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": 75846.54784012104,
-      "required_pages_per_second_upper_estimate": 0.006574186535463655
+      "required_token_positions_per_second": 75074.43834255643,
+      "required_pages_per_second_upper_estimate": 0.0065072620411083856
     },
     {
       "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": 9480.81848001513,
-      "required_pages_per_second_upper_estimate": 0.0037820402425463263
+      "required_token_positions_per_second": 9384.304792819554,
+      "required_pages_per_second_upper_estimate": 0.0037435394897157945
     },
     {
       "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": 18961.63696003026,
-      "required_pages_per_second_upper_estimate": 0.004386440997290232
+      "required_token_positions_per_second": 18768.609585639108,
+      "required_pages_per_second_upper_estimate": 0.004341787511390601
     },
     {
       "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": 28442.455440045393,
-      "required_pages_per_second_upper_estimate": 0.0017319963102678658
+      "required_token_positions_per_second": 28152.91437845866,
+      "required_pages_per_second_upper_estimate": 0.0017143647787217814
     },
     {
       "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": 28442.455440045393,
-      "required_pages_per_second_upper_estimate": 0.002189578953530537
+      "required_token_positions_per_second": 28152.91437845866,
+      "required_pages_per_second_upper_estimate": 0.0021672892811086335
     },
     {
       "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": 28442.455440045393,
-      "required_pages_per_second_upper_estimate": 0.35316010579044904
+      "required_token_positions_per_second": 28152.91437845866,
+      "required_pages_per_second_upper_estimate": 0.34956497483713894
     }
   ],
   "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": 0.06714246780467192,
-    "producer_p95_seconds": 0.1657440859999042,
-    "gpu_update_seconds": 0.9720679727627579,
-    "producer_batches_per_second": 14.893703384706658,
-    "gpu_batches_per_second": 1.028734644098864,
-    "producer_headroom": 14.477692056101626,
+    "producer_batch_seconds": 0.3224209352109497,
+    "producer_p95_seconds": 0.16911214900028426,
+    "gpu_update_seconds": 0.9820652891678953,
+    "producer_batches_per_second": 3.1015355728862084,
+    "gpu_batches_per_second": 1.018262238804205,
+    "producer_headroom": 3.0459104292510077,
     "sustainable": true,
-    "buffer_seconds": 60.26821431129099,
+    "buffer_seconds": 60.888047928409506,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 482311.33747013856,
-    "upload_bytes_per_second": 5432658.7250128,
+    "generation_bytes_per_second": 406665.7071276919,
+    "upload_bytes_per_second": 5640708.793518272,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.49278538778417,
+    "estimated_full_shards_per_hour": 87.26099405644482,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 6057442,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.1024301279999236,
+      "bytes": 7712415,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -33,3 +33,4 @@
 {"step": 43300, "loss": 0.6022576056420803, "lr": 0.0003, "grad_norm": 0.15580596171508607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6336267358, "tokens_per_second": 150087.74574274768, "data_wait_s": 1.1791999895649496e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 43400, "loss": 0.618283323943615, "lr": 0.0003, "grad_norm": 0.12760305932739904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6351024629, "tokens_per_second": 151044.35480064023, "data_wait_s": 9.823000254982617e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 43500, "loss": 0.635155625641346, "lr": 0.0003, "grad_norm": 0.13353834084727553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6365806091, "tokens_per_second": 151125.87165017365, "data_wait_s": 1.4281999938248191e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 43600, "loss": 0.6215451024472713, "lr": 0.0003, "grad_norm": 0.125585212272193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6380531863, "tokens_per_second": 150133.86798585454, "data_wait_s": 1.4137999642116483e-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 sha256:95d1e647d302d1adc7b1b49fc00ea57a4eea06996693042c6dd577b9f24af1d2
-size 74295
+oid sha256:030e1690133bcf8e58e4b17f5b372bf9f40c4d5c522afd879884e1916cea8e0c
+size 76413