[docs] 10/n towards Kimi K2.6: Kimi K2.7-Code INT4 QAT + LoRA end-to-end example - #2032
Draft
casper-hansen wants to merge 1 commit into
Draft
[docs] 10/n towards Kimi K2.6: Kimi K2.7-Code INT4 QAT + LoRA end-to-end example#2032casper-hansen wants to merge 1 commit into
casper-hansen wants to merge 1 commit into
Conversation
…end example 2-node (2x8 B300) DAPO example training Kimi-K2.7-Code at 262144-token context with INT4 fake-QAT + capacity-normalized expert LoRA and merge_lora=false adapter hot-loading, plus the docs page covering the full setup: dequantizing the INT4 release into BF16 masters (2/n), the CUDA-13 deploy stack (6/n), and the training/serving configuration (3/n, 4/n, 5/n). Notable pins the example carries: mla_prefill_backend=FLASHINFER (FA4 CuTe-DSL prefill is incompatible with the locked nvidia-cutlass-dsl), trainer.flash_attn=false (MLA + sample packing needs TE cuDNN fused attention), UV_PROJECT_ENVIRONMENT + --no-sync for Ray's uv hook (workers execute from Ray's copied working dir where no .venv exists, and concurrent implicit syncs race). Lands last in the series: depends on 2/n-6/n. Co-authored-by: Cursor <cursoragent@cursor.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
2-node (2x8 B300) DAPO example training Kimi-K2.7-Code at 262144-token context with INT4 fake-QAT + capacity-normalized expert LoRA and merge_lora=false adapter hot-loading, plus the docs page covering the full setup: dequantizing the INT4 release into BF16 masters (#2024), the CUDA-13 deploy stack (#2028), and the training/serving configuration (#2025, #2026, #2027).
Notable pins the example carries:
mla_prefill_backend=FLASHINFER(FA4 CuTe-DSL prefill is incompatible with the locked nvidia-cutlass-dsl),trainer.flash_attn=false(MLA + sample packing needs TE cuDNN fused attention),UV_PROJECT_ENVIRONMENT+--no-syncfor Ray's uv hook (workers execute from Ray's copied working dir where no .venv exists, and concurrent implicit syncs race).Why
Single runnable reference for the whole Kimi K2.x INT4+LoRA flow the series enables.
Lands last: draft until #2024, #2025, #2026, #2027, #2028 merge (the script and docs reference all of them). The diff itself is self-contained (example script + docs only). Follow-up to #1862.
Made with Cursor