fix: stack overflow in turbo_init_rotation + TURBO_D static_assert#23
Merged
TheTom merged 1 commit intoMar 31, 2026
Merged
Conversation
vonempalmeolmos
pushed a commit
to vonempalmeolmos/llama-cpp-turboquant
that referenced
this pull request
Mar 29, 2026
…m#23 Codex review found: 1. Stale duplicate code in dequantize_turbo3_0_t4 (compile would fail) 2. thread static is risky/non-portable in MSL Fixed: removed thread static caching, using plain thread locals. Speed unchanged (2.4 tok/s) — the static caching wasn't actually working on Metal. True optimization needs architectural change in flash attention kernel to dequantize once per block, not per chunk. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
vonempalmeolmos
pushed a commit
to vonempalmeolmos/llama-cpp-turboquant
that referenced
this pull request
Mar 29, 2026
…heTom#23 Root cause analysis: 8-32× redundant full-block dequantize per block from flash attention template. Four approaches documented with expected speedups and risk levels. Plan: D (reduce overhead) → A/B (eliminate redundant calls) Target: 2.4 tok/s → 20-40 tok/s Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
vonempalmeolmos
pushed a commit
to vonempalmeolmos/llama-cpp-turboquant
that referenced
this pull request
Mar 29, 2026
Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
vonempalmeolmos
pushed a commit
to vonempalmeolmos/llama-cpp-turboquant
that referenced
this pull request
Mar 29, 2026
…om#23 Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
vonempalmeolmos
pushed a commit
to vonempalmeolmos/llama-cpp-turboquant
that referenced
this pull request
Mar 29, 2026
…heTom#23 No-op dequant test: even returning all zeros from dequantize, turbo3 runs at 2.4 tok/s (same as with full WHT rotation). The bottleneck is NOT in the attention dequantize path. New hypothesis: the SET_ROWS (quantize) path is the bottleneck. The Metal quantize_turbo3_0 function does 3 WHT rotations per KV write, totaling ~3200 ops per block × 224 blocks per token. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
vonempalmeolmos
pushed a commit
to vonempalmeolmos/llama-cpp-turboquant
that referenced
this pull request
Mar 29, 2026
CRITICAL BUG: The #include "turbo-wht.h" caused Metal JIT compilation to fail at runtime. The model silently fell back to CPU for ALL ops. ALL previous benchmarks (2.4 tok/s) were measuring CPU, not Metal GPU. After inlining the header: - MoE gen: 2.4 → 10.7 tok/s (4.5× improvement, now actually on Metal) - MoE prompt: 4.2 → 60.9 tok/s (14.5× improvement) Remaining gap vs q8_0: 85 → 10.7 tok/s (8× slower, down from 35×) This is the SAME bug we hit with turbo-matrices.h earlier. Rule: NEVER use #include in ggml-metal.metal — always inline. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
vonempalmeolmos
pushed a commit
to vonempalmeolmos/llama-cpp-turboquant
that referenced
this pull request
Mar 29, 2026
…m#23 Previous 2.4 tok/s was CPU fallback. Real Metal numbers: MoE: 10.7 tok/s gen (8× slower than q8_0, was thought to be 35×) Qwopus: 5.3 tok/s gen (3.3× slower than q8_0) Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
vonempalmeolmos
pushed a commit
to vonempalmeolmos/llama-cpp-turboquant
that referenced
this pull request
Mar 29, 2026
vonempalmeolmos
pushed a commit
to vonempalmeolmos/llama-cpp-turboquant
that referenced
this pull request
Mar 29, 2026
…in) TheTom#23 Removing WHT rotation from dequant (quality broken, speed test only): gen: 10.7 → 49.1 tok/s (4.6× improvement, 57% of q8_0) prompt: 67.3 → 162.6 tok/s Confirms pre-rotate-queries would deliver ~49 tok/s. Remaining gap (49 vs 85) is block size + QJL overhead. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
vonempalmeolmos
pushed a commit
to vonempalmeolmos/llama-cpp-turboquant
that referenced
this pull request
Mar 29, 2026
…m#23 Instead of inverse-rotating every K during dequant, rotate Q once before attention. Math: <q, R^T*c[idx]> = <R*q, c[idx]>. Changes: - Store rotation matrix (R^T) in KV cache, filled after buffer clear - Apply ggml_mul_mat(R_T, q) in build_attn_mha after permute - Strip turbo_rotate_inverse from Metal dequant - Dynamic cast to access rotation from mctx Results: - MoE gen: 10.7 → 51.4 tok/s (4.8× speedup) - MoE prompt: 67.3 → 160.3 tok/s (2.4× speedup) - Now at 60% of q8_0 speed with 4.9× compression - Model produces coherent output Codex review: fixed buffer clear ordering (was zeroing rotation after init). Verified: rotation point is correct (after 4d reshape + permute, ne[0]=128). Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
vonempalmeolmos
pushed a commit
to vonempalmeolmos/llama-cpp-turboquant
that referenced
this pull request
Mar 29, 2026
…heTom#23 Full investigation log documenting every test, every dead end, and every breakthrough. 21× total improvement from CPU fallback to pre-rotate-queries. Key lessons: no #include in Metal, no-op testing, pre-rotate-queries, buffer clear ordering, codex+roast catch real bugs. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
vonempalmeolmos
pushed a commit
to vonempalmeolmos/llama-cpp-turboquant
that referenced
this pull request
Mar 29, 2026
Validated on real Qwen3 KV tensors: cosine sim 0.9508 → 0.9831 (+3.2%) MSE-only better on 99.3% of vectors including p1 tails. 3-bit index split: lower 2 bits in qs[], upper 1 bit in signs[]. No QJL stage in quantize or dequant. Results: - MoE gen: 51.4 → 62.2 tok/s (73% of q8_0, was 60%) - MoE prompt: 160 → 200 tok/s (90% of q8_0) - Qwopus gen: 14.6 → 15.5 tok/s (88% of q8_0, was 83%) - Qwopus prompt: 67 → 83 tok/s (100% of q8_0!) Codex verified: bit packing correct, quantize/dequant consistent. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
1. turbo_init_rotation() allocated float G[128*128] (64KB) on the stack then memcpy'd into the static turbo_rotation array. This segfaults on llama.cpp worker threads with reduced stack sizes (512KB macOS, 64KB some Linux). Fix: generate the Gaussian matrix directly into turbo_rotation, eliminating both the stack allocation and the memcpy. 2. TURBO_D and QK_TURBO3_GROUP are defined separately but must always match (both represent the rotation group size). Add static_assert to catch silent divergence between CPU reference and GPU kernels. Fixes: TheTom#29 (remaining items from PR TheTom#18 review) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ed76b24 to
f59394d
Compare
mihai-chiorean
pushed a commit
to mihai-chiorean/turbo3-cuda
that referenced
this pull request
Mar 31, 2026
…m#23 Codex review found: 1. Stale duplicate code in dequantize_turbo3_0_t4 (compile would fail) 2. thread static is risky/non-portable in MSL Fixed: removed thread static caching, using plain thread locals. Speed unchanged (2.4 tok/s) — the static caching wasn't actually working on Metal. True optimization needs architectural change in flash attention kernel to dequantize once per block, not per chunk. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
mihai-chiorean
pushed a commit
to mihai-chiorean/turbo3-cuda
that referenced
this pull request
Mar 31, 2026
…heTom#23 Root cause analysis: 8-32× redundant full-block dequantize per block from flash attention template. Four approaches documented with expected speedups and risk levels. Plan: D (reduce overhead) → A/B (eliminate redundant calls) Target: 2.4 tok/s → 20-40 tok/s Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
mihai-chiorean
pushed a commit
to mihai-chiorean/turbo3-cuda
that referenced
this pull request
Mar 31, 2026
Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
mihai-chiorean
pushed a commit
to mihai-chiorean/turbo3-cuda
that referenced
this pull request
Mar 31, 2026
…om#23 Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
mihai-chiorean
pushed a commit
to mihai-chiorean/turbo3-cuda
that referenced
this pull request
Mar 31, 2026
…heTom#23 No-op dequant test: even returning all zeros from dequantize, turbo3 runs at 2.4 tok/s (same as with full WHT rotation). The bottleneck is NOT in the attention dequantize path. New hypothesis: the SET_ROWS (quantize) path is the bottleneck. The Metal quantize_turbo3_0 function does 3 WHT rotations per KV write, totaling ~3200 ops per block × 224 blocks per token. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
mihai-chiorean
pushed a commit
to mihai-chiorean/turbo3-cuda
that referenced
this pull request
Mar 31, 2026
CRITICAL BUG: The #include "turbo-wht.h" caused Metal JIT compilation to fail at runtime. The model silently fell back to CPU for ALL ops. ALL previous benchmarks (2.4 tok/s) were measuring CPU, not Metal GPU. After inlining the header: - MoE gen: 2.4 → 10.7 tok/s (4.5× improvement, now actually on Metal) - MoE prompt: 4.2 → 60.9 tok/s (14.5× improvement) Remaining gap vs q8_0: 85 → 10.7 tok/s (8× slower, down from 35×) This is the SAME bug we hit with turbo-matrices.h earlier. Rule: NEVER use #include in ggml-metal.metal — always inline. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
mihai-chiorean
pushed a commit
to mihai-chiorean/turbo3-cuda
that referenced
this pull request
Mar 31, 2026
…m#23 Previous 2.4 tok/s was CPU fallback. Real Metal numbers: MoE: 10.7 tok/s gen (8× slower than q8_0, was thought to be 35×) Qwopus: 5.3 tok/s gen (3.3× slower than q8_0) Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
mihai-chiorean
pushed a commit
to mihai-chiorean/turbo3-cuda
that referenced
this pull request
Mar 31, 2026
mihai-chiorean
pushed a commit
to mihai-chiorean/turbo3-cuda
that referenced
this pull request
Mar 31, 2026
…in) TheTom#23 Removing WHT rotation from dequant (quality broken, speed test only): gen: 10.7 → 49.1 tok/s (4.6× improvement, 57% of q8_0) prompt: 67.3 → 162.6 tok/s Confirms pre-rotate-queries would deliver ~49 tok/s. Remaining gap (49 vs 85) is block size + QJL overhead. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
mihai-chiorean
pushed a commit
to mihai-chiorean/turbo3-cuda
that referenced
this pull request
Mar 31, 2026
…m#23 Instead of inverse-rotating every K during dequant, rotate Q once before attention. Math: <q, R^T*c[idx]> = <R*q, c[idx]>. Changes: - Store rotation matrix (R^T) in KV cache, filled after buffer clear - Apply ggml_mul_mat(R_T, q) in build_attn_mha after permute - Strip turbo_rotate_inverse from Metal dequant - Dynamic cast to access rotation from mctx Results: - MoE gen: 10.7 → 51.4 tok/s (4.8× speedup) - MoE prompt: 67.3 → 160.3 tok/s (2.4× speedup) - Now at 60% of q8_0 speed with 4.9× compression - Model produces coherent output Codex review: fixed buffer clear ordering (was zeroing rotation after init). Verified: rotation point is correct (after 4d reshape + permute, ne[0]=128). Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
mihai-chiorean
pushed a commit
to mihai-chiorean/turbo3-cuda
that referenced
this pull request
Mar 31, 2026
…heTom#23 Full investigation log documenting every test, every dead end, and every breakthrough. 21× total improvement from CPU fallback to pre-rotate-queries. Key lessons: no #include in Metal, no-op testing, pre-rotate-queries, buffer clear ordering, codex+roast catch real bugs. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
mihai-chiorean
pushed a commit
to mihai-chiorean/turbo3-cuda
that referenced
this pull request
Mar 31, 2026
Validated on real Qwen3 KV tensors: cosine sim 0.9508 → 0.9831 (+3.2%) MSE-only better on 99.3% of vectors including p1 tails. 3-bit index split: lower 2 bits in qs[], upper 1 bit in signs[]. No QJL stage in quantize or dequant. Results: - MoE gen: 51.4 → 62.2 tok/s (73% of q8_0, was 60%) - MoE prompt: 160 → 200 tok/s (90% of q8_0) - Qwopus gen: 14.6 → 15.5 tok/s (88% of q8_0, was 83%) - Qwopus prompt: 67 → 83 tok/s (100% of q8_0!) Codex verified: bit packing correct, quantize/dequant consistent. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Author
|
Hey @TheTom — gentle ping on this one. Just the stack overflow fix (64KB on worker thread stack) + static_assert guard. Rebased clean on current HEAD (post block-128, post signalnine merge). Happy to adjust anything if needed. |
Owner
|
Thank you for the ping. Will test this morning (cst) before noon. |
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
Full investigation log with all tests, results, and the root cause. Upstream TurboQuant activity tracked in #27. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
…in) #23 Removing WHT rotation from dequant (quality broken, speed test only): gen: 10.7 → 49.1 tok/s (4.6× improvement, 57% of q8_0) prompt: 67.3 → 162.6 tok/s Confirms pre-rotate-queries would deliver ~49 tok/s. Remaining gap (49 vs 85) is block size + QJL overhead. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
Instead of inverse-rotating every K during dequant, rotate Q once before attention. Math: <q, R^T*c[idx]> = <R*q, c[idx]>. Changes: - Store rotation matrix (R^T) in KV cache, filled after buffer clear - Apply ggml_mul_mat(R_T, q) in build_attn_mha after permute - Strip turbo_rotate_inverse from Metal dequant - Dynamic cast to access rotation from mctx Results: - MoE gen: 10.7 → 51.4 tok/s (4.8× speedup) - MoE prompt: 67.3 → 160.3 tok/s (2.4× speedup) - Now at 60% of q8_0 speed with 4.9× compression - Model produces coherent output Codex review: fixed buffer clear ordering (was zeroing rotation after init). Verified: rotation point is correct (after 4d reshape + permute, ne[0]=128). Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
…23 Full investigation log documenting every test, every dead end, and every breakthrough. 21× total improvement from CPU fallback to pre-rotate-queries. Key lessons: no #include in Metal, no-op testing, pre-rotate-queries, buffer clear ordering, codex+roast catch real bugs. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
Validated on real Qwen3 KV tensors: cosine sim 0.9508 → 0.9831 (+3.2%) MSE-only better on 99.3% of vectors including p1 tails. 3-bit index split: lower 2 bits in qs[], upper 1 bit in signs[]. No QJL stage in quantize or dequant. Results: - MoE gen: 51.4 → 62.2 tok/s (73% of q8_0, was 60%) - MoE prompt: 160 → 200 tok/s (90% of q8_0) - Qwopus gen: 14.6 → 15.5 tok/s (88% of q8_0, was 83%) - Qwopus prompt: 67 → 83 tok/s (100% of q8_0!) Codex verified: bit packing correct, quantize/dequant consistent. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
Codex review found: 1. Stale duplicate code in dequantize_turbo3_0_t4 (compile would fail) 2. thread static is risky/non-portable in MSL Fixed: removed thread static caching, using plain thread locals. Speed unchanged (2.4 tok/s) — the static caching wasn't actually working on Metal. True optimization needs architectural change in flash attention kernel to dequantize once per block, not per chunk. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
Root cause analysis: 8-32× redundant full-block dequantize per block from flash attention template. Four approaches documented with expected speedups and risk levels. Plan: D (reduce overhead) → A/B (eliminate redundant calls) Target: 2.4 tok/s → 20-40 tok/s Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
…23 No-op dequant test: even returning all zeros from dequantize, turbo3 runs at 2.4 tok/s (same as with full WHT rotation). The bottleneck is NOT in the attention dequantize path. New hypothesis: the SET_ROWS (quantize) path is the bottleneck. The Metal quantize_turbo3_0 function does 3 WHT rotations per KV write, totaling ~3200 ops per block × 224 blocks per token. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
CRITICAL BUG: The #include "turbo-wht.h" caused Metal JIT compilation to fail at runtime. The model silently fell back to CPU for ALL ops. ALL previous benchmarks (2.4 tok/s) were measuring CPU, not Metal GPU. After inlining the header: - MoE gen: 2.4 → 10.7 tok/s (4.5× improvement, now actually on Metal) - MoE prompt: 4.2 → 60.9 tok/s (14.5× improvement) Remaining gap vs q8_0: 85 → 10.7 tok/s (8× slower, down from 35×) This is the SAME bug we hit with turbo-matrices.h earlier. Rule: NEVER use #include in ggml-metal.metal — always inline. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
Previous 2.4 tok/s was CPU fallback. Real Metal numbers: MoE: 10.7 tok/s gen (8× slower than q8_0, was thought to be 35×) Qwopus: 5.3 tok/s gen (3.3× slower than q8_0) Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
Full investigation log with all tests, results, and the root cause. Upstream TurboQuant activity tracked in #27. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
…in) #23 Removing WHT rotation from dequant (quality broken, speed test only): gen: 10.7 → 49.1 tok/s (4.6× improvement, 57% of q8_0) prompt: 67.3 → 162.6 tok/s Confirms pre-rotate-queries would deliver ~49 tok/s. Remaining gap (49 vs 85) is block size + QJL overhead. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
Instead of inverse-rotating every K during dequant, rotate Q once before attention. Math: <q, R^T*c[idx]> = <R*q, c[idx]>. Changes: - Store rotation matrix (R^T) in KV cache, filled after buffer clear - Apply ggml_mul_mat(R_T, q) in build_attn_mha after permute - Strip turbo_rotate_inverse from Metal dequant - Dynamic cast to access rotation from mctx Results: - MoE gen: 10.7 → 51.4 tok/s (4.8× speedup) - MoE prompt: 67.3 → 160.3 tok/s (2.4× speedup) - Now at 60% of q8_0 speed with 4.9× compression - Model produces coherent output Codex review: fixed buffer clear ordering (was zeroing rotation after init). Verified: rotation point is correct (after 4d reshape + permute, ne[0]=128). Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
…23 Full investigation log documenting every test, every dead end, and every breakthrough. 21× total improvement from CPU fallback to pre-rotate-queries. Key lessons: no #include in Metal, no-op testing, pre-rotate-queries, buffer clear ordering, codex+roast catch real bugs. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TheTom
added a commit
that referenced
this pull request
Apr 2, 2026
Validated on real Qwen3 KV tensors: cosine sim 0.9508 → 0.9831 (+3.2%) MSE-only better on 99.3% of vectors including p1 tails. 3-bit index split: lower 2 bits in qs[], upper 1 bit in signs[]. No QJL stage in quantize or dequant. Results: - MoE gen: 51.4 → 62.2 tok/s (73% of q8_0, was 60%) - MoE prompt: 160 → 200 tok/s (90% of q8_0) - Qwopus gen: 14.6 → 15.5 tok/s (88% of q8_0, was 83%) - Qwopus prompt: 67 → 83 tok/s (100% of q8_0!) Codex verified: bit packing correct, quantize/dequant consistent. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
iamwavecut
pushed a commit
to iamwavecut/llama-cpp-turboquant
that referenced
this pull request
Apr 8, 2026
…m#23 Codex review found: 1. Stale duplicate code in dequantize_turbo3_0_t4 (compile would fail) 2. thread static is risky/non-portable in MSL Fixed: removed thread static caching, using plain thread locals. Speed unchanged (2.4 tok/s) — the static caching wasn't actually working on Metal. True optimization needs architectural change in flash attention kernel to dequantize once per block, not per chunk. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
iamwavecut
pushed a commit
to iamwavecut/llama-cpp-turboquant
that referenced
this pull request
Apr 8, 2026
…heTom#23 Root cause analysis: 8-32× redundant full-block dequantize per block from flash attention template. Four approaches documented with expected speedups and risk levels. Plan: D (reduce overhead) → A/B (eliminate redundant calls) Target: 2.4 tok/s → 20-40 tok/s Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
iamwavecut
pushed a commit
to iamwavecut/llama-cpp-turboquant
that referenced
this pull request
Apr 8, 2026
Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
iamwavecut
pushed a commit
to iamwavecut/llama-cpp-turboquant
that referenced
this pull request
Apr 8, 2026
…om#23 Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
iamwavecut
pushed a commit
to iamwavecut/llama-cpp-turboquant
that referenced
this pull request
Apr 8, 2026
…heTom#23 No-op dequant test: even returning all zeros from dequantize, turbo3 runs at 2.4 tok/s (same as with full WHT rotation). The bottleneck is NOT in the attention dequantize path. New hypothesis: the SET_ROWS (quantize) path is the bottleneck. The Metal quantize_turbo3_0 function does 3 WHT rotations per KV write, totaling ~3200 ops per block × 224 blocks per token. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
iamwavecut
pushed a commit
to iamwavecut/llama-cpp-turboquant
that referenced
this pull request
Apr 8, 2026
CRITICAL BUG: The #include "turbo-wht.h" caused Metal JIT compilation to fail at runtime. The model silently fell back to CPU for ALL ops. ALL previous benchmarks (2.4 tok/s) were measuring CPU, not Metal GPU. After inlining the header: - MoE gen: 2.4 → 10.7 tok/s (4.5× improvement, now actually on Metal) - MoE prompt: 4.2 → 60.9 tok/s (14.5× improvement) Remaining gap vs q8_0: 85 → 10.7 tok/s (8× slower, down from 35×) This is the SAME bug we hit with turbo-matrices.h earlier. Rule: NEVER use #include in ggml-metal.metal — always inline. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
iamwavecut
pushed a commit
to iamwavecut/llama-cpp-turboquant
that referenced
this pull request
Apr 8, 2026
…m#23 Previous 2.4 tok/s was CPU fallback. Real Metal numbers: MoE: 10.7 tok/s gen (8× slower than q8_0, was thought to be 35×) Qwopus: 5.3 tok/s gen (3.3× slower than q8_0) Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
iamwavecut
pushed a commit
to iamwavecut/llama-cpp-turboquant
that referenced
this pull request
Apr 8, 2026
iamwavecut
pushed a commit
to iamwavecut/llama-cpp-turboquant
that referenced
this pull request
Apr 8, 2026
…in) TheTom#23 Removing WHT rotation from dequant (quality broken, speed test only): gen: 10.7 → 49.1 tok/s (4.6× improvement, 57% of q8_0) prompt: 67.3 → 162.6 tok/s Confirms pre-rotate-queries would deliver ~49 tok/s. Remaining gap (49 vs 85) is block size + QJL overhead. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
iamwavecut
pushed a commit
to iamwavecut/llama-cpp-turboquant
that referenced
this pull request
Apr 8, 2026
…m#23 Instead of inverse-rotating every K during dequant, rotate Q once before attention. Math: <q, R^T*c[idx]> = <R*q, c[idx]>. Changes: - Store rotation matrix (R^T) in KV cache, filled after buffer clear - Apply ggml_mul_mat(R_T, q) in build_attn_mha after permute - Strip turbo_rotate_inverse from Metal dequant - Dynamic cast to access rotation from mctx Results: - MoE gen: 10.7 → 51.4 tok/s (4.8× speedup) - MoE prompt: 67.3 → 160.3 tok/s (2.4× speedup) - Now at 60% of q8_0 speed with 4.9× compression - Model produces coherent output Codex review: fixed buffer clear ordering (was zeroing rotation after init). Verified: rotation point is correct (after 4d reshape + permute, ne[0]=128). Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
iamwavecut
pushed a commit
to iamwavecut/llama-cpp-turboquant
that referenced
this pull request
Apr 8, 2026
…heTom#23 Full investigation log documenting every test, every dead end, and every breakthrough. 21× total improvement from CPU fallback to pre-rotate-queries. Key lessons: no #include in Metal, no-op testing, pre-rotate-queries, buffer clear ordering, codex+roast catch real bugs. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
iamwavecut
pushed a commit
to iamwavecut/llama-cpp-turboquant
that referenced
this pull request
Apr 8, 2026
Validated on real Qwen3 KV tensors: cosine sim 0.9508 → 0.9831 (+3.2%) MSE-only better on 99.3% of vectors including p1 tails. 3-bit index split: lower 2 bits in qs[], upper 1 bit in signs[]. No QJL stage in quantize or dequant. Results: - MoE gen: 51.4 → 62.2 tok/s (73% of q8_0, was 60%) - MoE prompt: 160 → 200 tok/s (90% of q8_0) - Qwopus gen: 14.6 → 15.5 tok/s (88% of q8_0, was 83%) - Qwopus prompt: 67 → 83 tok/s (100% of q8_0!) Codex verified: bit packing correct, quantize/dequant consistent. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
KGardevoir
pushed a commit
to KGardevoir/llama-cpp-turboquant
that referenced
this pull request
Apr 9, 2026
…m#23 Codex review found: 1. Stale duplicate code in dequantize_turbo3_0_t4 (compile would fail) 2. thread static is risky/non-portable in MSL Fixed: removed thread static caching, using plain thread locals. Speed unchanged (2.4 tok/s) — the static caching wasn't actually working on Metal. True optimization needs architectural change in flash attention kernel to dequantize once per block, not per chunk. Co-Authored-By: tturney@psyguard.ai Co-Authored-By: Claude Opus 4.6 (1M context) <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
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
Minimal resubmission per PR #18 review — only the two fixes still needed against current HEAD (
172fc85):Stack overflow fix:
turbo_init_rotation()allocatesfloat G[128*128](64KB) on the stack atggml-turbo-quant.c:71, thenmemcpys into the staticturbo_rotation. This segfaults on llama.cpp worker threads with reduced stack sizes (512KB macOS, 64KB some Linux). Fix generates the Gaussian matrix directly intoturbo_rotation, eliminating both the stack allocation and thememcpy.static_assertguard:TURBO_DandQK_TURBO3_GROUPare defined separately but must always match (both = rotation group size 128). Adds compile-time check to catch silent divergence between CPU reference path and GPU kernels.What changed from PR #18
Dropped fixes #2 (non-128-aligned head dims) and #4 (WHT dispatch assert) — both resolved by signalnine's PR #3 merge. This is the minimal diff TheTom requested.
Test plan
gcc -std=c11 -D_GNU_SOURCE