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": 76357.51005666674,
-      "required_pages_per_second_upper_estimate": 0.005282589719553327
+      "required_token_positions_per_second": 77501.12837690504,
+      "required_pages_per_second_upper_estimate": 0.005361707888507505
     },
     {
       "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": 9544.688757083342,
-      "required_pages_per_second_upper_estimate": 0.003996161866824596
+      "required_token_positions_per_second": 9687.64104711313,
+      "required_pages_per_second_upper_estimate": 0.004056013005476796
     },
     {
       "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": 19089.377514166685,
-      "required_pages_per_second_upper_estimate": 0.0039758139154622936
+      "required_token_positions_per_second": 19375.28209422626,
+      "required_pages_per_second_upper_estimate": 0.004035360299677896
     },
     {
       "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": 28634.066271250027,
-      "required_pages_per_second_upper_estimate": 0.0017165685512738108
+      "required_token_positions_per_second": 29062.92314133939,
+      "required_pages_per_second_upper_estimate": 0.001742277865809143
     },
     {
       "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": 28634.066271250027,
-      "required_pages_per_second_upper_estimate": 0.002226172704642752
+      "required_token_positions_per_second": 29062.92314133939,
+      "required_pages_per_second_upper_estimate": 0.002259514439950183
     },
     {
       "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": 28634.066271250027,
-      "required_pages_per_second_upper_estimate": 0.367287057262606
+      "required_token_positions_per_second": 29062.92314133939,
+      "required_pages_per_second_upper_estimate": 0.3727879727214811
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08469342544537994,
-    "producer_p95_seconds": 0.2953102920000674,
-    "gpu_update_seconds": 0.965563177024561,
-    "producer_batches_per_second": 11.80729194434242,
-    "gpu_batches_per_second": 1.0356650127043556,
-    "producer_headroom": 11.400686321835773,
+    "producer_batch_seconds": 0.10353483939829289,
+    "producer_p95_seconds": 0.309768326998892,
+    "gpu_update_seconds": 0.9513151813925149,
+    "producer_batches_per_second": 9.65858454807714,
+    "gpu_batches_per_second": 1.0511763288968239,
+    "producer_headroom": 9.188358111348947,
     "sustainable": true,
-    "buffer_seconds": 60.83048015254734,
+    "buffer_seconds": 60.884171609120955,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331883.15068931517,
-    "upload_bytes_per_second": 5613899.857854061,
+    "generation_bytes_per_second": 344605.1923345548,
+    "upload_bytes_per_second": 5588235.649889084,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.21439829358665,
+    "estimated_full_shards_per_hour": 73.94425227668269,
     "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": 1.3863541229984548,
-      "bytes": 10474994,
-      "pages": 40,
-      "shards": 3
-    },
-    {
-      "seconds": 1.2548714610020397,
-      "bytes": 2378266,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.0437052470006165,
       "bytes": 6102982,
@@ -562,6 +550,18 @@
       "bytes": 13287961,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 0.7943182929993782,
+      "bytes": 1190422,
+      "pages": 36,
+      "shards": 1
+    },
+    {
+      "seconds": 1.1334048999997322,
+      "bytes": 5665943,
+      "pages": 19,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -298,3 +298,4 @@
 {"step": 674800, "loss": 0.5367597937583923, "lr": 0.0001, "grad_norm": 0.1808463882216845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98773094831, "tokens_per_second": 152515.78546607023, "data_wait_s": 1.2541997421067208e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 674900, "loss": 0.5363932251930237, "lr": 0.0001, "grad_norm": 0.20466907236173054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98787737477, "tokens_per_second": 152134.1510004654, "data_wait_s": 8.256000000983477e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 675000, "loss": 0.5931280255317688, "lr": 0.0001, "grad_norm": 0.20633527339194027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98802239020, "tokens_per_second": 151837.96994796887, "data_wait_s": 1.0897001629928127e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 675100, "loss": 0.5807151794433594, "lr": 0.0001, "grad_norm": 0.1930958039307218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98816785728, "tokens_per_second": 141621.95821272343, "data_wait_s": 1.3183998817112297e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5606697878ced851bf92f926f4ef2561a76f1d13b17f8adb55de1493e67c13e3
-size 640263
+oid sha256:3dd66e6e9b60b45730c1873a98daa8976c14ba49e66fc266e8874430f51f328c
+size 642381