Repository navigation
Lean: ROCmI4 on Vulkan (our ROCmFPX fork); the bump rebases our commit - #58
Merged
Merged
Conversation
… the bump rebases our commit The ROCmFPX pin moves to our fork's 1bit/vulkan-rocmi4 (b44bf74): charlie12345/ROCmFPX fb08d7c plus Vulkan kernels for Q4_0_ROCMI4 (test-backend-ops passes on Vulkan0; Qwen3-0.6B PPL 28.33 on Vulkan vs 28.37 on ROCm). bump-rocmfpx.yml now syncs the fork, rebases our commit onto upstream main (tagging the old head) and opens the PR, like bump-hrx. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Docs7 for 1bit-monster/engine
Commit |
This was referenced Oct 2, 2026
bong-water-water-bong
added a commit
that referenced
this pull request
Oct 2, 2026
…mat SwiGLU, Hadamard Q4_0 (#278) * Pin llama.cpp bd5b297: HRX remainder prefill, 32K+ prompts, mixed-format SwiGLU, Hadamard Q4_0 llama.cpp fork commits since 6e42b51 (figures from each PR, Qwen3.8-27B on HRX0, balanced mode): - #61: q8_1 x4 prefill takes the 256-aligned head of a Q5_K/IQ4_XS remainder chunk. UD-Q4_K_XL pp400 67.4 -> 103.4, pp1862 172.1 -> 234.6 tok/s; KLD vs 6e42b51 0.00215, same top token 97.19%; MUL_MAT HRX0 287/287. - #59: V stays row-major past copy_transpose_f16's 32768-row range. pp512 @ d32768 failed before, 227.4 tok/s now; a 34,816-token wikitext chunk gives PPL 6.2150. - #60: K-quant SwiGLU decode stages separate gate/up codebooks. UD-IQ2_S tg128 3.12 -> 8.32; UD-Q4_K_XL unchanged; KLD 0.00013; test-backend-ops HRX0 996/996. - #58: llama loads onebit.hadamard_q4_0 files through llama-hadamard (needed by #277). Q4_0-H32 vs BF16: PPL 6.021, KLD 0.0292, same top token 92.5%. Docs: docs/hrx.md #55 section (remainders now covered for Q5_K/IQ4_XS) and "Our patches". Registry regenerated (tools/registry_build.py; no mapping changes). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * docs/hrx.md: the #61 figures were measured on 6e42b51 + #61 (a36d547) (review) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: bong-water-water-bong <bong-water-water-bong@1bit.gg> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
bong-water-water-bong
added a commit
that referenced
this pull request
Oct 2, 2026
The engine's H32 files (tools/hadamard_q4_0.py, onebit.hadamard_q4_0 = 32) ran on the lean ROCm route only. Our llama.cpp now reads the stamp (llama-hadamard, fork #58) and rotates the activations as a matmul, so serve accepts --device hrx for them: HRX0, no W4A4 or Hadamard variables. --device auto keeps ROCm for these files until HRX routes Q4_0 prompt matmuls to a fast kernel: Qwen3.8-27B-Q4_0-H32 on HRX0 runs pp512 at 63.5 tok/s today (PPL 6.021, KLD 0.0292, same top token 92.5% vs BF16). tests/hadamard_route.sh: --device hrx -> HRX0 with neither variable. ctest -LE gpu: 19/19. docs/hrx.md gets a section, docs/lean.md a pointer. Needs the third_party/llama.cpp pin that includes fork #58. Co-authored-by: bong-water-water-bong <bong-water-water-bong@1bit.gg> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
bong-water-water-bong
pushed a commit
that referenced
this pull request
Oct 3, 2026
…ds it as an h1) Fixup for 3745371 (owner decisions on stage 3): tools/site.py refuses a doc line that starts with '#<number>', so the site did not build. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This was referenced Oct 7, 2026
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
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.
Moves the lean option's ROCmFPX pin to 1bit-MONSTER/ROCmFPX
1bit/vulkan-rocmi4(b44bf74): upstreamfb08d7c(our previous pin) plus Vulkan kernels forQ4_0_ROCMI4, built by a pi agent (goal mugak664, audited and approved) and verified here:test-backend-ops -b Vulkan0: MUL_MAT 26/26, MUL_MAT_ID 73/73, GET_ROWS 4/4, CPY 17/17.--leanalready runs on Vulkan0, so ROCmI4 files now run on the GPU there; this also makes them usable in 1bit OS (Vulkan, no ROCm). The 27B row in docs/lean.md is not re-measured yet (noted).bump-rocmfpx.ymlnow works likebump-hrx.yml: sync the fork's main, rebase our commit onto charlie12345/ROCmFPX main (old head kept as tagrocmi4-<sha>), force-with-lease the branch, open the PR. Needs: theHRX_BUMP_TOKENfine-grained token must also have Contents write on1bit-MONSTER/ROCmFPX(it has it for the llama.cpp fork).🤖 Generated with Claude Code