From 99211f8ad49809479a6d4bbebfa5dc26326d15de Mon Sep 17 00:00:00 2001 From: bcloud Date: Fri, 13 Nov 2026 17:16:26 -0400 Subject: [PATCH] hrx: the q4_0 refusal's MESSAGE named a writer that has since merged I found this by content-verifying my own merged PR rather than trusting the merge: #338 landed, the guard and the corrected expiry are both in 1bit/hrx-vulkan-patched, and the message they carry still said 'cannot write a q4_0 V cache (no SET_ROWS writer: gpu-kernels#75)' gpu-kernels#75 merged the writer. So the refusal was true, correctly placed, and explained itself with a fact that had stopped being true -- the same stale-reason class a reviewer caught on #338's first revision. The refusal stays (the READ path is still absent: dispatch-flash-attention.cpp:238 requires value->type == GGML_TYPE_F16); only the reason changes to the one that is true today, plus the tense in the expiry comment. Measured on device, gpu-hold, gfx1151, LFM2.5-1.2B-Q4_K_M, -c 256: -ctv q4_0 rc 1 'cannot READ a q4_0 V cache (attention requires f16 for V; gpu-kernels#75 merged the SET_ROWS WRITER, the reader is still absent, engine#679)' -ctv q8_0 rc 0 Final estimate: PPL = 23.9396 +/- 0.20233 --- src/llama-context.cpp | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/src/llama-context.cpp b/src/llama-context.cpp index 82a7ef9a8466..ba0a26639c80 100644 --- a/src/llama-context.cpp +++ b/src/llama-context.cpp @@ -4153,7 +4153,7 @@ llama_context * llama_init_from_model( // exit code (measured on Qwen3-0.6B at gpu-kernels b0ed206, 2026-11-13). Refuse here instead, where the // message can name the fix. Gated on the device so every other backend keeps its q4_0 V support. // EXPIRY: remove this whole block when the READ path accepts a q4_0 V -- not when the writer lands. gpu-kernels#75 - // adds `set_rows_q4_0`, and the FA matchers still require value->type == GGML_TYPE_F16 + // merged `set_rows_q4_0`, and the FA matchers still require value->type == GGML_TYPE_F16 // (dispatch-flash-attention.cpp:238), so a run after #75 fails one node later, at attention, which is the same // graph-failure-with-no-output shape this refusal exists to prevent. Two halves, two conditions. if (params.type_v == GGML_TYPE_Q4_0) { @@ -4164,8 +4164,9 @@ llama_context * llama_init_from_model( on_hrx = name != nullptr && std::strncmp(name, "HRX", 3) == 0; } if (on_hrx) { - LLAMA_LOG_ERROR("%s: the HRX backend cannot write a q4_0 V cache (no SET_ROWS writer: gpu-kernels#75, " - "engine#679); use -ctv q8_0 or f16\n", __func__); + LLAMA_LOG_ERROR("%s: the HRX backend cannot READ a q4_0 V cache (attention requires f16 for V; " + "gpu-kernels#75 merged the SET_ROWS WRITER, the reader is still absent, engine#679); " + "use -ctv q8_0 or f16\n", __func__); return nullptr; } }