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": 76025.62015568998,
-      "required_pages_per_second_upper_estimate": 0.006589708069922536
+      "required_token_positions_per_second": 76010.8375361023,
+      "required_pages_per_second_upper_estimate": 0.006588426749922884
     },
     {
       "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": 9503.202519461247,
-      "required_pages_per_second_upper_estimate": 0.0037909695705525956
+      "required_token_positions_per_second": 9501.354692012788,
+      "required_pages_per_second_upper_estimate": 0.003790232444555923
     },
     {
       "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": 19006.405038922494,
-      "required_pages_per_second_upper_estimate": 0.004396797304450672
+      "required_token_positions_per_second": 19002.709384025577,
+      "required_pages_per_second_upper_estimate": 0.004395942379731575
     },
     {
       "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": 28509.60755838374,
-      "required_pages_per_second_upper_estimate": 0.0017360855219547351
+      "required_token_positions_per_second": 28504.064076038365,
+      "required_pages_per_second_upper_estimate": 0.0017357479529643057
     },
     {
       "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": 28509.60755838374,
-      "required_pages_per_second_upper_estimate": 0.002194748509489184
+      "required_token_positions_per_second": 28504.064076038365,
+      "required_pages_per_second_upper_estimate": 0.0021943217568728955
     },
     {
       "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": 28509.60755838374,
-      "required_pages_per_second_upper_estimate": 0.35399391035652855
+      "required_token_positions_per_second": 28504.064076038365,
+      "required_pages_per_second_upper_estimate": 0.35392507885864094
     }
   ],
   "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": 0.11026818247654546,
-    "producer_p95_seconds": 0.37229756499436917,
-    "gpu_update_seconds": 0.9697783437874657,
-    "producer_batches_per_second": 9.06879915439528,
-    "gpu_batches_per_second": 1.031163467823486,
-    "producer_headroom": 8.794725024090626,
+    "producer_batch_seconds": 0.11601098349211725,
+    "producer_p95_seconds": 0.3488710200035712,
+    "gpu_update_seconds": 0.9699669466868057,
+    "producer_batches_per_second": 8.619873480065346,
+    "gpu_batches_per_second": 1.0309629657131931,
+    "producer_headroom": 8.360992360285554,
     "sustainable": true,
-    "buffer_seconds": 60.12625731482288,
+    "buffer_seconds": 60.13795069458195,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 468251.1302116926,
+    "generation_bytes_per_second": 499583.660512152,
     "upload_bytes_per_second": 5165708.559907301,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.47579221499522,
+    "estimated_full_shards_per_hour": 107.19902383349819,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -359,3 +359,4 @@
 {"step": 190900, "loss": 0.5981983579695225, "lr": 0.0003, "grad_norm": 0.149478788079187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27940072447, "tokens_per_second": 150759.55153794054, "data_wait_s": 9.71899862634018e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 191000, "loss": 0.6219041235744953, "lr": 0.0003, "grad_norm": 0.13180663164012466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27954752847, "tokens_per_second": 150940.91277868982, "data_wait_s": 8.448005246464163e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 191100, "loss": 0.6132036913186312, "lr": 0.0003, "grad_norm": 0.1498338886749033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27969638045, "tokens_per_second": 152934.77459276258, "data_wait_s": 1.0142000974155962e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 191200, "loss": 0.6133407857269049, "lr": 0.0003, "grad_norm": 0.1524501252896246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27984303600, "tokens_per_second": 150681.225511353, "data_wait_s": 9.14699921850115e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:26dd0635adb10b8805d88c7ef5e0ab661b777529343b22f1ea7f908154c1b0f6
-size 770244
+oid sha256:ac93b7856b15352763fa0dbfe136ce0e1f8c3ccf46f9c10e32d50b5404a4a494
+size 772362