Files
project_6/MUH_PROJECT_CHECKPOINT.md

209 lines
11 KiB
Markdown
Raw Normal View History

# MUH Project Checkpoint
> **最后更新**: 2026-07-30
> **GitHub Project**: github.com/users/dylanyunlon/projects/6
> **代码仓库**: github.com/dylanyunlon/project_6
> **竞赛截止**: 2026-09-30
---
## 一、项目是什么
参加信创模盒 ModelHub XC 的"模型适配引擎竞赛-第一届"。目标是优化 vllm 引擎,让 Qwen3.6-35B-A3B 在天数智芯天垓1004×BI-V100 GPU上跑出最高的 Token 吞吐加权值。
**计分公式**:
```
Token吞吐加权值 = Output TPS × 16.796 + Input TPS × 2.799 + Cache TPS × 0.56
```
Output TPS 权重占 83%——decode 阶段优化收益最大。
**奖项**:
- 基础奖 200,000 积分1:1 兑现金): 通过全部功能/效果测试 + 性能达标(≥8000)
- 高级奖 +100,000: 加权值提升 ≥ 30%
- 特级奖 +50,000: 加权值提升 ≥ 50%
## 二、竞赛测评流程
参赛者提交的是 **Git 仓库地址**(在 dev.modelhub.org.cn 上)。平台自动执行:
1. **构建镜像**: 读取仓库根目录的 `Dockerfile`,基于基础镜像 `harbor.4pd.io/modelhubxc/enginex-iluvatar/bi100-3.2.3-x86-ubuntu20.04-py3.10-poc-llm-infer:v1.2.3` 构建
2. **启动服务**: 读取 `computility-run.yaml``command`,在 4×天垓100 容器里启动 vllm api server模型权重平台预挂载在 `/model`
3. **功能测试(门控)**: 50+ 个 OpenAI 兼容 API 测试用例,全部通过才进入下一步
4. **效果测试(门控)**: 标准 benchmark 偏差 ≤ ±4%
5. **性能测试(排名)**: 计算加权值
**你能改的**: Dockerfile + vllm 源码 + computility-run.yaml 启动参数。模型本身不能改。
## 三、muh 是什么
muh 是我们设计的 **tuning DSL领域特定语言**,用于:
1. 把 CCCL 的 tuning patternblock_threads / items_per_thread / load_algorithm / cache_modifier 等)抽象成硬件无关的参数空间
2. 针对天垓100 的硬件特性搜索最优参数组合
3. Codegen 输出实际的 vllm kernel 修改 + computility-run.yaml + Dockerfile
**为什么需要它**: CCCL 有 27 个 tuning_*.cuh 文件17000+ 行),每个算法都有针对不同 NVIDIA SM 架构的特化参数。天垓100 不是 NVIDIA GPU不能直接用这些参数但 tuning 的维度block size、warp 策略、shared memory 用量、prefetch 策略是通用的。muh 让迁移过程变成"改配置 + 跑 benchmark"而不是"手改 kernel + 祈祷"。
[MUH] Fix 7 structural discrepancies vs CCCL — read source, not grep Fixes found by reading all 17 muh files + 6 CCCL counterpart policy_selectors as full source code input: 1. topk: BLOCK_LOAD_DIRECT → BLOCK_LOAD_VECTORIZE (CCCL SM90+ uses VECTORIZE). bits_per_pass was wrong (muh: ks<=4→9, CCCL: ks>=2→11). items now computed dynamically (4*4/key_size) not hardcoded. 2. reduce: added determinism dispatch — three modes matching CCCL: gpu_to_gpu (BLOCK_REDUCE_RAKING, vec_size=1, LOAD_DEFAULT), run_to_run (WARP_REDUCTIONS, LOAD_LDG, default), not_guaranteed (WARP_REDUCTIONS_NONDETERMINISTIC). Added bi100_det_float32 and bi100_det_float64 tuning structs with SM90 benchmark reference values. 3. batch_memcpy: flat single-tier → SmallBuffer+LargeBuffer two-tier matching CCCL structure (128 threads small, 256 threads large, warp_threshold=128, block_threshold=8192). 4. transform: single BulkPolicy → three-policy structure (VectorizedPolicy + AsyncCopyPolicy + PrefetchPolicy) matching CCCL. items_per_thread computed from bytes_in_flight / (threads * elem_size). 5. compile_test: 17 checks → 33 checks. Now verifies exact values: reduce determinism modes, topk VECTORIZE + bits=11, batch_memcpy two-tier thresholds, transform three-policy structure. 6. gen_patch: added fallback extraction for inline policy_selector values (topk now generates SAMPLING_BLOCK_SIZE patch). 7. MUH_PROJECT_CHECKPOINT.md: 'PRD设计阶段还没有代码' → actual status. 7 files changed, 413 insertions, 265 deletions.
2026-07-30 14:37:38 +00:00
**muh 的状态**: v0.3 — 6个算法的C++ tuning headers已就绪(reduce/scan/topk/transform/batch_memcpy/for)compile_test 33项通过gen_patch.py从C++ headers提取bi100值生成vllm patches。参数值从CCCL SM100复制等BI-V100实测替换。
## 四、已完成的工作
### 4.1 Project 6 已有 16 个真实 GitHub Issue不是 Draft
都在 `dylanyunlon/project_6` 仓库里,已关联到 GitHub Project 6有 label 和 Priority
| # | 标题 | Labels | Priority |
|---|------|--------|----------|
| 1 | [FEA] 非流式基础对话 | 基本功能,vllm,天垓100,Qwen3.6 | P0 |
| 2 | [FEA] 流式对话 SSE | 基本功能,vllm | P0 |
| 3 | [FEA] Tool Calling | 基本功能,vllm,Qwen3.6 | P0 |
| 4 | [FEA] Reasoning/Thinking 分离 | 基本功能,thinking,Qwen3.6 | P0 |
| 5 | [FEA] Prefix Cache | 基本功能,性能测试,vllm | P0 |
| 6 | [FEA] 采样参数边界 | 采样参数,vllm | P1 |
| 7 | [FEA] max_tokens 边界 | max_tokens,vllm | P1 |
| 8 | [FEA] 结构化输出 | 结构化输出,vllm | P0 |
| 9 | [FEA] 多语言 Emoji | 多语言,Qwen3.6 | P1 |
| 10 | [FEA] 多模态 base64 PNG | 多模态,基本功能,Qwen3.6 | P0 |
| 11 | [FEA] 参数校验 | 参数校验,vllm | P1 |
| 12 | [FEA] 基础能力 | 基础能力,vllm,Qwen3.6 | P0 |
| 13 | [FEA] 输出截断 | 截断测试,vllm | P1 |
| 14 | [FEA] 效果测试 | 效果测试,Qwen3.6,天垓100 | P0 |
| 15 | [EPIC] 性能基准 | 性能测试,天垓100,vllm | P0 |
| 16 | [EPIC] 开发环境与代码提交 | infra,天垓100 | P1 |
这 16 个覆盖了竞赛功能测试的所有 50+ 用例。每个 issue 的 body 里都有 PND 级别的测试用例表(前置条件 + 原子步骤 + 二值判定标准)。
### 4.2 仓库里已有 NVIDIA CCCL 代码
`project_6/cccl_upstream/` 目录下包含完整的 CCCL
- `cub/` — GPU 原语reduce, scan, sort, topk, block/warp/device 三层)
- `thrust/` — 高层算法 + 60 个示例
- `libcudacxx/` — CUDA C++ 标准库
- `cudax/` — 实验性功能allocators, memory resources
- `cub/cub/device/dispatch/tuning/` — 27 个硬件特化 tuning 文件17000+ 行)
### 4.3 Label 体系已建立
仓库上已创建 16 个 label基本功能、thinking、采样参数、max_tokens、基础能力、结构化输出、多语言、多模态、参数校验、截断测试、效果测试、性能测试、infra、vllm、天垓100、Qwen3.6
### 4.4 Project 6 里有 15 个遗留 Draft Issue 需要清理
这些是早期用 addProjectV2DraftIssue 创建的,没有 repo 关联、没有 label。应该从 Project 面板里手动删除。
## 五、还没做的(下一步)
[MUH] Fix 7 structural discrepancies vs CCCL — read source, not grep Fixes found by reading all 17 muh files + 6 CCCL counterpart policy_selectors as full source code input: 1. topk: BLOCK_LOAD_DIRECT → BLOCK_LOAD_VECTORIZE (CCCL SM90+ uses VECTORIZE). bits_per_pass was wrong (muh: ks<=4→9, CCCL: ks>=2→11). items now computed dynamically (4*4/key_size) not hardcoded. 2. reduce: added determinism dispatch — three modes matching CCCL: gpu_to_gpu (BLOCK_REDUCE_RAKING, vec_size=1, LOAD_DEFAULT), run_to_run (WARP_REDUCTIONS, LOAD_LDG, default), not_guaranteed (WARP_REDUCTIONS_NONDETERMINISTIC). Added bi100_det_float32 and bi100_det_float64 tuning structs with SM90 benchmark reference values. 3. batch_memcpy: flat single-tier → SmallBuffer+LargeBuffer two-tier matching CCCL structure (128 threads small, 256 threads large, warp_threshold=128, block_threshold=8192). 4. transform: single BulkPolicy → three-policy structure (VectorizedPolicy + AsyncCopyPolicy + PrefetchPolicy) matching CCCL. items_per_thread computed from bytes_in_flight / (threads * elem_size). 5. compile_test: 17 checks → 33 checks. Now verifies exact values: reduce determinism modes, topk VECTORIZE + bits=11, batch_memcpy two-tier thresholds, transform three-policy structure. 6. gen_patch: added fallback extraction for inline policy_selector values (topk now generates SAMPLING_BLOCK_SIZE patch). 7. MUH_PROJECT_CHECKPOINT.md: 'PRD设计阶段还没有代码' → actual status. 7 files changed, 413 insertions, 265 deletions.
2026-07-30 14:37:38 +00:00
1. ~~muh 语言 PRD 设计~~ ✅ Done — muh是C++ header-only lib不是独立语言
2. ~~从 CCCL tuning_*.cuh 提取参数空间~~ ✅ Done — 6个算法的policy_selector已实现
3. **在BI-V100上跑benchmark** — 用实测数据替换bi100_*中的SM100复制值
4. **获取 enginex-vllm-bi100-qwen36 的实际代码** — 需要在 Phanthy Cloud 开发环境里操作
5. **设计 muh → vllm kernel 的 codegen 管道**
6. **实际在天垓100 上跑 benchmark**
## 六、参考项目
- **NVIDIA CCCL Project #6**: github.com/orgs/NVIDIA/projects/61990 itemsIssue-first 模式label 做模块分类)
- **pub/sub-loop Project #4**: github.com/users/dylanyunlon/projects/41632 itemsDraft-first 模式,已验证 1111 个有真实测试步骤154 个有"按AC验证"占位符)
- **PND 测试库**: 818 条车载软件测试用例,作为 PRD 测试用例质量基准
## 七、关键文件路径
```
project_6/
├── cccl_upstream/ # NVIDIA CCCL 完整代码
│ ├── cub/cub/device/dispatch/tuning/ # 27 个 tuning policy 文件
│ ├── cub/cub/warp/ # warp-level 原语
│ ├── cub/cub/block/ # block-level 原语
│ ├── thrust/examples/ # 60 个优化模式示例
│ └── cudax/...allocators/ # 内存分配器
├── Dockerfile # TODO: 待创建
├── computility-run.yaml # TODO: 待创建
└── muh/ # TODO: muh 语言实现
```
## 八、竞赛关键参数(来自 computility-run.yaml 参考)
```yaml
concurrency: 1
command:
- python3 -m vllm.entrypoints.openai.api_server
- --model /model
- --served-model-name llm
- --max-model-len 100000
- --gpu-memory-utilization 0.9
- -tp 4
- --max-num-seqs 1
- --max-num-batched-tokens 8192
- --enable-chunked-prefill
- --max-seq-len-to-capture 32768
- --enable-auto-tool-choice
- --tool-call-parser qwen3_coder
- --reasoning-parser qwen3
- --enable-prefix-caching
env:
- name: VLLM_ENGINE_ITERATION_TIMEOUT_S
value: 3600
```
基础镜像: `harbor.4pd.io/modelhubxc/enginex-iluvatar/bi100-3.2.3-x86-ubuntu20.04-py3.10-poc-llm-infer:v1.2.3`
## 九、CCCL Tuning 文件全量模型输入记录
**所有 27 个 tuning_*.cuh 文件的完整源码已在本 context 中作为模型输入读取。** 关键发现:
### policy_selector 统一模式
每个算法都有一个 `policy_selector` struct接受 `::cuda::compute_capability cc` 参数,内部按 SM 版本做 if-else 分支:
```
if (cc >= {10, 0}) → sm100 tuning (Blackwell)
if (cc >= {9, 0}) → sm90 tuning (Hopper)
if (cc >= {8, 0}) → sm80 tuning (Ampere)
if (cc >= {7, 0}) → sm70 tuning (Volta)
if (cc >= {6, 0}) → sm60 tuning (Pascal)
fallback → sm50 tuning
```
**muh 的核心工作就是给每个 policy_selector 添加一个 `cc == {iluvatar, 100}` 分支填入在天垓100 上跑出的最优 benchmark 数据。**
### 各算法提取的参数维度
| 算法 | 文件 | 行数 | 参数维度 |
|------|------|------|---------|
| reduce | tuning_reduce.cuh | 478 | threads, items, vec_size, reduce_algorithm, load_modifier, determinism |
| scan | tuning_scan.cuh | 1525 | threads, items, load_algo, load_mod, store_algo, scan_algo, delay_policy + lookahead variant |
| radix_sort | tuning_radix_sort.cuh | 2381 | histogram(threads,items,partitions,radix_bits) + exclusive_sum + onesweep(threads,items,store,rank,scan,partitions,radix_bits) + downsweep + upsweep + single_tile |
| reduce_by_key | tuning_reduce_by_key.cuh | 1735 | threads, items, load_algo, load_mod, scan_algo, delay_policy |
| select_if | tuning_select_if.cuh | 2729 | threads, items, load_algo, load_mod, scan_algo, delay_policy |
| histogram | tuning_histogram.cuh | 363 | threads, pixels_per_thread, vec_size, load_algo, load_mod, rle_compress, mem_preference, work_stealing |
| topk | tuning_topk.cuh | 121 | threads, items (simple, no SM-specific tuning yet) |
| batched_topk | tuning_batched_topk.cuh | 186 | worker_policy array × 6 tiers + multi_worker_policy |
| merge | tuning_merge.cuh | 180 | threads, items, load_mod, store_algo, bulk_copy_keys, bulk_copy_values |
| merge_sort | tuning_merge_sort.cuh | 193 | threads, items, load_algo, load_mod, store_algo |
| transform | tuning_transform.cuh | 549 | threads, items, load_algo, store_algo, load_mod |
| rle_encode | tuning_rle_encode.cuh | 626 | threads, items, load_algo, load_mod, scan_algo, delay_policy |
| rle_non_trivial | tuning_rle_non_trivial_runs.cuh | 691 | threads, items, load_algo, load_mod, store_time_slicing, scan_algo, delay |
| adjacent_diff | tuning_adjacent_difference.cuh | 118 | threads, items, load_algo, load_mod, store_algo (single policy, no SM branching) |
| for | tuning_for.cuh | 78 | threads, items (trivial, 256×2) |
| find | tuning_find.cuh | 90 | threads, items, vec_size, load_mod |
| batch_memcpy | tuning_batch_memcpy.cuh | 227 | small_buffer + large_buffer sub-policies |
| scan_by_key | tuning_scan_by_key.cuh | ~2000 | same as reduce_by_key pattern |
| unique_by_key | tuning_unique_by_key.cuh | ~1500 | same pattern |
| three_way_partition | tuning_three_way_partition.cuh | ~780 | same pattern |
| segmented_* | 4 files | ~1300 total | segmented variants of reduce/scan/sort |
### Benchmark 注释格式
每个 sm100 tuning 都有注释格式:
```
// ipt_22.tpb_384.ns_1904.dcid_6.l2w_830.trp_1.ld_0 1.148442 0.997167 1.139902 1.462651
```
- `ipt` = items_per_thread
- `tpb` = threads_per_block
- `ns` = delay nanoseconds
- `dcid` = delay constructor ID
- `l2w` = L2 cache window
- `trp` = transpose (0=DIRECT, 1=WARP_TRANSPOSE)
- `ld` = load modifier (0=DEFAULT, 1=LDG, 2=CA)
- 4 个数字 = 4 种 problem size 下的加速比 (vs 前代 SM)