From d4d1522b20809a350ffb094db20f40f17d3ab80f Mon Sep 17 00:00:00 2001 From: Molly Sophia Date: Tue, 22 Jul 2025 23:01:29 +0800 Subject: [PATCH 1/3] llama : add model type detection for rwkv7 7B&14B (#14816) Signed-off-by: Molly Sophia --- src/llama-model.cpp | 18 ++++++++++++++++-- 1 file changed, 16 insertions(+), 2 deletions(-) diff --git a/src/llama-model.cpp b/src/llama-model.cpp index 2d90ec1ac68..35e718aa989 100644 --- a/src/llama-model.cpp +++ b/src/llama-model.cpp @@ -1544,7 +1544,11 @@ void llama_model::load_hparams(llama_model_loader & ml) { ml.get_key(LLM_KV_TOKEN_SHIFT_COUNT, hparams.token_shift_count, false); switch (hparams.n_layer) { - case 12: type = LLM_TYPE_190M; break; + case 12: + switch (hparams.n_embd) { + case 768: type = LLM_TYPE_190M; break; + default: type = LLM_TYPE_UNKNOWN; + } break; case 24: switch (hparams.n_embd) { case 1024: type = LLM_TYPE_450M; break; @@ -1557,7 +1561,17 @@ void llama_model::load_hparams(llama_model_loader & ml) { case 3584: type = LLM_TYPE_7B; break; default: type = LLM_TYPE_UNKNOWN; } break; - case 32: type = LLM_TYPE_2_9B; break; // RWKV-7-World + case 32: + switch (hparams.n_embd) { + case 2560: type = LLM_TYPE_2_9B; break; + case 4096: type = LLM_TYPE_7B; break; + default: type = LLM_TYPE_UNKNOWN; + } break; + case 61: + switch (hparams.n_embd) { + case 4096: type = LLM_TYPE_14B; break; + default: type = LLM_TYPE_UNKNOWN; + } break; default: type = LLM_TYPE_UNKNOWN; } } break; From 84712b60439453fb393c2ca753cee682a4ad41f5 Mon Sep 17 00:00:00 2001 From: Jeff Bolz Date: Tue, 22 Jul 2025 10:35:21 -0500 Subject: [PATCH 2/3] vulkan: fix rms_norm_mul to handle broadcasting dim0 (#14817) --- ggml/src/ggml-vulkan/ggml-vulkan.cpp | 2 +- ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp | 10 ++++++++-- 2 files changed, 9 insertions(+), 3 deletions(-) diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp index c3f1369b663..1a7a381ce59 100644 --- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp +++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp @@ -10248,7 +10248,7 @@ static bool ggml_vk_can_fuse(const struct ggml_cgraph * cgraph, int node_idx, st } // if rms_norm is the B operand, then we don't handle broadcast if (rms_norm == mul->src[1] && - mul->src[0]->ne[1] != rms_norm->ne[1]) { + !ggml_are_same_shape(mul->src[0], rms_norm)) { return false; } // rms_norm shader assumes contiguous rows diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp b/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp index 6428ca7ba33..bdd7db2d698 100644 --- a/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +++ b/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp @@ -50,8 +50,14 @@ void main() { const FLOAT_TYPE scale = inversesqrt(mean + FLOAT_TYPE(p.param1)); if (do_multiply) { - [[unroll]] for (uint col = tid; col < ncols; col += BLOCK_SIZE) { - data_d[d_offset + col] = D_TYPE(scale * FLOAT_TYPE(data_a[a_offset + col]) * FLOAT_TYPE(data_b[b_offset + col])); + if (ncols > p.ne10) { + [[unroll]] for (uint col = tid; col < ncols; col += BLOCK_SIZE) { + data_d[d_offset + col] = D_TYPE(scale * FLOAT_TYPE(data_a[a_offset + col]) * FLOAT_TYPE(data_b[b_offset + fastmod(col, p.ne10)])); + } + } else { + [[unroll]] for (uint col = tid; col < ncols; col += BLOCK_SIZE) { + data_d[d_offset + col] = D_TYPE(scale * FLOAT_TYPE(data_a[a_offset + col]) * FLOAT_TYPE(data_b[b_offset + col])); + } } } else { [[unroll]] for (uint col = tid; col < ncols; col += BLOCK_SIZE) { From acd6cb1c41676f6bbb25c2a76fa5abeb1719301e Mon Sep 17 00:00:00 2001 From: Csaba Kecskemeti Date: Tue, 22 Jul 2025 09:29:43 -0700 Subject: [PATCH 3/3] ggml : model card yaml tab->2xspace (#14819) --- gguf-py/gguf/metadata.py | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/gguf-py/gguf/metadata.py b/gguf-py/gguf/metadata.py index e807f434689..67efedbdbc5 100644 --- a/gguf-py/gguf/metadata.py +++ b/gguf-py/gguf/metadata.py @@ -144,6 +144,10 @@ def load_model_card(model_path: Optional[Path] = None) -> dict[str, Any]: # Quick hack to fix the Norway problem # https://hitchdev.com/strictyaml/why/implicit-typing-removed/ yaml_content = yaml_content.replace("- no\n", "- \"no\"\n") + # yaml should use 2 spaces insted of tab + # this issue has came up with the Qwen/Qwen3-235B-A22B-Instruct-2507 model card + # (I've also sent a pr tp fix the modelcard too) + yaml_content = yaml_content.replace("\t", " ") if yaml_content: data = yaml.safe_load(yaml_content)