Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75314.04190882544,
-      "required_pages_per_second_upper_estimate": 0.006528030270962897
+      "required_token_positions_per_second": 75311.42165884901,
+      "required_pages_per_second_upper_estimate": 0.006527803154335902
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9414.25523860318,
-      "required_pages_per_second_upper_estimate": 0.003755487170337953
+      "required_token_positions_per_second": 9413.927707356126,
+      "required_pages_per_second_upper_estimate": 0.0037553565132264744
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18828.51047720636,
-      "required_pages_per_second_upper_estimate": 0.004355644528434999
+      "required_token_positions_per_second": 18827.855414712252,
+      "required_pages_per_second_upper_estimate": 0.004355492991255694
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28242.76571580954,
-      "required_pages_per_second_upper_estimate": 0.0017198362537529156
+      "required_token_positions_per_second": 28241.783122068377,
+      "required_pages_per_second_upper_estimate": 0.0017197764189493497
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28242.76571580954,
-      "required_pages_per_second_upper_estimate": 0.002174206285782325
+      "required_token_positions_per_second": 28241.783122068377,
+      "required_pages_per_second_upper_estimate": 0.0021741306429961366
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28242.76571580954,
-      "required_pages_per_second_upper_estimate": 0.3506806277339551
+      "required_token_positions_per_second": 28241.783122068377,
+      "required_pages_per_second_upper_estimate": 0.35066842720821956
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.1589884608906118,
+    "producer_batch_seconds": 0.1816695902267611,
     "producer_p95_seconds": 0.4476383849978447,
-    "gpu_update_seconds": 0.9789409535243708,
-    "producer_batches_per_second": 6.289764643284559,
-    "gpu_batches_per_second": 1.0215120701609353,
-    "producer_headroom": 6.1573081973408605,
+    "gpu_update_seconds": 0.9789750130329274,
+    "producer_batches_per_second": 5.504498572115421,
+    "gpu_batches_per_second": 1.0214765307461073,
+    "producer_headroom": 5.388766561376424,
     "sustainable": true,
-    "buffer_seconds": 60.69433911851099,
+    "buffer_seconds": 60.696450808041504,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 378010.75955980347,
-    "upload_bytes_per_second": 5555677.627573773,
+    "generation_bytes_per_second": 381799.3975817243,
+    "upload_bytes_per_second": 5524837.628749407,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.11230936141565,
+    "estimated_full_shards_per_hour": 81.92526288594051,
     "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": 1.0165252450024127,
-      "bytes": 7299775,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.2853216939984122,
       "bytes": 5894461,
@@ -562,6 +556,12 @@
       "bytes": 2686484,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2383149749948643,
+      "bytes": 6204112,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -418,3 +418,4 @@
 {"step": 236800, "loss": 0.6036998704075813, "lr": 0.0003, "grad_norm": 0.15617763721920008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34660451462, "tokens_per_second": 148662.44309618772, "data_wait_s": 1.107400021282956e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 236900, "loss": 0.6146441139280796, "lr": 0.0003, "grad_norm": 0.1577055018214105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34674944830, "tokens_per_second": 147353.71542987463, "data_wait_s": 1.2205004168208688e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 237000, "loss": 0.5927971564233303, "lr": 0.0003, "grad_norm": 0.1361256721305232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34689567793, "tokens_per_second": 147853.5303958649, "data_wait_s": 1.569400046719238e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 237100, "loss": 0.6155394297093153, "lr": 0.0003, "grad_norm": 0.14689367528609554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34704172945, "tokens_per_second": 147793.50264406175, "data_wait_s": 1.1158001143485308e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a0add15001f6e55e15ad4cf94c8b7b27daf72aff6ae863bd77a41e9778f8c828
-size 895989
+oid sha256:89cd8903e665660a43ad94917980205556771aea2a4cc3f9f150906823f0ad53
+size 898107