dylanyunlong
0 Followers · 0 Following
  • Joined on 2026-08-06
Repositories
10
 Projects Public Activity Starred Repositories
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-11 15:36:21 +08:00
e440207977 fix(build): bridge编译不依赖torch.utils.cpp_extension
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-11 15:32:07 +08:00
7e21571086 test: verify_build.sh for real hardware validation
c17fd30144 feat(CRITICAL): 接入ix_unified到qwen3_5.py MoE prefill路径
Compare 2 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-11 15:14:12 +08:00
14fe8fb0d9 fix(CRITICAL): docker build容错 + max_completion_tokens + extra=ignore + ix_unified bridge
651fb660f1 fix(build): Dockerfile tolerant of patch_ops.sh exit — matches working 26e6cb40
Compare 2 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-11 14:46:53 +08:00
f2a7785700 fix(build): set -e safe VLLM2 mirror — [[ ]] && without || true kills script
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-11 13:35:53 +08:00
b8f84bca64 fix(CRITICAL): restore VLLM2 mirror — paged_attn.py not deployed to corex vllm path
32ee28122e ref(upstream): 搬运 xllm ilu kernel+layer 完整源码 — 2089行 14个API声明
6cdf2ec87b ref(upstream): 搬运 3 大 GDN 上游仓库 — FLA naive ops + vllm GDN 子树 + xllm C++ 参考
5862708b32 feat(CRITICAL): import wudixzy/competition complete corex stack — 12 prebuilt .so + 13 CUDA kernels + 2615-line qwen3_5.py
81875fff52 feat(CRITICAL): rewrite corex_gdn/moe/fa2 to use real ixformer dispatch
Compare 10 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-11 09:53:30 +08:00
0478628f17 fix(PROVEN): _moe_C compiles and runs on real BI-V100 hardware
1cd8ca0649 fix(CRITICAL): align patch_ops.sh with comp 168 — keep base qwen3_5.py + upstream搬运
56146f8130 feat(CRITICAL): ix_bridge call chain + upstream xllm/ds_vllm sync
26e6cb4019 update subrizhi
Compare 4 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 23:27:41 +08:00
db8e677b45 fix(CRITICAL): copy_blocks Tensor→dict conversion for ixformer vllm_copy_cache
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 22:25:31 +08:00
96a4afba43 fix(CRITICAL): copy_blocks → vllm_copy_cache, swap_blocks → vllm_swap_blocks
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 21:10:45 +08:00
539d0fc6ff fix(CRITICAL): deploy paged_attn.py — NameError context_attention_fwd kills engine on first request
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 18:56:06 +08:00
b3f2e4d970 fix: computility-run.yaml — remove limit-mm-per-prompt, set max-model-len=80000
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 18:29:56 +08:00
7185de5eef build(CRITICAL): complete .so build + deploy pipeline for competition submission
70c898ac8b fix: ex_engine.python subpackage + flash_qla_sm70 deploy + vllm v0.5.5 MoE kernels
35111e7a28 feat: implement topk_softmax + moe_align_block_size + invoke_fused_moe_kernel
accf9539e6 fix(CRITICAL): CoreXGDN interface mismatch + engine death protection
2aedf7377b ref(upstream): add Deep-Spark/vllm latest + xllm ILU kernel sources
Compare 8 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 17:25:16 +08:00
35f9da0c80 fix(NO-FALLBACK): eliminate all silent fallbacks — crash or succeed
f87689a4ef fix(CRITICAL): engine death on image request + stop overwriting base corex modules
Compare 2 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 16:30:34 +08:00
ff3562b941 fix(CRITICAL): always deploy our qwen3_5.py — base image GDN also has NaN
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 16:20:47 +08:00
5dde115245 fix(build): restore working Dockerfile format from c2807549
cd61968f01 fix(build): install gcc + ninja-build before compilation
b869eddbb4 fix(build): add upstream_ref to dockerignore + clean debug files
3aa0c3cffb fix(build): restore strict error handling — find real build failures
98fdcff9e9 fix(build): remove set -euo pipefail + bulletproof Dockerfile
Compare 7 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 15:59:31 +08:00
e7db38d76f debug: topk kernel mismatch diagnostic
7ec50ff3cf fix: total_mem → total_memory (corex torch API)
570ee94172 test: single-card BI-V100 verification — CUDA kernel + GDN NaN + ixformer ops
Compare 3 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 15:51:14 +08:00
8b6f3fd242 fix(MoE): robust CUDA kernel loading + no-GPU precompile
c0cc4e7dc9 feat(MoE): wire CUDA topk_softmax kernel into _custom_ops dispatch
c17c490e06 fix(GDN): remove pre-cumsum clamp — match xllm reference, fix 99.98% NaN
a0d76bc06e fix: remove ix_moe_bridge — nm -D confirms libixformer.so has NO MoE symbols
c280754903 fix(CRITICAL): conditional qwen3_5.py deploy + ix_moe_bridge topk_softmax
Compare 5 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 15:14:22 +08:00
d646a96c09 debug: deep probe MoE kernel dispatch in base image
04197138c2 docs: update PROJECT_SUMMARY — comp 168 analysis + three critical fixes
af08856d5c fix(CRITICAL): max_model_len 256000→80000 + topk_softmax silent fallback + deploy _custom_ops
Compare 3 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 14:51:01 +08:00
4a91c31ffc debug: probe base image vllm FusedMoE actual dispatch chain
f265cb8ad3 fix(yaml): align launch params with comp 168 proven config
Compare 2 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 14:40:11 +08:00
905bf4db2c feat(moe): wire silu_and_mul through C++ bridge in corex_moe.py
7127d18491 refactor(bridge): rewrite ix_full_bridge.cpp for actual base image symbols
b9fd2755d9 debug: probe all ixformer symbol locations
3e7fc565ff debug: probe script to find silu_and_mul symbol location
a54dbda3bb fix(bridge): link against libixformer.so for silu_and_mul symbol
Compare 8 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-10 14:20:17 +08:00
d7fa7b0682 docs: update PROJECT_SUMMARY.md — corex rewrite + dispatch chain analysis
d86b39d1ae refactor(corex): rewrite 3 dlopen modules to use real ixformer::infer dispatch chain
33b7327c1d fix: add error logging to all corex module imports + dtype guard
7d4edd4ac7 fix(GDN): force fp16 cast before norm+out_proj — ixformer matmul requires kHalf
d841c44e55 fix(GDN): dtype guard on _ix_matmul/_ix_bmm — ixformer.matmul requires kHalf
Compare 25 commits »