Skip to content

About

TensorFold deployment for GLM-5.3-Flash on 4x NVIDIA CMP 170HX (GA100 sm_80): bit-exact 4-rank TP, shm all-gather, W4A16 affine - complete tested solution (199.4 tok/s structured, exactness maintained)

Topics

Resources

Contributing

Stars

3 stars

Watchers

0 watching

Forks

Repository files navigation

CMP170HX-TensorFold — AX300 Engine

NVIDIA CMP 170HX (GA100, sm_80) 上的 GLM-5.3-Flash 推理引擎,支持 512K 上下文。

中文 | English


引擎名称:AX300

  • AX = Axis(主线/轴心)—— 本项目在 TensorFold python-0.6 HEAD 之上建立的自研主线
  • 300 = tok/s 冲刺目标
  • AX300 = 以 300 tok/s 为目标的 CMP 170HX 主线引擎

代码出处声明(自检通过)

本项目全部通信、量化和优化代码均为自研:

组件 出处 自研?
通信内核(4 个 CUDA 内核) shm_allgather.py ✅ 100% 自研
fp8-over-NCCL / wide-NB / fp8 gather 本项目新增 ✅ 100% 自研
基础框架 TensorFold python-0.6 (ashhart, Apache/MIT) 框架(类比 PyTorch)

实测性能(2026-10-13,AX300 生产引擎)

Decode(三个场景 × 上下文长度,全打出来)

上下文 structured coding prose exactness
32K 241.6 181.7 117.1 PASS
128K 241.2 181.3 116.7 PASS
256K 241.5 181.5 117.3 PASS
512K 241.1 181.5 117.0 PASS

冷 Prefill(唯一 prompt,非缓存)

上下文 (tok) prefill (tok/s) 时延
5,931 1,561 3.8s
30,684 1,522 20.2s
129,830 1,483 87.6s
259,661 1,438 180.5s
519,322 1,394 372.6s

显存

上下文 每卡 余量
32K 46.35 GiB 10.2 GiB
512K 55.37 GiB 1.2 GiB

快速启动

~/tensorfold-poc/ax300_prod.sh start     # ~15 分钟加载,:8000 服务
~/tensorfold-poc/ax300_prod.sh status    # 健康检查
~/tensorfold-poc/ax300_prod.sh stop      # 停止

# 基准(decode 三数 + exactness)
TF_BENCH_BASE=http://127.0.0.1:8000 venv/bin/python tf4_ab_bench.py --profile structured --tag <名> --exactness
# 冷 prefill(带上下文长度列)
TF_BENCH_BASE=http://127.0.0.1:8000 venv/bin/python prefill_probe.py 6000 31000 131000

环境开关(全部默认关,生产脚本 ax300_prod.sh 已开启)

变量 作用 增益
TF_SHM_AG=1 自研 shm one-shot all-gather +21.5%
TF_BF16_GATHER=1 bf16 通信载荷 +35.8%
TF_FP8_GATHER=1 fp8 通信载荷 +20.7%
TF_FP8_WIDE=1 wide-NB 内核 prefill 4.5×
TF_FP8_NCCL=1 fp8-over-NCCL prefill 1.7×
TF_CONTEXT=N 上下文(最大 524288) —

文件结构

cmp170hx/patches/       构建补丁(sm_80 准入等)
cmp170hx/results/       实验收据(TEST-MATRIX.md 为总表)
cmp170hx/scripts/       基准和测试脚本
docs/                   设计文档、PP4 规格、长上下文评估
tools/                  重合检查器等工具

致谢

  • TensorFold 引擎框架:ashhart/TensorFold(Apache/MIT)
  • host-shm collective 思想参考
  • copy-drafts 思想:BobClawblaw(上游 PR #468)

许可

见 LICENSE 和 THIRD_PARTY_NOTICES。


English

GLM-5.3-Flash inference on 4× NVIDIA CMP 170HX (sm_80, PCIe-only, no P2P). 512K context support. All communication, quantization, and optimization code is self-developed.

Performance: 241 tok/s structured / 181 coding / 117 prose decode (stable across 32K-512K context); 1,394-1,586 tok/s cold prefill. Exactness contract drafted == serial maintained.

Engine name: AX300 = "Axis 300" — the mainline branch targeting 300 tok/s on CMP 170HX.

Framework: TensorFold python-0.6 HEAD (ashhart, Apache/MIT), used as the engine framework (analogous to PyTorch for applications). Our sm_80 admission (PR #464) and GLM 4-rank (PR #465) were relanded upstream under our name.

About

TensorFold deployment for GLM-5.3-Flash on 4x NVIDIA CMP 170HX (GA100 sm_80): bit-exact 4-rank TP, shm all-gather, W4A16 affine - complete tested solution (199.4 tok/s structured, exactness maintained)

Topics

Resources

Contributing

Stars

3 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages