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": 76681.25282947421,
-      "required_pages_per_second_upper_estimate": 0.005304986995762874
+      "required_token_positions_per_second": 77315.4303183706,
+      "required_pages_per_second_upper_estimate": 0.005348860865939234
     },
     {
       "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": 9585.156603684276,
-      "required_pages_per_second_upper_estimate": 0.004013104909131674
+      "required_token_positions_per_second": 9664.428789796326,
+      "required_pages_per_second_upper_estimate": 0.004046294518065303
     },
     {
       "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": 19170.313207368552,
-      "required_pages_per_second_upper_estimate": 0.003992670685938479
+      "required_token_positions_per_second": 19328.85757959265,
+      "required_pages_per_second_upper_estimate": 0.004025691297576497
     },
     {
       "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": 28755.46981105283,
-      "required_pages_per_second_upper_estimate": 0.001723846508112517
+      "required_token_positions_per_second": 28993.286369388978,
+      "required_pages_per_second_upper_estimate": 0.0017381032476599101
     },
     {
       "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": 28755.46981105283,
-      "required_pages_per_second_upper_estimate": 0.0022356112958646828
+      "required_token_positions_per_second": 28993.286369388978,
+      "required_pages_per_second_upper_estimate": 0.0022541004872308275
     },
     {
       "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": 28755.46981105283,
-      "required_pages_per_second_upper_estimate": 0.3688442915182313
+      "required_token_positions_per_second": 28993.286369388978,
+      "required_pages_per_second_upper_estimate": 0.3718947469810415
     }
   ],
   "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": 0.3163721971327931,
-    "producer_p95_seconds": 0.23021320200132322,
-    "gpu_update_seconds": 0.9614866382524848,
-    "producer_batches_per_second": 3.160834008369778,
-    "gpu_batches_per_second": 1.0400560550872695,
-    "producer_headroom": 3.0390996647815847,
+    "producer_batch_seconds": 0.11076392227344911,
+    "producer_p95_seconds": 0.2394828310007142,
+    "gpu_update_seconds": 0.9536000730565912,
+    "producer_batches_per_second": 9.028210445015155,
+    "gpu_batches_per_second": 1.0486576377817194,
+    "producer_headroom": 8.609302139936732,
     "sustainable": true,
-    "buffer_seconds": 60.57365820990654,
+    "buffer_seconds": 60.07680460256525,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430696.48260590306,
-    "upload_bytes_per_second": 5733144.966716273,
+    "generation_bytes_per_second": 446005.5643779642,
+    "upload_bytes_per_second": 5683393.084092235,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.41743906624622,
+    "estimated_full_shards_per_hour": 95.70241163734622,
     "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": 1.260675972000172,
-      "bytes": 10735591,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 1.1190538959999685,
       "bytes": 7646532,
@@ -562,6 +556,12 @@
       "bytes": 15814359,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 1.3585308149995399,
+      "bytes": 7486305,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -145,3 +145,4 @@
 {"step": 774500, "loss": 0.5222188234329224, "lr": 3e-05, "grad_norm": 0.24253370722559092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113364531091, "tokens_per_second": 151652.01044864577, "data_wait_s": 1.0929999916697852e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 774600, "loss": 0.5660772323608398, "lr": 3e-05, "grad_norm": 0.2558079389723242, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113379147847, "tokens_per_second": 151113.6816932582, "data_wait_s": 9.150999176199548e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 774700, "loss": 0.5596816539764404, "lr": 3e-05, "grad_norm": 0.2507133016162925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113393855270, "tokens_per_second": 153358.90246631383, "data_wait_s": 1.1791000360972248e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 774800, "loss": 0.5517721772193909, "lr": 3e-05, "grad_norm": 0.24762378204733163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113408521251, "tokens_per_second": 152852.20293922134, "data_wait_s": 1.1960999472648837e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5027eed000fb27555b91f7a576097a35750f5eba4056fd7585504bd8c3be5c55
-size 313077
+oid sha256:4f7dc55d7b7f647bcf875fe7a47a635d670d32ed9c8096192e338f283e51530a
+size 315195