diff --git a/crates/inference-core/src/engine.rs b/crates/inference-core/src/engine.rs index 8aba11e..55a9c9b 100644 --- a/crates/inference-core/src/engine.rs +++ b/crates/inference-core/src/engine.rs @@ -232,7 +232,7 @@ impl InferenceEngine { LLAMA_FLASH_ATTN_TYPE_DISABLED }; cparams.offload_kqv = true; - cparams.no_perf = false; + cparams.no_perf = true; // 6. Init context let ctx_ptr = unsafe { (lib.init_from_model)(model_ptr, cparams) }; diff --git a/crates/inference-core/src/session.rs b/crates/inference-core/src/session.rs index a0defd6..e3286d6 100644 --- a/crates/inference-core/src/session.rs +++ b/crates/inference-core/src/session.rs @@ -112,7 +112,8 @@ impl GenerationSession { debug!(n_tokens = n_prompt_tokens, "Prompt tokenised"); // ── 2. Build sampler chain ───────────────────────────────────────── - let sc_params = unsafe { (lib.sampler_chain_default_params)() }; + let mut sc_params = unsafe { (lib.sampler_chain_default_params)() }; + sc_params.no_perf = true; let sampler_chain = unsafe { (lib.sampler_chain_init)(sc_params) }; if sampler_chain.is_null() { bail!("sampler_chain_init returned NULL");