server : fixes + clean-up

Author: ggerganov

tools/server/README.md

@@ -587,7 +587,7 @@ These words will not be included in the completion, so make sure to add them to
   - `word`: Stopped due to encountering a stopping word from `stop` JSON array provided
 - `stopping_word`: The stopping word encountered which stopped the generation (or "" if not stopped due to a stopping word)
 - `timings`: Hash of timing information about the completion such as the number of tokens `predicted_per_second`
-- `tokens_cached`: Number of tokens from the prompt which could be re-used from previous completion (`n_past`)
+- `tokens_cached`: Number of tokens from the prompt which could be re-used from previous completion
 - `tokens_evaluated`: Number of tokens evaluated in total from the prompt
 - `truncated`: Boolean indicating if the context size was exceeded during generation, i.e. the number of tokens provided in the prompt (`tokens_evaluated`) plus tokens generated (`tokens predicted`) exceeded the context size (`n_ctx`)
 
@@ -1045,7 +1045,7 @@ Available metrics:
 - `llamacpp:kv_cache_tokens`: KV-cache tokens.
 - `llamacpp:requests_processing`: Number of requests processing.
 - `llamacpp:requests_deferred`: Number of requests deferred.
-- `llamacpp:n_past_max`: High watermark of the context size observed.
+- `llamacpp:n_tokens_max`: High watermark of the context size observed.
 
 ### POST `/slots/{id_slot}?action=save`: Save the prompt cache of the specified slot to a file.
 

tools/server/server.cpp

@@ -3707,9 +3707,9 @@ struct server_context {
                                 n_past = slot.prompt.tokens.get_common_prefix(input_tokens);
 
                                 // if there is an alora invoked, don't cache after the invocation start
-                                if (slot.alora_invocation_start >= 0) {
-                                    SLT_DBG(slot, "only caching to alora invocation start (n_past=%d, alora_invocation_start=%d)\n", n_past, slot.alora_invocation_start);
-                                    n_past = std::min(n_past, slot.alora_invocation_start);
+                                if (slot.alora_invocation_start > 0) {
+                                    SLT_DBG(slot, "only caching to alora invocation start (n_past = %d, alora_invocation_start = %d)\n", n_past, slot.alora_invocation_start);
+                                    n_past = std::min(n_past, slot.alora_invocation_start - 1);
                                 }
 
                                 // reuse chunks from the cached prompt by shifting their KV cache in the new position
@@ -3769,7 +3769,7 @@ struct server_context {
                             const auto n_swa = std::max(1, llama_model_n_swa(model));
 
                             // the largest pos_min required for a checkpoint to be useful
-                            const auto pos_min_thold = std::max(0, n_past - n_swa - 1);
+                            const auto pos_min_thold = std::max(0, n_past - n_swa);
 
                             if (n_past > 0 && n_past < slot.prompt.n_tokens()) {
                                 const auto pos_min = llama_memory_seq_pos_min(llama_get_memory(ctx), slot.id);