model: add Clover support (#31306)

* model : add AR-only Clover MoE support

Add GGUF conversion, model loading, and causal hybrid attention for Clover, reusing shared attention, recurrent-state, MoE, and LoRA helpers.

* Update tests/test-llama-archs.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>

* convert : filter Clover tensors using filter_tensors

* clover : use the standard ssm_a tensor name

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
This commit is contained in:
Maphist0andSigbjørn Skjæret authored and GitHub committed 2026-11-02 18:37:27 +01:00
1 parent 1b6935b227
commit e4b6c8ca6a
11 files changed
+560 -6

No files matched your search

+4
View File
@@ -38,6 +38,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
{ LLM_ARCH_QWEN3, "qwen3" },
{ LLM_ARCH_FLINT, "flint" },
{ LLM_ARCH_QWEN3MOE, "qwen3moe" },
{ LLM_ARCH_CLOVER_MOE, "clover_moe" },
{ LLM_ARCH_QWEN3NEXT, "qwen3next" },
{ LLM_ARCH_QWEN3VL, "qwen3vl" },
{ LLM_ARCH_QWEN3VLMOE, "qwen3vlmoe" },
@@ -1292,6 +1293,7 @@ bool llm_arch_is_hybrid(const llm_arch & arch) {
case LLM_ARCH_GLM5_NEXT:
case LLM_ARCH_QWEN35:
case LLM_ARCH_QWEN35MOE:
case LLM_ARCH_CLOVER_MOE:
case LLM_ARCH_QWEN4EXP:
case LLM_ARCH_DEEPSEEK4:
case LLM_ARCH_DEEPSEEK41:
@@ -1319,6 +1321,7 @@ bool llm_arch_supports_rs_rollback(const llm_arch & arch) {
case LLM_ARCH_KIMI_K3:
case LLM_ARCH_QWEN35:
case LLM_ARCH_QWEN35MOE:
case LLM_ARCH_CLOVER_MOE:
case LLM_ARCH_QWEN4EXP:
case LLM_ARCH_DEEPSEEK4:
case LLM_ARCH_DEEPSEEK41:
@@ -1354,6 +1357,7 @@ bool llm_arch_supports_mixed_batch(const llm_arch & arch) {
bool llm_arch_supports_sm_tensor(const llm_arch & arch) {
switch (arch) {
case LLM_ARCH_CLOVER_MOE:
case LLM_ARCH_GROK:
case LLM_ARCH_MPT:
case LLM_ARCH_PLAMO2: