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": 75875.2931364603,
-      "required_pages_per_second_upper_estimate": 0.0052492288341160365
+      "required_token_positions_per_second": 75925.02189202065,
+      "required_pages_per_second_upper_estimate": 0.005252669184812312
     },
     {
       "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": 9484.411642057537,
-      "required_pages_per_second_upper_estimate": 0.003970925097492588
+      "required_token_positions_per_second": 9490.627736502582,
+      "required_pages_per_second_upper_estimate": 0.003973527646429916
     },
     {
       "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": 18968.823284115075,
-      "required_pages_per_second_upper_estimate": 0.003950705648571384
+      "required_token_positions_per_second": 18981.255473005163,
+      "required_pages_per_second_upper_estimate": 0.00395329494565832
     },
     {
       "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": 28453.234926172612,
-      "required_pages_per_second_upper_estimate": 0.0017057279882498962
+      "required_token_positions_per_second": 28471.88320950775,
+      "required_pages_per_second_upper_estimate": 0.0017068459243615576
     },
     {
       "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": 28453.234926172612,
-      "required_pages_per_second_upper_estimate": 0.00221211386295601
+      "required_token_positions_per_second": 28471.88320950775,
+      "required_pages_per_second_upper_estimate": 0.002213563684960187
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28453.234926172612,
-      "required_pages_per_second_upper_estimate": 0.3649675469295239
+      "required_token_positions_per_second": 28471.88320950775,
+      "required_pages_per_second_upper_estimate": 0.3652067470851804
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.14460936013267656,
-    "producer_p95_seconds": 0.3409924560037325,
-    "gpu_update_seconds": 0.9716997055603009,
-    "producer_batches_per_second": 6.915181694203733,
-    "gpu_batches_per_second": 1.0291245271329794,
-    "producer_headroom": 6.71948001615375,
+    "producer_batch_seconds": 0.14120304146075568,
+    "producer_p95_seconds": 0.3439147459939704,
+    "gpu_update_seconds": 0.9710632695615784,
+    "producer_batches_per_second": 7.0820004275752675,
+    "gpu_batches_per_second": 1.0297990165475892,
+    "producer_headroom": 6.8770704902377355,
     "sustainable": true,
-    "buffer_seconds": 60.24538174473866,
+    "buffer_seconds": 60.20592271281786,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 423409.24078064074,
-    "upload_bytes_per_second": 4869530.744304102,
+    "generation_bytes_per_second": 407412.26425123675,
+    "upload_bytes_per_second": 5607008.950839485,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.85376660885255,
+    "estimated_full_shards_per_hour": 87.42118783619715,
     "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": 13.443922794998798,
-      "bytes": 11916771,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.9173718489983003,
       "bytes": 2617778,
@@ -562,6 +556,12 @@
       "bytes": 974526,
       "pages": 28,
       "shards": 1
+    },
+    {
+      "seconds": 1.0208734549960354,
+      "bytes": 7425154,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -399,3 +399,4 @@
 {"step": 597000, "loss": 0.5525102615356445, "lr": 0.0003, "grad_norm": 0.17048468384412183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87382905317, "tokens_per_second": 153100.68204656694, "data_wait_s": 9.212999430019408e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 597100, "loss": 0.5419397354125977, "lr": 0.0003, "grad_norm": 0.158056373317174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87397593914, "tokens_per_second": 152742.58909844302, "data_wait_s": 1.333500404143706e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 597200, "loss": 0.5880463719367981, "lr": 0.0003, "grad_norm": 0.1496795230774184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87412159123, "tokens_per_second": 151386.53012126498, "data_wait_s": 9.612995199859142e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 597300, "loss": 0.5437367558479309, "lr": 0.0003, "grad_norm": 0.1642466124584293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87426797447, "tokens_per_second": 152611.62927840132, "data_wait_s": 1.680700370343402e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa68d03c443307fcc4227e6d595197a352c01778e89ab2c2c91c9a7ccc8255bf
-size 855747
+oid sha256:cc10b454afb18014453d950d8c21ea6725d969905500939ac1e2073430ac00f3
+size 857865