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": 75011.84687498615,
-      "required_pages_per_second_upper_estimate": 0.006501836771336018
+      "required_token_positions_per_second": 76276.8363679701,
+      "required_pages_per_second_upper_estimate": 0.006611482854501308
     },
     {
       "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": 9376.480859373269,
-      "required_pages_per_second_upper_estimate": 0.003740418405685842
+      "required_token_positions_per_second": 9534.604545996262,
+      "required_pages_per_second_upper_estimate": 0.0038034963084395495
     },
     {
       "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": 18752.961718746537,
-      "required_pages_per_second_upper_estimate": 0.004338167652778056
+      "required_token_positions_per_second": 19069.209091992525,
+      "required_pages_per_second_upper_estimate": 0.004411325916814841
     },
     {
       "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": 28129.442578119804,
-      "required_pages_per_second_upper_estimate": 0.0017129354692281642
+      "required_token_positions_per_second": 28603.813637988787,
+      "required_pages_per_second_upper_estimate": 0.0017418221779415812
     },
     {
       "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": 28129.442578119804,
-      "required_pages_per_second_upper_estimate": 0.0021654823569444465
+      "required_token_positions_per_second": 28603.813637988787,
+      "required_pages_per_second_upper_estimate": 0.0022020007542762994
     },
     {
       "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": 28129.442578119804,
-      "required_pages_per_second_upper_estimate": 0.3492735336319928
+      "required_token_positions_per_second": 28603.813637988787,
+      "required_pages_per_second_upper_estimate": 0.3551636345777566
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08469986458588608,
+    "producer_batch_seconds": 0.10900843958592077,
     "producer_p95_seconds": 0.19523060299979988,
-    "gpu_update_seconds": 0.9828847451639767,
-    "producer_batches_per_second": 11.806394318210451,
-    "gpu_batches_per_second": 1.0174132876924118,
-    "producer_headroom": 11.604324870759703,
+    "gpu_update_seconds": 0.9665843984971512,
+    "producer_batches_per_second": 9.173601638539163,
+    "gpu_batches_per_second": 1.0345708057721639,
+    "producer_headroom": 8.867060221839855,
     "sustainable": true,
-    "buffer_seconds": 60.93885420016656,
+    "buffer_seconds": 60.89481710532052,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268629.69447363494,
-    "upload_bytes_per_second": 5042123.475565554,
+    "generation_bytes_per_second": 290860.4127439957,
+    "upload_bytes_per_second": 5078302.965840554,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.641679054801806,
+    "estimated_full_shards_per_hour": 62.411873690985715,
     "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": [
@@ -454,6 +454,12 @@
       "bytes": 1199165,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.0709359819993551,
+      "bytes": 7553738,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -84,3 +84,4 @@
 {"step": 163400, "loss": 0.6147342901676893, "lr": 0.0003, "grad_norm": 0.17651489611382884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23914424884, "tokens_per_second": 149734.31000742893, "data_wait_s": 1.2913998943986371e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 163500, "loss": 0.6141842287033796, "lr": 0.0003, "grad_norm": 0.14708701026552712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23929135948, "tokens_per_second": 150572.25310601245, "data_wait_s": 2.3628001144970767e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 163600, "loss": 0.6215760614722967, "lr": 0.0003, "grad_norm": 0.1403461007080842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23943861458, "tokens_per_second": 151748.93567929196, "data_wait_s": 1.136699938797392e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 163700, "loss": 0.5640713144093752, "lr": 0.0003, "grad_norm": 0.1453051579154165, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23958602611, "tokens_per_second": 140797.66320512534, "data_wait_s": 1.2819000403396785e-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 sha256:818e1d1ba33fc57cc4c1bad4903f97a32398dd9840d092d8e77f4ad86e5d974a
-size 183096
+oid sha256:570ac75fd5a017093019894a7365a09b9d8cd2ad996ca0b4de6e51cf05d0493a
+size 185214