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": 76661.70461350658,
-      "required_pages_per_second_upper_estimate": 0.005303634604824123
+      "required_token_positions_per_second": 76672.7035295612,
+      "required_pages_per_second_upper_estimate": 0.005304395535357783
     },
     {
       "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": 9582.713076688322,
-      "required_pages_per_second_upper_estimate": 0.004012081855405115
+      "required_token_positions_per_second": 9584.08794119515,
+      "required_pages_per_second_upper_estimate": 0.004012657482463688
     },
     {
       "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": 19165.426153376644,
-      "required_pages_per_second_upper_estimate": 0.003991652841472245
+      "required_token_positions_per_second": 19168.1758823903,
+      "required_pages_per_second_upper_estimate": 0.00399222553751056
     },
     {
       "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": 28748.139230064964,
-      "required_pages_per_second_upper_estimate": 0.0017234070509754434
+      "required_token_positions_per_second": 28752.26382358545,
+      "required_pages_per_second_upper_estimate": 0.0017236543140591069
     },
     {
       "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": 28748.139230064964,
-      "required_pages_per_second_upper_estimate": 0.0022350413754366944
+      "required_token_positions_per_second": 28752.26382358545,
+      "required_pages_per_second_upper_estimate": 0.0022353620444407426
     },
     {
       "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": 28748.139230064964,
-      "required_pages_per_second_upper_estimate": 0.3687502626962836
+      "required_token_positions_per_second": 28752.26382358545,
+      "required_pages_per_second_upper_estimate": 0.3688031685533209
     }
   ],
   "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": 0.36552698448480214,
-    "producer_p95_seconds": 0.30872122300206684,
-    "gpu_update_seconds": 0.9617318108396236,
-    "producer_batches_per_second": 2.7357761326690175,
-    "gpu_batches_per_second": 1.039790915439271,
-    "producer_headroom": 2.6310829341235964,
+    "producer_batch_seconds": 0.17591175734384024,
+    "producer_p95_seconds": 0.3247485059982864,
+    "gpu_update_seconds": 0.9615938476927989,
+    "producer_batches_per_second": 5.684668353607441,
+    "gpu_batches_per_second": 1.039940097785932,
+    "producer_headroom": 5.466342115002868,
     "sustainable": true,
-    "buffer_seconds": 60.589104082896284,
+    "buffer_seconds": 60.58041240464633,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294840.11493745906,
-    "upload_bytes_per_second": 4661587.228182904,
+    "generation_bytes_per_second": 304457.1096605911,
+    "upload_bytes_per_second": 4623001.23625818,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.26582513897733,
+    "estimated_full_shards_per_hour": 65.32940833438205,
     "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": 1.344746707000013,
-      "bytes": 12817447,
-      "pages": 46,
-      "shards": 4
-    },
     {
       "seconds": 1.0572433240013197,
       "bytes": 864410,
@@ -562,6 +556,12 @@
       "bytes": 6924568,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.8365363409975544,
+      "bytes": 11520166,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -377,3 +377,4 @@
 {"step": 797700, "loss": 0.5641277432441711, "lr": 3e-05, "grad_norm": 0.26915051255483546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116760208303, "tokens_per_second": 152140.50883850653, "data_wait_s": 9.659001079853624e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 797800, "loss": 0.50200355052948, "lr": 3e-05, "grad_norm": 0.2549789617319407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116774866499, "tokens_per_second": 152610.73590216186, "data_wait_s": 6.841000868007541e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 797900, "loss": 0.5144842863082886, "lr": 3e-05, "grad_norm": 0.27186701635641053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116789626837, "tokens_per_second": 153466.22160680825, "data_wait_s": 8.831004379317164e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 798000, "loss": 0.5630819797515869, "lr": 3e-05, "grad_norm": 0.2660672224736182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116804324569, "tokens_per_second": 152752.5646681396, "data_wait_s": 1.0050003766082227e-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 sha256:4332d91593ca44b81c5cb5174cb9c346e8ed13f4ce289523711d63a3ca6be80b
-size 808368
+oid sha256:702e7105856dde19e1b1908ec198d3f9f7bfefa084cc9db97b838d44660c2df5
+size 810486