Refresh llama.cpp upstream canary patch queue - #839
Merged
Conversation
📝 WalkthroughWalkthroughThe patch series expands the Skippy ABI, adds runtime-slice graph filtering across many model builders, and introduces new APIs for activation frames, sampling, state/KV persistence, GGUF composition, and backend device enumeration. ChangesSkippy ABI and runtime-slice evolution
ChangesSkippy ABI and runtime-slice evolution
✨ Finishing Touches🧪 Generate unit tests (beta)
|
i386
changed the title
i386
marked this pull request as ready for review
June 13, 2026 02:24
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (4)
third_party/llama.cpp/patches/0011-Add-native-stage-KV-page-ABI.patch (1)
276-437:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winSilent failure on payload size mismatch at end of import.
Line 436 returns the result of
src == static_cast<const char *>(input) + input_byteswithout setting an error message if bytes consumed don't match. If there's any miscalculation in the V-transposed copy loop, this silently returnsfalsewith no diagnostic.Proposed fix
- return src == static_cast<const char *>(input) + input_bytes; + if (src != static_cast<const char *>(input) + input_bytes) { + error = "native KV page import consumed unexpected number of bytes"; + return false; + } + return true;🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@third_party/llama.cpp/patches/0011-Add-native-stage-KV-page-ABI.patch` around lines 276 - 437, The function llama_kv_cache::stage_import_kv_page currently returns the boolean of (src == input + input_bytes) without setting error on mismatch; change this by checking the final consumed byte count after the K/V copy loops (compare src - static_cast<const char *>(input) to input_bytes) and if they differ set a descriptive error on the error string (e.g. "imported payload size mismatch: consumed X bytes, expected Y bytes") and return false; keep the successful path returning true. Place this check just before the current return and reference the local symbols src, input, input_bytes and error, so any mismatch (including V-transposed loop mistakes) yields a clear error message instead of silent failure.third_party/llama.cpp/patches/0020-Add-staged-batched-verification-frame-ABI.patch (1)
180-270:⚠️ Potential issue | 🟡 MinorFix missing
token_idsargument toskippy_decode_activation_framein patch 0020
0018-Support-more-runtime-slice-model-families.patchaddsconst llama_token * token_idstoskippy_decode_activation_frame(...).- In
0020-Add-staged-batched-verification-frame-ABI.patch, the decode branch callsskippy_decode_activation_frame(session, input_desc, input_payload, token_count, false, out_error)(omitstoken_ids).0025-Fix-staged-verification-activation-handoff.patchcorrects this by passingtoken_ids(..., input_payload, token_ids, token_count, false, out_error), so compilation relies on0025being applied.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@third_party/llama.cpp/patches/0020-Add-staged-batched-verification-frame-ABI.patch` around lines 180 - 270, In skippy_verify_tokens_frame the call to skippy_decode_activation_frame is missing the new token_ids parameter introduced earlier; update the decode branch inside skippy_verify_tokens_frame so the call passes token_ids (i.e., change skippy_decode_activation_frame(session, input_desc, input_payload, token_count, false, out_error) to include token_ids as the argument before token_count), ensuring the function signature matches the updated skippy_decode_activation_frame declaration and compilation succeeds.third_party/llama.cpp/patches/0057-Support-RWKV7-activation-sideband.patch (1)
350-353:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winDefensive check should verify sufficient space for sideband data.
The condition at line 353 checks
desc->payload_bytes >= hidden_bytes, but when the RWKV7 flag is set, the sideband data spans anotherhidden_bytesafter the hidden portion. If validation is bypassed or buggy, this could allow reading beyond the payload buffer.The copy logic at lines 507-515 correctly checks
input_desc->payload_bytes < hidden_bytes * 2. For consistency and defense-in-depth, this check should also be>= hidden_bytes * 2:Suggested fix
if (desc != nullptr && payload != nullptr && (desc->flags & SKIPPY_ACTIVATION_FLAG_RWKV7_V_FIRST) != 0 && - desc->payload_bytes >= hidden_bytes) { + desc->payload_bytes >= hidden_bytes * 2) {🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@third_party/llama.cpp/patches/0057-Support-RWKV7-activation-sideband.patch` around lines 350 - 353, The conditional that validates the payload for RWKV7 sideband (checking desc != nullptr, payload != nullptr, (desc->flags & SKIPPY_ACTIVATION_FLAG_RWKV7_V_FIRST) != 0, and desc->payload_bytes >= hidden_bytes) is insufficient because RWKV7 includes an extra hidden_bytes of sideband data; update the check on desc->payload_bytes to require at least hidden_bytes * 2 (matching the input_desc->payload_bytes < hidden_bytes * 2 check used later) so the branch that handles RWKV7 activation-sideband cannot read past the payload buffer.third_party/llama.cpp/patches/0075-Carry-activation-frame-position-sideband.patch (1)
106-118:⚠️ Potential issue | 🟠 Major | ⚡ Quick winPosition fallback may leave uninitialized data for unusual
n_pos_per_embdvalues.When
positions == nullptrandn_pos_per_embdis neither 1 nor 4, the else branch (lines 114-117) only fillsn_tokenspositions butpos_storagewas allocated forn_tokens * n_pos_per_embdelements. This could leave uninitialized positions in the batch for models withn_pos_per_embdvalues like 2 or 3.Consider zero-initializing
pos_storageat allocation or extending the fallback logic.🛡️ Proposed fix: zero-initialize or handle all cases
- std::vector<llama_pos> pos_storage(expected_position_count); + std::vector<llama_pos> pos_storage(expected_position_count, 0);🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@third_party/llama.cpp/patches/0075-Carry-activation-frame-position-sideband.patch` around lines 106 - 118, The fallback branch that runs when positions == nullptr only fills the first n_tokens entries leaving pos_storage (allocated for n_tokens * n_pos_per_embd) partially uninitialized for n_pos_per_embd != 1 or 4; update the fallback to fully initialize all n_pos_per_embd slots by replacing the simple loop with a nested loop that for each token i and each slot k sets pos_storage[k * n_tokens + i] = session->n_past + i (or alternatively zero-initialize pos_storage at allocation with memset/{}), ensuring pos_storage, n_pos_per_embd, n_tokens and session->n_past are used to locate and fill every element when positions == nullptr.
🧹 Nitpick comments (3)
third_party/llama.cpp/patches/0007-Implement-GGUF-slice-writer-ABI.patch (1)
69-71: 💤 Low valuePotential overflow on large GGUF files with 32-bit
long.
std::fseektakes alongoffset, which may overflow on 32-bit platforms or Windows (wherelongis 32 bits) when seeking beyond 2GB. This affects bothskippy_skip(here) andskippy_write_slice_gguf(line 426).Consider using platform-specific 64-bit seek functions (
fseeko/_fseeki64) for large file support, or document the limitation.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@third_party/llama.cpp/patches/0007-Implement-GGUF-slice-writer-ABI.patch` around lines 69 - 71, The skippy_skip implementation (and the seek in skippy_write_slice_gguf) uses std::fseek with a cast to long which can overflow on 32-bit platforms; change these seeks to use a 64-bit seek API (fseeko on POSIX or _fseeki64 on Windows) via conditional compilation: detect _WIN32 to call _fseeki64 with a 64-bit offset type, otherwise use fseeko with off_t, and keep a fallback to std::fseek for platforms that lack 64-bit variants; update skippy_skip and the seek in skippy_write_slice_gguf to use this wrapper so large GGUF files seek correctly.third_party/llama.cpp/patches/0016-Add-batched-token-verification-ABI.patch (1)
56-74: 💤 Low valueReturning token 0 on null logits could mask errors.
When
llama_get_logits_ithreturnsnullptr, the function returns token0which may be a valid token (BOS/padding). This could silently produce incorrect verification results if logits computation fails for a batch position.Consider returning a sentinel value like
-1(which is commonly used for invalid tokens) or propagating the error to the caller.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@third_party/llama.cpp/patches/0016-Add-batched-token-verification-ABI.patch` around lines 56 - 74, The helper skippy_greedy_sample_ith currently returns token 0 when llama_get_logits_ith returns nullptr which can mask errors; change the function to return a sentinel for "invalid" (e.g. use int32_t as the return type or an existing LLAMA_TOKEN_INVALID constant) and return -1 (or that constant) on nullptr from llama_get_logits_ith, and then update all callers of skippy_greedy_sample_ith to check for the sentinel and handle the error path instead of treating it as a valid token; reference skippy_greedy_sample_ith and llama_get_logits_ith when making the change.third_party/llama.cpp/patches/0037-Apply-chat-grammar-during-stage-sampling.patch (1)
182-211: 💤 Low valueSilent success on JSON parse exceptions may mask configuration errors.
When
json::parse(metadata_json)throws, the code clears chat sampling and returnsskippy_success(out_error). This silently ignores malformed metadata JSON, making it difficult for callers to diagnose why grammar sampling isn't active. Consider returning an error status or at minimum logging/reporting the parse failure.Potential fix to report parse errors
} catch (const std::exception & e) { skippy_clear_chat_sampling(session); - return skippy_success(out_error); + skippy_set_error(out_error, SKIPPY_STATUS_INVALID_ARGUMENT, + ("failed to parse chat sampling metadata: " + std::string(e.what())).c_str()); + return SKIPPY_STATUS_INVALID_ARGUMENT; } catch (...) { skippy_clear_chat_sampling(session); - return skippy_success(out_error); + skippy_set_error(out_error, SKIPPY_STATUS_INVALID_ARGUMENT, "failed to parse chat sampling metadata"); + return SKIPPY_STATUS_INVALID_ARGUMENT; }🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@third_party/llama.cpp/patches/0037-Apply-chat-grammar-during-stage-sampling.patch` around lines 182 - 211, The code currently swallows JSON parse exceptions and returns skippy_success, hiding malformed metadata; update the JSON parsing path inside skippy_init_chat_grammar_sampler (the json::parse(metadata_json) handling) to catch parse exceptions and propagate an error instead of returning success — create and set a descriptive skippy_error (or call an existing helper to populate out_error) and return false (or appropriate failure code) so callers know grammar metadata failed to parse, and include the parse exception message in the error for diagnostics.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Outside diff comments:
In `@third_party/llama.cpp/patches/0011-Add-native-stage-KV-page-ABI.patch`:
- Around line 276-437: The function llama_kv_cache::stage_import_kv_page
currently returns the boolean of (src == input + input_bytes) without setting
error on mismatch; change this by checking the final consumed byte count after
the K/V copy loops (compare src - static_cast<const char *>(input) to
input_bytes) and if they differ set a descriptive error on the error string
(e.g. "imported payload size mismatch: consumed X bytes, expected Y bytes") and
return false; keep the successful path returning true. Place this check just
before the current return and reference the local symbols src, input,
input_bytes and error, so any mismatch (including V-transposed loop mistakes)
yields a clear error message instead of silent failure.
In
`@third_party/llama.cpp/patches/0020-Add-staged-batched-verification-frame-ABI.patch`:
- Around line 180-270: In skippy_verify_tokens_frame the call to
skippy_decode_activation_frame is missing the new token_ids parameter introduced
earlier; update the decode branch inside skippy_verify_tokens_frame so the call
passes token_ids (i.e., change skippy_decode_activation_frame(session,
input_desc, input_payload, token_count, false, out_error) to include token_ids
as the argument before token_count), ensuring the function signature matches the
updated skippy_decode_activation_frame declaration and compilation succeeds.
In `@third_party/llama.cpp/patches/0057-Support-RWKV7-activation-sideband.patch`:
- Around line 350-353: The conditional that validates the payload for RWKV7
sideband (checking desc != nullptr, payload != nullptr, (desc->flags &
SKIPPY_ACTIVATION_FLAG_RWKV7_V_FIRST) != 0, and desc->payload_bytes >=
hidden_bytes) is insufficient because RWKV7 includes an extra hidden_bytes of
sideband data; update the check on desc->payload_bytes to require at least
hidden_bytes * 2 (matching the input_desc->payload_bytes < hidden_bytes * 2
check used later) so the branch that handles RWKV7 activation-sideband cannot
read past the payload buffer.
In
`@third_party/llama.cpp/patches/0075-Carry-activation-frame-position-sideband.patch`:
- Around line 106-118: The fallback branch that runs when positions == nullptr
only fills the first n_tokens entries leaving pos_storage (allocated for
n_tokens * n_pos_per_embd) partially uninitialized for n_pos_per_embd != 1 or 4;
update the fallback to fully initialize all n_pos_per_embd slots by replacing
the simple loop with a nested loop that for each token i and each slot k sets
pos_storage[k * n_tokens + i] = session->n_past + i (or alternatively
zero-initialize pos_storage at allocation with memset/{}), ensuring pos_storage,
n_pos_per_embd, n_tokens and session->n_past are used to locate and fill every
element when positions == nullptr.
---
Nitpick comments:
In `@third_party/llama.cpp/patches/0007-Implement-GGUF-slice-writer-ABI.patch`:
- Around line 69-71: The skippy_skip implementation (and the seek in
skippy_write_slice_gguf) uses std::fseek with a cast to long which can overflow
on 32-bit platforms; change these seeks to use a 64-bit seek API (fseeko on
POSIX or _fseeki64 on Windows) via conditional compilation: detect _WIN32 to
call _fseeki64 with a 64-bit offset type, otherwise use fseeko with off_t, and
keep a fallback to std::fseek for platforms that lack 64-bit variants; update
skippy_skip and the seek in skippy_write_slice_gguf to use this wrapper so large
GGUF files seek correctly.
In `@third_party/llama.cpp/patches/0016-Add-batched-token-verification-ABI.patch`:
- Around line 56-74: The helper skippy_greedy_sample_ith currently returns token
0 when llama_get_logits_ith returns nullptr which can mask errors; change the
function to return a sentinel for "invalid" (e.g. use int32_t as the return type
or an existing LLAMA_TOKEN_INVALID constant) and return -1 (or that constant) on
nullptr from llama_get_logits_ith, and then update all callers of
skippy_greedy_sample_ith to check for the sentinel and handle the error path
instead of treating it as a valid token; reference skippy_greedy_sample_ith and
llama_get_logits_ith when making the change.
In
`@third_party/llama.cpp/patches/0037-Apply-chat-grammar-during-stage-sampling.patch`:
- Around line 182-211: The code currently swallows JSON parse exceptions and
returns skippy_success, hiding malformed metadata; update the JSON parsing path
inside skippy_init_chat_grammar_sampler (the json::parse(metadata_json)
handling) to catch parse exceptions and propagate an error instead of returning
success — create and set a descriptive skippy_error (or call an existing helper
to populate out_error) and return false (or appropriate failure code) so callers
know grammar metadata failed to parse, and include the parse exception message
in the error for diagnostics.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: a62febfe-267a-4cfb-a27f-e2671dc7ea69
📒 Files selected for processing (83)
third_party/llama.cpp/patches/0001-Draft-skippy-ABI-header.patchthird_party/llama.cpp/patches/0002-Implement-skippy-model-info-ABI.patchthird_party/llama.cpp/patches/0003-Implement-single-stage-runtime-ABI-baseline.patchthird_party/llama.cpp/patches/0004-Implement-runtime-slice-tensor-filtering.patchthird_party/llama.cpp/patches/0005-Add-activation-frame-runtime-ABI.patchthird_party/llama.cpp/patches/0006-Execute-runtime-slice-activation-frames.patchthird_party/llama.cpp/patches/0007-Implement-GGUF-slice-writer-ABI.patchthird_party/llama.cpp/patches/0008-Add-GGUF-package-part-composer-ABI.patchthird_party/llama.cpp/patches/0009-Support-Qwen35MoE-runtime-slice-packages.patchthird_party/llama.cpp/patches/0010-Handle-Qwen35MoE-recurrent-only-runtime-slices.patchthird_party/llama.cpp/patches/0011-Add-native-stage-KV-page-ABI.patchthird_party/llama.cpp/patches/0012-Add-stage-session-reset-ABI.patchthird_party/llama.cpp/patches/0013-Expose-stage-token-EOG-ABI.patchthird_party/llama.cpp/patches/0014-Position-stage-token-batches-from-session-offset.patchthird_party/llama.cpp/patches/0015-Optimize-native-KV-page-cell-lookup.patchthird_party/llama.cpp/patches/0016-Add-batched-token-verification-ABI.patchthird_party/llama.cpp/patches/0017-Expose-stage-chat-template-ABI.patchthird_party/llama.cpp/patches/0018-Support-more-runtime-slice-model-families.patchthird_party/llama.cpp/patches/0019-Add-stage-sampling-config-ABI.patchthird_party/llama.cpp/patches/0020-Add-staged-batched-verification-frame-ABI.patchthird_party/llama.cpp/patches/0021-Add-recurrent-state-checkpoint-ABI.patchthird_party/llama.cpp/patches/0022-Add-stage-logit-bias-sampling-ABI.patchthird_party/llama.cpp/patches/0023-Add-stage-session-trim-ABI.patchthird_party/llama.cpp/patches/0024-Add-native-stage-session-checkpoint-ABI.patchthird_party/llama.cpp/patches/0025-Fix-staged-verification-activation-handoff.patchthird_party/llama.cpp/patches/0026-Strip-split-metadata-from-stage-GGUF-artifacts.patchthird_party/llama.cpp/patches/0027-Expose-stage-chat-template-thinking-toggle.patchthird_party/llama.cpp/patches/0028-Load-stage-models-from-ordered-GGUF-parts.patchthird_party/llama.cpp/patches/0029-Support-Qwen3MoE-runtime-slice-execution.patchthird_party/llama.cpp/patches/0030-Expose-stage-KV-cache-type-config.patchthird_party/llama.cpp/patches/0031-Expose-stage-generation-signal-ABI.patchthird_party/llama.cpp/patches/0032-Expose-selected-backend-device-ABI.patchthird_party/llama.cpp/patches/0033-Expose-external-media-prefill-ABI.patchthird_party/llama.cpp/patches/0034-Add-shared-execution-lanes-to-skippy-ABI.patchthird_party/llama.cpp/patches/0035-Expose-stage-batch-and-flash-attention-config.patchthird_party/llama.cpp/patches/0036-Expose-tool-aware-chat-template-ABI.patchthird_party/llama.cpp/patches/0037-Apply-chat-grammar-during-stage-sampling.patchthird_party/llama.cpp/patches/0038-Add-resident-prefix-cache-ABI.patchthird_party/llama.cpp/patches/0039-Restore-exact-prefix-state-cache-ABI.patchthird_party/llama.cpp/patches/0040-Add-borrowed-resident-prefix-session-ABI.patchthird_party/llama.cpp/patches/0041-Compute-stage-generation-signals-lazily.patchthird_party/llama.cpp/patches/0042-Avoid-heap-batch-allocation-for-single-token-decode.patchthird_party/llama.cpp/patches/0043-Default-stage-threads-like-llama-server.patchthird_party/llama.cpp/patches/0044-Allow-borrowing-resident-prefix-cache-sequences.patchthird_party/llama.cpp/patches/0045-Preserve-resident-prefix-lanes.patchthird_party/llama.cpp/patches/0046-Remap-recurrent-state-imports-to-active-session.patchthird_party/llama.cpp/patches/0047-Support-initial-dense-runtime-slice-families.patchthird_party/llama.cpp/patches/0048-Support-additional-dense-runtime-slice-families.patchthird_party/llama.cpp/patches/0049-Avoid-rescaling-staged-activation-inputs.patchthird_party/llama.cpp/patches/0050-Support-more-decoder-runtime-slice-families.patchthird_party/llama.cpp/patches/0051-Support-LFM2-runtime-slice-execution.patchthird_party/llama.cpp/patches/0052-Support-Mamba-runtime-slice-execution.patchthird_party/llama.cpp/patches/0053-Support-Jamba-runtime-slice-execution.patchthird_party/llama.cpp/patches/0054-Support-RWKV6-runtime-slice-execution.patchthird_party/llama.cpp/patches/0055-Support-Qwen2MoE-runtime-slice-execution.patchthird_party/llama.cpp/patches/0056-Expose-skippy-session-native-sequence-id.patchthird_party/llama.cpp/patches/0057-Support-RWKV7-activation-sideband.patchthird_party/llama.cpp/patches/0058-Support-Phi2-runtime-slice-execution.patchthird_party/llama.cpp/patches/0059-Support-Granite-runtime-slice-variants.patchthird_party/llama.cpp/patches/0060-Support-Hunyuan-Dense-runtime-slice-execution.patchthird_party/llama.cpp/patches/0061-Support-Hunyuan-MoE-runtime-slice-execution.patchthird_party/llama.cpp/patches/0062-Support-PhiMoE-runtime-slice-execution.patchthird_party/llama.cpp/patches/0063-Allow-tied-output-embeddings-in-final-runtime-slices.patchthird_party/llama.cpp/patches/0064-Support-Qwen35-and-Hunyuan-VL-runtime-slices.patchthird_party/llama.cpp/patches/0065-Support-broad-llama-family-runtime-slices.patchthird_party/llama.cpp/patches/0066-Fix-BitNet-tied-output-runtime-slices.patchthird_party/llama.cpp/patches/0067-Fix-StarCoder-positional-embeddings-in-runtime-slice.patchthird_party/llama.cpp/patches/0068-Support-DeepSeek-OCR-and-Qwen3-VL-MoE-runtime-slices.patchthird_party/llama.cpp/patches/0069-Avoid-double-counting-optional-filtered-tensors.patchthird_party/llama.cpp/patches/0070-Skip-recurrent-layers-in-native-KV-page-export.patchthird_party/llama.cpp/patches/0071-Expose-external-decode-stage-filter-ABI.patchthird_party/llama.cpp/patches/0072-Support-external-media-prefill-in-staged-runtime-ABI.patchthird_party/llama.cpp/patches/0073-Expose-model-info-tensor-element-counts.patchthird_party/llama.cpp/patches/0074-Pad-activation-frame-decode-batches-to-native-input-.patchthird_party/llama.cpp/patches/0075-Carry-activation-frame-position-sideband.patchthird_party/llama.cpp/patches/0076-Support-Gemma3n-runtime-slices.patchthird_party/llama.cpp/patches/0077-Support-Llama4-and-Mistral4-runtime-slices.patchthird_party/llama.cpp/patches/0078-Contain-chat-grammar-sampler-exceptions.patchthird_party/llama.cpp/patches/0079-Trigger-lazy-chat-grammar-before-trailing-whitespace.patchthird_party/llama.cpp/patches/0080-Expose-skippy-backend-device-enumeration-ABI.patchthird_party/llama.cpp/patches/0081-Add-min-p-sampler-to-skippy-sampling-chain.patchthird_party/llama.cpp/patches/0082-Compact-KV-cache-cells-during-optimized-memory-updates.patchthird_party/llama.cpp/upstream.txt
michaelneale
added a commit
that referenced
this pull request
Jun 14, 2026
* origin/main: (29 commits) MoA: don't let small-model consensus pre-empt a still-running large model (#837) fix(console): render thinking traces as markdown Add bounded direct path repair (#846) Fix skippy smoke PR gate (#850) Stabilize skippy smoke chain startup (#849) fix(ci): switch back to auto-assign workflow fix(website): polish longform visual explainer (#843) fix: gemma thinking Carry GLM llama MTP patches (#840) Refresh llama.cpp canary patch queue (#839) Add transport-aware Skippy stage ordering (#814) Share Skippy stage wire byte accounting (#818) Report Skippy artifact cold-start costs (#815) fix: debug output capturing for TUI / panics (#827) fix(hero): visual corrections for iPhone SE size devices (#838) Add Skippy stage role metadata (#816) Add Skippy request cache epoch telemetry (#817) Consolidate agent skills and fix stale docs (Windows deploy, repo map, design docs) (#836) feature(version): normalize version markers for different build types (#831) fix(website): fix visual regressions (#835) ... # Conflicts: # AGENTS.md
This was referenced Jun 18, 2026
This was referenced Jun 30, 2026
Merged
This was referenced Jul 26, 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
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.
The llama.cpp upstream canary can apply the Skippy patch queue to the current ggml-org/llama.cpp master again.
What changed
third_party/llama.cpp/upstream.txtfrome22b0de60dff987d8905ba3cd9636134efb9b672toc34b92235b2d6a07963f896085f9ca077ff400b4.n_layer/is_recrnaming and Qwen35/Qwen35MoE staged range typing.Root cause
The scheduled canary was failing in
Prepare requested llama.cpp upstreamwhile applying patch0006to newer llama.cpp. The patch queue had drifted behind upstream graph changes, sogit am --3waycould no longer apply the carried Skippy graph patches to current master.Validation
LLAMA_WORKDIR=/tmp/mesh-llm-canary-validate.4GsGPQ scripts/prepare-llama.sh latestLLAMA_WORKDIR=/tmp/mesh-llm-canary-validate.4GsGPQ scripts/build-llama.shcargo check -p skippy-ffi -p skippy-runtime -p skippy-server -p skippy-model-package -p skippy-correctness -p llama-spec-benchgit diff --checkreports whitespace inside regenerated mail-format patch files; those patch artifacts intentionally preserve upstream patch whitespace.Summary by CodeRabbit
Release Notes
New Features
Performance Improvements
Enhancements