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": 76839.76630804983,
-      "required_pages_per_second_upper_estimate": 0.0053159533260126795
+      "required_token_positions_per_second": 77125.41820844865,
+      "required_pages_per_second_upper_estimate": 0.005335715387285995
     },
     {
       "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": 9604.970788506229,
-      "required_pages_per_second_upper_estimate": 0.004021400694549396
+      "required_token_positions_per_second": 9640.677276056082,
+      "required_pages_per_second_upper_estimate": 0.00403635025525027
     },
     {
       "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": 19209.941577012458,
-      "required_pages_per_second_upper_estimate": 0.004000924230264919
+      "required_token_positions_per_second": 19281.354552112163,
+      "required_pages_per_second_upper_estimate": 0.004015797669691389
     },
     {
       "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": 28814.912365518685,
-      "required_pages_per_second_upper_estimate": 0.0017274099984892195
+      "required_token_positions_per_second": 28922.031828168245,
+      "required_pages_per_second_upper_estimate": 0.0017338316467131118
     },
     {
       "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": 28814.912365518685,
-      "required_pages_per_second_upper_estimate": 0.002240232693014238
+      "required_token_positions_per_second": 28922.031828168245,
+      "required_pages_per_second_upper_estimate": 0.0022485607600665194
     },
     {
       "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": 28814.912365518685,
-      "required_pages_per_second_upper_estimate": 0.36960675678247695
+      "required_token_positions_per_second": 28922.031828168245,
+      "required_pages_per_second_upper_estimate": 0.37098077023342757
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0537852284999758,
-    "producer_p95_seconds": 0.17454174200065609,
-    "gpu_update_seconds": 0.9595031783988672,
-    "producer_batches_per_second": 18.592465401545144,
-    "gpu_batches_per_second": 1.042206031738957,
-    "producer_headroom": 17.83952964705354,
+    "producer_batch_seconds": 0.22782729545310332,
+    "producer_p95_seconds": 0.18474830699960876,
+    "gpu_update_seconds": 0.9559494355120853,
+    "producer_batches_per_second": 4.389289694244924,
+    "gpu_batches_per_second": 1.0460804335998353,
+    "producer_headroom": 4.195939005512449,
     "sustainable": true,
-    "buffer_seconds": 60.448700239128634,
+    "buffer_seconds": 60.22481443726138,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 451372.6946431653,
-    "upload_bytes_per_second": 4386396.1666391175,
+    "generation_bytes_per_second": 438039.28333854047,
+    "upload_bytes_per_second": 4331069.976195898,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.85407285186022,
+    "estimated_full_shards_per_hour": 93.9930331718174,
     "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": [
@@ -460,6 +460,12 @@
       "bytes": 7567101,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.065770677000728,
+      "bytes": 897206,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -55,3 +55,4 @@
 {"step": 848000, "loss": 0.5419650077819824, "lr": 3e-05, "grad_norm": 0.29961385374063215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124123678252, "tokens_per_second": 150977.06386969503, "data_wait_s": 1.1256000107096042e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 848100, "loss": 0.5064298510551453, "lr": 3e-05, "grad_norm": 0.2911187229154403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124138304065, "tokens_per_second": 152643.5207272398, "data_wait_s": 1.2854000488005113e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 848200, "loss": 0.5381395816802979, "lr": 3e-05, "grad_norm": 0.28488509242539006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124152970850, "tokens_per_second": 152565.69994112532, "data_wait_s": 1.4092000128584914e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 848300, "loss": 0.5027939081192017, "lr": 3e-05, "grad_norm": 0.2984285320279911, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124167642263, "tokens_per_second": 152824.67051482873, "data_wait_s": 7.983000614331104e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:08ee548cf4c3d554def585d73e663422e41de9013764fb7a946d2f69e286b8cc
-size 121674
+oid sha256:6efab3e62975efdd2939aded1a94ad98a60c809f22c0e16aa2f70663256b2b36
+size 123792