Skip to content

Remove unused DFlash runtime warning sources - #139

Closed
FujitsuPolycom wants to merge 1 commit into
codex/glm53-dflash7-draft-loader-integrationfrom
codex/glm53-dflash7-safe-warning-cleanup
Closed

Remove unused DFlash runtime warning sources#139
FujitsuPolycom wants to merge 1 commit into
codex/glm53-dflash7-draft-loader-integrationfrom
codex/glm53-dflash7-safe-warning-cleanup

Conversation

@FujitsuPolycom

Copy link
Copy Markdown
Owner

Result

The DFlash7 derived-image contract removes an unused Python distribution without hiding runtime warnings. The build proves that deep_ep has exactly one owner, deep_ep==2.0.0+local, uninstalls only that distribution, and verifies that the module and distribution metadata are absent.

The content-addressed removal receipt has SHA-256 65514f44829e7d176b0b2cacc9559ed22724e525b7041a8bcd4d2e02d1f372e3. Prepared-context hashes, OCI labels, profile image-label requirements, launch attestation, and composed-image verification bind that receipt and its module-absence postcondition.

Runtime profile

Both safetensors and fastsafetensors profiles:

  • leave OMP_NUM_THREADS unset so vLLM selects the Torch thread count;
  • select language-model-only serving;
  • set VLLM_ALLREDUCE_USE_SYMM_MEM=0 and VLLM_ALLREDUCE_USE_FLASHINFER=0;
  • set pass_config.fuse_allreduce_rms=false;
  • retain --disable-custom-all-reduce so PYNCCL remains the collective path;
  • retain the pinned NCCL preload and VLLM_NCCL_SO_PATH;
  • retain B12X attention, MoE, and linear backends.

ModelOpt experimental-quantization and FP8 KV accuracy warnings remain visible and are recorded as allowed limitations. The profiles do not apply a general warning filter or lower the logging level.

Compatibility

Target and draft checkpoints, loader selection, DFlash depth, draft TP, FP8 KV, block size, sequence limits, topology, SparkCache identity, cache wire values, digest salts, and logical geometry are unchanged. Cache namespace impact: none.

Validation

  • Focused DFlash image, shared-verifier, profile-resolution, and four-rank plan tests: 34 passed.
  • Maintained CPU-only suite: 1,944 passed, 9 skipped. One PR Integrate the DFlash7 artifact with draft-loader separation #136 base README assertion still fails because commit 2614bfd omits the sentence that its existing test requires.
  • Repository-wide Ruff: passed.
  • Bash syntax, Python compilation, JSON parsing, and diff checks: passed.

No image was built or published. No serving host or container was changed.

Attest deep_ep module ownership by the exact deep_ep 2.0.0+local distribution, uninstall only that distribution in the derived image, and verify the module and distribution metadata are absent. Carry the content-addressed removal receipt through prepared-context hashes, OCI labels, profile attestation, and composed-image verification.

DFlash profiles leave Torch thread selection unset, select language-model-only serving, disable unsupported symmetric-memory and FlashInfer all-reduce candidates, and disable all-reduce RMS fusion. PYNCCL, the pinned NCCL library, and B12X attention, MoE, and linear bindings remain unchanged. ModelOpt and FP8 KV warnings remain visible.

Compatibility: model, draft, loader, KV, topology, SparkCache identity, cache wire values, digest salts, and 256-token geometry are unchanged. Cache namespace impact: none.

Validation: 34 focused GPU-free tests passed; repository-wide Ruff passed; Bash syntax, Python compilation, JSON parsing, and diff checks passed. The maintained suite passed 1,944 tests with 9 skipped and retained one unrelated PR #136 base README assertion failure. No image was built and no service was modified.
@FujitsuPolycom

Copy link
Copy Markdown
Owner Author

The resulting runtime, operator contract, and evidence are consolidated in retained draft stack #146#147#150. Independent page-base research remains in #149. Closing this superseded draft and deleting only its remote head branch.

@FujitsuPolycom
FujitsuPolycom deleted the codex/glm53-dflash7-safe-warning-cleanup branch August 31, 2026 01:55
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant