owen/longhaul.cpp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

progress

Browse Source
This commit is contained in:
Owen Qwen
2026-07-30 21:20:05 -05:00
parent d5458aa44e
commit 22f42628dd
25 changed files with 1038 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-model.cpp
+6 -3
View File
@@ -312,6 +312,8 @@ static llama_model * llama_model_mapping(llm_arch arch, const llama_model_params
return new llama_model_kimi_linear(params);
case LLM_ARCH_STEP35:
return new llama_model_step35(params);
case LLM_ARCH_INKLING:
return new llama_model_inkling(params);
default:
throw std::runtime_error(std::string("unsupported model architecture: '") + llm_arch_name(arch) + "'");
}
@@ -1356,8 +1358,8 @@ bool llama_model_base::load_tensors(llama_model_loader & ml) {
}
if (params.load_mode == LLAMA_LOAD_MODE_LONGHAUL) {
if (arch != LLM_ARCH_QWEN35MOE && arch != LLM_ARCH_LAGUNA) {
throw std::runtime_error("longhaul currently requires a qwen35moe or laguna model");
if (arch != LLM_ARCH_QWEN35MOE && arch != LLM_ARCH_LAGUNA && arch != LLM_ARCH_INKLING) {
throw std::runtime_error("longhaul currently requires a qwen35moe, laguna, or inkling model");
}
if (hparams.n_layer_nextn != 0) {
throw std::runtime_error("longhaul does not support MTP tensors");
@@ -2165,7 +2167,7 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
// layer filters, so pick the right one here
llama_memory_hybrid::layer_filter_cb filter_attn = nullptr;
llama_memory_hybrid::layer_filter_cb filter_recr = nullptr;
if (arch == LLM_ARCH_FALCON_H1) {
if (arch == LLM_ARCH_FALCON_H1 || arch == LLM_ARCH_INKLING) {
filter_attn = [&](uint32_t) { return true; };
filter_recr = [&](uint32_t) { return true; };
} else if (arch == LLM_ARCH_NEMOTRON_H || arch == LLM_ARCH_NEMOTRON_H_MOE) {
@@ -2506,6 +2508,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
case LLM_ARCH_NEMOTRON_H:
case LLM_ARCH_NEMOTRON_H_MOE:
case LLM_ARCH_KIMI_LINEAR:
case LLM_ARCH_INKLING:
return LLAMA_ROPE_TYPE_NONE;
// use what we call a normal RoPE, operating on pairs of consecutive head values