelLinearWithShardedLoRA(QKVParallelLinearWithLoRA): """ Differs from QKVParallelLinearWithLoRA by slicing the LoRA A's also. Based on S-LoRA, slicing happens along the rank dim. """ def slice_lora_a(self, lora_a: torch.Tensor) -> torch.Tensor: shard_size = self.lora_a_stacked[0].shape[2] start_idx = self.tp_rank * shard_size lora_a = lora_a[start_idx : start_idx + shard_size, :] return lora_a def apply(self, x: torch.Tensor, bias: torch.Tensor | None = None) -> torch.Tensor: return _mcp_apply(x, bias, self) @classmethod @_fully_sharded_can_replace def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: # specifying kwargs so they can be easily accessed in decorator return super().can_replace_layer( source_layer=source_layer, lora_config=lora_config, packed_modules_list=packed_modules_list, model_config=model_config, decorate=False, ) class MergedQKVParallelLinearWithShardedLoRA(MergedQKVParallelLinearWithLoRA): """ Differs from MergedQKVParallelLinearWithLoRA by slicing the LoRA A's also. Based on S-LoRA, slicing happens along the rank dim. """ def slice_lora_a( self, lora_a: list[torch.Tensor | None] ) -> list[torch.Tensor | None]: # NOTE: lora_a contains 3 subloras, and each sublora could be None. shard_size = [self.lora_a_stacked[i].shape[2] for i in range(3)] start_idx = [self.tp_rank * shard_size[i] for i in range(3)] lora_a = [ lora_a[0][start_idx[0] : start_idx[0] + shard_size[0], :] if lora_a[0] is not None else None, lora_a[1][start_idx[1] : start_idx[1] + shard_size[1], :] if lora_a[1] is not None else None, lora_a[2][start_idx[2] : start_idx[2] + shard_size[2], :] if lora_a[2] is not None else None, ] return lora_a def apply(self, x: torch.Tensor, bias: torch.Tensor | None = None) -> torch.Tensor: return _mcp_apply(x, bias, self) @classmethod @_fully_sharded_can_replace def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: # specifying kwargs so they can be easily accessed in decorator return super().can_replace_layer( source_layer=source_layer, lora_config=lora_config, packed_modules_list=packed_modules_list, model_config=model_config, decorate=False, ) class MergedColumnParallelLinearVariableSliceWithLoRA( MergedColumnParallelLinearWithLoRA ): """MergedColumnParallelLinear with variable number of slices (3+). This handles cases where the checkpoint has a single weight for the whole module (not split into slices), but the layer itself has multiple slices. """ @classmethod @_not_fully_sharded_can_replace def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: # Support MergedColumnParallelLinear with 3 or more slices # (2 slices are handled by MergedColumnParallelLinearWithLoRA) if type(source_layer) is not MergedColumnParallelLinear: return False # If packed_modules_list has 3+ items, use this class if len(packed_modules_list) >= 3: return True # If packed_modules_list has exactly 2 items, let # MergedColumnParallelLinearWithLoRA handle it if len(packed_modules_list) == 2: return False # If packed_modules_list is empty or has 1 item, # check the layer's output_sizes. # This handles cases where the checkpoint has a single weight # but the layer has multiple slices (3+) return ( hasattr(source_layer, "output_sizes") and len(source_layer.output_sizes) >= 3 ) def set_lora( self, index: int, lora_a: torch.Tensor | list[torch.Tensor], lora_b: torch.Tensor | list[torch.Tensor], ): """Override to handle single tensor weights that need to be split into slices.""" self.reset_lora(index) # Handle case where checkpoint has single tensor weights # lora_a shape: (rank, input_size) - same for all slices, duplicate it if isinstance(lora_a, torch.Tensor): lora_a = [lora_a] * self.n_slices # lora_b shape: (total_output_size, rank) - # split along dim 0 based on output_sizes if isinstance(lora_b, torch.Tensor): output_sizes = self.base_layer.output_sizes lora_b_list = [] start_idx = 0 for output_size in output_sizes: end_idx = start_idx + output_size lora_b_list.append(lora_b[start_idx:end_idx, :]) start_idx = end_idx lora_b = lora_b_list # Now call parent's set_lora which expects lists super().set_lora(index, lora_a, lora_b)
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/lora/layers/column_parallel_linear.py", "license": "Apache License 2.0", "lines": 563, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/lora/layers/logits_processor.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import torch import torch.nn as nn from transformers import PretrainedConfig from vllm.config.lora import LoRAConfig from vllm.distributed import ( get_tensor_model_parallel_rank, get_tensor_model_parallel_world_size, ) from vllm.model_executor.layers.logits_processor import LogitsProcessor from vllm.model_executor.layers.vocab_parallel_embedding import VocabParallelEmbedding from vllm.platforms import current_platform from .base import BaseLayerWithLoRA class LogitsProcessorWithLoRA(BaseLayerWithLoRA): """ LoRA wrapper for LogitsProcessor, with extra logic to handle the application of the LoRA adapter and added LoRA vocabulary. Args: base_layer: LogitsProcessor layer hidden_size: hidden size of the model dtype: data type of the model device: device of the model sharded_to_full_mapping: index mapping from sharded vocab to full vocab received from base_layer.get_sharded_to_full_mapping(). If None, no reindexing will be done. """ def __init__( self, base_layer: LogitsProcessor, hidden_size: int, dtype: torch.dtype, device: torch.device, sharded_to_full_mapping: list[int] | None, ) -> None: super().__init__() self.base_layer = base_layer self.hidden_size = hidden_size self.dtype = dtype self.device = device self.tp_size = get_tensor_model_parallel_world_size() self.tp_rank = get_tensor_model_parallel_rank() self.sharded_to_full_mapping = sharded_to_full_mapping @property def logits_as_input(self): return self.base_layer.logits_as_input @property def vocab_size(self): return self.base_layer.vocab_size @property def scale(self): return self.base_layer.scale @property def soft_cap(self): return self.base_layer.soft_cap @property def use_all_gather(self): return self.base_layer.use_all_gather @property def org_vocab_size(self): return self.base_layer.org_vocab_size @property def include_gpu_probs_tensor(self): return self.base_layer.include_gpu_probs_tensor @property def should_modify_greedy_probs_inplace(self): return self.base_layer.should_modify_greedy_probs_inplace def create_lora_weights( self, max_loras: int, lora_config: LoRAConfig, model_config: PretrainedConfig | None = None, ) -> None: # TODO: Verify if this condition can be further relaxed if self.base_layer.vocab_size > 258048: raise ValueError("When using LoRA, vocab size must be <= 258048") self.lora_a_stacked = torch.zeros( ( max_loras, 1, lora_config.max_lora_rank, self.hidden_size, ), dtype=lora_config.lora_dtype, device=self.device, ) self.lora_b_stacked = torch.zeros( ( max_loras, 1, self.base_layer.vocab_size, lora_config.max_lora_rank, ), dtype=lora_config.lora_dtype, device=self.device, ) if self.sharded_to_full_mapping is not None: self.sharded_to_full_mapping_gpu = torch.tensor( self.sharded_to_full_mapping, device=self.device, dtype=torch.long ) else: self.sharded_to_full_mapping_gpu = None def reset_lora(self, index: int): self.lora_a_stacked[index] = 0 self.lora_b_stacked[index] = 0 def set_lora( self, index: int, lora_a: torch.Tensor | list[torch.Tensor], lora_b: torch.Tensor | list[torch.Tensor], ): assert isinstance(lora_a, torch.Tensor) assert isinstance(lora_b, torch.Tensor) self.reset_lora(index) self.lora_a_stacked[index, 0, : lora_a.shape[0], : lora_a.shape[1]].copy_( lora_a, non_blocking=True ) self.lora_b_stacked[index, 0, : lora_b.shape[0], : lora_b.shape[1]].copy_( lora_b, non_blocking=True ) def _get_logits( self, hidden_states: torch.Tensor, lm_head: VocabParallelEmbedding, embedding_bias: torch.Tensor | None = None, ) -> torch.Tensor | None: # Get the logits for the next tokens. if hasattr(lm_head, "base_layer"): actual_lm_head = lm_head.base_layer else: actual_lm_head = lm_head logits = actual_lm_head.quant_method.apply(actual_lm_head, hidden_states) if embedding_bias is not None: logits += embedding_bias # Gather logits for TP logits = self.base_layer._gather_logits(logits) if logits is None: return None if self.sharded_to_full_mapping_gpu is not None: # Reindex full logits tensor to ensure 1:1 mapping between # index and token_id # Example for: # org_vocab_size = 4 # added_vocab_size = 2 # pad_to_size = 8 # tp_size = 2 # indices: [0, 1, 2, 3, 4, 5, 6, 7] # token_id: [0, 1, 4, -1, 2, 3, 5, -1] # Therefore, the mapping is expected to be: # [0, 1, 4, 6, 2, 3, 5, 7] so that when we reindex, # we get: # indices: [0, 1, 2, 3, 4, 5, 6, 7] # token_id: [0, 1, 2, 3, 4, 5, -1, -1] logits = logits[:, self.sharded_to_full_mapping_gpu] lora_output: torch.Tensor | None = self.punica_wrapper.add_lora_logits( logits, hidden_states, self.lora_a_stacked, self.lora_b_stacked, 1.0 ) if not current_platform.can_update_inplace(): logits = lora_output # Remove paddings in vocab (if any). logits = logits[:, : self.base_layer.vocab_size] return logits def forward(self, *args, **kwargs): return type(self.base_layer).forward(self, *args, **kwargs) @classmethod def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: # Special handling for the LogitsProcessor. return False
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/lora/layers/logits_processor.py", "license": "Apache License 2.0", "lines": 174, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/lora/layers/replicated_linear.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import torch import torch.nn as nn from transformers import PretrainedConfig from vllm.config.lora import LoRAConfig from vllm.model_executor.layers.linear import ReplicatedLinear from .base_linear import BaseLinearLayerWithLoRA class ReplicatedLinearWithLoRA(BaseLinearLayerWithLoRA): def __init__(self, base_layer: ReplicatedLinear) -> None: super().__init__( base_layer, ) # To ensure interface compatibility, set to 1 always. self.output_size = self.base_layer.output_size self.n_slices = 1 def forward( self, input_: torch.Tensor ) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor | None]: """Forward of ReplicatedLinearWithLoRA Args: input_: Tensor whose last dimension is `input_size`. Returns: - output - bias """ bias = self.base_layer.bias if not self.base_layer.skip_bias_add else None # Matrix multiply. output = self.apply(input_, bias) output_bias = self.base_layer.bias if self.base_layer.skip_bias_add else None if not self.base_layer.return_bias: return output return output, output_bias # ReplicatedLinear should always be replaced, regardless of the fully # sharded LoRAs setting, because it is, by definition, copied per GPU. @classmethod def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: return type(source_layer) is ReplicatedLinear def slice_lora_a( self, lora_a: torch.Tensor | list[torch.Tensor | None] ) -> torch.Tensor | list[torch.Tensor | None]: """Slice lora a if splitting for tensor parallelism.""" return lora_a def slice_lora_b( self, lora_b: torch.Tensor | list[torch.Tensor | None] ) -> torch.Tensor | list[torch.Tensor | None]: """Slice lora b if splitting with tensor parallelism.""" return lora_b
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/lora/layers/replicated_linear.py", "license": "Apache License 2.0", "lines": 54, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/lora/layers/row_parallel_linear.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import torch import torch.nn as nn from transformers import PretrainedConfig from vllm.config.lora import LoRAConfig from vllm.distributed import ( split_tensor_along_last_dim, tensor_model_parallel_all_reduce, ) from vllm.model_executor.layers.linear import RowParallelLinear from vllm.platforms import current_platform from .base_linear import BaseLinearLayerWithLoRA from .utils import _fully_sharded_can_replace, _not_fully_sharded_can_replace class RowParallelLinearWithLoRA(BaseLinearLayerWithLoRA): def __init__(self, base_layer: RowParallelLinear) -> None: super().__init__(base_layer) # reset input_size self.input_size = self.base_layer.input_size_per_partition self.output_size = self.base_layer.output_size # There is only one LoRA layer. self.n_slices = 1 def slice_lora_a(self, lora_a: torch.Tensor) -> torch.Tensor: shard_size = self.input_size start_idx = self.tp_rank * shard_size end_idx = (self.tp_rank + 1) * shard_size lora_a = lora_a[:, start_idx:end_idx] return lora_a def slice_lora_b(self, lora_b: torch.Tensor) -> torch.Tensor: return lora_b def forward( self, input_: torch.Tensor ) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor | None]: """Forward of RowParallelLinear Args: input_: tensor whose last dimension is `input_size`. If `input_is_parallel` is set, then the last dimension is `input_size // tp_size`. Returns: - output - bias """ # set up backprop all-reduce. if self.base_layer.input_is_parallel: input_parallel = input_ else: # TODO: simplify code below splitted_input = split_tensor_along_last_dim( input_, num_partitions=self.tp_size ) input_parallel = splitted_input[self.tp_rank].contiguous() # Matrix multiply. bias_ = ( None if (self.tp_rank > 0 or self.base_layer.skip_bias_add) else self.base_layer.bias ) output_parallel = self.apply(input_parallel, bias_) if self.base_layer.reduce_results and self.tp_size > 1: output = tensor_model_parallel_all_reduce(output_parallel) else: output = output_parallel output_bias = self.base_layer.bias if self.base_layer.skip_bias_add else None if not self.base_layer.return_bias: return output return output, output_bias @classmethod @_not_fully_sharded_can_replace def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: return type(source_layer) is RowParallelLinear # The following layer is based on the tensor parallelism strategy given in # Y. Sheng et al., S-LoRA: Serving Thousands of Concurrent LoRA Adapters. 2023, # https://arxiv.org/abs/2311.03285. class RowParallelLinearWithShardedLoRA(RowParallelLinearWithLoRA): """ Differs from RowParallelLinearWithLoRA by slicing the LoRA B's also. Based on S-LoRA, slicing happens along the output dim. This yields a combined partial sum from the row parallel base layer and column partitioned output from the LoRA. """ def slice_lora_b(self, lora_b: torch.Tensor) -> torch.Tensor: shard_size = self.lora_b_stacked[0].shape[2] start_idx = self.tp_rank * shard_size end_idx = (self.tp_rank + 1) * shard_size lora_b = lora_b[start_idx:end_idx, :] return lora_b def apply(self, x: torch.Tensor, bias: torch.Tensor | None = None) -> torch.Tensor: output = self.base_layer.quant_method.apply(self.base_layer, x, bias) x = x.view(-1, x.shape[-1]) output, out_orig_shape = output.view(-1, output.shape[-1]), output.shape buffer = torch.zeros( (self.n_slices, x.shape[0], self.lora_a_stacked[0].shape[2]), dtype=torch.float32, device=x.device, ) shrunk_buffer: torch.Tensor | None = self.punica_wrapper.add_shrink( buffer, x, self.lora_a_stacked, 1.0 ) if not current_platform.can_update_inplace(): buffer = shrunk_buffer if self.tp_size > 1: buffer = tensor_model_parallel_all_reduce(buffer) # following S-LoRA, allows the fusing of all_gather and all_reduce # by adding the column partitioned lora output to a slice of output # tensor, which is a partial sum due to row parallel. All that # remains is a standard all_reduce. User should be aware though that # the output is not the same as a normal row_parallel, it should be # reduced before being used # NOTE offset are based on the rank. shard_size = self.lora_b_stacked[0].shape[2] offset_start = self.tp_rank * shard_size lora_output: torch.Tensor | None = self.punica_wrapper.add_expand( output, buffer, self.lora_b_stacked, self.output_slices, offset_start=offset_start, add_input=True, ) if not current_platform.can_update_inplace(): output = lora_output output = output.view(*out_orig_shape) return output @classmethod @_fully_sharded_can_replace def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: # specifying kwargs so they can be easily accessed in decorator return super().can_replace_layer( source_layer=source_layer, lora_config=lora_config, packed_modules_list=packed_modules_list, model_config=model_config, decorate=False, )
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/lora/layers/row_parallel_linear.py", "license": "Apache License 2.0", "lines": 147, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/lora/layers/utils.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from dataclasses import dataclass from enum import Enum import torch import torch.nn as nn from vllm.model_executor.layers.fused_moe.fused_moe import try_get_optimal_moe_config from vllm.utils.math_utils import next_power_of_2 class LoRAMappingType(Enum): LANGUAGE = 1 TOWER = 2 CONNECTOR = 3 @dataclass class LoRAMapping: index_mapping: tuple[int, ...] prompt_mapping: tuple[int, ...] is_prefill: bool = False type: LoRAMappingType = LoRAMappingType.LANGUAGE def __post_init__(self): self.index_mapping = tuple(self.index_mapping) self.prompt_mapping = tuple(self.prompt_mapping) def _get_lora_device(base_layer: nn.Module) -> torch.device: # code borrowed from https://github.com/fmmoret/vllm/blob/fm-support-lora-on-quantized-models/vllm/lora/layers.py#L34 """Returns the device for where to place the LoRA tensors.""" # unquantizedLinear if hasattr(base_layer, "weight"): return base_layer.weight.device # Compressed Tensor elif hasattr(base_layer, "weight_packed"): return base_layer.weight_packed.device # GPTQ/AWQ elif hasattr(base_layer, "qweight"): return base_layer.qweight.device # MoE layer elif hasattr(base_layer, "w2_weight"): return base_layer.w2_weight.device # MoE Compressed Tensor elif hasattr(base_layer, "w2_weight_packed"): return base_layer.w2_weight_packed.device # MoE GPTQ/AWQ/GGUF elif hasattr(base_layer, "w2_qweight"): return base_layer.w2_qweight.device else: raise ValueError(f"Unsupported base layer: {base_layer}") def _not_fully_sharded_can_replace(can_replace): """ decorator which adds the condition of not using fully sharded loras intended to wrap can_replace_layer() """ def dec(*args, **kwargs): decorate = kwargs.pop("decorate") if "decorate" in kwargs else True condition = not kwargs["lora_config"].fully_sharded_loras if decorate else True return can_replace(*args, **kwargs) and condition return dec def _fully_sharded_can_replace(can_replace): """ decorator which adds the condition of fully sharded loras intended to wrap can_replace_layer() """ def dec(*args, **kwargs): return ( can_replace(*args, **kwargs) and kwargs["lora_config"].fully_sharded_loras ) return dec def try_get_optimal_moe_lora_config( op_type: str, w1_shape: tuple[int, ...], w2_shape: tuple[int, ...], rank: int, top_k: int, dtype: str | None, M: int, block_shape: list[int] | None = None, ) -> dict[str, int | None]: config = try_get_optimal_moe_config( w1_shape, w2_shape, top_k, dtype, M, block_shape ).copy() if op_type in [ "fused_moe_lora_w13_shrink", "fused_moe_lora_w2_shrink", ]: config["BLOCK_SIZE_N"] = min( config.get("BLOCK_SIZE_N", 64), next_power_of_2(rank) ) elif op_type in [ "fused_moe_lora_w13_expand", "fused_moe_lora_w2_expand", ]: config["BLOCK_SIZE_K"] = max( 16, min(config.get("BLOCK_SIZE_K", 32), next_power_of_2(rank)) ) return config
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/lora/layers/utils.py", "license": "Apache License 2.0", "lines": 92, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/lora/layers/vocal_parallel_embedding.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import torch import torch.nn as nn import torch.nn.functional as F from transformers import PretrainedConfig from vllm.config.lora import LoRAConfig from vllm.model_executor.layers.vocab_parallel_embedding import VocabParallelEmbedding from vllm.platforms import current_platform from .base import BaseLayerWithLoRA class VocabParallelEmbeddingWithLoRA(BaseLayerWithLoRA): def __init__(self, base_layer: VocabParallelEmbedding) -> None: super().__init__() self.base_layer = base_layer self.embeddings_slice: tuple[int, int] | None self.embeddings_weights: torch.Tensor | None def create_lora_weights( self, max_loras: int, lora_config: LoRAConfig, model_config: PretrainedConfig | None = None, ) -> None: if self.base_layer.num_added_embeddings_per_partition > 0: # We can start adding lora weights self.embeddings_weights = self.base_layer.weight.data[ self.base_layer.num_org_embeddings_per_partition : self.base_layer.num_org_embeddings_per_partition # noqa: E501 + self.base_layer.num_added_embeddings_per_partition ] self.embeddings_slice = ( self.base_layer.shard_indices.added_vocab_start_index - self.base_layer.org_vocab_size, self.base_layer.shard_indices.added_vocab_end_index - self.base_layer.org_vocab_size, ) self.base_layer.weight.data[ self.base_layer.num_org_embeddings_per_partition : ].fill_(0) else: self.embeddings_slice = None self.embeddings_weights = None self.lora_a_stacked = torch.zeros( ( max_loras, self.base_layer.org_vocab_size, lora_config.max_lora_rank, ), dtype=lora_config.lora_dtype, device=self.base_layer.weight.device, ) self.lora_b_stacked = torch.zeros( ( max_loras, 1, self.base_layer.embedding_dim, lora_config.max_lora_rank, ), dtype=lora_config.lora_dtype, device=self.base_layer.weight.device, ) self.lora_a_stacked_2d = self.lora_a_stacked.view( self.lora_a_stacked.shape[0] * self.lora_a_stacked.shape[1], self.lora_a_stacked.shape[2], ) def reset_lora(self, index: int): self.lora_a_stacked[index] = 0 self.lora_b_stacked[index] = 0 def set_lora( self, index: int, lora_a: torch.Tensor | list[torch.Tensor], lora_b: torch.Tensor | list[torch.Tensor], ): assert isinstance(lora_a, torch.Tensor) assert isinstance(lora_b, torch.Tensor) self.reset_lora(index) # NOTE self.lora_a_stacked is row-major, and lora_a is col-major, # so we need transpose here self.lora_a_stacked[index, : lora_a.shape[1], : lora_a.shape[0]].copy_( lora_a.T, non_blocking=True ) self.lora_b_stacked[index, 0, : lora_b.shape[0], : lora_b.shape[1]].copy_( lora_b, non_blocking=True ) def forward(self, x: torch.Tensor) -> torch.Tensor: # NB: Don't use torch.narrow here. torch.narrow triggers some # Dynamic Shape specialization in torch.compile num_tokens = x.shape[0] indices_1 = self.punica_wrapper._embeddings_indices[1][:num_tokens] full_lora_a_embeddings = F.embedding( x + indices_1, self.lora_a_stacked_2d, ) full_output = self.base_layer.forward(x) full_output_org = full_output if full_output.ndim == 3: full_output = full_output.view( full_output.shape[0] * full_output.shape[1], -1 ) if full_lora_a_embeddings.ndim == 3: full_lora_a_embeddings = full_lora_a_embeddings.view( full_lora_a_embeddings.shape[0] * full_lora_a_embeddings.shape[1], -1, ) lora_output: torch.Tensor | None = self.punica_wrapper.add_lora_embedding( full_output, full_lora_a_embeddings, self.lora_b_stacked, add_input=True ) if not current_platform.can_update_inplace(): full_output = lora_output return full_output.view_as(full_output_org) @classmethod def can_replace_layer( cls, source_layer: nn.Module, lora_config: LoRAConfig, packed_modules_list: list, model_config: PretrainedConfig | None = None, ) -> bool: return type(source_layer) is VocabParallelEmbedding @property def weight(self): return self.base_layer.weight
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/lora/layers/vocal_parallel_embedding.py", "license": "Apache License 2.0", "lines": 121, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:tests/models/language/generation_ppl_test/ppl_utils.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project # Adapted from https://huggingface.co/docs/transformers/perplexity from typing import cast import torch from datasets import load_dataset import tests.ci_envs as ci_envs from tests.models.utils import ( GenerateModelInfo, TokensTextLogprobsPromptLogprobs, get_vllm_extra_kwargs, ) from vllm.logprobs import Logprob # See #24485 PPL_TOL = 0.01 MAX_LENGTH = 1024 @torch.inference_mode def wikitext_ppl_test( hf_runner, vllm_runner, model_info: GenerateModelInfo, max_length=MAX_LENGTH, vllm_extra_kwargs=None, atol=PPL_TOL, ): vllm_extra_kwargs = get_vllm_extra_kwargs(model_info, vllm_extra_kwargs) dataset = load_dataset("wikitext", "wikitext-2-raw-v1", split="test") with vllm_runner( model_info.name, gpu_memory_utilization=0.7, max_model_len=max_length, max_num_seqs=1, **vllm_extra_kwargs, ) as vllm_model: # Use max_num_seqs=1 to avoid OOM, # and avoid batch different requests together. model_config = vllm_model.llm.llm_engine.model_config # Confirm whether vllm is using the correct architecture if model_info.architecture: assert model_info.architecture in model_config.architectures max_length = min(model_config.max_model_len - 1, max_length) stride = max_length tokenizer = vllm_model.llm.get_tokenizer() tokens = tokenizer.encode("\n\n".join(dataset["text"])) n_tokens = len(tokens) chunks = [] for begin_loc in range(0, n_tokens, stride): end_loc = min(begin_loc + max_length, n_tokens) chunks.append(tokens[begin_loc:end_loc]) outputs = vllm_model.generate_greedy_logprobs( prompts=chunks, max_tokens=1, num_logprobs=None, num_prompt_logprobs=0, use_tqdm=False, ) nll_sum = torch.tensor(0.0, dtype=torch.float32, device="cpu") n_tokens = 0 for output in outputs: output = cast(TokensTextLogprobsPromptLogprobs, output) token_datas = cast(list[dict[int, Logprob] | None], output[3]) assert token_datas[0] is None token_log_probs = [] for token_data in token_datas[1:]: assert token_data is not None assert len(token_data) == 1 token_log_prob = list(token_data.values())[0].logprob token_log_probs.append(token_log_prob) neg_log_likelihood = -torch.tensor( token_log_probs, dtype=torch.float32, device="cpu" ).sum() nll_sum += neg_log_likelihood n_tokens += len(token_log_probs) vllm_ppl = float(torch.exp(nll_sum / n_tokens)) vllm_dtype = model_config.dtype head_dtype = model_config.head_dtype # Accelerate ppl test by setting Transformers ppl score to a constant if model_info.hf_ppl is None: with hf_runner( model_info.name, dtype=ci_envs.VLLM_CI_HF_DTYPE or model_info.hf_dtype, ) as hf_model: nll_sum = torch.tensor(0.0, dtype=torch.float32, device="cpu") n_tokens = 0 for chunk in chunks: inputs = hf_model.wrap_device({"input_ids": torch.tensor([chunk])}) input_ids = inputs["input_ids"] outputs = hf_model.model(input_ids, labels=input_ids) neg_log_likelihood = outputs.loss neg_log_likelihood = neg_log_likelihood.to(torch.float32).cpu() num_loss_tokens = len(chunk) - 1 nll_sum += neg_log_likelihood * num_loss_tokens n_tokens += num_loss_tokens hf_ppl = float(torch.exp(nll_sum / n_tokens)) hf_dtype = next(hf_model.model.parameters()).dtype else: hf_ppl = model_info.hf_ppl hf_dtype = "Constant" differ = (vllm_ppl - hf_ppl) / hf_ppl print("Model:", model_info.name) print("VLLM:", f"dtype:{vllm_dtype}", f"head_dtype:{head_dtype}", vllm_ppl) print("Transformers:", hf_dtype, hf_ppl) print("Difference (%):", differ * 100) # PPL the smaller, the better # We are not concerned that the vllm PPL is less than Transformers, # so we only perform one-sided testing. assert differ < atol
{ "repo_id": "vllm-project/vllm", "file_path": "tests/models/language/generation_ppl_test/ppl_utils.py", "license": "Apache License 2.0", "lines": 107, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/models/language/generation_ppl_test/test_gemma.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import pytest from tests.models.utils import GenerateModelInfo from .ppl_utils import wikitext_ppl_test MODELS = [ GenerateModelInfo("google/gemma-2b"), GenerateModelInfo("google/gemma-2-2b"), GenerateModelInfo("google/gemma-3-4b-it"), ] @pytest.mark.parametrize("model_info", MODELS) def test_ppl(hf_runner, vllm_runner, model_info: GenerateModelInfo): wikitext_ppl_test(hf_runner, vllm_runner, model_info)
{ "repo_id": "vllm-project/vllm", "file_path": "tests/models/language/generation_ppl_test/test_gemma.py", "license": "Apache License 2.0", "lines": 13, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/models/language/generation_ppl_test/test_gpt.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import pytest from tests.models.utils import GenerateModelInfo from .ppl_utils import wikitext_ppl_test MODELS = [GenerateModelInfo("openai-community/gpt2-large")] @pytest.mark.parametrize("model_info", MODELS) def test_ppl(hf_runner, vllm_runner, model_info: GenerateModelInfo): wikitext_ppl_test(hf_runner, vllm_runner, model_info)
{ "repo_id": "vllm-project/vllm", "file_path": "tests/models/language/generation_ppl_test/test_gpt.py", "license": "Apache License 2.0", "lines": 9, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:tests/models/language/generation_ppl_test/test_qwen.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import pytest from tests.models.utils import GenerateModelInfo from .ppl_utils import wikitext_ppl_test MODELS = [ GenerateModelInfo("Qwen/Qwen3-0.6B"), GenerateModelInfo("Qwen/Qwen3-0.6B-FP8"), # transformers: # Loading a GPTQ quantized model requires optimum, gptqmodel # GenerateModelInfo("Qwen/Qwen3-0.6B-GPTQ-Int8"), ] @pytest.mark.parametrize("model_info", MODELS) def test_ppl(hf_runner, vllm_runner, model_info: GenerateModelInfo): wikitext_ppl_test(hf_runner, vllm_runner, model_info)
{ "repo_id": "vllm-project/vllm", "file_path": "tests/models/language/generation_ppl_test/test_qwen.py", "license": "Apache License 2.0", "lines": 15, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
test
vllm-project/vllm:vllm/config/load.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from typing import TYPE_CHECKING, Any from pydantic import Field, field_validator from vllm.config.utils import config from vllm.logger import init_logger from vllm.utils.hashing import safe_hash if TYPE_CHECKING: from vllm.model_executor.model_loader import LoadFormats from vllm.model_executor.model_loader.tensorizer import TensorizerConfig else: LoadFormats = Any TensorizerConfig = Any logger = init_logger(__name__) @config class LoadConfig: """Configuration for loading the model weights.""" load_format: str | LoadFormats = "auto" """The format of the model weights to load:\n - "auto" will try to load the weights in the safetensors format and fall back to the pytorch bin format if safetensors format is not available.\n - "pt" will load the weights in the pytorch bin format.\n - "safetensors" will load the weights in the safetensors format.\n - "npcache" will load the weights in pytorch format and store a numpy cache to speed up the loading.\n - "dummy" will initialize the weights with random values, which is mainly for profiling.\n - "tensorizer" will use CoreWeave's tensorizer library for fast weight loading. See the Tensorize vLLM Model script in the Examples section for more information.\n - "runai_streamer" will load the Safetensors weights using Run:ai Model Streamer.\n - "runai_streamer_sharded" will load weights from pre-sharded checkpoint files using Run:ai Model Streamer.\n - "bitsandbytes" will load the weights using bitsandbytes quantization.\n - "sharded_state" will load weights from pre-sharded checkpoint files, supporting efficient loading of tensor-parallel models.\n - "gguf" will load weights from GGUF format files (details specified in https://github.com/ggml-org/ggml/blob/master/docs/gguf.md).\n - "mistral" will load weights from consolidated safetensors files used by Mistral models. - Other custom values can be supported via plugins.""" download_dir: str | None = None """Directory to download and load the weights, default to the default cache directory of Hugging Face.""" safetensors_load_strategy: str = "lazy" """Specifies the loading strategy for safetensors weights. - "lazy" (default): Weights are memory-mapped from the file. This enables on-demand loading and is highly efficient for models on local storage. - "eager": The entire file is read into CPU memory upfront before loading. This is recommended for models on network filesystems (e.g., Lustre, NFS) as it avoids inefficient random reads, significantly speeding up model initialization. However, it uses more CPU RAM. - "torchao": Weights are loaded in upfront and then reconstructed into torchao tensor subclasses. This is used when the checkpoint was quantized using torchao and saved using safetensors. Needs torchao >= 0.14.0 """ model_loader_extra_config: dict | TensorizerConfig = Field(default_factory=dict) """Extra config for model loader. This will be passed to the model loader corresponding to the chosen load_format.""" device: str | None = None """Device to which model weights will be loaded, default to device_config.device""" ignore_patterns: list[str] | str = Field(default_factory=lambda: ["original/**/*"]) """The list of patterns to ignore when loading the model. Default to "original/**/*" to avoid repeated loading of llama's checkpoints.""" use_tqdm_on_load: bool = True """Whether to enable tqdm for showing progress bar when loading model weights.""" pt_load_map_location: str | dict[str, str] = "cpu" """ pt_load_map_location: the map location for loading pytorch checkpoint, to support loading checkpoints can only be loaded on certain devices like "cuda", this is equivalent to {"": "cuda"}. Another supported format is mapping from different devices like from GPU 1 to GPU 0: {"cuda:1": "cuda:0"}. Note that when passed from command line, the strings in dictionary needs to be double quoted for json parsing. For more details, see original doc for `map_location` in https://pytorch.org/docs/stable/generated/torch.load.html """ def compute_hash(self) -> str: """ WARNING: Whenever a new field is added to this config, ensure that it is included in the factors list if it affects the computation graph. Provide a hash that uniquely identifies all the configs that affect the structure of the computation graph from input ids/embeddings to the final hidden states, excluding anything before input ids/embeddings and after the final hidden states. """ # no factors to consider. # this config will not affect the computation graph. factors: list[Any] = [] hash_str = safe_hash(str(factors).encode(), usedforsecurity=False).hexdigest() return hash_str @field_validator("load_format", mode="after") def _lowercase_load_format(cls, load_format: str) -> str: return load_format.lower() @field_validator("ignore_patterns", mode="after") def _validate_ignore_patterns( cls, ignore_patterns: list[str] | str ) -> list[str] | str: if ignore_patterns != ["original/**/*"] and len(ignore_patterns) > 0: logger.info( "Ignoring the following patterns when downloading weights: %s", ignore_patterns, ) return ignore_patterns
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/config/load.py", "license": "Apache License 2.0", "lines": 109, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/model_executor/models/nano_nemotron_vl.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project # -------------------------------------------------------- # Adapted from # https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/internvl.py # under Apache-2.0 License # LICENSE is in root directory. # -------------------------------------------------------- import copy import math import warnings from abc import ABC, abstractmethod from collections.abc import Iterable, Mapping, Sequence from dataclasses import dataclass from functools import cached_property from typing import Annotated, Any, Literal, TypeAlias, TypeVar import einops import numpy.typing as npt import regex as re import torch import torch.nn as nn import torchvision.transforms as T from PIL import Image from transformers import BatchFeature, PretrainedConfig, TensorType from vllm.config import VllmConfig from vllm.config.multimodal import BaseDummyOptions, VideoDummyOptions from vllm.logger import init_logger from vllm.model_executor.layers.activation import ReLUSquaredActivation from vllm.model_executor.layers.layernorm import RMSNorm from vllm.model_executor.model_loader.weight_utils import default_weight_loader from vllm.model_executor.models.interfaces import ( HasInnerState, IsHybrid, MultiModalEmbeddings, SupportsMultiModal, SupportsMultiModalPruning, ) from vllm.model_executor.models.internvl import ( calculate_internvl_targets, get_internvl_target_ratios, ) from vllm.model_executor.models.module_mapping import MultiModelKeys from vllm.model_executor.models.nemotron_h import NemotronHForCausalLM from vllm.model_executor.models.parakeet import ParakeetExtractor, ProjectedParakeet from vllm.model_executor.models.radio import RadioModel, calc_seq_lens from vllm.model_executor.models.utils import ( init_vllm_registered_model, maybe_prefix, ) from vllm.multimodal import MULTIMODAL_REGISTRY from vllm.multimodal.evs import ( compute_retained_tokens_count, compute_retention_mask, ) from vllm.multimodal.inputs import ( AudioItem, MultiModalDataDict, MultiModalFieldConfig, MultiModalKwargsItems, VideoItem, ) from vllm.multimodal.parse import ( AudioProcessorItems, ImageEmbeddingItems, ImageProcessorItems, ImageSize, MultiModalDataItems, MultiModalDataParser, ) from vllm.multimodal.processing import BaseDummyInputsBuilder from vllm.multimodal.processing.processor import ( BaseMultiModalProcessor, BaseProcessingInfo, PromptReplacement, PromptUpdate, PromptUpdateDetails, _seq2tokens, ) from vllm.renderers import TokenizeParams from vllm.sequence import IntermediateTensors from vllm.tokenizers import TokenizerLike, cached_tokenizer_from_config from vllm.transformers_utils.configs.radio import RadioConfig from vllm.utils.tensor_schema import TensorSchema, TensorShape from .utils import _merge_multimodal_embeddings logger = init_logger(__name__) # Configure PIL to handle large images without warnings # This prevents DecompressionBombWarning for legitimate large images Image.MAX_IMAGE_PIXELS = None # Disable the limit entirely # Alternative: Set a specific higher limit # Image.MAX_IMAGE_PIXELS = 300000000 # ~300M pixels class NanoNemotronVLAudioFeatureInputs(TensorSchema): """ Dimensions: - b: Number of audio clips - t: Audio feature length - f: Feature size (mel bins) """ type: Literal["audio_features"] = "audio_features" input_audio_features: Annotated[torch.Tensor, TensorShape("b", "t", "f")] feature_attention_mask: Annotated[torch.Tensor, TensorShape("b", "t")] audio_feature_lengths: Annotated[torch.Tensor, TensorShape("b")] MAX_AUDIO_LEN_S = 10 * 60 # 10 minutes IMG_START = "<img>" IMG_END = "</img>" IMG_CONTEXT = "<image>" AUDIO_START = "<so_start>" AUDIO_END = "<so_end>" AUDIO_CONTEXT = "<so_embedding>" # Profiling # MAX_FRAMES = 16 DEFAULT_NUM_TILES = 12 class NanoNemotronVLImagePixelInputs(TensorSchema): """ Dimensions: - bn: Batch size * number of images - bnp: Batch size * number of images * (1 + num_patches) - c: Number of channels (3) - h: Height of each image patch - w: Width of each image patch """ type: Literal["pixel_values"] = "pixel_values" pixel_values_flat: Annotated[torch.Tensor, TensorShape("bnp", 3, "h", "w")] num_patches: Annotated[torch.Tensor, TensorShape("bn")] class NanoNemotronVLImagePixelInputsDynamic(TensorSchema): """ Dynamic-resolution image inputs. imgs_sizes: per-image (height, width) in pixels. num_tokens_per_image: per-image number of embedding tokens (post downsample). """ type: Literal["pixel_values_dynamic"] = "pixel_values_dynamic" pixel_values_flat: Annotated[torch.Tensor, TensorShape("bn", "h", "w")] imgs_sizes: list[tuple[int, int]] num_tokens_per_image: list[int] class NanoNemotronVLImageEmbeddingInputs(TensorSchema): """ Dimensions: - n: Number of images - f: Total image feature size - h: Hidden size (must match the hidden size of language model backbone) """ type: Literal["image_embeds"] data: Annotated[torch.Tensor | list[torch.Tensor], TensorShape("n", "f", "h")] NanoNemotronVLImageInputs: TypeAlias = ( NanoNemotronVLImagePixelInputs | NanoNemotronVLImagePixelInputsDynamic | NanoNemotronVLImageEmbeddingInputs ) class NanoNemotronVLVideoPixelInputs(TensorSchema): """ Dimensions: - bvf: Batch size * number of videos * num_frames - bn: Batch size * number of videos - f: Number of frames - c: Number of channels (3) - h: Height of each video frame - w: Width of each video frame """ type: Literal["pixel_values_videos"] pixel_values_flat: Annotated[torch.Tensor, TensorShape("bvf", 3, "h", "w")] num_patches: Annotated[torch.Tensor, TensorShape("bn")] frames_indices: Annotated[torch.Tensor, TensorShape("bvf")] frame_duration_ms: Annotated[torch.Tensor, TensorShape("bn")] class NanoNemotronVLVideoEmbeddingInputs(TensorSchema): """ Dimensions: - n: Number of videos - f: Total video feature size - h: Hidden size (must match the hidden size of language model backbone) """ type: Literal["video_embeds"] data: Annotated[torch.Tensor | list[torch.Tensor], TensorShape("n", "f", "h")] NanoNemotronVLVideoInputs: TypeAlias = ( NanoNemotronVLVideoPixelInputs | NanoNemotronVLVideoEmbeddingInputs ) def dynamic_preprocess( image, *, image_size=512, max_num_tiles=12, use_thumbnail=True, idx=0 ): orig_width, orig_height = image.size target_ratios = get_internvl_target_ratios(1, max_num_tiles) blocks, target_width, target_height = calculate_internvl_targets( orig_width=orig_width, orig_height=orig_height, target_ratios=target_ratios, image_size=image_size, use_thumbnail=False, ) # resize the image resized_img = image.resize((target_width, target_height)) processed_images = [] for i in range(blocks): box = ( (i % (target_width // image_size)) * image_size, (i // (target_width // image_size)) * image_size, ((i % (target_width // image_size)) + 1) * image_size, ((i // (target_width // image_size)) + 1) * image_size, ) # split the image split_img = resized_img.crop(box) processed_images.append(split_img) assert len(processed_images) == blocks if use_thumbnail and len(processed_images) != 1: thumbnail_img = image.resize((image_size, image_size)) processed_images.append(thumbnail_img) processed_images = [ img.convert("RGB") if img.mode != "RGB" else img for img in processed_images ] processed_images = [ T.Resize((image_size, image_size), interpolation=T.InterpolationMode.BICUBIC)( img ) for img in processed_images ] processed_images = [T.ToTensor()(img) for img in processed_images] return processed_images def image_to_pixel_values( image: Image.Image, *, input_size: int, max_num: int, use_thumbnail: bool, idx: int, ) -> torch.Tensor: images = dynamic_preprocess( image, image_size=input_size, max_num_tiles=max_num, use_thumbnail=use_thumbnail, idx=idx, ) pixel_values = torch.stack(images) return pixel_values def video_to_pixel_values( video: npt.NDArray, *, input_size: int, max_num_tiles: int = 1, use_thumbnail: bool, ) -> torch.Tensor: assert max_num_tiles == 1, "Video modality always uses one tile" # Convert each frame to a single resized tile tensor consistent # with image path frames_tensors: list[torch.Tensor] = [] for frame in video: pil_frame = dynamic_preprocess( Image.fromarray(frame, mode="RGB"), image_size=input_size, max_num_tiles=max_num_tiles, use_thumbnail=use_thumbnail, idx=0, ) # dynamic_preprocess returns tensors already; take the single tile assert len(pil_frame) >= 1 frames_tensors.append(pil_frame[-1]) return torch.stack(frames_tensors) def input_conditioner(x, norm_mean, norm_std): return (x - norm_mean) / norm_std def calculate_timestamps( indices: list[int] | torch.Tensor, frame_duration_ms: int, ): if not isinstance(indices, list): indices = indices.tolist() timestamps = [int(i) * frame_duration_ms / 1000.0 for i in indices] return timestamps class DynamicResolutionImageTiler: CONV_MERGING = False PIXEL_SHUFFLE = True USE_THUMBNAIL = False def __init__( self, *, max_model_len: int, patch_size: int, min_num_patches: int, max_num_patches: int, downsample_ratio: int, norm_mean: Sequence[float], norm_std: Sequence[float], factor_max: float = 1.0, use_thumbnail: bool = False, ) -> None: assert use_thumbnail is False, "use_thumbnail is not supported" self._patch_size: int = patch_size self._max_model_len = max_model_len self._min_num_patches = min_num_patches self._max_num_patches = max_num_patches if max_num_patches > 0 else float("inf") self._factor_max = factor_max self.norm_mean = torch.tensor(norm_mean).reshape(3, 1, 1) self.norm_std = torch.tensor(norm_std).reshape(3, 1, 1) self._transform = T.Compose( [ T.Lambda(lambda img: img.convert("RGB") if img.mode != "RGB" else img), T.ToTensor(), ] ) assert downsample_ratio < 1 reduction_factor = 1 / downsample_ratio assert reduction_factor == 2.0 self._downsample_ratio = int(reduction_factor) ** ( self.PIXEL_SHUFFLE + self.CONV_MERGING ) assert self._downsample_ratio == 2 def _get_num_embeddings(self, width: int, height: int) -> int: num_patches = (width // self._patch_size) * (height // self._patch_size) num_tokens = num_patches // (self._downsample_ratio**2) return num_tokens def width_and_height_for_max_num_tokens_available( self, target_num_tokens_post_shuffle: int, ) -> tuple[int, int]: """ TODO: optimize this so it squeezes closer to target number of tokens. Calculate image dimensions that produce approximately `target` tokens after pixel_shuffle. With pixel_shuffle enabled, each 2x2 patch grid becomes 1 token, so we need 4*B patches to get B tokens. Examples: >>> PATCH_SIZE = 16 >>> DOWNSAMPLE_RATIO = 0.5 >>> tiler = DynamicResolutionImageTiler( ... max_model_len=16384, ... patch_size=PATCH_SIZE, ... downsample_ratio=DOWNSAMPLE_RATIO, ... min_num_patches=4, ... max_num_patches=0, ... ) >>> width, height = tiler.width_and_height_for_max_num_tokens_available( ... target_num_tokens_post_shuffle=8192, ... ) >>> assert width, height == (2880, 2880) >>> assert (width // PATCH_SIZE) * ( ... height // PATCH_SIZE ... ) // 2**2 == 8100 # tokens post-shuffle >>> assert tiler._get_num_embeddings(width=width, height=height) == 8100 """ side_pixels = ( math.isqrt(target_num_tokens_post_shuffle) * self._downsample_ratio * self._patch_size ) assert isinstance(side_pixels, int) and side_pixels % self._patch_size == 0 return side_pixels, side_pixels def max_num_tokens_available(self, text_prompt_length: int) -> int: return self._max_model_len - text_prompt_length - 4 def _images_to_pixel_values_lst( self, text_prompt_length: int, images: list[Image.Image], ) -> tuple[list[torch.Tensor], list[int]]: num_tokens_available = self.max_num_tokens_available(text_prompt_length) params_per_image = self.compute_params(images, num_tokens_available) feature_sizes = [] images = [] for param in params_per_image: for t in self.apply_params(param): assert t.ndim == 3, f"{t.ndim=}: expected 3 dim tensor" images.append(t) feature_sizes.append(param.num_embeddings) return images, feature_sizes feature_size_cache: dict[Image.Image, int] = {} @classmethod def get_cached_feature_size(cls, image: Image.Image) -> int: feature_size = cls.feature_size_cache[id(image)] # hard assert that we only use the feature size once del cls.feature_size_cache[id(image)] return feature_size @dataclass class DynamicResolutionParams: media: Image.Image num_tiles: int num_embeddings: int patch_size: tuple[int, int] def apply_params(self, params: DynamicResolutionParams) -> list[torch.Tensor]: resized_img = params.media.resize( ( params.patch_size[0] * self._patch_size, params.patch_size[1] * self._patch_size, ) ) processed_images = [resized_img] return [self._transform(img) for img in processed_images] def process_media( self, media: Image.Image, num_tokens_available: int, ) -> tuple[DynamicResolutionParams, int]: """Process a single media item and return its parameters. Args: media: The media item to process num_tokens_available: Number of tokens available for this media Returns: DynamicResolutionParams for the media """ current_num_tokens_available = num_tokens_available assert isinstance(media, Image.Image), ( "Dynamic resolution is only supported for image media" ) orig_width, orig_height = media.width, media.height closest_patch_height = round(orig_height / self._patch_size + 0.5) closest_patch_width = round(orig_width / self._patch_size + 0.5) patches = closest_patch_height * closest_patch_width factor = min( math.sqrt(current_num_tokens_available / patches), self._factor_max ) target_patch_height = math.floor(factor * closest_patch_height) target_patch_width = math.floor(factor * closest_patch_width) # Consider self._min_num_patches if > current_num_tokens_available. if ( current_num_tokens_available > self._min_num_patches and target_patch_height * target_patch_width < self._min_num_patches ): up_factor = math.sqrt( self._min_num_patches / (target_patch_height * target_patch_width) ) target_patch_height = math.ceil(up_factor * target_patch_height) target_patch_width = math.ceil(up_factor * target_patch_width) # Round patch grid to be divisible by 2 (pixel-shuffle OR conv-merging) # or by 4 when BOTH are enabled (two successive 2x reductions) if self.PIXEL_SHUFFLE or self.CONV_MERGING: required_divisor = 4 if (self.PIXEL_SHUFFLE and self.CONV_MERGING) else 2 rem_h = target_patch_height % required_divisor if rem_h != 0: inc_h = required_divisor - rem_h if ( target_patch_height + inc_h ) * target_patch_width <= current_num_tokens_available: target_patch_height += inc_h else: target_patch_height = max( required_divisor, target_patch_height - rem_h ) rem_w = target_patch_width % required_divisor if rem_w != 0: inc_w = required_divisor - rem_w if ( target_patch_height * (target_patch_width + inc_w) <= current_num_tokens_available ): target_patch_width += inc_w else: target_patch_width = max( required_divisor, target_patch_width - rem_w ) # Calculate embeddings for the main dynamic resolution image num_embeddings = self._get_num_embeddings( target_patch_width * self._patch_size, target_patch_height * self._patch_size, ) token_count = target_patch_width * target_patch_height # Add thumbnail embeddings if enabled and image area is below threshold num_tiles = 1 # Base dynamic resolution image return self.DynamicResolutionParams( media=media, num_tiles=num_tiles, num_embeddings=num_embeddings, patch_size=(target_patch_width, target_patch_height), ), token_count def compute_params( self, media_list: list[Image.Image], num_tokens_available: int | None = None, ) -> list[DynamicResolutionParams]: """Compute parameters for all media with iterative token budgeting. Args: media_list: List of media items to process num_tokens_available: Total number of tokens available across all media Returns: List of ImageTilingParams for each media item """ num_tokens_available = ( num_tokens_available * (4 if self.PIXEL_SHUFFLE else 1) * (4 if self.CONV_MERGING else 1) ) # When the number of available token is too small, # allow self._min_num_patches per media and let the sample be truncated. num_tokens_available = max( num_tokens_available, self._min_num_patches * len(media_list) ) # Clip the number of tokens available per media to >min and <max patches. num_tokens_available_per_media = [ max(min(num_tokens_available, self._max_num_patches), self._min_num_patches) for _ in range(len(media_list)) ] # prevent infinite loop in any case for _ in range(10): # Step 1: Process each media with current token budget params = [] token_counts = [] for media, tokens_for_media in zip( media_list, num_tokens_available_per_media ): param, token_count = self.process_media(media, tokens_for_media) params.append(param) token_counts.append(token_count) self.feature_size_cache[id(param.media)] = param.num_embeddings # Step 2: Check if total tokens is within budget total_tokens = sum(token_counts) if total_tokens <= num_tokens_available: # We're within budget, return the params return params # Step 3: We're over budget, need to scale down # Calculate scaling factor to get under budget scaling_factor = num_tokens_available / total_tokens # Recalculate token budgets for each media based on scaling # Each media gets a proportional share of the total budget scaled_down_num_tokens_available_per_media = [ max(self._min_num_patches, int(token_count * scaling_factor)) for token_count in token_counts ] scaled_down = any( [ scaled_down_num_tokens_available_per_media[i] < num_tokens_available_per_media[i] for i in range(len(num_tokens_available_per_media)) ] ) # If there wasn't scaling down, we're stuck with min_num_patches per media, # else try with the scaled down num_tokens_available_per_media. if not scaled_down: num_tokens_available_per_media = [self._min_num_patches] * len( media_list ) else: num_tokens_available_per_media = ( scaled_down_num_tokens_available_per_media ) ctx = f"{params=} {total_tokens=} {num_tokens_available=}" raise ValueError( f"Should be unreachable - `return params` above must be reached: {ctx}" ) @staticmethod def stack(images: list[torch.Tensor], patch_size: int) -> torch.Tensor: assert len(images) > 0, "No images to stack" def rearrange_img(x): py = x.shape[-2] // patch_size px = x.shape[-1] // patch_size x = einops.rearrange( x, "c (py yy) (px xx) -> (py px) (c yy xx)", py=py, yy=patch_size, px=px, xx=patch_size, ) return x imgs = [rearrange_img(img) for img in images] pixel_values_flat = torch.cat(imgs, dim=0).unsqueeze(0) return pixel_values_flat class BaseNanoNemotronVLProcessor(ABC): """ This model doesn't define its own HF processor, so we implement our own one here. The code to insert image tokens is based on: https://huggingface.co/OpenGVLab/InternVL2-1B/blob/main/modeling_internvl_chat.py#L252 """ def __init__( self, config: PretrainedConfig, tokenizer: TokenizerLike, *args, max_model_len: int, max_num_tiles: int | None = None, **kwargs, ) -> None: super().__init__() self.config = config self.tokenizer = tokenizer self.max_num_tiles = max_num_tiles or DEFAULT_NUM_TILES image_size: int = config.force_image_size patch_size: int = config.patch_size downsample_ratio: int = config.downsample_ratio self.num_image_token = int( (image_size // patch_size) ** 2 * (downsample_ratio**2) ) self.image_size = image_size self.use_thumbnail: bool = config.use_thumbnail self.norm_mean = torch.Tensor(config.norm_mean).reshape(1, 3, 1, 1) self.norm_std = torch.Tensor(config.norm_std).reshape(1, 3, 1, 1) self.dynamic_tiler: DynamicResolutionImageTiler | None = None if self.use_dynamic_resolution(config): self.dynamic_tiler = DynamicResolutionImageTiler( max_model_len=max_model_len, patch_size=patch_size, downsample_ratio=downsample_ratio, min_num_patches=config.vision_config.args["min_num_patches"], max_num_patches=config.vision_config.args["max_num_patches"], norm_mean=config.norm_mean, norm_std=config.norm_std, ) @staticmethod def use_dynamic_resolution(config: PretrainedConfig) -> bool: return "min_num_patches" in config.vision_config.args @property @abstractmethod def image_token_id(self) -> int: raise NotImplementedError @abstractmethod def get_image_repl( self, feature_size: int, num_patches: int | None, ) -> PromptUpdateDetails[str]: raise NotImplementedError def get_num_image_tokens( self, *, image_width: int, image_height: int, max_num_tiles: int, ) -> int: target_ratios = get_internvl_target_ratios(1, max_num_tiles) num_patches, _, _ = calculate_internvl_targets( orig_width=image_width, orig_height=image_height, target_ratios=target_ratios, image_size=self.image_size, use_thumbnail=self.use_thumbnail, ) return num_patches * self.num_image_token def _images_to_pixel_values_lst( self, images: list[Image.Image], max_num_tiles: int, ) -> list[torch.Tensor]: return [ image_to_pixel_values( image, input_size=self.image_size, max_num=max_num_tiles, use_thumbnail=self.use_thumbnail, idx=idx, ) for idx, image in enumerate(images) ] def _preprocess_image( self, text: list[str], images: list[Image.Image], max_num_tiles: int, ) -> tuple[list[str], dict[str, Any]]: if len(images) == 0: image_inputs = {} return text, image_inputs if tiler := self.dynamic_tiler: sans_images = text[0].replace("<image>", "") text_prompt_length = len( self.tokenizer(sans_images, add_special_tokens=False).input_ids ) pixel_values_lst, num_tokens_per_image = tiler._images_to_pixel_values_lst( text_prompt_length=text_prompt_length, images=images, ) imgs_sizes = [(pv.shape[-2], pv.shape[-1]) for pv in pixel_values_lst] normalized = [ input_conditioner(img, tiler.norm_mean, tiler.norm_std) for img in pixel_values_lst ] image_num_patches = torch.tensor([1] * len(num_tokens_per_image)) image_inputs = { "pixel_values_flat": normalized, "imgs_sizes": imgs_sizes, "num_tokens_per_image": num_tokens_per_image, } else: pixel_values_lst = self._images_to_pixel_values_lst(images, max_num_tiles) image_num_patches = torch.tensor([len(item) for item in pixel_values_lst]) pixel_values_flat = input_conditioner( torch.cat(pixel_values_lst), self.norm_mean, self.norm_std ) image_inputs = { "pixel_values_flat": pixel_values_flat, "image_num_patches": image_num_patches, } num_tokens_per_image = [ self.num_image_token * len(item) for item in pixel_values_lst ] assert len(text) == 1, ( "hf_processor is called on the output of get_dummy_text, " "which should be a single string" ) parts = [x for x in re.split(r"(<image>)", text[0]) if x] assert parts.count("<image>") == len(pixel_values_lst), ( "the number of <image> tokens in the text should be the " "same as the number of images" ) for i, (feature_size, num_patches) in enumerate( zip(num_tokens_per_image, image_num_patches, strict=True) ): image_repl = self.get_image_repl(feature_size, num_patches) parts[i] = parts[i].replace("<image>", image_repl.full) text = ["".join(parts)] return text, image_inputs def _make_batch_input(self, input_item: Any | list[Any] | None = None): if input_item is None: input_item = [] if not isinstance(input_item, list): input_item = [input_item] return input_item @abstractmethod def __call__( self, text: str | list[str] | None = None, images: Image.Image | list[Image.Image] | None = None, return_tensors: str | TensorType | None = None, max_num_tiles: int | None = None, ) -> BatchFeature: raise NotImplementedError class NanoNemotronVLProcessor(BaseNanoNemotronVLProcessor): """ HF Processor with extended video processing logic. Code for video processing is adapted from video example: https://huggingface.co/OpenGVLab/InternVL3-1B#inference-with-transformers """ def __init__( self, config: PretrainedConfig, tokenizer: TokenizerLike, *, max_model_len: int, max_num_tiles: int | None = None, video_token: str | None = None, video_pruning_rate: float | None = None, ) -> None: super().__init__( config=config, tokenizer=tokenizer, max_model_len=max_model_len, max_num_tiles=max_num_tiles, ) # add extra video token for video processing self.video_token = video_token self.video_pruning_rate = video_pruning_rate self.audio_extractor: ParakeetExtractor | None = None raw_sound_config = getattr(config, "sound_config", None) if raw_sound_config is not None: self.audio_extractor = ParakeetExtractor(raw_sound_config) # Pre-tokenize special tokens for video processing # to avoid repeated tokenization self._img_start_token_ids = tokenizer.encode( IMG_START, add_special_tokens=False ) self._img_end_token_ids = tokenizer.encode(IMG_END, add_special_tokens=False) self._img_context_token_ids = tokenizer.encode( IMG_CONTEXT, add_special_tokens=False ) @property def supports_video(self) -> bool: return self.video_token_id is not None @property def video_token_id(self) -> int | None: if self.video_token is None: return None return self.tokenizer.get_vocab().get(self.video_token, None) @property def image_token_id(self) -> int: return self.tokenizer.convert_tokens_to_ids(IMG_CONTEXT) def _videos_to_pixel_values_lst( self, videos: list[npt.NDArray], max_num_tiles: int, ) -> list[torch.Tensor]: return [ video_to_pixel_values( video, input_size=self.image_size, max_num_tiles=max_num_tiles, use_thumbnail=self.use_thumbnail, ) for video in videos ] def _preprocess_video( self, text: list[str], videos: list[tuple[npt.NDArray, dict[str, Any]]], max_num_tiles: int, ): if len(videos) == 0 or not self.supports_video: video_inputs = {} else: videos_lst = [v[0] for v in videos] video_metadata_lst = [v[1] for v in videos] pixel_values_lst_video = self._videos_to_pixel_values_lst( videos_lst, max_num_tiles=max_num_tiles, ) # We use frame duration in milliseconds (as integer) to ensure # we have consistent timestamps calculation. At preprocessing # fps parameter is given in fp32, while at inference it is bf16 # which leads to inaccurate timestamp calculation and causes # timestamp values to differ.In rare cases this causes # mismatching number of output tokens for tokenized frame prefixes frame_duration_ms_lst = [ int(1000.0 / metadata["fps"]) for metadata in video_metadata_lst ] frames_indices_lst = [ metadata["frames_indices"] for metadata in video_metadata_lst ] video_inputs = { "pixel_values_flat_video": input_conditioner( torch.cat(pixel_values_lst_video), self.norm_mean, self.norm_std ), "video_num_patches": torch.tensor( [len(item) for item in pixel_values_lst_video] ), "frames_indices": frames_indices_lst, "frame_duration_ms": torch.tensor(frame_duration_ms_lst), } image_size: int = self.config.force_image_size patch_size: int = self.config.patch_size downsample_ratio = self.config.downsample_ratio tokens_in_single_frame = int( (image_size * image_size // patch_size**2) * (downsample_ratio**2) ) for pixel_values, video_metadata, frames_indices, frame_duration_ms in zip( pixel_values_lst_video, video_metadata_lst, frames_indices_lst, frame_duration_ms_lst, ): num_frames = pixel_values.shape[0] if ( self.video_pruning_rate is not None and self.video_pruning_rate > 0.0 ): # Start of EVS-specific code num_tokens = compute_retained_tokens_count( tokens_per_frame=tokens_in_single_frame, num_frames=num_frames, q=self.video_pruning_rate, ) # Here we just need placeholders that won't actually be replaced - # we just need to make sure the total number of tokens is correct # assign all tokens to the first frame tokens_per_frame = [num_tokens] + [0] * (num_frames - 1) # End of EVS-specific code else: tokens_per_frame = [tokens_in_single_frame] * num_frames video_repl = self.get_video_repl( tokens_per_frame=tokens_per_frame, frames_indices=frames_indices, frame_duration_ms=frame_duration_ms, tokenizer=self.tokenizer, img_start_token_ids=self._img_start_token_ids, img_end_token_ids=self._img_end_token_ids, img_context_token_ids=self._img_context_token_ids, ) # video_repl.full is a list of token IDs # Convert token IDs back to text for the HF processor flow video_repl_text = self.tokenizer.decode( video_repl.full, skip_special_tokens=False ) text = [t.replace("<video>", video_repl_text, 1) for t in text] return text, video_inputs def _preprocess_audio( self, text: list[str], audios: list[npt.NDArray], ): if len(audios) == 0: return text, {} assert self.audio_extractor is not None extractor = self.audio_extractor parts = [x for x in re.split(f"({re.escape(AUDIO_CONTEXT)})", text[0]) if x] token_count = parts.count(AUDIO_CONTEXT) if token_count != len(audios): raise ValueError( "Number of audio tokens in text does not match the number " f"of audios (tokens={token_count}, audios={len(audios)})." ) audio_index = 0 for idx, part in enumerate(parts): if part == AUDIO_CONTEXT: audio_repl = self.get_audio_repl(audios[audio_index]) parts[idx] = audio_repl.full audio_index += 1 text = ["".join(parts)] audio_inputs = extractor( audios, sampling_rate=extractor.sampling_rate, return_tensors="pt", ) input_audio_features = audio_inputs.input_features feature_attention_mask = audio_inputs.attention_mask audio_feature_lengths = feature_attention_mask.sum(dim=1) audio_inputs = { "input_audio_features": input_audio_features, "feature_attention_mask": feature_attention_mask, "audio_feature_lengths": audio_feature_lengths, } return text, audio_inputs def __call__( self, text: str | list[str] | None = None, images: Image.Image | list[Image.Image] | None = None, videos: list[tuple[npt.NDArray, dict[str, Any]]] | None = None, audios: AudioItem | list[AudioItem] | None = None, return_tensors: str | TensorType | None = None, max_num_tiles: int | None = None, ) -> BatchFeature: # Use default if not provided if max_num_tiles is None: max_num_tiles = self.max_num_tiles text, images, videos, audios = [ self._make_batch_input(x) for x in (text, images, videos, audios) ] text, image_inputs = self._preprocess_image( text=text, images=images, max_num_tiles=max_num_tiles, ) text, video_inputs = self._preprocess_video( text=text, videos=videos, max_num_tiles=1, ) text, audio_inputs = self._preprocess_audio( text=text, audios=audios, ) text_inputs = self.tokenizer(text, add_special_tokens=False) combined_inputs = {**text_inputs, **video_inputs, **audio_inputs} if self.dynamic_tiler is None: batch = BatchFeature( {**combined_inputs, **image_inputs}, tensor_type=return_tensors, ) else: batch = BatchFeature(combined_inputs, tensor_type=return_tensors) # allow images to be exempt from the BatchFeature validation: # We will .stack() them in _parse_and_validate_image_input batch.update(image_inputs) return batch def get_image_repl( self, feature_size: int, num_patches: int | None, ) -> PromptUpdateDetails[str]: repl_features = IMG_CONTEXT * feature_size repl_full = IMG_START + repl_features + IMG_END return PromptUpdateDetails.select_text(repl_full, IMG_CONTEXT) def get_audio_repl( self, audio: npt.NDArray, ) -> PromptUpdateDetails[str]: assert self.audio_extractor is not None num_tokens = self.audio_extractor.audio_token_count(len(audio)) repl_full = f"{AUDIO_START}{AUDIO_CONTEXT * num_tokens}{AUDIO_END}" return PromptUpdateDetails.select_text(repl_full, AUDIO_CONTEXT) @classmethod def get_video_repl( cls, *, tokens_per_frame: list[int], frames_indices: list[int], frame_duration_ms: int, tokenizer: TokenizerLike, img_start_token_ids: list[int], img_end_token_ids: list[int], img_context_token_ids: list[int], ) -> PromptUpdateDetails[list[int]]: """ Build prompt replacement for a video. The replacement returned is not actually used to replace the placeholder tokens - it's just used to make sure we allocate the correct number of tokens. Actual replacement is done in embed_multimodal of NemotronH_Nano_VL_V2 (specifically in _process_video_input -> _create_final_video_embeddings). There, we create the final embeddings with text embeddings for indicator tokens and video embeddings for video tokens. This is a single function that handles all cases - non EVS, EVS dummy, EVS real. The differentiation is done via tokens_per_frame parameter. - non EVS case - constant value same value across all frames - EVS dummy - Doesn't matter how tokens are distributed between frames - just make sure the total number of tokens is correct. - EVS real (called from get_real_video_repl_for_evs) - different value per frame Args: tokens_per_frame (list[int]): number of tokens per frame frames_indices (list[int]): frame indices frame_duration_ms (int): duration of each frame in milliseconds tokenizer (TokenizerLike): tokenizer to use for tokenizing frame separators img_start_token_ids (list[int]): pre-tokenized IMG_START tokens img_end_token_ids (list[int]): pre-tokenized IMG_END tokens img_context_token_ids (list[int]): pre-tokenized IMG_CONTEXT tokens """ # TODO: Add support of frame_duration_ms to be None # At preprocessing step we should allow absent / metadata without # frames_indices field. timestamps_enabled = frame_duration_ms is not None if timestamps_enabled: timestamps = calculate_timestamps(frames_indices, frame_duration_ms) assert len(timestamps) == len(tokens_per_frame), ( "timestamps and tokens_per_frame must have the same length" ) frame_separators = [ f"Frame {i + 1} sampled at {timestamp:.2f} seconds: " for i, timestamp in enumerate(timestamps) ] else: frame_separators = [ f"Frame {i + 1}: " for i, _ in enumerate(tokens_per_frame) ] # Tokenize frame separator independently frame_separators_tokenized = [ _seq2tokens(tokenizer, sep) for sep in frame_separators ] # Tokenize each component independently to avoid tokenizer merging tokens # across boundaries. This ensures consistent tokenization regardless of # num_tokens_per_frame values. all_token_ids = [] for i, num_tokens in enumerate(tokens_per_frame): frame_sep_token_ids = frame_separators_tokenized[i] all_token_ids.extend(frame_sep_token_ids) # Add pre-tokenized special tokens all_token_ids.extend(img_start_token_ids) all_token_ids.extend(img_context_token_ids * num_tokens) all_token_ids.extend(img_end_token_ids) return PromptUpdateDetails.from_seq(all_token_ids) class BaseNanoNemotronVLProcessingInfo(BaseProcessingInfo): """Basic image-only ProcessingInfo for InternVL-style models.""" @abstractmethod def get_hf_processor( self, **kwargs: object, ) -> BaseNanoNemotronVLProcessor: raise NotImplementedError def get_default_tok_params(self) -> TokenizeParams: return super().get_default_tok_params().with_kwargs(add_special_tokens=False) def get_supported_mm_limits(self) -> Mapping[str, int | None]: return {"image": None} def get_image_size_with_most_features(self, max_num_tiles: int) -> ImageSize: processor = self.get_hf_processor() base_size = processor.image_size target_ratios = get_internvl_target_ratios(1, max_num_tiles) largest_feature_size, largest_feature_pinpoint = 0, None for wr, hr in target_ratios: width, height = base_size * wr, base_size * hr feat_size = processor.get_num_image_tokens( image_width=width, image_height=height, max_num_tiles=max_num_tiles ) if feat_size > largest_feature_size: largest_feature_size = feat_size largest_feature_pinpoint = ImageSize(width=width, height=height) if largest_feature_size == 0 or largest_feature_pinpoint is None: raise ValueError("Cannot have a largest feature size of 0!") return largest_feature_pinpoint def get_max_image_tokens(self) -> int: processor = self.get_hf_processor() # Use default max_num_tiles for max tokens calculation max_num_tiles = processor.max_num_tiles target_width, target_height = self.get_image_size_with_most_features( max_num_tiles ) return processor.get_num_image_tokens( image_width=target_width, image_height=target_height, max_num_tiles=max_num_tiles, ) _I = TypeVar("_I", bound=BaseNanoNemotronVLProcessingInfo) class NanoNemotronVLProcessingInfo(BaseNanoNemotronVLProcessingInfo): """ProcessingInfo extended for video processing""" @property def supports_video(self): return self.get_hf_processor().supports_video @property def audio_extractor(self) -> ParakeetExtractor | None: return self.get_hf_processor().audio_extractor def get_data_parser(self): target_sr = None target_channels = None if extractor := self.audio_extractor: target_sr = extractor.sampling_rate target_channels = 1 return MultiModalDataParser( video_needs_metadata=True, target_sr=target_sr, target_channels=target_channels, expected_hidden_size=self._get_expected_hidden_size(), ) def get_supported_mm_limits(self): video_limit = {"video": None} if self.supports_video else {} audio_limit = {"audio": None} if self.audio_extractor is not None else {} return {**super().get_supported_mm_limits(), **video_limit, **audio_limit} def get_video_token(self) -> str | None: return IMG_CONTEXT def get_video_pruning_rate(self) -> float | None: return self.ctx.get_mm_config().video_pruning_rate def get_num_frames_with_most_features( self, seq_len: int, mm_counts: Mapping[str, int], ) -> int: max_images = mm_counts.get("image", 0) max_videos = mm_counts.get("video", 0) processor = self.get_hf_processor() # we get the CustomProcessor here max_image_tokens = self.get_max_image_tokens() * max_images max_total_frames = (seq_len - max_image_tokens) // processor.num_image_token max_frames_per_video = max_total_frames // max(max_videos, 1) return max(max_frames_per_video, 1) def get_hf_processor(self, **kwargs: object) -> NanoNemotronVLProcessor: return self.ctx.init_processor( NanoNemotronVLProcessor, config=self.get_hf_config(), tokenizer=self.get_tokenizer(), video_token=self.get_video_token(), video_pruning_rate=self.get_video_pruning_rate(), max_model_len=self.ctx.model_config.max_model_len, **kwargs, ) class NanoNemotronBaseVLMultiModalProcessor(BaseMultiModalProcessor[_I]): """Basic image-only MultiModalProcessor for InternVL-style models.""" @cached_property def is_dynamic_tiler(self) -> bool: return self.info.get_hf_processor().dynamic_tiler is not None def _get_mm_fields_config( self, hf_inputs: BatchFeature, hf_processor_mm_kwargs: Mapping[str, object], ) -> Mapping[str, MultiModalFieldConfig]: if self.is_dynamic_tiler: pixel_values_flat = MultiModalFieldConfig.batched("image") else: image_num_patches = hf_inputs.get("image_num_patches", torch.empty(0)) pixel_values_flat = MultiModalFieldConfig.flat_from_sizes( "image", image_num_patches ) return dict( pixel_values_flat=pixel_values_flat, image_num_patches=MultiModalFieldConfig.batched("image"), image_embeds=MultiModalFieldConfig.batched("image"), num_tokens_per_image=MultiModalFieldConfig.batched("image"), imgs_sizes=MultiModalFieldConfig.batched("image"), ) def _get_prompt_updates( self, mm_items: MultiModalDataItems, hf_processor_mm_kwargs: Mapping[str, object], out_mm_kwargs: MultiModalKwargsItems, ) -> Sequence[PromptUpdate]: hf_processor = self.info.get_hf_processor(**hf_processor_mm_kwargs) out_mm_data = out_mm_kwargs.get_data() if "image_num_patches" in out_mm_data: image_num_patches = out_mm_data["image_num_patches"] assert isinstance(image_num_patches, torch.Tensor) image_num_patches = image_num_patches.tolist() elif "image_embeds" in out_mm_data: # to compute num_patches (similar to Qwen2-VL) image_num_patches = [None] * len(out_mm_data["image_embeds"]) else: image_num_patches = [] def get_replacement_custom(item_idx: int): images = mm_items.get_items( "image", (ImageEmbeddingItems, ImageProcessorItems) ) if isinstance(images, ImageEmbeddingItems): feature_size = images.get_feature_size(item_idx) elif tiler := hf_processor.dynamic_tiler: image = images.get(item_idx) feature_size = tiler.get_cached_feature_size(image) else: image_size = images.get_image_size(item_idx) # Extract max_num_tiles from kwargs, default to 12 max_num_tiles = hf_processor_mm_kwargs.get( "max_num_tiles", hf_processor.max_num_tiles ) feature_size = hf_processor.get_num_image_tokens( image_width=image_size.width, image_height=image_size.height, max_num_tiles=max_num_tiles, ) num_patches = None local_image_num_patches = image_num_patches if isinstance(local_image_num_patches, torch.Tensor): local_image_num_patches = local_image_num_patches.tolist() if isinstance(local_image_num_patches, (list, tuple)) and item_idx < len( local_image_num_patches ): num_patches = int(local_image_num_patches[item_idx]) return hf_processor.get_image_repl(feature_size, num_patches) return [ PromptReplacement( modality="image", target="<image>", replacement=get_replacement_custom, ) ] class NanoNemotronVLMultiModalProcessor( NanoNemotronBaseVLMultiModalProcessor[NanoNemotronVLProcessingInfo] ): """MultiModalProcessor extended for video support""" def _get_mm_fields_config( self, hf_inputs: BatchFeature, hf_processor_mm_kwargs: Mapping[str, object], ) -> Mapping[str, MultiModalFieldConfig]: image_fields = super()._get_mm_fields_config(hf_inputs, hf_processor_mm_kwargs) if self.info.supports_video: video_num_patches = hf_inputs.get("video_num_patches", torch.empty(0)) video_fields = dict( pixel_values_flat_video=MultiModalFieldConfig.flat_from_sizes( "video", video_num_patches ), video_num_patches=MultiModalFieldConfig.batched("video"), frames_indices=MultiModalFieldConfig.batched("video"), frame_duration_ms=MultiModalFieldConfig.batched("video"), ) else: video_fields = {} if self.info.audio_extractor is not None: audio_fields = dict( input_audio_features=MultiModalFieldConfig.batched("audio"), feature_attention_mask=MultiModalFieldConfig.batched("audio"), audio_feature_lengths=MultiModalFieldConfig.batched("audio"), ) else: audio_fields = {} return image_fields | video_fields | audio_fields def _get_prompt_updates( self, mm_items: MultiModalDataItems, hf_processor_mm_kwargs: Mapping[str, object], out_mm_kwargs: MultiModalKwargsItems, ) -> Sequence[PromptUpdate]: prompt_repl = super()._get_prompt_updates( mm_items=mm_items, hf_processor_mm_kwargs=hf_processor_mm_kwargs, out_mm_kwargs=out_mm_kwargs, ) hf_processor = self.info.get_hf_processor(**hf_processor_mm_kwargs) out_mm_data = out_mm_kwargs.get_data() if "video_num_patches" in out_mm_data: video_num_patches = out_mm_data["video_num_patches"] assert isinstance(video_num_patches, torch.Tensor) video_num_patches = video_num_patches.tolist() else: video_num_patches = [] def get_video_replacement_internvl(item_idx: int): feature_size = hf_processor.num_image_token video, metadata = mm_items["video"][item_idx] num_patches = video_num_patches[item_idx] if num_patches is not None: assert isinstance(num_patches, int) video_pruning_rate = self.info.ctx.get_mm_config().video_pruning_rate if video_pruning_rate is not None and video_pruning_rate > 0.0: # Start of EVS-specific code num_tokens = compute_retained_tokens_count( tokens_per_frame=feature_size, num_frames=num_patches, q=video_pruning_rate, ) # Here we just need placeholders that won't actually be replaced - # we just need to make sure the total number of tokens is correct # assign all tokens to the first frame tokens_per_frame = [num_tokens] + [0] * (num_patches - 1) # End of EVS-specific code else: tokens_per_frame = [feature_size] * num_patches frame_duration_ms = int(1000 / metadata["fps"]) return hf_processor.get_video_repl( tokens_per_frame=tokens_per_frame, frames_indices=metadata["frames_indices"], frame_duration_ms=frame_duration_ms, tokenizer=hf_processor.tokenizer, img_start_token_ids=hf_processor._img_start_token_ids, img_end_token_ids=hf_processor._img_end_token_ids, img_context_token_ids=hf_processor._img_context_token_ids, ) if self.info.supports_video: prompt_repl = [ *prompt_repl, PromptReplacement( modality="video", target="<video>", replacement=get_video_replacement_internvl, ), ] def get_audio_replacement(item_idx: int): audios = mm_items.get_items("audio", AudioProcessorItems) return hf_processor.get_audio_repl(audios.get(item_idx)) if self.info.audio_extractor is not None: prompt_repl = [ *prompt_repl, PromptReplacement( modality="audio", target=AUDIO_CONTEXT, replacement=get_audio_replacement, ), ] return prompt_repl class NanoNemotronVLDummyInputsBuilder(BaseDummyInputsBuilder[_I]): """Basic image-only DummyInputsBuilder for InternVL-style models.""" def get_dummy_text(self, mm_counts: Mapping[str, int]) -> str: num_images = mm_counts.get("image", 0) return "<image>" * num_images def get_dummy_mm_data( self, seq_len: int, mm_counts: Mapping[str, int], mm_options: Mapping[str, BaseDummyOptions], ) -> MultiModalDataDict: num_images = mm_counts.get("image", 0) processor = self.info.get_hf_processor() if tiler := processor.dynamic_tiler: budget = tiler.max_num_tokens_available(text_prompt_length=num_images) target_width, target_height = ( tiler.width_and_height_for_max_num_tokens_available(budget) ) else: max_num_tiles = 12 target_width, target_height = self.info.get_image_size_with_most_features( max_num_tiles ) image_overrides = mm_options.get("image") return { "image": self._get_dummy_images( width=target_width, height=target_height, num_images=num_images, overrides=image_overrides, ) } class NanoNemotronVLDummyInputsBuilder( NanoNemotronVLDummyInputsBuilder[NanoNemotronVLProcessingInfo] ): """DummyInputsBuilder extended for video support""" def get_dummy_text(self, mm_counts: Mapping[str, int]) -> str: num_videos = mm_counts.get("video", 0) num_audios = mm_counts.get("audio", 0) return ( super().get_dummy_text(mm_counts) + "<video>" * num_videos + AUDIO_CONTEXT * num_audios ) def _get_dummy_videos( self, *, width: int, height: int, num_frames: int, num_videos: int, overrides: VideoDummyOptions | None = None, ) -> list[VideoItem]: video = super()._get_dummy_videos( width=width, height=height, num_frames=num_frames, num_videos=1, overrides=overrides, )[0] video_items = [] for _ in range(num_videos): video_metadata = { "total_num_frames": num_frames, "fps": 2, "duration": num_frames / 2.0, "video_backend": "opencv_dynamic", "frames_indices": [i for i in range(num_frames)], "do_sample_frames": False, } video_item = (video.copy(), video_metadata) video_items.append(video_item) return video_items def get_dummy_mm_data( self, seq_len: int, mm_counts: Mapping[str, int], mm_options: Mapping[str, BaseDummyOptions], ) -> MultiModalDataDict: dummy_image = super().get_dummy_mm_data(seq_len, mm_counts, mm_options) if self.info.supports_video: config = self.info.get_hf_config() image_size: int = config.force_image_size target_num_frames = self.info.get_num_frames_with_most_features( seq_len, mm_counts ) num_videos = mm_counts.get("video", 0) video_overrides = mm_options.get("video") dummy_video = { "video": self._get_dummy_videos( width=image_size, height=image_size, num_frames=target_num_frames, num_videos=num_videos, overrides=video_overrides, ) } else: dummy_video = {} if extractor := self.info.audio_extractor: num_audios = mm_counts.get("audio", 0) audio_overrides = mm_options.get("audio") if mm_options else None tokens_per_audio = max(1, seq_len // max(num_audios, 1)) max_audio_num_samples = MAX_AUDIO_LEN_S * extractor.sampling_rate calculated_max_audio_num_samples = extractor.audio_length(tokens_per_audio) audio_len = min(max_audio_num_samples, calculated_max_audio_num_samples) dummy_audio = { "audio": self._get_dummy_audios( length=audio_len, num_audios=num_audios, overrides=audio_overrides, ) } else: dummy_audio = {} return {**dummy_image, **dummy_video, **dummy_audio} @MULTIMODAL_REGISTRY.register_processor( NanoNemotronVLMultiModalProcessor, info=NanoNemotronVLProcessingInfo, dummy_inputs=NanoNemotronVLDummyInputsBuilder, ) class NemotronH_Nano_VL_V2( nn.Module, HasInnerState, IsHybrid, SupportsMultiModal, SupportsMultiModalPruning ): @classmethod def get_placeholder_str(cls, modality: str, i: int) -> str | None: if modality.startswith("image"): return "<image>" if modality.startswith("video"): return "<video>" if modality.startswith("audio"): return AUDIO_CONTEXT return None def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): super().__init__() model_config = vllm_config.model_config config = model_config.hf_config multimodal_config = model_config.multimodal_config image_size = config.force_image_size patch_size = config.patch_size self.patch_size = patch_size self.template = config.template self.num_image_token = int( (image_size // patch_size) ** 2 * (config.downsample_ratio**2) ) self.downsample_ratio = config.downsample_ratio self.ps_version = config.ps_version self.image_tag_type = config.image_tag_type self.video_pruning_rate = multimodal_config.video_pruning_rate with self._mark_language_model(vllm_config): self.language_model = init_vllm_registered_model( vllm_config=vllm_config, hf_config=config.text_config, prefix=maybe_prefix(prefix, "language_model"), ) llm_dtype = self.language_model.config.dtype assert isinstance(llm_dtype, torch.dtype) self.llm_dtype = llm_dtype with self._mark_tower_model(vllm_config, {"image", "video", "audio"}): self.vision_model = self.get_vit_model_from_radio_config(config).to( llm_dtype ) # Construct the vision projection. vit_hidden_size = config.vit_hidden_size vision_projection_hidden_size = config.projector_hidden_size llm_hidden_size = config.text_config.hidden_size mlp1 = nn.Sequential( RMSNorm( hidden_size=vit_hidden_size * int(1 / self.downsample_ratio) ** 2, eps=1e-5, ), nn.Linear( vit_hidden_size * int(1 / self.downsample_ratio) ** 2, vision_projection_hidden_size, bias=False, ), ReLUSquaredActivation(), nn.Linear(vision_projection_hidden_size, llm_hidden_size, bias=False), ) self.mlp1 = mlp1.to(llm_dtype) self.sound_encoder: ProjectedParakeet | None = None if getattr(config, "sound_config", None) is not None: logger.info_once( "Found sound config, initializing sound encoder for Nemotron AVLM", scope="global", ) self.sound_encoder = ProjectedParakeet( config.sound_config, dtype=llm_dtype, llm_hidden_size=llm_hidden_size, max_model_len=model_config.max_model_len, ) self.config = config self.model_config = vllm_config.model_config # Pre-tokenize special tokens for video processing # to avoid repeated tokenization tokenizer = cached_tokenizer_from_config(model_config) self._img_start_token_ids = tokenizer.encode( IMG_START, add_special_tokens=False ) self._img_end_token_ids = tokenizer.encode(IMG_END, add_special_tokens=False) self._img_context_token_ids = tokenizer.encode( IMG_CONTEXT, add_special_tokens=False ) self.dynamic_resolution = BaseNanoNemotronVLProcessor.use_dynamic_resolution( config ) if self.dynamic_resolution: logger.info_once( "Dynamic resolution is enabled for NanoNemotronVLProcessor", scope="global", ) def pixel_shuffle(self, x, scale_factor=0.5): n, w, h, c = x.size() # N, W, H, C --> N, W, H * scale, C // scale x = x.view( n, w, int(h * scale_factor), int(c / scale_factor), ) # N, W, H * scale, C // scale --> N, H * scale, W, C // scale x = x.permute(0, 2, 1, 3).contiguous() # N, H * scale, W, C // scale --> # N, H * scale, W * scale, C // (scale ** 2) x = x.view( n, int(h * scale_factor), int(w * scale_factor), int(c / (scale_factor * scale_factor)), ) if self.ps_version == "v1": warnings.warn( "In ps_version 'v1', the height and width have not " "been swapped back, which results in a transposed image.", stacklevel=2, ) else: x = x.permute(0, 2, 1, 3).contiguous() return x def pixel_shuffle_dynamic_res( self, x: torch.Tensor, *, imgs_sizes: list[tuple[int, int]] ) -> torch.Tensor: scale_factor = self.downsample_ratio patch_dim = self.patch_size seq_lens = calc_seq_lens(imgs_sizes, patch_dim) splits = torch.split(x, seq_lens, dim=-2) out = [] for i, sv in enumerate(splits): h = imgs_sizes[i][0] // patch_dim w = imgs_sizes[i][1] // patch_dim sv = sv.reshape(sv.shape[0], h, w, -1) n, h, w, c = sv.size() sv = sv.view(n, h, int(w * scale_factor), int(c / scale_factor)) sv = sv.permute(0, 2, 1, 3).contiguous() sv = sv.view( n, int(w * scale_factor), int(h * scale_factor), int(c / (scale_factor * scale_factor)), ) if self.ps_version == "v2": sv = sv.permute(0, 2, 1, 3).contiguous() sv = sv.reshape(sv.shape[0], -1, sv.shape[-1]) out.append(sv) x = torch.cat(out, dim=-2) return x def extract_feature_dynamic( self, pixel_values: torch.Tensor, imgs_sizes: list[tuple[int, int]] ): """Dynamic resolution extract_feature for images.""" _, vit_embeds = self.vision_model(pixel_values, imgs_sizes=imgs_sizes) vit_embeds = vit_embeds.to(dtype=torch.bfloat16) vit_embeds = self.pixel_shuffle_dynamic_res(vit_embeds, imgs_sizes=imgs_sizes) vit_embeds = self.mlp1(vit_embeds) return vit_embeds def extract_feature(self, pixel_values: torch.Tensor): # Process images in a micro-batch of at most 128 frames per call # This is done on purpose to ensure peak GPU ram usage of huge batch # (namely for really long videos with EVS ON) won't cause any problems # as we don't support chunked prefill for video media micro_batch_size = 128 n = pixel_values.shape[0] vit_embeds_list = [] for i in range(0, n, micro_batch_size): _, vit_embeds = self.vision_model(pixel_values[i : i + micro_batch_size]) vit_embeds = vit_embeds.to(dtype=torch.bfloat16) h = w = int(vit_embeds.shape[1] ** 0.5) vit_embeds = vit_embeds.reshape(vit_embeds.shape[0], h, w, -1) vit_embeds = self.pixel_shuffle( vit_embeds, scale_factor=self.downsample_ratio ) vit_embeds = vit_embeds.reshape( vit_embeds.shape[0], -1, vit_embeds.shape[-1] ) vit_embeds = self.mlp1(vit_embeds) vit_embeds_list.append(vit_embeds) vit_embeds = torch.cat(vit_embeds_list, dim=0) return vit_embeds def _parse_and_validate_image_input( self, **kwargs: object ) -> NanoNemotronVLImageInputs | None: if image_embeds := kwargs.pop("image_embeds", None): return NanoNemotronVLImageEmbeddingInputs( type="image_embeds", data=image_embeds, ) if self.dynamic_resolution: pixel_values_flat = DynamicResolutionImageTiler.stack( kwargs.pop("pixel_values_flat"), self.patch_size ) return NanoNemotronVLImagePixelInputsDynamic( pixel_values_flat=pixel_values_flat, **kwargs ) else: return NanoNemotronVLImagePixelInputs( num_patches=kwargs.pop("image_num_patches"), **kwargs ) def _process_image_input_dynamic( self, image_input: NanoNemotronVLImagePixelInputsDynamic ) -> tuple[torch.Tensor, ...]: image_embeds = self.extract_feature_dynamic( image_input.pixel_values_flat, image_input.imgs_sizes ) num_tokens_per_image = image_input.num_tokens_per_image if len(num_tokens_per_image) == 1: return (image_embeds.view(-1, self.config.text_config.hidden_size),) image_embeds = image_embeds.view(-1, self.config.text_config.hidden_size) return image_embeds.split(num_tokens_per_image) def _process_image_input( self, image_input: NanoNemotronVLImagePixelInputs ) -> tuple[torch.Tensor, ...]: image_embeds = self.extract_feature(image_input["pixel_values_flat"]) num_patches = image_input["num_patches"] # Only one image in the current batch if len(num_patches) == 1: return (image_embeds.view(-1, self.config.text_config.hidden_size),) # NOTE: Image embeddings are split into separate tensors for each image # by the size of each embedding. feature_size = image_embeds.shape[1] image_embeds = image_embeds.view(-1, self.config.text_config.hidden_size) image_feature_sizes = [ num_patches * feature_size for num_patches in num_patches ] return image_embeds.split(image_feature_sizes) def _process_video_input( self, video_input: NanoNemotronVLVideoPixelInputs ) -> tuple[torch.Tensor, ...]: """Process video input and create final embeddings with video content and indicator tokens.""" # Get video embeddings using the same processing as images video_embeddings = self._process_image_input(video_input) final_video_embeddings: tuple[torch.Tensor, ...] = () image_rows = image_cols = self.config.force_image_size downsample_ratio = self.config.downsample_ratio patch_size = self.config.patch_size rows = int(image_rows * downsample_ratio // patch_size) cols = int(image_cols * downsample_ratio // patch_size) video_pruning_rate = self.video_pruning_rate video_num_frames = video_input["num_patches"].tolist() video_frames_indices = video_input["frames_indices"].split(video_num_frames) # Calculate video feature dimensions (number of frames and # their feature size (AKA tokens per frame)) # TODO: Maybe this can be optimized to avoid the loop? for i, single_video_embeddings in enumerate(video_embeddings): num_frames = video_num_frames[i] frames_indices = video_frames_indices[i].tolist() frame_duration_ms = video_input["frame_duration_ms"][i].item() assert single_video_embeddings.shape[0] % num_frames == 0 if video_pruning_rate is not None and video_pruning_rate > 0.0: # Start of EVS-specific code retention_mask = compute_retention_mask( single_video_embeddings, video_size_thw=(num_frames, rows, cols), spatial_merge_size=1, q=video_pruning_rate, ) # apply retention mask single_video_embeddings = single_video_embeddings[retention_mask] # calculate the actual number of retained tokens per frame retention_mask_thw = retention_mask.reshape(num_frames, rows, cols) num_tokens_per_frame = ( retention_mask_thw.sum(dim=(1, 2)).long().tolist() ) # End of EVS-specific code else: feature_size = single_video_embeddings.shape[0] // num_frames num_tokens_per_frame = [feature_size] * num_frames final_video_embeddings += ( self._create_final_video_embeddings( single_video_embeddings, num_tokens_per_frame, frames_indices, frame_duration_ms, ), ) return final_video_embeddings def _process_audio_input( self, audio_input: NanoNemotronVLAudioFeatureInputs ) -> tuple[torch.Tensor, ...]: assert self.sound_encoder is not None input_audio_features = audio_input.input_audio_features feature_attention_mask = audio_input.feature_attention_mask target_device = next(self.sound_encoder.parameters()).device # When cross-request batching combines audio clips with different # time dimensions, _reduce_data returns a list instead of a stacked # tensor. Pad to the max time dim and stack; the attention mask # already marks valid positions so zero-padding is safe. if isinstance(input_audio_features, list): feature_sizes = [f.shape[-2] for f in input_audio_features] max_t = max(feature_sizes) padded_feats = [ torch.nn.functional.pad(feat, (0, 0, 0, max_t - feat_size)) for feat, feat_size in zip( input_audio_features, feature_sizes, strict=True ) ] padded_masks = [ torch.nn.functional.pad(mask, (0, max_t - mask.shape[-1])) for mask in feature_attention_mask ] input_audio_features = torch.stack(padded_feats) feature_attention_mask = torch.stack(padded_masks) input_audio_features = input_audio_features.to( dtype=self.llm_dtype, device=target_device ) feature_attention_mask = feature_attention_mask.to(device=target_device) sound_embeds = self.sound_encoder(input_audio_features, feature_attention_mask) valid_input_lens = feature_attention_mask.sum(dim=1) valid_output_lens = self.sound_encoder.encoder._get_subsampling_output_length( valid_input_lens ) truncated_embeds = [] for i in range(sound_embeds.shape[0]): valid_len = valid_output_lens[i].item() truncated_embeds.append(sound_embeds[i, :valid_len]) return tuple(truncated_embeds) def _create_final_video_embeddings( self, video_embeddings: torch.Tensor, num_tokens_per_frame: list[int], frames_indices: list[int], frame_duration_ms: int, ) -> torch.Tensor: """Create final embeddings that combine video embeddings with text embeddings of indicator tokens. These final embeddings contain: - Actual video embeddings in positions corresponding to video content - Text embeddings for indicator tokens (<img>, </img>, and frame separation text) in their respective positions These embeddings will replace the placeholder embeddings to create input_embeds for the LLM. """ device = video_embeddings.device tokenizer = cached_tokenizer_from_config(self.model_config) # Generate video replacement token IDs using get_video_repl # This tokenizes each frame separator independently, then uses pre-tokenized # special tokens to ensure consistent tokenization regardless of # num_tokens_per_frame values. video_repl = NanoNemotronVLProcessor.get_video_repl( tokens_per_frame=num_tokens_per_frame, frames_indices=frames_indices, frame_duration_ms=frame_duration_ms, tokenizer=tokenizer, img_start_token_ids=self._img_start_token_ids, img_end_token_ids=self._img_end_token_ids, img_context_token_ids=self._img_context_token_ids, ) # video_repl.full is a list of token IDs repl_token_ids = torch.tensor(video_repl.full, device=device) # Get embedding token IDs for image context (use pre-tokenized version) embed_token_ids = torch.tensor(self._img_context_token_ids, device=device) # Create mask for video embedding positions is_video_embed = torch.isin(repl_token_ids, embed_token_ids) # Create final video embeddings, merging text embeddings for indicator # tokens with video embeddings text_embeddings = self.get_language_model().embed_input_ids(repl_token_ids) final_video_embeddings = _merge_multimodal_embeddings( inputs_embeds=text_embeddings, multimodal_embeddings=video_embeddings, is_multimodal=is_video_embed, ) return final_video_embeddings def _parse_and_validate_video_input( self, **kwargs: object ) -> NanoNemotronVLVideoPixelInputs | None: pixel_values_flat_video = kwargs.pop("pixel_values_flat_video", None) video_num_patches = kwargs.pop("video_num_patches", None) video_embeds = kwargs.pop("video_embeds", None) frames_indices = kwargs.pop("frames_indices", None) frame_duration_ms = kwargs.pop("frame_duration_ms", None) if pixel_values_flat_video is None and video_embeds is None: return None if video_embeds is not None: return NanoNemotronVLVideoEmbeddingInputs( type="video_embeds", data=video_embeds, ) if pixel_values_flat_video is not None: if torch.is_tensor(frames_indices): frames_indices = frames_indices.flatten() else: frames_indices = torch.cat([f.flatten() for f in frames_indices], dim=0) frame_duration_ms = frame_duration_ms.flatten() expected_h = expected_w = self.config.force_image_size num_frames = video_num_patches[0].item() resolve_bindings = {"h": expected_h, "w": expected_w, "f": num_frames} return NanoNemotronVLVideoPixelInputs( type="pixel_values_videos", pixel_values_flat=pixel_values_flat_video, num_patches=video_num_patches, frames_indices=frames_indices, frame_duration_ms=frame_duration_ms, resolve_bindings=resolve_bindings, ) raise AssertionError("This line should be unreachable.") def _parse_and_validate_multimodal_inputs(self, **kwargs: object) -> dict: modalities = {} # Preserve the order of modalities if there are multiple of them # from the order of kwargs. for input_key in kwargs: if ( input_key in ("pixel_values_flat", "image_embeds") and "images" not in modalities ): modalities["images"] = self._parse_and_validate_image_input(**kwargs) if input_key in ("pixel_values_flat_video",) and "videos" not in modalities: modalities["videos"] = self._parse_and_validate_video_input(**kwargs) if ( input_key in ( "input_audio_features", "feature_attention_mask", "audio_feature_lengths", ) and "audios" not in modalities ): modalities["audios"] = NanoNemotronVLAudioFeatureInputs( **kwargs, validate=False ) return modalities def embed_multimodal(self, **kwargs: object) -> MultiModalEmbeddings: # Validate the multimodal input keyword arguments modalities = self._parse_and_validate_multimodal_inputs(**kwargs) if modalities is None: return [] # # The result multimodal_embeddings is tuple of tensors, with each # tensor corresponding to a multimodal data item (image or video). multimodal_embeddings: tuple[torch.Tensor, ...] = () # NOTE: It is important to iterate over the keys in this dictionary # to preserve the order of the modalities. for modality in modalities: if modality == "images": image_input = modalities["images"] if image_input["type"] == "image_embeds": image_embeddings = image_input["data"] elif self.dynamic_resolution: assert image_input["type"] == "pixel_values_dynamic" image_embeddings = self._process_image_input_dynamic(image_input) else: image_embeddings = self._process_image_input(image_input) multimodal_embeddings += tuple(image_embeddings) if modality == "videos": video_input = modalities["videos"] video_embeddings = self._process_video_input(video_input) multimodal_embeddings += tuple(video_embeddings) if modality == "audios": audio_input = modalities["audios"] audio_embeddings = self._process_audio_input(audio_input) multimodal_embeddings += tuple(audio_embeddings) return multimodal_embeddings def forward( self, input_ids: torch.Tensor | None, positions: torch.Tensor, intermediate_tensors: IntermediateTensors | None = None, inputs_embeds: torch.Tensor | None = None, **kwargs: object, ) -> torch.Tensor | IntermediateTensors: if intermediate_tensors is not None: inputs_embeds = None hidden_states = self.language_model( input_ids=input_ids, positions=positions, intermediate_tensors=intermediate_tensors, inputs_embeds=inputs_embeds, **kwargs, ) return hidden_states def get_mm_mapping(self) -> MultiModelKeys: """ Get the module prefix in multimodal models """ return MultiModelKeys.from_string_field( language_model="language_model", connector=["mlp1", "sound_encoder.projection"], tower_model=["vision_model", "sound_encoder.encoder"], ) def compute_logits( self, hidden_states: torch.Tensor, ) -> torch.Tensor | None: return self.language_model.compute_logits(hidden_states) def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]): adapter_dict = dict(self.mlp1.named_parameters()) def is_llm(name: str) -> bool: return name.startswith("language_model") def is_adapter_weights(weight: tuple[str, torch.Tensor]): return weight[0].startswith("mlp1") def is_vision_weights(name: str) -> bool: return name.startswith("vision_model.radio_model.") def is_sound_weights(name: str) -> bool: return name.startswith("sound") # Separate weights by component llm_weights = [] vision_weights = [] sound_weights = [] for name, w in weights: if is_llm(name): # Strip 'language_model.' prefix for LLM weights llm_weights.append((".".join(name.split(".")[1:]), w)) elif is_adapter_weights((name, w)): # Load vision-language adapter weights directly trimmed_name = ".".join(name.split(".")[1:]) param = adapter_dict[trimmed_name] with torch.no_grad(): default_weight_loader(param, w) elif is_vision_weights(name): # Convert: vision_model.radio_model.* → radio_model.* hf_key = name[len("vision_model.") :] # Remove "vision_model." prefix vision_weights.append((hf_key, w)) elif is_sound_weights(name): assert self.sound_encoder is not None sound_weights.append((name, w)) self.language_model.load_weights(llm_weights) self.vision_model.load_weights(vision_weights) if self.sound_encoder is not None: assert len(sound_weights) > 0 self.sound_encoder.load_weights(sound_weights) def print_architecture(self, detailed: bool = True, save_to_file: str = None): """ Print model architecture with parameter names, shapes, and sizes. Args: detailed: If True, show detailed parameter breakdown save_to_file: If provided, save output to this file path """ import sys from io import StringIO # Capture output if saving to file original_stdout = sys.stdout if save_to_file: sys.stdout = StringIO() try: print("=" * 100) print("NemotronH_Nano_VL_V2 Model Architecture") print("=" * 100) total_params = 0 param_groups = { "language_model": [], "vision_model": [], "mlp1": [], "other": [], } for name, param in self.named_parameters(): param_size = param.numel() total_params += param_size # Group parameters by main component if name.startswith("language_model"): param_groups["language_model"].append( (name, param.shape, param_size, param.dtype) ) elif name.startswith("vision_model"): param_groups["vision_model"].append( (name, param.shape, param_size, param.dtype) ) elif name.startswith("mlp1"): param_groups["mlp1"].append( (name, param.shape, param_size, param.dtype) ) else: param_groups["other"].append( (name, param.shape, param_size, param.dtype) ) if detailed: print( f"{name:<70} | Shape: {str(param.shape):<25} | " f"Size: {param_size:>12,} | Dtype: {param.dtype}" ) print("=" * 100) print("Summary by Component:") print("-" * 60) for component, params in param_groups.items(): if params: # Only show components that have parameters component_total = sum(size for _, _, size, _ in params) percentage = ( (component_total / total_params) * 100 if total_params > 0 else 0 ) print( f"{component:<20} | Parameters: {len(params):>4} | " f"Total Size: {component_total:>15,} | " f"{percentage:>6.2f}%" ) print("-" * 60) print(f"{'Total Parameters':<20} | {total_params:>15,}") # Estimate memory usage (assuming bfloat16 = 2 bytes per parameter) memory_mb = total_params * 2 / (1024**2) memory_gb = memory_mb / 1024 print(f"{'Est. Memory (MB)':<20} | {memory_mb:>15.2f}") print(f"{'Est. Memory (GB)':<20} | {memory_gb:>15.2f}") print("=" * 100) # Save to file if requested if save_to_file: output = sys.stdout.getvalue() sys.stdout = original_stdout with open(save_to_file, "w") as f: f.write(output) print(f"Architecture saved to: {save_to_file}") print(output) # Also print to console finally: if save_to_file and sys.stdout != original_stdout: sys.stdout = original_stdout def get_vit_model_from_radio_config(self, hf_config): hf_config_vision = hf_config.vision_config model_name = hf_config_vision.args.get("model") if model_name is None: raise ValueError(f"Unsupported vit model type: {model_name}") preferred_resolution = getattr(hf_config_vision, "preferred_resolution", None) image_size = preferred_resolution[0] if preferred_resolution else 224 patch_size = getattr(hf_config_vision, "patch_size", 16) radio_config = RadioConfig( model_name=model_name, image_size=image_size, patch_size=patch_size, norm_mean=hf_config.norm_mean, norm_std=hf_config.norm_std, **hf_config_vision.args, ) return RadioModel(config=radio_config) def copy_inputs_before_cuda_graphs(self, input_buffers, **kwargs): return self.language_model.mamba_cache.copy_inputs_before_cuda_graphs( input_buffers, **kwargs ) def get_seqlen_agnostic_capture_inputs(self, batch_size: int): return self.language_model.mamba_cache.get_seqlen_agnostic_capture_inputs( batch_size ) @classmethod def get_mamba_state_shape_from_config(cls, vllm_config: "VllmConfig"): text_config = vllm_config.model_config.hf_config.text_config temp_vllm_config = copy.deepcopy(vllm_config) temp_vllm_config.model_config.hf_config = text_config return NemotronHForCausalLM.get_mamba_state_shape_from_config(temp_vllm_config) @classmethod def get_mamba_state_dtype_from_config(cls, vllm_config: "VllmConfig"): text_config = vllm_config.model_config.hf_config.text_config temp_vllm_config = copy.deepcopy(vllm_config) temp_vllm_config.model_config.hf_config = text_config return NemotronHForCausalLM.get_mamba_state_dtype_from_config(temp_vllm_config) @classmethod def get_mamba_state_copy_func(cls): return NemotronHForCausalLM.get_mamba_state_copy_func()
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/model_executor/models/nano_nemotron_vl.py", "license": "Apache License 2.0", "lines": 2053, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license
vllm-project/vllm:vllm/transformers_utils/runai_utils.py
# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import hashlib import os import shutil import signal from vllm import envs from vllm.assets.base import get_cache_dir from vllm.logger import init_logger from vllm.utils.import_utils import PlaceholderModule logger = init_logger(__name__) SUPPORTED_SCHEMES = ["s3://", "gs://"] try: from runai_model_streamer import list_safetensors as runai_list_safetensors from runai_model_streamer import pull_files as runai_pull_files except ImportError: runai_model_streamer = PlaceholderModule("runai_model_streamer") # type: ignore[assignment] runai_pull_files = runai_model_streamer.placeholder_attr("pull_files") runai_list_safetensors = runai_model_streamer.placeholder_attr("list_safetensors") def list_safetensors(path: str = "") -> list[str]: """ List full file names from object path and filter by allow pattern. Args: path: The object storage path to list from. Returns: list[str]: List of full object storage paths allowed by the pattern """ return runai_list_safetensors(path) def is_runai_obj_uri(model_or_path: str) -> bool: return model_or_path.lower().startswith(tuple(SUPPORTED_SCHEMES)) class ObjectStorageModel: """ A class representing an ObjectStorage model mirrored into a temporary directory. Attributes: dir: The temporary created directory. Methods: pull_files(): Pull model from object storage to the temporary directory. """ def __init__(self, url: str) -> None: if envs.VLLM_ASSETS_CACHE_MODEL_CLEAN: for sig in (signal.SIGINT, signal.SIGTERM): existing_handler = signal.getsignal(sig) signal.signal(sig, self._close_by_signal(existing_handler)) dir_name = os.path.join( get_cache_dir(), "model_streamer", hashlib.sha256(str(url).encode()).hexdigest()[:8], ) os.makedirs(dir_name, exist_ok=True) self.dir = dir_name logger.debug("Init object storage, model cache path is: %s", dir_name) def _close(self) -> None: if os.path.exists(self.dir): shutil.rmtree(self.dir) def _close_by_signal(self, existing_handler=None): def new_handler(signum, frame): self._close() if existing_handler: existing_handler(signum, frame) return new_handler def pull_files( self, model_path: str = "", allow_pattern: list[str] | None = None, ignore_pattern: list[str] | None = None, ) -> None: """ Pull files from object storage into the temporary directory. Args: model_path: The object storage path of the model. allow_pattern: A list of patterns of which files to pull. ignore_pattern: A list of patterns of which files not to pull. """ if not model_path.endswith("/"): model_path = model_path + "/" runai_pull_files(model_path, self.dir, allow_pattern, ignore_pattern)
{ "repo_id": "vllm-project/vllm", "file_path": "vllm/transformers_utils/runai_utils.py", "license": "Apache License 2.0", "lines": 77, "canary_id": -1, "canary_value": "", "pii_type": "", "provider": "", "regex_pattern": "", "repetition": -1, "template": "" }
license