Repository navigation
Conversation
New ggml-sycl ops behind an in-tree extension hook (src/sycl/vla_sycl_ext.cpp), built as vla_sycl_ops on oneDNN: - BF16 activations: BF16 matmul (BF16 or F32 result), norm, rms_norm, scale. - FLASH_ATTN_EXT on oneDNN's fused SDPA, operands read in place (F16/BF16, arbitrary strides, batched); default on SYCL for pi0 and pi0.5. - Fused elementwise nodes (layers/fused_ew.h): adaRMS, gated residual, geglu, LayerNorm, RoPE + K/V cache write, GEMM with bias/GELU/residual epilogue. - W8A8 linear with per-token INT8 activations (layers/qlinear.h), weights quantized at load; opt-in for pi0.5 (VLA_PI05_W8A8, VLA_PI05_W4). pi0.5 uses them for an in-place F16 K/V cache, fused projections and fused SigLIP blocks with all camera views in one pass. No CUDA, BitVLA or Evo-1 files change. Test: test_bf16_sycl_ops.
…buckets - patch_ggml_openvino.py: a graph fingerprint in front of the naive_compute cache. On a hit the cached decoder is already bound to these tensors, so naive_key and update_io() are skipped. GGML_OPENVINO_EXEC_MODE=performance lets a model opt out of the ACCURACY hint that pins non-MUL_MAT graphs to f32. - pi0.5 on OpenVINO: the language segment is padded to 16-token buckets with the pad masked as keys, so a prompt whose length changes every step compiles one model per bucket instead of one per length; performance exec mode.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Depends on #<PR 1>; review only the second commit (3 files).
On the OpenVINO GPU plugin, π0.5 on main runs out of device memory after 88 of 100 LIBERO episodes, at 1025 ms per request. With this PR it completes all 100 at LIBERO-object 95/100, and it runs in 386 ms with Q8_0 language-model weights (two views, Core Ultra 7 356H).
Changes
scripts/patch_ggml_openvino.py(two hunks on top of the existing ones)naive_computecache. On a hit, the cached decoder is already bound to these tensors, sonaive_keyandupdate_io()are skipped. π0.5: 988 → 590 ms per request.GGML_OPENVINO_EXEC_MODE=performance. The naive path compiles every graph that doesn't start with a MUL_MAT under the ACCURACY hint, which pins it to f32 and overridesGGML_OPENVINO_GPU_PRECISION. A model whose numerics hold at f16 can opt out. π0.5's vision tower: 217 → 64 ms.π0.5 on OpenVINO (
src/models/pi05.cpp)VLA_PI05_LANG_BUCKEToverrides the bucket size; 0 turns bucketing off.src/backend.h: anis_openvinoflag onBackend, alongsideis_cudaandis_sycl.Testing
cteston OpenVINO: 11/11 pass.tests/predict_checkmatch the build the LIBERO result was measured on, bit for bit.-Wall -Wextra(first-party code)ctestpassesvla_predict_checkdiff), or the change ismeant to move it and a LIBERO sweep is below
Archs and backends tested: