Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75679.64397203029,
-      "required_pages_per_second_upper_estimate": 0.006559719731191535
+      "required_token_positions_per_second": 75618.69681316367,
+      "required_pages_per_second_upper_estimate": 0.006554436985930138
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9459.955496503786,
-      "required_pages_per_second_upper_estimate": 0.00377371768649425
+      "required_token_positions_per_second": 9452.337101645458,
+      "required_pages_per_second_upper_estimate": 0.0037706785948801095
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18919.910993007572,
-      "required_pages_per_second_upper_estimate": 0.004376788429171472
+      "required_token_positions_per_second": 18904.674203290917,
+      "required_pages_per_second_upper_estimate": 0.004373263666028864
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28379.866489511358,
-      "required_pages_per_second_upper_estimate": 0.0017281849715591881
+      "required_token_positions_per_second": 28357.011304936375,
+      "required_pages_per_second_upper_estimate": 0.001726793210730458
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28379.866489511358,
-      "required_pages_per_second_upper_estimate": 0.002184760682861123
+      "required_token_positions_per_second": 28357.011304936375,
+      "required_pages_per_second_upper_estimate": 0.0021830012274853404
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28379.866489511358,
-      "required_pages_per_second_upper_estimate": 0.3523829604965588
+      "required_token_positions_per_second": 28357.011304936375,
+      "required_pages_per_second_upper_estimate": 0.3520991755955198
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13342304237528424,
-    "producer_p95_seconds": 0.4142885940018459,
-    "gpu_update_seconds": 0.9742117712293734,
-    "producer_batches_per_second": 7.494957259236082,
-    "gpu_batches_per_second": 1.0264708655060533,
-    "producer_headroom": 7.301675586808834,
+    "producer_batch_seconds": 0.14770106192952426,
+    "producer_p95_seconds": 0.40922035800031153,
+    "gpu_update_seconds": 0.9749969664534799,
+    "producer_batches_per_second": 6.770432026258222,
+    "gpu_batches_per_second": 1.0256442167584048,
+    "producer_headroom": 6.601150687181253,
     "sustainable": true,
-    "buffer_seconds": 60.40112981622115,
+    "buffer_seconds": 60.44981192011575,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 444381.0714056959,
+    "generation_bytes_per_second": 444815.581026319,
     "upload_bytes_per_second": 5635392.595787868,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.35383326175841,
+    "estimated_full_shards_per_hour": 95.44706891147783,
     "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": [

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -359,3 +359,4 @@
 {"step": 230900, "loss": 0.596820717677474, "lr": 0.0003, "grad_norm": 0.15527798252055194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33797560295, "tokens_per_second": 150714.67299316105, "data_wait_s": 9.115996363107115e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 231000, "loss": 0.642324585467577, "lr": 0.0003, "grad_norm": 0.1480826859223639, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33812281985, "tokens_per_second": 150122.08909107908, "data_wait_s": 1.1200994777027518e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 231100, "loss": 0.5924375485628843, "lr": 0.0003, "grad_norm": 0.15358562835226264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33826888084, "tokens_per_second": 148428.65143981823, "data_wait_s": 1.4299002941697836e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 231200, "loss": 0.6378389857709408, "lr": 0.0003, "grad_norm": 0.15674925291480424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33841347076, "tokens_per_second": 146559.54507288084, "data_wait_s": 1.1919000826310366e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0c2bec7d1829c30cbcfed3a33c7c2f8d90b16c4ddcd02825fabc3212f4aaa7e3
-size 770244
+oid sha256:0493686522c5fd955a21556831be2b0e22cd97c0a1e5e02aaec190183d96b7bf
+size 772362