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": 76991.03476619159,
-      "required_pages_per_second_upper_estimate": 0.005326418428938106
+      "required_token_positions_per_second": 76437.48036187985,
+      "required_pages_per_second_upper_estimate": 0.0052881222508246424
     },
     {
       "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": 9623.879345773948,
-      "required_pages_per_second_upper_estimate": 0.004029317312621814
+      "required_token_positions_per_second": 9554.68504523498,
+      "required_pages_per_second_upper_estimate": 0.004000347103927453
     },
     {
       "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": 19247.758691547897,
-      "required_pages_per_second_upper_estimate": 0.004008800537918275
+      "required_token_positions_per_second": 19109.37009046996,
+      "required_pages_per_second_upper_estimate": 0.003979977841866574
     },
     {
       "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": 28871.638037321845,
-      "required_pages_per_second_upper_estimate": 0.0017308106158987334
+      "required_token_positions_per_second": 28664.055135704944,
+      "required_pages_per_second_upper_estimate": 0.0017183663379075384
     },
     {
       "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": 28871.638037321845,
-      "required_pages_per_second_upper_estimate": 0.0022446428644870725
+      "required_token_positions_per_second": 28664.055135704944,
+      "required_pages_per_second_upper_estimate": 0.0022285042069470472
     },
     {
       "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": 28871.638037321845,
-      "required_pages_per_second_upper_estimate": 0.37033437279308684
+      "required_token_positions_per_second": 28664.055135704944,
+      "required_pages_per_second_upper_estimate": 0.36767172221629973
     }
   ],
   "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": 0.12369976728868437,
-    "producer_p95_seconds": 0.28683057699890924,
-    "gpu_update_seconds": 0.9576179905088839,
-    "producer_batches_per_second": 8.084089581722896,
-    "gpu_batches_per_second": 1.04425774151193,
-    "producer_headroom": 7.741469620343283,
+    "producer_batch_seconds": 0.387383661484364,
+    "producer_p95_seconds": 0.29978638700049487,
+    "gpu_update_seconds": 0.9645529869763853,
+    "producer_batches_per_second": 2.5814201770106484,
+    "gpu_batches_per_second": 1.0367496793874762,
+    "producer_headroom": 2.48991654237673,
     "sustainable": true,
-    "buffer_seconds": 60.32993340205968,
+    "buffer_seconds": 60.76683817951227,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 372123.489399354,
-    "upload_bytes_per_second": 3202257.5440629544,
+    "generation_bytes_per_second": 317952.2221512336,
+    "upload_bytes_per_second": 3241180.539628136,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.8490382336184,
+    "estimated_full_shards_per_hour": 68.2251453247333,
     "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": 1.75038528100049,
-      "bytes": 5975710,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 2.091957274999004,
       "bytes": 2323345,
@@ -562,6 +556,12 @@
       "bytes": 7499768,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.6402012090002245,
+      "bytes": 10865294,
+      "pages": 22,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -285,3 +285,4 @@
 {"step": 828500, "loss": 0.518890917301178, "lr": 3e-05, "grad_norm": 0.27430080897680775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121269045734, "tokens_per_second": 152040.20547331744, "data_wait_s": 9.425999451195821e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 828600, "loss": 0.48698189854621887, "lr": 3e-05, "grad_norm": 0.2820801254666777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121283689201, "tokens_per_second": 151651.99115701005, "data_wait_s": 1.44199984788429e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 828700, "loss": 0.5343151688575745, "lr": 3e-05, "grad_norm": 0.28000622198113484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121298329142, "tokens_per_second": 128172.62825832651, "data_wait_s": 2.0457999198697507e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 828800, "loss": 0.5465978980064392, "lr": 3e-05, "grad_norm": 0.2814293850838141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121313035457, "tokens_per_second": 152555.76097589397, "data_wait_s": 7.995000487426296e-06, "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:e861bc5152dff24ccad586da90c85b1c40b4344ae405712fd4a531106595c0a5
-size 611946
+oid sha256:c27ba905577ac93a6b8e9ae860bf4975385f08c8c8201f4c12f2d38afbc5b1d3
+size 614064