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": 76607.15001672419,
-      "required_pages_per_second_upper_estimate": 0.005299860391234613
+      "required_token_positions_per_second": 76705.35142333739,
+      "required_pages_per_second_upper_estimate": 0.00530665419240277
     },
     {
       "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": 9575.893752090524,
-      "required_pages_per_second_upper_estimate": 0.004009226746599708
+      "required_token_positions_per_second": 9588.168927917173,
+      "required_pages_per_second_upper_estimate": 0.004014366106383505
     },
     {
       "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": 19151.787504181048,
-      "required_pages_per_second_upper_estimate": 0.003988812270520169
+      "required_token_positions_per_second": 19176.337855834347,
+      "required_pages_per_second_upper_estimate": 0.00399392546133324
     },
     {
       "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": 28727.68125627157,
-      "required_pages_per_second_upper_estimate": 0.001722180626684047
+      "required_token_positions_per_second": 28764.50678375152,
+      "required_pages_per_second_upper_estimate": 0.0017243882608271451
     },
     {
       "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": 28727.68125627157,
-      "required_pages_per_second_upper_estimate": 0.0022334508579593747
+      "required_token_positions_per_second": 28764.50678375152,
+      "required_pages_per_second_upper_estimate": 0.0022363138807425637
     },
     {
       "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": 28727.68125627157,
-      "required_pages_per_second_upper_estimate": 0.3684878497745228
+      "required_token_positions_per_second": 28764.50678375152,
+      "required_pages_per_second_upper_estimate": 0.3689602081008648
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.12567245335151256,
-    "producer_p95_seconds": 0.22773471100299503,
-    "gpu_update_seconds": 0.9624166932708548,
-    "producer_batches_per_second": 7.957193269737057,
-    "gpu_batches_per_second": 1.039050971363989,
-    "producer_headroom": 7.658135634377439,
-    "sustainable": true,
-    "buffer_seconds": 60.63225167606385,
-    "buffer_drain_seconds": null,
+    "producer_batch_seconds": 1.0123949462810344,
+    "producer_p95_seconds": 0.2638948749990959,
+    "gpu_update_seconds": 0.9611845670727018,
+    "producer_batches_per_second": 0.9877568074331402,
+    "gpu_batches_per_second": 1.0403829131854572,
+    "producer_headroom": 0.949416599325737,
+    "sustainable": false,
+    "buffer_seconds": 60.55462772558022,
+    "buffer_drain_seconds": 1197.1244898208397,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315972.9663681307,
-    "upload_bytes_per_second": 3843073.938856984,
+    "generation_bytes_per_second": 248602.8116479169,
+    "upload_bytes_per_second": 3853812.398988319,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.80044310839597,
+    "estimated_full_shards_per_hour": 53.34437620237475,
     "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": 2.0304724819998228,
-      "bytes": 10476032,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.5808353909997095,
       "bytes": 7239609,
@@ -562,6 +556,12 @@
       "bytes": 1024497,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.8518995710001036,
+      "bytes": 10973445,
+      "pages": 22,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -187,3 +187,4 @@
 {"step": 818700, "loss": 0.565858781337738, "lr": 3e-05, "grad_norm": 0.27960762824885566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119834757197, "tokens_per_second": 153248.86429755297, "data_wait_s": 8.236998837674037e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 818800, "loss": 0.5714428424835205, "lr": 3e-05, "grad_norm": 0.2975220904316392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119849479219, "tokens_per_second": 152612.43304282203, "data_wait_s": 1.0782998288050294e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 818900, "loss": 0.5745950937271118, "lr": 3e-05, "grad_norm": 0.2938284744034004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119863964638, "tokens_per_second": 98068.47779072159, "data_wait_s": 1.6866000805748627e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 819000, "loss": 0.5096691250801086, "lr": 3e-05, "grad_norm": 0.2746947830905528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119878522906, "tokens_per_second": 150749.14686199775, "data_wait_s": 1.6498997865710407e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1d677f09f67e2188ce8a84ff01891749a6840299683ece1162bd35f91b1582c5
-size 402816
+oid sha256:a6eeae8f14e81362af9a8f51aa7ab22476c4405b7be8633aead32d0b63ceea8e
+size 404934