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": 76105.306219887,
-      "required_pages_per_second_upper_estimate": 0.00526514167293138
+      "required_token_positions_per_second": 75682.50741479248,
+      "required_pages_per_second_upper_estimate": 0.005235891470566553
     },
     {
       "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": 9513.163277485875,
-      "required_pages_per_second_upper_estimate": 0.003982962806844011
+      "required_token_positions_per_second": 9460.31342684906,
+      "required_pages_per_second_upper_estimate": 0.003960835678012756
     },
     {
       "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": 19026.32655497175,
-      "required_pages_per_second_upper_estimate": 0.003962682063427599
+      "required_token_positions_per_second": 18920.62685369812,
+      "required_pages_per_second_upper_estimate": 0.003940667603140874
     },
     {
       "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": 28539.489832457624,
-      "required_pages_per_second_upper_estimate": 0.0017108988381780666
+      "required_token_positions_per_second": 28380.94028054718,
+      "required_pages_per_second_upper_estimate": 0.0017013940346321838
     },
     {
       "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": 28539.489832457624,
-      "required_pages_per_second_upper_estimate": 0.0022188198025244327
+      "required_token_positions_per_second": 28380.94028054718,
+      "required_pages_per_second_upper_estimate": 0.0022064932722491723
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28539.489832457624,
-      "required_pages_per_second_upper_estimate": 0.3660739322540453
+      "required_token_positions_per_second": 28380.94028054718,
+      "required_pages_per_second_upper_estimate": 0.36404022883938353
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.061444202125017,
-    "producer_p95_seconds": 0.20464719800111197,
-    "gpu_update_seconds": 0.9687629373303042,
-    "producer_batches_per_second": 16.274928559823387,
-    "gpu_batches_per_second": 1.0322442792410889,
-    "producer_headroom": 15.76654759645536,
+    "producer_batch_seconds": 0.08980410083589163,
+    "producer_p95_seconds": 0.24404907700045442,
+    "gpu_update_seconds": 0.9741749119901587,
+    "producer_batches_per_second": 11.135348950571911,
+    "gpu_batches_per_second": 1.0265097034341428,
+    "producer_headroom": 10.847777583903097,
     "sustainable": true,
-    "buffer_seconds": 60.06330211447886,
+    "buffer_seconds": 60.39884454338984,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 411367.8092132365,
-    "upload_bytes_per_second": 4885662.4577949,
+    "generation_bytes_per_second": 413953.8999537634,
+    "upload_bytes_per_second": 4940252.442675524,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.26995570466825,
+    "estimated_full_shards_per_hour": 88.82487057647397,
     "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": 1.064991151999493,
-      "bytes": 1003408,
-      "pages": 31,
-      "shards": 1
-    },
     {
       "seconds": 1.010816134000379,
       "bytes": 5780671,
@@ -562,6 +556,12 @@
       "bytes": 8102539,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.1965843840007437,
+      "bytes": 6380308,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -118,3 +118,4 @@
 {"step": 731800, "loss": 0.5290740132331848, "lr": 5e-05, "grad_norm": 0.22209223910479947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107114684526, "tokens_per_second": 149706.74671501736, "data_wait_s": 2.137299998139497e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 731900, "loss": 0.5578677654266357, "lr": 5e-05, "grad_norm": 0.2192783896648435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107129246802, "tokens_per_second": 142523.7032786384, "data_wait_s": 1.9320001229061745e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 732000, "loss": 0.5463915467262268, "lr": 5e-05, "grad_norm": 0.27102495449617353, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107143868285, "tokens_per_second": 148884.5458037416, "data_wait_s": 2.929699985543266e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 732100, "loss": 0.52305668592453, "lr": 5e-05, "grad_norm": 0.2194483695417114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107158494165, "tokens_per_second": 150453.51858312136, "data_wait_s": 2.0274999769753776e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:994ffb24accd8d078dccee8b066c0e8c7ffc9f139065a15d03740a772a414191
-size 255891
+oid sha256:084465f604c1daf9ebc9505cc2d4e52b662f6896a5a5a58e8c2d8b113c553bcf
+size 258009