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": 76660.2071508742,
-      "required_pages_per_second_upper_estimate": 0.005303531006884622
+      "required_token_positions_per_second": 76449.47441857518,
+      "required_pages_per_second_upper_estimate": 0.0052889520274968715
     },
     {
       "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": 9582.525893859274,
-      "required_pages_per_second_upper_estimate": 0.004012003485863415
+      "required_token_positions_per_second": 9556.184302321897,
+      "required_pages_per_second_upper_estimate": 0.004000974811561697
     },
     {
       "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": 19165.05178771855,
-      "required_pages_per_second_upper_estimate": 0.0039915748709783505
+      "required_token_positions_per_second": 19112.368604643794,
+      "required_pages_per_second_upper_estimate": 0.003980602353292846
     },
     {
       "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": 28747.577681577826,
-      "required_pages_per_second_upper_estimate": 0.0017233733870010745
+      "required_token_positions_per_second": 28668.55290696569,
+      "required_pages_per_second_upper_estimate": 0.0017186359724269211
     },
     {
       "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": 28747.577681577826,
-      "required_pages_per_second_upper_estimate": 0.0022349977175118102
+      "required_token_positions_per_second": 28668.55290696569,
+      "required_pages_per_second_upper_estimate": 0.0022288538888789655
     },
     {
       "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": 28747.577681577826,
-      "required_pages_per_second_upper_estimate": 0.36874305975523436
+      "required_token_positions_per_second": 28668.55290696569,
+      "required_pages_per_second_upper_estimate": 0.36772941479670207
     }
   ],
   "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": 0.09983346711706531,
-    "producer_p95_seconds": 0.26933502800238784,
-    "gpu_update_seconds": 0.9617505970848298,
-    "producer_batches_per_second": 10.016681067756508,
-    "gpu_batches_per_second": 1.0397706048024387,
-    "producer_headroom": 9.633548997723132,
+    "producer_batch_seconds": 0.12694775575769768,
+    "producer_p95_seconds": 0.2854320159967756,
+    "gpu_update_seconds": 0.9644016595369302,
+    "producer_batches_per_second": 7.877256230576281,
+    "gpu_batches_per_second": 1.0369123591929141,
+    "producer_headroom": 7.596838981365389,
     "sustainable": true,
-    "buffer_seconds": 60.59028761634428,
+    "buffer_seconds": 60.7573045508266,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340706.53901590715,
-    "upload_bytes_per_second": 4929080.673748696,
+    "generation_bytes_per_second": 362868.32461869327,
+    "upload_bytes_per_second": 4916137.6064509135,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.10769203050529,
+    "estimated_full_shards_per_hour": 77.86309532089804,
     "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": 1.608177831996727,
-      "bytes": 5937731,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.4116140810001525,
       "bytes": 4972441,
@@ -562,6 +556,12 @@
       "bytes": 864410,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.8016076449966931,
+      "bytes": 5772859,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -300,3 +300,4 @@
 {"step": 790000, "loss": 0.5306972861289978, "lr": 3e-05, "grad_norm": 0.2637444456124192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115633471654, "tokens_per_second": 152705.55062787078, "data_wait_s": 8.73099997988902e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 790100, "loss": 0.5542400479316711, "lr": 3e-05, "grad_norm": 0.260805359336468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115648197682, "tokens_per_second": 145252.18146208223, "data_wait_s": 1.3286000466905534e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 790200, "loss": 0.5087929964065552, "lr": 3e-05, "grad_norm": 0.2584321778377075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115662904098, "tokens_per_second": 153086.08349205996, "data_wait_s": 9.760999091668054e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 790300, "loss": 0.5574849247932434, "lr": 3e-05, "grad_norm": 0.28539266337026326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115677722134, "tokens_per_second": 154485.94780169296, "data_wait_s": 8.484999852953479e-06, "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 sha256:1ab77a1e4680842645642ff94d45d6b4ad7c90e0704c97ca1896e7d15b456d12
-size 644499
+oid sha256:581789c8b53dbb1d0d9d09844780a0b1972e37faece70a9f9ac0e0914c65c33e
+size 646617