model: support vllm decision 2 model (#30378)

* models: support pplx-decider

* vllm d2

* update

* clarify naming rule

* stricter validation

* fix ci

* rm redundant filter_tensors
This commit is contained in:
Xuan-Son Nguyen authored and GitHub committed 2026-10-12 13:39:55 +02:00
1 parent ccdbe29053
commit 097cf33d94
23 files changed
+549 -19

No files matched your search

+5
View File
@@ -346,6 +346,11 @@ static bool tensor_allows_quantization(const llama_model_quantize_params * param
quantize &= name.find("attn_v_b.weight") == std::string::npos;
}
// the decision head of vllm-decision2 is small and computed in F32
if (arch == LLM_ARCH_VLLM_DECISION2) {
quantize &= name.compare(0, 9, "decision.") != 0;
}
// do not quantize RWKV's small yet 2D weights
quantize &= name.find("time_mix_first.weight") == std::string::npos;
quantize &= name.find("time_mix_w0.weight") == std::string::npos;