common: avoid deprecated draft warmup API

Anbeeld · Anbeeld · commit 275647528c36 · 2026-06-03T11:12:17.000+02:00
diff --git a/common/speculative.cpp b/common/speculative.cpp
@@ -4310,7 +4310,6 @@ llama_context * common_speculative_create_ctx_dft(const common_params_speculativ
         if (eos != LLAMA_TOKEN_NULL) { tmp[n_tmp++] = eos; }
         if (n_tmp == 0) { tmp[n_tmp++] = 0; }
 
-        llama_set_warmup(ctx_dft, true);
         int ret = llama_decode(ctx_dft, llama_batch_get_one(tmp, n_tmp));
         if (ret != 0) {
             LOG_WRN("%s: draft warmup decode failed: %d (non-fatal)\n", __func__, ret);
@@ -4322,7 +4321,6 @@ llama_context * common_speculative_create_ctx_dft(const common_params_speculativ
         }
         llama_synchronize(ctx_dft);
         llama_perf_context_reset(ctx_dft);
-        llama_set_warmup(ctx_dft, false);
 
         LOG_INF("%s: draft model warmup complete\n", __func__);
     }

Original file line number	Diff line number	Diff line change
`@@ -4310,7 +4310,6 @@ llama_context * common_speculative_create_ctx_dft(const common_params_speculativ`
`4310`	`4310`	`if (eos != LLAMA_TOKEN_NULL) { tmp[n_tmp++] = eos; }`
`4311`	`4311`	`if (n_tmp == 0) { tmp[n_tmp++] = 0; }`
`4312`	`4312`
`4313`		`- llama_set_warmup(ctx_dft, true);`
`4314`	`4313`	`int ret = llama_decode(ctx_dft, llama_batch_get_one(tmp, n_tmp));`
`4315`	`4314`	`if (ret != 0) {`
`4316`	`4315`	`LOG_WRN("%s: draft warmup decode failed: %d (non-fatal)\n", __func__, ret);`
`@@ -4322,7 +4321,6 @@ llama_context * common_speculative_create_ctx_dft(const common_params_speculativ`
`4322`	`4321`	`}`
`4323`	`4322`	`llama_synchronize(ctx_dft);`
`4324`	`4323`	`llama_perf_context_reset(ctx_dft);`
`4325`		`- llama_set_warmup(ctx_dft, false);`
`4326`	`4324`
`4327`	`4325`	`LOG_INF("%s: draft model warmup complete\n", __func__);`
`4328`	`4326`	`}`