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": 76238.26196636575,
-      "required_pages_per_second_upper_estimate": 0.005274339859972614
+      "required_token_positions_per_second": 76259.65606142364,
+      "required_pages_per_second_upper_estimate": 0.005275819953109859
     },
     {
       "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": 9529.78274579572,
-      "required_pages_per_second_upper_estimate": 0.00398992103117138
+      "required_token_positions_per_second": 9532.457007677955,
+      "required_pages_per_second_upper_estimate": 0.003991040688776534
     },
     {
       "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": 19059.56549159144,
-      "required_pages_per_second_upper_estimate": 0.003969604857355775
+      "required_token_positions_per_second": 19064.91401535591,
+      "required_pages_per_second_upper_estimate": 0.003970718813805857
     },
     {
       "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": 28589.348237387156,
-      "required_pages_per_second_upper_estimate": 0.0017138877734242158
+      "required_token_positions_per_second": 28597.371023033862,
+      "required_pages_per_second_upper_estimate": 0.0017143687271736513
     },
     {
       "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": 28589.348237387156,
-      "required_pages_per_second_upper_estimate": 0.002222696074203757
+      "required_token_positions_per_second": 28597.371023033862,
+      "required_pages_per_second_upper_estimate": 0.002223319810499284
     },
     {
       "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": 28589.348237387156,
-      "required_pages_per_second_upper_estimate": 0.366713462338697
+      "required_token_positions_per_second": 28597.371023033862,
+      "required_pages_per_second_upper_estimate": 0.36681637001877687
     }
   ],
   "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": 0.12863128952349712,
-    "producer_p95_seconds": 0.29422660500131315,
-    "gpu_update_seconds": 0.9670734628306032,
-    "producer_batches_per_second": 7.774158244890561,
-    "gpu_batches_per_second": 1.0340476069656812,
-    "producer_headroom": 7.5181821344794,
+    "producer_batch_seconds": 0.10395503071109147,
+    "producer_p95_seconds": 0.30262544600191177,
+    "gpu_update_seconds": 0.9668021573637245,
+    "producer_batches_per_second": 9.619544077469117,
+    "gpu_batches_per_second": 1.0343377829511669,
+    "producer_headroom": 9.30019596695258,
     "sustainable": true,
-    "buffer_seconds": 60.925628158328,
+    "buffer_seconds": 60.90853591391465,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380971.5812132714,
-    "upload_bytes_per_second": 4857321.401894118,
+    "generation_bytes_per_second": 396831.3862172863,
+    "upload_bytes_per_second": 4771805.45814758,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.74763276384932,
+    "estimated_full_shards_per_hour": 85.15077772034589,
     "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": 1.142808556996897,
-      "bytes": 7320475,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.333093295001163,
       "bytes": 10300175,
@@ -562,6 +556,12 @@
       "bytes": 7384470,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.4279891010010033,
+      "bytes": 1106558,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -315,3 +315,4 @@
 {"step": 791500, "loss": 0.5506086349487305, "lr": 3e-05, "grad_norm": 0.28517336271728827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115853113635, "tokens_per_second": 153999.3772493908, "data_wait_s": 1.79190028575249e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 791600, "loss": 0.5225836038589478, "lr": 3e-05, "grad_norm": 0.28814563517960295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115867706785, "tokens_per_second": 151674.07777134108, "data_wait_s": 1.2565997167257592e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 791700, "loss": 0.5417950749397278, "lr": 3e-05, "grad_norm": 0.2542495182302116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115882165321, "tokens_per_second": 149714.58733290606, "data_wait_s": 1.6931000573094934e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 791800, "loss": 0.5130705237388611, "lr": 3e-05, "grad_norm": 0.26144605561041123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115896740397, "tokens_per_second": 151697.27504640946, "data_wait_s": 9.578001481713727e-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 sha256:3ce4461185589c62d5d5f73c81f93ef85a4991697ac75dfa26c4e2f256cae56e
-size 676269
+oid sha256:3700f2e14ed65b02c3f37bfe61b3b39d90de1786ceb60832dbc3c7b3ffa6a02c
+size 678387