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": 76776.10637041909,
-      "required_pages_per_second_upper_estimate": 0.0053115491838159785
+      "required_token_positions_per_second": 76653.32013675013,
+      "required_pages_per_second_upper_estimate": 0.0053030545472674455
     },
     {
       "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": 9597.013296302386,
-      "required_pages_per_second_upper_estimate": 0.00401806905873498
+      "required_token_positions_per_second": 9581.665017093766,
+      "required_pages_per_second_upper_estimate": 0.004011643054738848
     },
     {
       "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": 19194.026592604772,
-      "required_pages_per_second_upper_estimate": 0.003997609558719161
+      "required_token_positions_per_second": 19163.330034187533,
+      "required_pages_per_second_upper_estimate": 0.003991216275123534
     },
     {
       "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": 28791.039888907162,
-      "required_pages_per_second_upper_estimate": 0.00172597887996752
+      "required_token_positions_per_second": 28744.9950512813,
+      "required_pages_per_second_upper_estimate": 0.001723218562258246
     },
     {
       "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": 28791.039888907162,
-      "required_pages_per_second_upper_estimate": 0.0022383767129616193
+      "required_token_positions_per_second": 28744.9950512813,
+      "required_pages_per_second_upper_estimate": 0.0022347969293660453
     },
     {
       "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": 28791.039888907162,
-      "required_pages_per_second_upper_estimate": 0.36930054628477266
+      "required_token_positions_per_second": 28744.9950512813,
+      "required_pages_per_second_upper_estimate": 0.3687099325468029
     }
   ],
   "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": 0.08787092653903983,
-    "producer_p95_seconds": 0.3050311380029598,
-    "gpu_update_seconds": 0.960298763319502,
-    "producer_batches_per_second": 11.380328390593604,
-    "gpu_batches_per_second": 1.0413425885744778,
-    "producer_headroom": 10.928515279656857,
+    "producer_batch_seconds": 0.1261778453828697,
+    "producer_p95_seconds": 0.30013623900231323,
+    "gpu_update_seconds": 0.9618370067789452,
+    "producer_batches_per_second": 7.925321572623424,
+    "gpu_batches_per_second": 1.0396771936950702,
+    "producer_headroom": 7.622867579172717,
     "sustainable": true,
-    "buffer_seconds": 60.49882208912863,
+    "buffer_seconds": 60.59573142707355,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358330.5274270928,
-    "upload_bytes_per_second": 4864238.359603431,
+    "generation_bytes_per_second": 435560.0160494135,
+    "upload_bytes_per_second": 4922534.627713278,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.88938967809284,
+    "estimated_full_shards_per_hour": 93.46104012595943,
     "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": 1.0435017019990482,
-      "bytes": 1045686,
-      "pages": 30,
-      "shards": 1
-    },
     {
       "seconds": 1.3271499310030777,
       "bytes": 10074712,
@@ -562,6 +556,12 @@
       "bytes": 5901441,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0388182249989768,
+      "bytes": 6175327,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -320,3 +320,4 @@
 {"step": 792000, "loss": 0.5425643920898438, "lr": 3e-05, "grad_norm": 0.24929407403225676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115926006994, "tokens_per_second": 151874.6013311929, "data_wait_s": 1.0699997801566496e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 792100, "loss": 0.513585090637207, "lr": 3e-05, "grad_norm": 0.2581473778282717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115940671783, "tokens_per_second": 152120.09961518695, "data_wait_s": 1.7366000975016505e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 792200, "loss": 0.5342683792114258, "lr": 3e-05, "grad_norm": 0.2635454912110246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115955309528, "tokens_per_second": 151981.81773728828, "data_wait_s": 9.202998626278713e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 792300, "loss": 0.5348979234695435, "lr": 3e-05, "grad_norm": 0.2578812605807442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115969991276, "tokens_per_second": 152120.4504045221, "data_wait_s": 1.0077998013002798e-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:f7989008a5819721eac87a8cd2697e9775c6067c499403e21e93048b26694c02
-size 686859
+oid sha256:64c5197992b43f34b4221d93430d42cf970be6fc5814a781f279e0036b2885dd
+size 688977