Apply HauhauCS FastMTP d2t vocab trimming patch

src/models/qwen35.cpp

@@ -36,12 +36,22 @@ void llama_model_qwen35::load_arch_tensors(llama_model_loader & ml) {
     const auto nf = nextn_flags(ml);
     const int trunk_flags = nf.trunk;
     const int mtp_flags   = nf.mtp;
+    const bool mtp_only   = (nf.trunk & TENSOR_NOT_REQUIRED) != 0;
+
+    int64_t n_vocab_out = n_vocab;
+    const ggml_tensor * d2t_meta = ml.get_tensor_meta("d2t");
+    if (mtp_only && d2t_meta) {
+        n_vocab_out = d2t_meta->ne[0];
+        d2t = create_tensor(tn(LLM_TENSOR_D2T), { n_vocab_out }, 0);
+        LLAMA_LOG_INFO("%s: QWEN35 MTP using d2t draft-vocab trim (n_vocab_out = %lld)\n",
+                __func__, (long long) n_vocab_out);
+    }
 
     tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), { n_embd, n_vocab }, 0);
 
     // output
     output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), { n_embd }, 0);
-    output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), { n_embd, n_vocab }, TENSOR_NOT_REQUIRED);
+    output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), { n_embd, n_vocab_out }, TENSOR_NOT_REQUIRED);
 
     // optional projection of the embeddings output
     cls_out   = create_tensor(tn(LLM_TENSOR_CLS_OUT, "weight"), { n_embd, hparams.n_embd_out() }, TENSOR_NOT_REQUIRED);
@@ -49,6 +59,7 @@ void llama_model_qwen35::load_arch_tensors(llama_model_loader & ml) {
 
     // if output is NULL, init from the input tok embed
     if (output == NULL) {
+        GGML_ASSERT(!d2t && "d2t draft-vocab trim requires output.weight");
         output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), { n_embd, n_vocab }, TENSOR_DUPLICATED);
     }
 
@@ -667,5 +678,21 @@ llama_model_qwen35::graph_mtp::graph_mtp(const llama_model & model, const llm_gr
     cur = build_lora_mm(head_w, cur, head_s);
     cb(cur, "result_output", -1);
 
+    if (model.d2t) {
+        const int64_t n_draft_vocab = cur->ne[0];
+        const int64_t n_outputs     = cur->ne[1];
+        const int64_t n_vocab_full  = (int64_t) model.vocab.n_tokens();
+
+        GGML_ASSERT(model.d2t->ne[0] == n_draft_vocab);
+
+        ggml_tensor * logits = ggml_fill(ctx0,
+                ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, 1, n_vocab_full, n_outputs), -INFINITY);
+        cur = ggml_set_rows(ctx0, logits,
+                ggml_reshape_3d(ctx0, cur,       1,             n_draft_vocab, n_outputs),
+                ggml_reshape_3d(ctx0, model.d2t, n_draft_vocab, 1,             1));
+        cur = ggml_reshape_2d(ctx0, cur, n_vocab_full, n_outputs);
+        cb(cur, "result_output_d2t", -1);
+    }
+
     res->t_logits = cur;
 }