SOURCE  C:\Users\marco\.codex\sessions\2026\05\31\rollout-2026-05-31T22-02-05-019e80ea-b4d5-72b0-8da6-27bbf39ec069.jsonl
        line 519
        2026-05-31 22:31 -04:00
NOTE    Literal terminal output, unedited. [ ... ] markers are the only
        editorial additions. No image was rendered for this extract, so
        there is no amber marking: every [ ... ] line is ours.
----------------------------------------------------------------------------------------------------
validation_summary.json   3840 2026-05-31 10:31:32 PM





{

    "generated_at":  "2026-05-31T22:31:32.4959235-04:00",

    "objective":  "F51 Nitro 20GB kit safe5 validation on local RTX 5060 Ti 16GB",

    "dirty_diff_archive":  "C:\\Users\\marco\\Desktop\\F51-Nitro-MoE\\evidencias\\patches\\llamacpp_dirty_diff_20260531_220558.patch",

    "source_worktree":  "C:\\Users\\marco\\Desktop\\F51-Nitro-MoE\\build_worktrees\\llama_clean_20260531_220628",

    "source_head":  "579fe01",

    "applied_patch":  "F51_Nitro_20GB_DualGPU\\patches\\f51_llama_cpp_swa_checkpoint_seqid_retarget_2026-05-30.patch",

    "rejected_patch":  "F51_Nitro_20GB_DualGPU\\patches\\f51_arch_aware_moe_pool_2026-05-31.patch failed git apply --check: corrupt patch at line 62",

    "clean_build":  {

                        "path":  "C:\\Users\\marco\\Desktop\\F51-Nitro-MoE\\build_worktrees\\llama_clean_20260531_220628\\build-f51-swa\\bin\\llama-server.exe",

                        "version":  "version: 3 (579fe01), MSVC 19.50.35721.0, CUDA 12.8, CMAKE_CUDA_ARCHITECTURES=120a",

                        "sha256":  "66ADA34304FB7AF62A9A666005B4D7CC7E9F136A586AA4C1046A4FD76CB5CB21"

                    },

    "packaged_runtime":  {

                             "path":  "C:\\Users\\marco\\Desktop\\F51-Nitro-MoE\\F51_Nitro_20GB_DualGPU\\bin\\llama-server.exe",

                             "version":  "version: 9049 (8f18e1350), MSVC 19.50.35721.0",

                             "sha256":  "7E1A64FF492E12DCCCBA8904CDECF933EEF55C9F0F121639D7F3472361FB4256"

                         },

    "safe5":  {

                  "command":  "profiles\\Start-F51-20GB.ps1 -Profile safe5 -Port 8080",

                  "model":  "gpt-oss-120b-mxfp4-00001-of-00003.gguf",

                  "args":  "--n-cpu-moe 31 --no-mmap -ngl 99 -c 4096 -b 2048 -ub 512 -t 16 --host 127.0.0.1 --port 8080 -fa on --swa-full -fit on -fitc 4096 -sm none",

                  "health":  "200 {\"status\":\"ok\"}",

                  "load_time_to_health_seconds_approx":  150,

                  "completion":  {

                                     "prompt":  "Explain MoE in one sentence.",

                                     "n_predict":  64,

                                     "wall_seconds":  35.82,

                                     "tokens_predicted":  64,

                                     "tokens_evaluated":  7,

                                     "prompt_per_second":  1.8820174043592361,

                                     "predicted_per_second":  2.0019705646514185,

                                     "stop_type":  "limit"

                                 },

                  "gpu_during_completion":  {

                                                "memory_used_mib":  16025,

                                                "memory_free_mib":  26,

                                                "utilization_percent":  0,

                                                "note":  "sample after completion/health; nvidia-smi under WDDM reports some process memory as N/A"

                                            },

                  "f51_evidence":  {

                                       "cache_log":  "f51_moe_cache.log showed CPU expert tensors, capacity=16, expert_size=4406400, hits/misses active",

                                       "scheduler_log":  "f51_scheduler.log showed skip backend=CUDA0 for MoE MUL_MAT_ID tensors",

                                       "stderr_log":  "server_stderr.log showed model loaded, server listening, prompt/eval timings, F51 L3 lookahead accuracy=70.0% preds=40"

                                   }

              },

    "post_stop":  {

                      "gpu_memory_used_mib":  208,

                      "gpu_memory_free_mib":  15843,

                      "server_process":  "stopped"

                  }