diff --git a/findings.md b/findings.md new file mode 100644 index 0000000..2497c6f --- /dev/null +++ b/findings.md @@ -0,0 +1,109 @@ +# Findings: CUDA Graph for Offload Mode + +## Discovery 1: 为什么 Offload Mode 不使用 CUDA Graph + +**位置**: `nanovllm/engine/model_runner.py:421` + +```python +use_eager = is_prefill or self.enforce_eager or input_ids.size(0) > 512 or context.is_chunked_prefill +``` + +**原因**: `run_chunked_offload_decode()` 设置 `is_chunked_prefill=True`，强制使用 eager mode。 + +--- + +## Discovery 2: 当前 CUDA Graph 架构 + +**文件**: `model_runner.py:682-717` + +```python +def capture_cudagraph(self): + # 为不同 batch size 捕获完整 model forward + for bs in [1, 2, 4, 8, 16, ...]: + with torch.cuda.graph(graph): + outputs[:bs] = self.model(input_ids[:bs], positions[:bs]) +``` + +**特点**: +- 捕获完整的 `model()` 调用（包含所有层） +- 使用 graph pool 共享内存 +- 只用于 decode（prefill 始终 eager） + +--- + +## Discovery 3: Offload Decode 的 Attention 流程 + +**文件**: `nanovllm/kvcache/sparse/full_policy.py:304-379` + +**Ring Buffer Pipeline**: +``` +1. 预加载前 N 个 blocks 到 GPU slots +2. 对每个 block: + a. wait_slot_layer() # 等待 H2D + b. get_kv_for_slot() # 获取 KV + c. flash_attn_with_lse() # ⭐ 可 graph + d. record_slot_compute_done() + e. load_next_block() # 启动下一个 H2D + f. merge_attention_outputs() # ⭐ 可 graph（但动态） +``` + +**关键**: H2D 传输不能 graph，但 attention 计算可以。 + +--- + +## Discovery 4: 验证 Graph 复用可行性 + +**测试**: `tests/test_chunk_attention_graph_reuse.py` + +**结论**: +- 只需 2 个 graph（causal + non-causal） +- 通过 `copy_()` 更新 static tensors +- 可复用于所有层和所有 chunk pairs + +**测试结果**: +``` +Layer 0: max_diff=3.91e-03 ✅ +Layer 1: max_diff=7.81e-03 ✅ +Layer 2: max_diff=3.91e-03 ✅ +✅ PASSED +``` + +--- + +## Discovery 5: Chunk Size 和 Block Size 关系 + +**观察**: +- Prefilled blocks 的 KV size = `block_size` +- Decode buffer 的 KV size = `1` 到 `block_size`（动态） + +**Graph 策略**: +- Prefilled blocks: 固定 size = block_size，适合 graph +- Decode buffer: 动态 size，建议保持 eager + +--- + +## Discovery 6: 使用的 Triton 算子 + +**文件**: `nanovllm/ops/chunked_attention.py` + +| 算子 | 功能 | 可 Graph | +|------|------|----------| +| `flash_attn_with_lse()` | Attention + LSE | ✅ | +| `merge_attention_outputs()` | 合并两个 attention 输出 | ✅ | + +这两个算子是纯 GPU 计算，可以被 CUDA Graph 捕获。 + +--- + +## Discovery 7: 数据依赖分析 + +**Attention 输入**: +- `q`: 来自当前层的 QKV projection，shape 固定 +- `k, v`: 来自 GPU slot（H2D 传输后），shape = [1, block_size, heads, dim] + +**依赖链**: +``` +H2D(block) → wait() → get_kv() → copy_to_static() → graph.replay() → clone_output() +``` + +**关键**: Graph 只封装 attention 计算，不包含数据传输。 diff --git a/progress.md b/progress.md new file mode 100644 index 0000000..dc78479 --- /dev/null +++ b/progress.md @@ -0,0 +1,55 @@ +# Progress: CUDA Graph for Offload Mode + +## Session: 2026-01-22 + +### 调研阶段 ✅ 完成 + +**完成的调研**: + +1. ✅ 分析 `model_runner.py` 中的 CUDA Graph 实现 + - `capture_cudagraph()`: 为不同 batch size 捕获完整 model forward + - `run_model()`: 通过 `is_chunked_prefill` 决定 eager/graph + +2. ✅ 分析 offload decode 流程 + - `run_chunked_offload_decode()` 设置 `is_chunked_prefill=True` + - 导致永远使用 eager mode + +3. ✅ 分析 ring buffer pipeline + - `_decode_ring_buffer_pipeline()` 包含 H2D 传输 + attention 计算 + - H2D 不能 graph，attention 可以 graph + +4. ✅ 验证 graph 复用策略 + - 创建 `test_chunk_attention_graph_reuse.py` + - 确认 2 个 graph 可复用于所有层 + +### 计划编写 ✅ 完成 + +- ✅ 创建 `task_plan.md` +- ✅ 创建 `findings.md` +- ✅ 创建 `progress.md` + +### 下一步: 实现 + +**Phase 1**: 添加 graph 捕获到 OffloadEngine +- [ ] 在 `offload_engine.py` 添加 `capture_attention_graphs()` +- [ ] 添加 `attention_graph_causal` 和 `attention_graph_non_causal` 属性 + +**Phase 2**: 修改 ring buffer pipeline +- [ ] 在 `_decode_ring_buffer_pipeline()` 使用 graph replay +- [ ] 保持 H2D 和 merge 为 eager + +**Phase 3**: 测试 +- [ ] 运行 needle test 验证正确性 +- [ ] 对比性能 + +--- + +## 文件清单 + +| 文件 | 状态 | 说明 | +|------|------|------| +| `tests/test_chunk_attention_graph.py` | ✅ 已提交 | 预分配 chunk pair graphs 测试 | +| `tests/test_chunk_attention_graph_reuse.py` | 待提交 | Graph 复用验证 | +| `task_plan.md` | ✅ 创建 | 实现计划 | +| `findings.md` | ✅ 创建 | 调研发现 | +| `progress.md` | ✅ 创建 | 进度日志 | diff --git a/task_plan.md b/task_plan.md index 5255c1a..e2dcc8d 100644 --- a/task_plan.md +++ b/task_plan.md @@ -1,90 +1,357 @@ -# Task Plan: XAttention BSA 集成到 nanovllm +# Task Plan: CUDA Graph 优化 Offload Mode Decode -## Goal +## 目标 -使用 `--sparse-policy XATTN_BSA` 运行 `test_ruler.py`，通过 `niah_single_1` 的前 5 个 sample。 +为 nanovllm 的 CPU offload 模式添加 CUDA Graph 支持，加速 decode 阶段的计算。 -**验收标准**: -```bash -CUDA_VISIBLE_DEVICES=X PYTHONPATH=/home/zijie/Code/nano-vllm:$PYTHONPATH \ - python tests/test_ruler.py \ - --model ~/models/Llama-3.1-8B-Instruct \ - --enable-offload \ - --sparse-policy XATTN_BSA \ - --task niah_single_1 \ - --sample-ids 0,1,2,3,4 -# 期望: 5/5 PASS +## 问题分析 + +### Transformer 层的完整结构 + +``` +Qwen3DecoderLayer.forward: +├── input_layernorm (RMSNorm) # ✅ 纯 GPU +├── self_attn: +│ ├── qkv_proj (Linear) # ✅ 纯 GPU +│ ├── q_norm, k_norm (RMSNorm) # ✅ 纯 GPU +│ ├── rotary_emb # ✅ 纯 GPU +│ ├── attn._chunked_decode_attention: # ⚠️ 包含 CPU→GPU +│ │ ├── H2D transfer # ❌ 不能 graph +│ │ ├── flash_attn_with_lse # ✅ 可以 graph +│ │ └── merge # ✅ 纯 GPU +│ └── o_proj (Linear) # ✅ 纯 GPU +├── post_attention_layernorm # ✅ 纯 GPU +└── mlp (FFN: gate, up, down) # ✅ 纯 GPU ``` -## 当前状态 +**核心问题**：H2D 传输被嵌在 attention 中间，打断了整层的 graph 捕获。 -- `XAttentionBSAPolicy.compute_chunked_prefill` 实现 = `FullAttentionPolicy`（无 sparse） -- `xattn_estimate_chunked` 已实现并验证 -- BSA kernel (`block_sparse_attn`) 可用 +### 可能的方案 -## Phases +| 方案 | 描述 | 优点 | 缺点 | +|------|------|------|------| +| A. 分段 Graph | 将层拆分为 pre/post attention 两段 | 覆盖面广 | 改动大，需拆分层执行 | +| B. 只 Graph Attention | 只优化 flash_attn_with_lse | 改动小 | 优化效果有限 | +| C. 重构执行流程 | 完全重写 model forward | 最优效果 | 工作量巨大 | -- [ ] Phase 1: 理解当前代码路径 -- [ ] Phase 2: 实现 sparse mask 估计 -- [ ] Phase 3: 实现 BSA sparse 计算 -- [ ] Phase 4: 测试验证 +### 推荐：方案 A（分段 Graph） -## Phase 1: 理解当前代码路径 +将每层拆分为两个 graph： +1. **pre_attention_graph**: `norm → qkv_proj → q/k_norm → rotary` +2. **post_attention_graph**: `o_proj → norm → FFN` -### 1.1 确认 XATTN_BSA policy 是否被正确加载 -- [ ] 检查 `test_ruler.py` 如何解析 `--sparse-policy XATTN_BSA` -- [ ] 检查 `KVCacheManager` 如何实例化 sparse_policy -- [ ] 运行 baseline 测试（`--sparse-policy FULL`）确认基础功能正常 +中间的 `_chunked_decode_attention` 保持 eager（包含 H2D），但内部的 `flash_attn_with_lse` 使用 graph。 -### 1.2 确认数据流 -- [ ] `compute_chunked_prefill` 的输入参数含义 -- [ ] `offload_engine` 提供的数据访问接口 -- [ ] 当前 chunk 的 K/V 如何获取 +--- -## Phase 2: 实现 sparse mask 估计 +## 当前状态分析 -### 2.1 调用 xattn_estimate_chunked -- [ ] 在 `compute_chunked_prefill` 中加载历史 K -- [ ] 拼接历史 K + 当前 K -- [ ] 调用 `xattn_estimate_chunked(q, k_full, q_start_pos=...)` -- [ ] 获取 block mask +### 现有 CUDA Graph 实现 -### 2.2 处理参数对齐 -- [ ] BSA block_size = 128 -- [ ] chunk_size 与 kvcache_block_size 的关系 -- [ ] q_start_pos 计算 +**文件**: `nanovllm/engine/model_runner.py` -## Phase 3: 实现 BSA sparse 计算 +| 方法 | 行号 | 功能 | +|------|------|------| +| `capture_cudagraph()` | 682-717 | 为不同 batch size 捕获完整 model forward | +| `run_model()` | 415-436 | 决定使用 eager 还是 graph replay | -### 3.1 方案选择 -- 选项 A: 历史 + 当前分开计算，然后 merge -- 选项 B: 全部一起用 BSA 计算 +**关键逻辑** (`run_model`): +```python +use_eager = is_prefill or self.enforce_eager or input_ids.size(0) > 512 or context.is_chunked_prefill +``` -### 3.2 实现 -- [ ] 构造 BSA 需要的输入格式 -- [ ] 调用 `block_sparse_attn_func` -- [ ] 处理输出格式 +**问题**: `run_chunked_offload_decode` 设置 `is_chunked_prefill=True`，导致**永远使用 eager mode**。 -## Phase 4: 测试验证 +### Offload Decode 流程 -### 4.1 单元测试 -- [ ] 验证 sparse mask 与 `test_xattn_estimate_chunked.py` 一致 +**文件**: `nanovllm/kvcache/sparse/full_policy.py` -### 4.2 集成测试 -- [ ] 运行验收命令 -- [ ] 5/5 PASS +`_decode_ring_buffer_pipeline()` (L304-379): +``` +for block in cpu_blocks: + 1. wait_slot_layer(slot) # 等待 H2D 完成 + 2. k, v = get_kv_for_slot(slot) # 获取 KV + 3. o, lse = flash_attn_with_lse() # ⭐ 纯 GPU 计算 + 4. record_slot_compute_done(slot) # 标记计算完成 + 5. load_next_block() # 启动下一个 H2D + 6. merge_attention_outputs() # ⭐ 纯 GPU 计算 +``` -## Key Questions +**可 Graph 化的部分**: +- `flash_attn_with_lse()` - 纯 GPU 计算 +- 不可 Graph 化: H2D 传输、动态 merge -1. 历史 K 如何高效加载？（全量 vs 按需） -2. BSA causal mask 如何处理？（历史 non-causal + 当前 causal） +## 验证结果 -## Status +**测试文件**: `tests/test_chunk_attention_graph_reuse.py` -**Currently in Phase 1** - 等待用户确认后开始 +| 测试 | 结果 | +|------|------| +| 2 个 Graph 复用于所有层和所有 chunk | ✅ PASSED | +| copy_() 更新 static tensors | ✅ 有效 | +| Eager merge | ✅ 用户已接受 | -## 待讨论 +**结论**: 只需 2 个 graph（causal + non-causal），通过 copy_() 复用。 -请确认： -1. 这个 goal 和验收标准是否正确？ -2. 我使用哪个 GPU 运行测试？ +--- + +## 修改计划（方案 A：分段 Graph） + +### 架构设计 + +``` +每层执行流程（Offload Decode）: +┌─────────────────────────────────────────────────────────────┐ +│ PRE-ATTENTION GRAPH (可复用于所有层) │ +│ input_layernorm → qkv_proj → q/k_norm → rotary → split Q │ +└─────────────────────────────────────────────────────────────┘ + ↓ +┌─────────────────────────────────────────────────────────────┐ +│ CHUNKED ATTENTION (Eager + 部分 Graph) │ +│ for block in cpu_blocks: │ +│ H2D transfer (eager) │ +│ flash_attn_with_lse (GRAPH - 2个可复用) │ +│ merge (eager) │ +│ decode_buffer attention (eager) │ +└─────────────────────────────────────────────────────────────┘ + ↓ +┌─────────────────────────────────────────────────────────────┐ +│ POST-ATTENTION GRAPH (可复用于所有层) │ +│ o_proj → post_layernorm → gate_proj → up_proj → SiLU │ +│ → down_proj → residual │ +└─────────────────────────────────────────────────────────────┘ +``` + +**总共需要的 Graph 数量**: +- 1 个 pre_attention_graph（所有层复用） +- 2 个 attention_graph（causal + non-causal，所有层复用） +- 1 个 post_attention_graph（所有层复用） +- **总计: 4 个 graph** + +--- + +### Phase 1: 拆分 DecoderLayer 执行 + +**目标**: 将 `Qwen3DecoderLayer.forward` 拆分为可独立调用的三段 + +**修改文件**: `nanovllm/models/qwen3.py` + +**新增方法**: +```python +class Qwen3DecoderLayer: + def forward_pre_attention(self, positions, hidden_states, residual): + """Pre-attention: norm → qkv → rotary → 返回 q, k, v""" + if residual is None: + hidden_states, residual = self.input_layernorm(hidden_states), hidden_states + else: + hidden_states, residual = self.input_layernorm(hidden_states, residual) + + qkv = self.self_attn.qkv_proj(hidden_states) + q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1) + q = q.view(-1, self.num_heads, self.head_dim) + k = k.view(-1, self.num_kv_heads, self.head_dim) + v = v.view(-1, self.num_kv_heads, self.head_dim) + q = self.self_attn.q_norm(q) + k = self.self_attn.k_norm(k) + q, k = self.self_attn.rotary_emb(positions, q, k) + return q, k, v, hidden_states, residual + + def forward_post_attention(self, attn_output, hidden_states, residual): + """Post-attention: o_proj → norm → FFN""" + output = self.self_attn.o_proj(attn_output.flatten(1, -1)) + hidden_states, residual = self.post_attention_layernorm(output, residual) + hidden_states = self.mlp(hidden_states) + return hidden_states, residual +``` + +**状态**: `pending` + +--- + +### Phase 2: 捕获 Pre/Post Attention Graph + +**目标**: 捕获 pre_attention 和 post_attention 的 graph + +**修改文件**: `nanovllm/engine/model_runner.py` + +**新增方法**: `capture_offload_layer_graphs()` + +```python +def capture_offload_layer_graphs(self): + """捕获 offload mode 的 layer graphs""" + # 获取任意一层作为模板（所有层结构相同） + layer = self.model.model.layers[0] + + # Static tensors + static_hidden = torch.zeros(1, self.hidden_size, ...) + static_residual = torch.zeros(1, self.hidden_size, ...) + static_positions = torch.zeros(1, ...) + + # Pre-attention graph + self.pre_attn_graph = torch.cuda.CUDAGraph() + with torch.cuda.graph(self.pre_attn_graph): + static_q, static_k, static_v, _, _ = layer.forward_pre_attention( + static_positions, static_hidden, static_residual + ) + + # Post-attention graph + self.post_attn_graph = torch.cuda.CUDAGraph() + with torch.cuda.graph(self.post_attn_graph): + _, _ = layer.forward_post_attention( + static_attn_output, static_hidden, static_residual + ) +``` + +**状态**: `pending` + +--- + +### Phase 3: 捕获 Attention Graph + +**目标**: 捕获 2 个 attention graph（causal + non-causal） + +**修改文件**: `nanovllm/kvcache/offload_engine.py` + +```python +class OffloadEngine: + def capture_attention_graphs(self): + """捕获 attention graphs（复用于所有层）""" + self.attn_graph_causal = self._capture_attn_graph(causal=True) + self.attn_graph_non_causal = self._capture_attn_graph(causal=False) + + def _capture_attn_graph(self, causal: bool): + static_q = torch.zeros(1, 1, num_heads, head_dim, ...) + static_k = torch.zeros(1, block_size, num_kv_heads, head_dim, ...) + static_v = torch.zeros(1, block_size, num_kv_heads, head_dim, ...) + + graph = torch.cuda.CUDAGraph() + with torch.cuda.graph(graph): + output, lse = flash_attn_with_lse(static_q, static_k, static_v, + self.scale, causal) + return AttentionGraph(graph, static_q, static_k, static_v, output, lse) +``` + +**状态**: `pending` + +--- + +### Phase 4: 修改 Offload Decode 执行流程 + +**目标**: 使用 graph replay 执行 offload decode + +**修改文件**: `nanovllm/engine/model_runner.py` + +**修改方法**: `run_chunked_offload_decode()` + +```python +def run_chunked_offload_decode_with_graph(self, seqs): + """使用 graph 加速的 offload decode""" + seq = seqs[0] + + # 准备输入 + input_ids = torch.tensor([seq.last_token], ...) + positions = torch.tensor([len(seq) - 1], ...) + + # Embedding + hidden_states = self.model.model.embed_tokens(input_ids) + residual = None + + for layer_id, layer in enumerate(self.model.model.layers): + # Phase 1: Pre-attention (GRAPH) + self.pre_attn_vars["hidden"].copy_(hidden_states) + self.pre_attn_vars["residual"].copy_(residual) if residual else None + self.pre_attn_vars["positions"].copy_(positions) + self.pre_attn_graph.replay() + q = self.pre_attn_vars["q"].clone() + k = self.pre_attn_vars["k"].clone() + v = self.pre_attn_vars["v"].clone() + + # Phase 2: Chunked Attention (Eager + Graph) + attn_output = self._chunked_attention_with_graph(q, k, v, layer_id, ...) + + # Phase 3: Post-attention (GRAPH) + self.post_attn_vars["attn_output"].copy_(attn_output) + self.post_attn_graph.replay() + hidden_states = self.post_attn_vars["hidden"].clone() + residual = self.post_attn_vars["residual"].clone() + + # LM head + logits = self.model.compute_logits(hidden_states) + return logits +``` + +**状态**: `pending` + +--- + +### Phase 5: 修改 Ring Buffer Pipeline + +**目标**: 在 attention 内部使用 graph + +**修改文件**: `nanovllm/kvcache/sparse/full_policy.py` + +**修改**: `_decode_ring_buffer_pipeline()` 中的 `flash_attn_with_lse` 调用 + +```python +# 当前：eager +prev_o, prev_lse = flash_attn_with_lse(q, k, v, scale, causal=False) + +# 修改为：graph replay +graph = offload_engine.attn_graph_non_causal +graph.static_q.copy_(q) +graph.static_k.copy_(k) +graph.static_v.copy_(v) +graph.graph.replay() +prev_o = graph.static_output.clone() +prev_lse = graph.static_lse.clone() +``` + +**状态**: `pending` + +--- + +### Phase 6: 添加配置开关 + +**修改文件**: `nanovllm/config.py` + +```python +enable_offload_graph: bool = True # 默认启用 +``` + +**状态**: `pending` + +--- + +## 文件修改清单 + +| 文件 | 修改类型 | 说明 | +|------|----------|------| +| `nanovllm/engine/model_runner.py` | 新增方法 | `capture_offload_attention_graph()` | +| `nanovllm/kvcache/offload_engine.py` | 新增属性+方法 | Graph 存储和访问 | +| `nanovllm/kvcache/sparse/full_policy.py` | 修改方法 | 使用 graph replay | +| `nanovllm/config.py` | 新增配置 | `enable_offload_graph` | + +--- + +## 风险和注意事项 + +1. **Graph 捕获时机**: 需要在 KV cache 分配后、第一次 decode 前捕获 +2. **Chunk size 匹配**: Graph 的 chunk_size 必须和 block_size 一致 +3. **多 GPU**: Graph 需要在每个 GPU 上分别捕获 +4. **内存**: 2 个 graph 的额外内存开销很小 + +--- + +## 测试计划 + +1. **单元测试**: 验证 graph replay 结果正确 +2. **集成测试**: 运行 `test_needle.py --enable-offload --input-len 32768` +3. **性能测试**: 对比 eager vs graph 的 decode 延迟 + +--- + +## 预期收益 + +- Decode 阶段 attention 计算加速（减少 kernel launch overhead） +- 与现有 ring buffer pipeline 兼容 +- 内存开销极小（只有 2 个额外 graph） diff --git a/tests/test_chunk_attention_graph_reuse.py b/tests/test_chunk_attention_graph_reuse.py new file mode 100644 index 0000000..a2afb29 --- /dev/null +++ b/tests/test_chunk_attention_graph_reuse.py @@ -0,0 +1,156 @@ +#!/usr/bin/env python3 +""" +Test: Reuse a single CUDA Graph across all layers and all chunk pairs. + +Key insight: LLM layers have identical computation structure. +We only need 2 graphs (causal + non-causal), reused for all (layer, Q_i, K_j) combinations. + +Usage: + CUDA_VISIBLE_DEVICES=0 python tests/test_chunk_attention_graph_reuse.py +""" + +from dataclasses import dataclass + +import torch + +from nanovllm.ops.chunked_attention import flash_attn_with_lse, merge_attention_outputs + + +@dataclass +class ReusableChunkGraph: + """A single graph that can be reused with copy_() updates.""" + graph: torch.cuda.CUDAGraph + static_q: torch.Tensor + static_k: torch.Tensor + static_v: torch.Tensor + static_output: torch.Tensor + static_lse: torch.Tensor + + +def capture_reusable_graph( + chunk_size: int, + num_heads: int, + num_kv_heads: int, + head_dim: int, + scale: float, + device: torch.device, + dtype: torch.dtype, + causal: bool, +) -> ReusableChunkGraph: + """Capture ONE graph to be reused for all chunk pairs.""" + static_q = torch.zeros(1, chunk_size, num_heads, head_dim, dtype=dtype, device=device) + static_k = torch.zeros(1, chunk_size, num_kv_heads, head_dim, dtype=dtype, device=device) + static_v = torch.zeros(1, chunk_size, num_kv_heads, head_dim, dtype=dtype, device=device) + + static_q.normal_() + static_k.normal_() + static_v.normal_() + + # Warmup + with torch.inference_mode(): + for _ in range(3): + _ = flash_attn_with_lse(static_q, static_k, static_v, scale, causal) + torch.cuda.synchronize() + + # Capture + graph = torch.cuda.CUDAGraph() + with torch.inference_mode(): + with torch.cuda.graph(graph): + static_output, static_lse = flash_attn_with_lse(static_q, static_k, static_v, scale, causal) + + torch.cuda.synchronize() + + return ReusableChunkGraph( + graph=graph, + static_q=static_q, + static_k=static_k, + static_v=static_v, + static_output=static_output, + static_lse=static_lse, + ) + + +def replay_with_copy(graph: ReusableChunkGraph, q: torch.Tensor, k: torch.Tensor, v: torch.Tensor): + """Replay graph after updating static tensors with copy_().""" + graph.static_q.copy_(q) + graph.static_k.copy_(k) + graph.static_v.copy_(v) + graph.graph.replay() + return graph.static_output.clone(), graph.static_lse.clone() + + +def main(): + device = torch.device("cuda") + dtype = torch.bfloat16 + + chunk_size = 64 + num_chunks = 4 + num_layers = 3 # Simulate multiple layers + num_heads = 8 + num_kv_heads = 8 + head_dim = 64 + scale = 1.0 / (head_dim ** 0.5) + seq_len = chunk_size * num_chunks + + print(f"Device: {torch.cuda.get_device_name()}") + print(f"Chunk size: {chunk_size}, Num chunks: {num_chunks}, Num layers: {num_layers}") + print(f"Only 2 graphs (causal + non-causal) for ALL layer × chunk combinations") + + # Capture only 2 graphs + graph_causal = capture_reusable_graph( + chunk_size, num_heads, num_kv_heads, head_dim, scale, device, dtype, causal=True + ) + graph_non_causal = capture_reusable_graph( + chunk_size, num_heads, num_kv_heads, head_dim, scale, device, dtype, causal=False + ) + print("2 graphs captured (causal + non-causal)") + + all_pass = True + + for layer_id in range(num_layers): + # Different Q/K/V for each layer (simulating different layer outputs) + full_q = torch.randn(1, seq_len, num_heads, head_dim, dtype=dtype, device=device) + full_k = torch.randn(1, seq_len, num_kv_heads, head_dim, dtype=dtype, device=device) + full_v = torch.randn(1, seq_len, num_kv_heads, head_dim, dtype=dtype, device=device) + + # Reference: full causal attention + with torch.inference_mode(): + full_output, _ = flash_attn_with_lse(full_q, full_k, full_v, scale, causal=True) + + # Chunked with graph reuse + chunked_output = torch.zeros_like(full_output) + + for q_idx in range(num_chunks): + q_chunk = full_q[:, q_idx*chunk_size:(q_idx+1)*chunk_size] + acc_out, acc_lse = None, None + + for k_idx in range(q_idx + 1): + k_chunk = full_k[:, k_idx*chunk_size:(k_idx+1)*chunk_size] + v_chunk = full_v[:, k_idx*chunk_size:(k_idx+1)*chunk_size] + + # Reuse graph with copy_() + graph = graph_causal if k_idx == q_idx else graph_non_causal + out, lse = replay_with_copy(graph, q_chunk, k_chunk, v_chunk) + + if acc_out is None: + acc_out, acc_lse = out, lse + else: + with torch.inference_mode(): + acc_out, acc_lse = merge_attention_outputs(acc_out, acc_lse, out, lse) + + chunked_output[:, q_idx*chunk_size:(q_idx+1)*chunk_size] = acc_out + + torch.cuda.synchronize() + + # Compare + max_diff = (full_output - chunked_output).abs().max().item() + status = "✅" if max_diff < 1e-2 else "❌" + print(f"Layer {layer_id}: max_diff={max_diff:.2e} {status}") + if max_diff >= 1e-2: + all_pass = False + + print("✅ PASSED - Single graph reuse across layers works!" if all_pass else "❌ FAILED") + + +if __name__ == "__main__": + main()