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": 76880.64195959603,
-      "required_pages_per_second_upper_estimate": 0.0053187811984312236
+      "required_token_positions_per_second": 76592.02181592688,
+      "required_pages_per_second_upper_estimate": 0.005298813787201193
     },
     {
       "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": 9610.080244949504,
-      "required_pages_per_second_upper_estimate": 0.004023539917264612
+      "required_token_positions_per_second": 9574.00272699086,
+      "required_pages_per_second_upper_estimate": 0.004008435013879573
     },
     {
       "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": 19220.160489899008,
-      "required_pages_per_second_upper_estimate": 0.004003052560328332
+      "required_token_positions_per_second": 19148.00545398172,
+      "required_pages_per_second_upper_estimate": 0.00398802456920302
     },
     {
       "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": 28830.240734848514,
-      "required_pages_per_second_upper_estimate": 0.001728328910825478
+      "required_token_positions_per_second": 28722.00818097258,
+      "required_pages_per_second_upper_estimate": 0.0017218405344821568
     },
     {
       "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": 28830.240734848514,
-      "required_pages_per_second_upper_estimate": 0.0022414244063072605
+      "required_token_positions_per_second": 28722.00818097258,
+      "required_pages_per_second_upper_estimate": 0.0022330098013080995
     },
     {
       "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": 28830.240734848514,
-      "required_pages_per_second_upper_estimate": 0.3698033726459193
+      "required_token_positions_per_second": 28722.00818097258,
+      "required_pages_per_second_upper_estimate": 0.368415081655861
     }
   ],
   "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": 0.0643160757656176,
-    "producer_p95_seconds": 0.222205889000179,
-    "gpu_update_seconds": 0.9589930328462544,
-    "producer_batches_per_second": 15.548212295231247,
-    "gpu_batches_per_second": 1.042760443245389,
-    "producer_headroom": 14.910627264341235,
+    "producer_batch_seconds": 0.08020106483607492,
+    "producer_p95_seconds": 0.2137567339996167,
+    "gpu_update_seconds": 0.962606786607488,
+    "producer_batches_per_second": 12.468662380529816,
+    "gpu_batches_per_second": 1.038845782008557,
+    "producer_headroom": 12.002419027415478,
     "sustainable": true,
-    "buffer_seconds": 60.41656106931403,
+    "buffer_seconds": 60.644227556271744,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358111.89799081435,
-    "upload_bytes_per_second": 5133620.597961741,
+    "generation_bytes_per_second": 357194.90096659993,
+    "upload_bytes_per_second": 5098008.182377474,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.8424768905003,
+    "estimated_full_shards_per_hour": 76.64571067570208,
     "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": 1.678345239999544,
-      "bytes": 6141170,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0750775129999965,
       "bytes": 1235188,
@@ -562,6 +556,12 @@
       "bytes": 13497203,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 2.2471259019985155,
+      "bytes": 6049971,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -176,3 +176,4 @@
 {"step": 777600, "loss": 0.48386308550834656, "lr": 3e-05, "grad_norm": 0.2531764229079415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113818792820, "tokens_per_second": 148077.30706260807, "data_wait_s": 2.9916001949459314e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 777700, "loss": 0.48820388317108154, "lr": 3e-05, "grad_norm": 0.25169611315464974, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113833433442, "tokens_per_second": 151999.86615718072, "data_wait_s": 9.237999620381743e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 777800, "loss": 0.5450183153152466, "lr": 3e-05, "grad_norm": 0.2716514554770019, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113847999148, "tokens_per_second": 150377.72094061942, "data_wait_s": 1.0497999028302729e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 777900, "loss": 0.549085795879364, "lr": 3e-05, "grad_norm": 0.24472834686400965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113862549260, "tokens_per_second": 151264.83046431563, "data_wait_s": 1.7368998669553548e-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 sha256:3e84c2f4e549a1aa97ee22a421b1e80e828e1f933c2266f64b1dad768642f938
-size 379518
+oid sha256:8206244439c5aa4d5e582cd40c5bbbed9165795d5b9c63ffeb3d782ebf37ac18
+size 381636