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": 76406.9928599857,
-      "required_pages_per_second_upper_estimate": 0.006622764489350607
+      "required_token_positions_per_second": 76311.37671389087,
+      "required_pages_per_second_upper_estimate": 0.006614476724144014
     },
     {
       "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": 9550.874107498212,
-      "required_pages_per_second_upper_estimate": 0.0038099864797742984
+      "required_token_positions_per_second": 9538.92208923636,
+      "required_pages_per_second_upper_estimate": 0.003805218640991048
     },
     {
       "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": 19101.748214996423,
-      "required_pages_per_second_upper_estimate": 0.004418853270252787
+      "required_token_positions_per_second": 19077.84417847272,
+      "required_pages_per_second_upper_estimate": 0.004413323492099702
     },
     {
       "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": 28652.622322494637,
-      "required_pages_per_second_upper_estimate": 0.0017447943707486107
+      "required_token_positions_per_second": 28616.766267709078,
+      "required_pages_per_second_upper_estimate": 0.0017426109251343508
     },
     {
       "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": 28652.622322494637,
-      "required_pages_per_second_upper_estimate": 0.002205758181920646
+      "required_token_positions_per_second": 28616.766267709078,
+      "required_pages_per_second_upper_estimate": 0.002202997883567341
     },
     {
       "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": 28652.622322494637,
-      "required_pages_per_second_upper_estimate": 0.3557696750871604
+      "required_token_positions_per_second": 28616.766267709078,
+      "required_pages_per_second_upper_estimate": 0.3553244628892196
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10399550307033678,
-    "producer_p95_seconds": 0.3144326110050315,
-    "gpu_update_seconds": 0.9649378576527033,
-    "producer_batches_per_second": 9.615800399788975,
-    "gpu_batches_per_second": 1.0363361661781914,
-    "producer_headroom": 9.278649837388382,
+    "producer_batch_seconds": 0.11136818915599633,
+    "producer_p95_seconds": 0.32149912499880884,
+    "gpu_update_seconds": 0.966146899385965,
+    "producer_batches_per_second": 8.979224746119144,
+    "gpu_batches_per_second": 1.0350392891966536,
+    "producer_headroom": 8.67525014735274,
     "sustainable": true,
-    "buffer_seconds": 60.791085032120314,
+    "buffer_seconds": 60.8672546613158,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 440874.71227738465,
-    "upload_bytes_per_second": 4566402.222583162,
+    "generation_bytes_per_second": 442680.12105962256,
+    "upload_bytes_per_second": 4553000.701032992,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.60145021668582,
+    "estimated_full_shards_per_hour": 94.9888489135886,
     "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": 1.1491751960020338,
-      "bytes": 6275126,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.3218376940021699,
       "bytes": 6023489,
@@ -562,6 +556,12 @@
       "bytes": 1254870,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.428139841002121,
+      "bytes": 6304931,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -330,3 +330,4 @@
 {"step": 188000, "loss": 0.6003498584032059, "lr": 0.0003, "grad_norm": 0.15873858486695672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27514881760, "tokens_per_second": 150405.78853102526, "data_wait_s": 9.98800533125177e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 188100, "loss": 0.6589372772723436, "lr": 0.0003, "grad_norm": 0.15250371718394956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27529558837, "tokens_per_second": 150646.49554816974, "data_wait_s": 9.374998626299202e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 188200, "loss": 0.6110364906489849, "lr": 0.0003, "grad_norm": 0.15254910905981836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27544352460, "tokens_per_second": 152036.56392652812, "data_wait_s": 8.128998160827905e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 188300, "loss": 0.6326652783900499, "lr": 0.0003, "grad_norm": 0.19723724846528282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27559049208, "tokens_per_second": 151113.9929722391, "data_wait_s": 1.4320998161565512e-05, "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 sha256:173dca8983f247019ec959a88d4444668233c19faacbc6c682747a03ed0d4d05
-size 708039
+oid sha256:1cd87e2103688b4cad99283ef51c90cb777d7cd0bbf642f3eced4757ad483fe2
+size 710157