llama.cpp : split llama_context_params into model and context params (#3301)

* llama.cpp : split llama_context_params into model and context params

ggml-ci

* fix metal build

* fix freq_base/scale default to model value

* llama-bench : keep the same model between tests when possible

* move n_threads to llama_context_params, add n_threads_batch

* fix mpi build

* remove kv_size(), cuda scratch fixes

* remove low-vram option

* add n_threads_batch to system info, refactor to get_system_info()

* add documentation about --threads-batch to the READMEs

* llama-bench fix

* main : fix rope freq/scale warning

* llama.cpp : add llama_get_model
common : add llama_tokenize from model

* remove duplicated ctx/model functions

ggml-ci

* cuda : print total VRAM used

1 files changed, 21 insertions, 20 deletions

diff --git a/examples/main/main.cpp b/examples/main/main.cpp
index 1ed543cb..fd506773 100644
--- a/examples/main/main.cpp
+++ b/examples/main/main.cpp
@@ -140,12 +140,17 @@ int main(int argc, char ** argv) {
         return 0;
     }
 
-    if (params.rope_freq_base != 10000.0) {
-        LOG_TEE("%s: warning: changing RoPE frequency base to %g (default 10000.0)\n", __func__, params.rope_freq_base);
+    if (params.n_ctx != 0 && params.n_ctx < 8) {
+        LOG_TEE("%s: warning: minimum context size is 8, using minimum size.\n", __func__);
+        params.n_ctx = 8;
+    }
+
+    if (params.rope_freq_base != 0.0) {
+        LOG_TEE("%s: warning: changing RoPE frequency base to %g.\n", __func__, params.rope_freq_base);
     }
 
-    if (params.rope_freq_scale != 1.0) {
-        LOG_TEE("%s: warning: scaling RoPE frequency by %g (default 1.0)\n", __func__, params.rope_freq_scale);
+    if (params.rope_freq_scale != 0.0) {
+        LOG_TEE("%s: warning: scaling RoPE frequency by %g.\n", __func__, params.rope_freq_scale);
     }
 
     LOG_TEE("%s: build = %d (%s)\n", __func__, BUILD_NUMBER, BUILD_COMMIT);
@@ -184,20 +189,19 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
-    const int n_ctx_train = llama_n_ctx_train(ctx);
-    if (params.n_ctx > n_ctx_train) {
+    const int n_ctx_train = llama_n_ctx_train(model);
+    const int n_ctx = llama_n_ctx(ctx);
+    LOG("n_ctx: %d\n", n_ctx);
+
+    if (n_ctx > n_ctx_train) {
         LOG_TEE("%s: warning: model was trained on only %d context tokens (%d specified)\n",
-                __func__, n_ctx_train, params.n_ctx);
-    } else if (params.n_ctx < 8) {
-        LOG_TEE("%s: warning: minimum context size is 8, using minimum size.\n", __func__);
-        params.n_ctx = 8;
+                __func__, n_ctx_train, n_ctx);
     }
 
     // print system information
     {
         LOG_TEE("\n");
-        LOG_TEE("system_info: n_threads = %d / %d | %s\n",
-                params.n_threads, std::thread::hardware_concurrency(), llama_print_system_info());
+        LOG_TEE("%s\n", get_system_info(params).c_str());
     }
 
     std::string path_session = params.path_prompt_cache;
@@ -211,7 +215,7 @@ int main(int argc, char ** argv) {
         if (fp != NULL) {
             std::fclose(fp);
 
-            session_tokens.resize(params.n_ctx);
+            session_tokens.resize(n_ctx);
             size_t n_token_count_out = 0;
             if (!llama_load_session_file(ctx, path_session.c_str(), session_tokens.data(), session_tokens.capacity(), &n_token_count_out)) {
                 LOG_TEE("%s: error: failed to load session file '%s'\n", __func__, path_session.c_str());
@@ -226,7 +230,7 @@ int main(int argc, char ** argv) {
         }
     }
 
-    const bool add_bos = llama_vocab_type(ctx) == LLAMA_VOCAB_TYPE_SPM;
+    const bool add_bos = llama_vocab_type(model) == LLAMA_VOCAB_TYPE_SPM;
     LOG("add_bos: %d\n", add_bos);
 
     std::vector<llama_token> embd_inp;
@@ -267,9 +271,6 @@ int main(int argc, char ** argv) {
         LOG("guidance_offset:     %s", log_tostr(guidance_offset));
     }
 
-    const int n_ctx = llama_n_ctx(ctx);
-    LOG("n_ctx: %d\n", n_ctx);
-
     if ((int) embd_inp.size() > n_ctx - 4) {
         LOG_TEE("%s: error: prompt is too long (%d tokens, max %d)\n", __func__, (int) embd_inp.size(), n_ctx - 4);
         return 1;
@@ -466,7 +467,7 @@ int main(int argc, char ** argv) {
     std::vector<llama_token> embd;
     std::vector<llama_token> embd_guidance;
 
-    const int n_vocab = llama_n_vocab(ctx);
+    const int n_vocab = llama_n_vocab(model);
 
     std::vector<llama_token_data> candidates;
     candidates.reserve(n_vocab);
@@ -576,7 +577,7 @@ int main(int argc, char ** argv) {
 
                 for (int i = 0; i < input_size; i += params.n_batch) {
                     int n_eval = std::min(input_size - i, params.n_batch);
-                    if (llama_decode(ctx_guidance, llama_batch_get_one(input_buf + i, n_eval, n_past_guidance, 0), params.n_threads)) {
+                    if (llama_decode(ctx_guidance, llama_batch_get_one(input_buf + i, n_eval, n_past_guidance, 0))) {
                         LOG_TEE("%s : failed to eval\n", __func__);
                         return 1;
                     }
@@ -593,7 +594,7 @@ int main(int argc, char ** argv) {
 
                 LOG("eval: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, embd));
 
-                if (llama_decode(ctx, llama_batch_get_one(&embd[i], n_eval, n_past, 0), params.n_threads)) {
+                if (llama_decode(ctx, llama_batch_get_one(&embd[i], n_eval, n_past, 0))) {
                     LOG_TEE("%s : failed to eval\n", __func__);
                     return 1;
                 }