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": 75767.43285380199,
-      "required_pages_per_second_upper_estimate": 0.00656732904896462
+      "required_token_positions_per_second": 75578.31874178846,
+      "required_pages_per_second_upper_estimate": 0.006550937117040622
     },
     {
       "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": 9470.929106725249,
-      "required_pages_per_second_upper_estimate": 0.0037780952236816856
+      "required_token_positions_per_second": 9447.289842723558,
+      "required_pages_per_second_upper_estimate": 0.0037686651678329177
     },
     {
       "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": 18941.858213450498,
-      "required_pages_per_second_upper_estimate": 0.004381865532363055
+      "required_token_positions_per_second": 18894.579685447115,
+      "required_pages_per_second_upper_estimate": 0.004370928476983147
     },
     {
       "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": 28412.787320175747,
-      "required_pages_per_second_upper_estimate": 0.0017301896774243996
+      "required_token_positions_per_second": 28341.86952817067,
+      "required_pages_per_second_upper_estimate": 0.0017258711559681895
     },
     {
       "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": 28412.787320175747,
-      "required_pages_per_second_upper_estimate": 0.0021872950195363596
+      "required_token_positions_per_second": 28341.86952817067,
+      "required_pages_per_second_upper_estimate": 0.002181835571594796
     },
     {
       "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": 28412.787320175747,
-      "required_pages_per_second_upper_estimate": 0.3527917270344779
+      "required_token_positions_per_second": 28341.86952817067,
+      "required_pages_per_second_upper_estimate": 0.35191116540435663
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3636384550858338,
-    "producer_p95_seconds": 0.3613100360016688,
-    "gpu_update_seconds": 0.9730829886009573,
-    "producer_batches_per_second": 2.7499841835043504,
-    "gpu_batches_per_second": 1.0276615784207084,
-    "producer_headroom": 2.6759628278897765,
+    "producer_batch_seconds": 0.12213896339062558,
+    "producer_p95_seconds": 0.3787307909951778,
+    "gpu_update_seconds": 0.9755178631571572,
+    "producer_batches_per_second": 8.187395506230013,
+    "gpu_batches_per_second": 1.0250965541149692,
+    "producer_headroom": 7.9869505690600135,
     "sustainable": true,
-    "buffer_seconds": 60.331145293259354,
+    "buffer_seconds": 60.48210751574374,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 401941.5965801479,
-    "upload_bytes_per_second": 5663747.688653589,
+    "generation_bytes_per_second": 425995.04122085054,
+    "upload_bytes_per_second": 5528238.864574658,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.24730990460709,
+    "estimated_full_shards_per_hour": 91.40861918896806,
     "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": 1.0528832150012022,
-      "bytes": 11948549,
-      "pages": 44,
-      "shards": 3
-    },
     {
       "seconds": 1.0517111139997724,
       "bytes": 6085549,
@@ -562,6 +556,12 @@
       "bytes": 12346007,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 0.916572713998903,
+      "bytes": 942236,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -352,3 +352,4 @@
 {"step": 75200, "loss": 0.6243664398789406, "lr": 0.0003, "grad_norm": 0.1403619385012886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11004993501, "tokens_per_second": 150818.62802642008, "data_wait_s": 8.654002158436924e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 75300, "loss": 0.6142660826444626, "lr": 0.0003, "grad_norm": 0.13490544914266286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11019686667, "tokens_per_second": 149725.26537174347, "data_wait_s": 1.0880001354962587e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 75400, "loss": 0.6059148088097572, "lr": 0.0003, "grad_norm": 0.13393706547614445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11034406397, "tokens_per_second": 150402.01004284364, "data_wait_s": 8.873001206666231e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 75500, "loss": 0.6612876560539007, "lr": 0.0003, "grad_norm": 0.1428170348202804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11049113914, "tokens_per_second": 151177.2646087947, "data_wait_s": 1.1476004146970809e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:55ebec7e4b32647f17a13abdf8623411fec26073869f9e3328e16ddd19750998
-size 755418
+oid sha256:bd14101843851504d19a94067249da97ab982dcfe72ab077fbebf5b565adb124
+size 757536