Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76760.54557534613,
-      "required_pages_per_second_upper_estimate": 0.005310472651906812
+      "required_token_positions_per_second": 76494.20415156282,
+      "required_pages_per_second_upper_estimate": 0.005292046534212236
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9595.068196918266,
-      "required_pages_per_second_upper_estimate": 0.004017254686241143
+      "required_token_positions_per_second": 9561.775518945353,
+      "required_pages_per_second_upper_estimate": 0.004003315737204058
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19190.136393836532,
-      "required_pages_per_second_upper_estimate": 0.003996799332907177
+      "required_token_positions_per_second": 19123.551037890706,
+      "required_pages_per_second_upper_estimate": 0.003982931359237582
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28785.204590754798,
-      "required_pages_per_second_upper_estimate": 0.0017256290627532703
+      "required_token_positions_per_second": 28685.32655683606,
+      "required_pages_per_second_upper_estimate": 0.0017196415271247692
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28785.204590754798,
-      "required_pages_per_second_upper_estimate": 0.0022379230441970373
+      "required_token_positions_per_second": 28685.32655683606,
+      "required_pages_per_second_upper_estimate": 0.0022301579663768977
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28785.204590754798,
-      "required_pages_per_second_upper_estimate": 0.3692256973455291
+      "required_token_positions_per_second": 28685.32655683606,
+      "required_pages_per_second_upper_estimate": 0.3679445691670971
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07010391683584771,
-    "producer_p95_seconds": 0.2553814689999854,
-    "gpu_update_seconds": 0.9604934337996665,
-    "producer_batches_per_second": 14.264538204642069,
-    "gpu_batches_per_second": 1.041131531783666,
-    "producer_headroom": 13.70099528174319,
+    "producer_batch_seconds": 0.0865716622965067,
+    "producer_p95_seconds": 0.2721079570001166,
+    "gpu_update_seconds": 0.9638377288548298,
+    "producer_batches_per_second": 11.551123929849174,
+    "gpu_batches_per_second": 1.0375190450244522,
+    "producer_headroom": 11.133409054266503,
     "sustainable": true,
-    "buffer_seconds": 60.51108632937898,
+    "buffer_seconds": 60.72177691785428,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 508829.1653785763,
-    "upload_bytes_per_second": 5543435.224251064,
+    "generation_bytes_per_second": 538655.0950930445,
+    "upload_bytes_per_second": 5590308.97750657,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.18289395349471,
+    "estimated_full_shards_per_hour": 115.58284415811062,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.1130494770004589,
-      "bytes": 2612508,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.9836850789997698,
       "bytes": 10345276,
@@ -562,6 +556,12 @@
       "bytes": 5944567,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1342636209992634,
+      "bytes": 6342296,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -213,3 +213,4 @@
 {"step": 506300, "loss": 0.6083144545555115, "lr": 0.0003, "grad_norm": 0.19012964500297008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74105556554, "tokens_per_second": 153883.2677622616, "data_wait_s": 1.4766999811399728e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 506400, "loss": 0.5746796727180481, "lr": 0.0003, "grad_norm": 0.15227731313016363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74120257019, "tokens_per_second": 153270.9278461322, "data_wait_s": 1.2759999663103372e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 506500, "loss": 0.5777621865272522, "lr": 0.0003, "grad_norm": 0.151934990963674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74135012738, "tokens_per_second": 153225.66126304475, "data_wait_s": 1.8749000446405262e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 506600, "loss": 0.5785301327705383, "lr": 0.0003, "grad_norm": 0.16547915430050822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74149480947, "tokens_per_second": 150849.97201876153, "data_wait_s": 8.549002814106643e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6cafb504d1665ce324e1931a600b94e152b0a0a6c0b0631f51f6385da449b4d2
-size 458667
+oid sha256:319fe383fc75434a6c542c4383a5b35c3a0d545d6459006e307f5b7e02d09521
+size 460785