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": 76955.18599713506,
-      "required_pages_per_second_upper_estimate": 0.005323938327914171
+      "required_token_positions_per_second": 76890.99296579718,
+      "required_pages_per_second_upper_estimate": 0.005319497305057863
     },
     {
       "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": 9619.398249641883,
-      "required_pages_per_second_upper_estimate": 0.004027441171247246
+      "required_token_positions_per_second": 9611.374120724648,
+      "required_pages_per_second_upper_estimate": 0.004024081636032466
     },
     {
       "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": 19238.796499283766,
-      "required_pages_per_second_upper_estimate": 0.004006933949618561
+      "required_token_positions_per_second": 19222.748241449295,
+      "required_pages_per_second_upper_estimate": 0.00400359152073266
     },
     {
       "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": 28858.19474892565,
-      "required_pages_per_second_upper_estimate": 0.0017300047112860943
+      "required_token_positions_per_second": 28834.12236217394,
+      "required_pages_per_second_upper_estimate": 0.0017285616084567357
     },
     {
       "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": 28858.19474892565,
-      "required_pages_per_second_upper_estimate": 0.0022435977079450446
+      "required_token_positions_per_second": 28834.12236217394,
+      "required_pages_per_second_upper_estimate": 0.0022417261857583383
     },
     {
       "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": 28858.19474892565,
-      "required_pages_per_second_upper_estimate": 0.370161936723819
+      "required_token_positions_per_second": 28834.12236217394,
+      "required_pages_per_second_upper_estimate": 0.36985316199348317
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.09047782585938258,
-    "producer_p95_seconds": 0.13472667099995306,
-    "gpu_update_seconds": 0.9580640868406815,
-    "producer_batches_per_second": 11.052431803060392,
-    "gpu_batches_per_second": 1.0437715114628778,
-    "producer_headroom": 10.58893798276796,
+    "producer_batch_seconds": 0.07400748915624966,
+    "producer_p95_seconds": 0.12748016799992,
+    "gpu_update_seconds": 0.9588639339434185,
+    "producer_batches_per_second": 13.512146019286396,
+    "gpu_batches_per_second": 1.0429008377522404,
+    "producer_headroom": 12.956309488070856,
     "sustainable": true,
-    "buffer_seconds": 60.358037470962934,
+    "buffer_seconds": 60.408427838435365,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 64,
+    "upload_pages_per_commit": 68,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409352.2020105325,
-    "upload_bytes_per_second": 4474637.642745559,
+    "generation_bytes_per_second": 407706.1314415831,
+    "upload_bytes_per_second": 4713398.971472693,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.83745331990224,
+    "estimated_full_shards_per_hour": 87.48424489436741,
     "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": [
@@ -244,6 +244,12 @@
       "bytes": 5027383,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.0457134279999991,
+      "bytes": 7722680,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -10,3 +10,4 @@
 {"step": 558100, "loss": 0.597959041595459, "lr": 0.0003, "grad_norm": 0.1546086724713371, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81687934063, "tokens_per_second": 151342.1559679288, "data_wait_s": 1.2883999943369417e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 558200, "loss": 0.48611176013946533, "lr": 0.0003, "grad_norm": 0.16909264014352998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81702614267, "tokens_per_second": 152473.31878895906, "data_wait_s": 1.3196000054449541e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 558300, "loss": 0.5838193297386169, "lr": 0.0003, "grad_norm": 0.15871918018532938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81717098421, "tokens_per_second": 150401.55621710682, "data_wait_s": 1.6968999943856033e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 558400, "loss": 0.5963554978370667, "lr": 0.0003, "grad_norm": 0.17721178056017542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81731761513, "tokens_per_second": 152104.66692934177, "data_wait_s": 1.2724000043817796e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a296ebfe8acfdd4efc268a2e60c6d44c5f1c8470693fa48835a6e9243e552c7d
-size 25581
+oid sha256:2990b5b56e4f423494093de565a29a5194c73e273247a3fc196773f817037b8f
+size 27699