[cuda] GGUF Q6_K real packed INT6 (W6A8 dp4a) + GGUF CI export#20229
Draft
Gasoonjia wants to merge 4 commits into
Draft
[cuda] GGUF Q6_K real packed INT6 (W6A8 dp4a) + GGUF CI export#20229Gasoonjia wants to merge 4 commits into
Gasoonjia wants to merge 4 commits into
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20229
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ❌ 1 New Failure, 5 PendingAs of commit eaea4a7 with merge base c5bf380 ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
Block-sparse early-exit in _sdpa_fwd_kernel_body: skip KV blocks that are entirely masked (sliding-window via HAS_MASK sum==0, causal via start_n>max_seq_pos). Exact (skipped blocks are x1,+0 no-ops). Prefill +46-88% all lengths; decode safe; SDPA nsys 58.1%->18.5%. Numerically bf16-exact vs dense+mask (unit test).
Add a genuine 6-bit packed weight path for GGUF Q6_K on the CUDA backend, parallel to the int4/int8 plain_mm paths: - int6_plain_mm CUDA shim (W6A8 dp4a; ql/qh planes; spread2; -32 symmetric offset) - CudaPackedInt6Tensor (ql/qh + per-group bf16 scale; symmetric, no zero tensor) - int6_dispatch: F.linear routing (M<=4 -> executorch_cuda::int6_plain_mm op, M>4 -> dequant) - backend fallback-kernel + custom_ops_to_c_shims registration; CMake build - GGUF Q6_K: gguf_loader returns the native torchao IntxUnpackedToInt8Tensor and the backend packer (pack_cuda.pack_linear_for_cuda) repacks a symmetric Q6_K weight into CudaPackedInt6Tensor -- mirroring Int4Tensor -> CudaCoalescedInt4Tensor, so the loader stays backend-agnostic; dequantize_weight handles the tied embedding - tests: int6 gtest, test_int6_dispatch.py, pack round-trip; fix stale int4/int6 type asserts CI (export_model_artifact.sh, gemma4_31b): download the Q4_K_M GGUF from unsloth/gemma-4-31B-it-GGUF (tokenizer from unsloth/gemma-4-31B-it) and run the inference sanity check + export via the GGUF loader (--gguf) instead of the prequantized HF checkpoint. Signed-off-by: gasoonjia <gasoonjia@icloud.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.
Add a genuine 6-bit packed weight path for GGUF Q6_K on the CUDA backend, parallel to the int4/int8 plain_mm paths:
CI (export_model_artifact.sh, gemma4_31b): download the Q4_K_M GGUF from unsloth/gemma-4-31B-it-GGUF (tokenizer from unsloth/gemma-4-31B-it) and run the inference sanity check + export via the GGUF loader (--gguf) instead of the prequantized HF checkpoint.