Rebase nm-shared patches onto antirez/ds4 main (0a7ad77) - #2
Open
salekseev wants to merge 25 commits into
Open
Conversation
Requires C23 extension Add check-metal target to ensure they exist. Signed-off-by: Evan Wies <evan@neomantra.net>
Signed-off-by: Evan Wies <evan@neomantra.net>
Signed-off-by: Evan Wies <evan@neomantra.net>
Signed-off-by: Evan Wies <evan@neomantra.net>
This is a hack, but a clean one. We rename stderr via preprocessor macros in a well-scoped way. There is a small trampoline injected which either provides stderr, or the client-assigned file. Signed-off-by: Evan Wies <evan@neomantra.net>
…lushing Signed-off-by: Evan Wies <evan@neomantra.net>
Signed-off-by: Evan Wies <evan@neomantra.net>
Signed-off-by: Evan Wies <evan@neomantra.net>
Signed-off-by: Evan Wies <evan@neomantra.net>
The nm-shared patches were rebased onto antirez/ds4 main (0a7ad77); this merge ties in the previous branch line so the rebased series can be adopted by fast-forward/merge without conflicts. Strategy 'ours': the tree is exactly the rebased result, unchanged. Co-Authored-By: Claude Fable 5 <noreply@anthropic.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
4 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Rebases the nm-shared shared-library patch set onto upstream antirez/ds4 main at antirez/ds4@0a7ad77 ("Fix batched server session recovery race") — 15 upstream commits since the previous base (antirez/ds4@80ebbc3). The result is the full upstream history with the 9 nm-shared patches as a clean linear series on top.
Brings in: Metal prefill/decode kernel optimizations, CUDA tensor parallelism + session batching, ROCm GLM 5.2 support, DSpark speculative decoding, distributed CLI unification, warning-free release builds.
(Replaces the earlier merge-commit PR #1, which was auto-closed by the history rewrite; the final tree here is bit-identical to that already-reviewed result.)
Patch updates during the rebase
build: Add shared library targetsnow compiles upstream's new sources into the shared libraries as PIC objects:ds4_tp,ds4_layer_pack,ds4_gpu_args(plus a-DDS4_NO_GPUvariant for shared-cpu, since ds4_gpu_args.c is GPU-define-sensitive).logging: add ds4_set_stderr ...additionally addsds4_stderr.oto upstream's new explicit test link rules (they list objects directly instead of using $(CORE_OBJS)).Add ROCM shared library supportgainsds4_rocm_compat/ds4_rocm_unavailablePIC objects for shared-rocm..metalfiles upstream, so the#embedmanifest (ds4_metal_sources.inc) is unchanged;make check-metalpasses.Validation
libds4.dyliblinks cleanly with 417 exportedds4_*symbols including the newds4_tp_*/ds4_layer_pack_*APIs.make testgreen against the real DeepSeek-V4-Flash model, including upstream's new DSpark coverage;test-metal-session-batchgreen.ds4.hagainst the dylib).ABI note
ds4_engine_optionsgrew andds4_tp_optionsis new — an ABI break for existing shared-library consumers. The soname is unversioned, so consumers must rebuild against the new headers.🤖 Generated with Claude Code
Mergeability
The tip commit is an
-s oursmerge of the previous nm-shared line: it ties the old history in as an ancestor so this PR merges without conflicts, while the tree remains exactly the rebased result.git log --first-parentstays readable: upstream history, the 9 patches, one knot-tying merge.