refactor: allow narrowing embeddings/rerank model ctx size

Signed-off-by: thxCode <thxcode0824@gmail.com>

llama-box/patches/llama.cpp/embedding.patch

@@ -12,19 +12,19 @@ index ab5e376e..658fd56a 100644
  }
  #endif
 diff --git a/src/llama.cpp b/src/llama.cpp
-index af5e686e..60721c87 100644
+index af5e686e..7fc47b78 100644
 --- a/src/llama.cpp
 +++ b/src/llama.cpp
 @@ -19693,10 +19693,10 @@ struct llama_context * llama_new_context_with_model(
      cparams.rope_freq_scale  = params.rope_freq_scale == 0.0f ? hparams.rope_freq_scale_train : params.rope_freq_scale;
 
      // this is necessary due to kv_self.n being padded later during inference
 -    cparams.n_ctx            = GGML_PAD(cparams.n_ctx, llama_kv_cache_get_padding(cparams));
-+    cparams.n_ctx            = hparams.causal_attn ? GGML_PAD(cparams.n_ctx, llama_kv_cache_get_padding(cparams)) : hparams.n_ctx_train;
++    cparams.n_ctx            = hparams.causal_attn ? GGML_PAD(cparams.n_ctx, llama_kv_cache_get_padding(cparams)) : std::min(cparams.n_ctx, hparams.n_ctx_train);
 
      // with causal attention, the batch size is limited by the context size
 -    cparams.n_batch          = hparams.causal_attn ? std::min(cparams.n_ctx, params.n_batch) : params.n_batch;
-+    cparams.n_batch          = hparams.causal_attn ? std::min(cparams.n_ctx, params.n_batch) : hparams.n_ctx_train;
++    cparams.n_batch          = hparams.causal_attn ? std::min(cparams.n_ctx, params.n_batch) : cparams.n_ctx;
 
      // the batch has to be at least GGML_KQ_MASK_PAD because we will be padding the KQ_mask
      // this is required by GPU kernels in order to avoid out-of-bounds accesses (e.g. ggml_flash_attn_ext)
@@ -33,7 +33,7 @@ index af5e686e..60721c87 100644
      }
 
 -    cparams.n_ubatch         = std::min(cparams.n_batch, params.n_ubatch == 0 ? params.n_batch : params.n_ubatch);
-+    cparams.n_ubatch         = hparams.causal_attn ? std::min(cparams.n_batch, params.n_ubatch == 0 ? params.n_batch : params.n_ubatch) : hparams.n_ctx_train;
++    cparams.n_ubatch         = hparams.causal_attn ? std::min(cparams.n_batch, params.n_ubatch == 0 ? params.n_batch : params.n_ubatch) : cparams.n_ctx;
 
      cparams.n_ctx_orig_yarn  = params.yarn_orig_ctx    != 0 ? params.yarn_orig_ctx    :
                                 hparams.n_ctx_orig_yarn != 0 ? hparams.n_ctx_orig_yarn :