diff --git a/src/llama-context.cpp b/src/llama-context.cpp index cbd3e7330e7f..82a7ef9a8466 100644 --- a/src/llama-context.cpp +++ b/src/llama-context.cpp @@ -4146,6 +4146,30 @@ llama_context * llama_init_from_model( } } + // 1BIT / engine#679 item 3a: the HRX add-on has no q4_0 SET_ROWS *writer* (gpu-kernels#75, open), so a + // `-ctv q4_0` run cannot work end to end: the matcher accepts the type (llama.cpp#335) and the hook then + // declines, which surfaced as "unsupported HRX node 28: SET_ROWS output=...q4_0" mid-graph -- and + // llama-perplexity then exits 0 having produced no output, which reads as a pass to any harness keyed on the + // 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 + // (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) { + bool on_hrx = false; + for (uint32_t il = 0; il < model->hparams.n_layer() && !on_hrx; ++il) { + ggml_backend_dev_t dev = model->dev_layer(il); + const char * name = dev ? ggml_backend_dev_name(dev) : nullptr; + 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__); + return nullptr; + } + } + if (params.flash_attn_type != LLAMA_FLASH_ATTN_TYPE_DISABLED && ggml_is_quantized(params.type_v)) { const uint32_t blck_size = ggml_blck_size(params.type_v); for (uint32_t il = 0; il < model->hparams.n_layer(); ++il) {