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": 76109.23530259746,
-      "required_pages_per_second_upper_estimate": 0.006596955618935824
+      "required_token_positions_per_second": 75490.23793852492,
+      "required_pages_per_second_upper_estimate": 0.006543302496252493
     },
     {
       "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": 9513.654412824682,
-      "required_pages_per_second_upper_estimate": 0.003795138987085002
+      "required_token_positions_per_second": 9436.279742315615,
+      "required_pages_per_second_upper_estimate": 0.0037642730741645185
     },
     {
       "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": 19027.308825649365,
-      "required_pages_per_second_upper_estimate": 0.004401633027615853
+      "required_token_positions_per_second": 18872.55948463123,
+      "required_pages_per_second_upper_estimate": 0.00436583449106669
     },
     {
       "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": 28540.963238474047,
-      "required_pages_per_second_upper_estimate": 0.0017379949183616954
+      "required_token_positions_per_second": 28308.839226946846,
+      "required_pages_per_second_upper_estimate": 0.001723859783920256
     },
     {
       "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": 28540.963238474047,
-      "required_pages_per_second_upper_estimate": 0.0021971623565406138
+      "required_token_positions_per_second": 28308.839226946846,
+      "required_pages_per_second_upper_estimate": 0.00217929280757286
     },
     {
       "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": 28540.963238474047,
-      "required_pages_per_second_upper_estimate": 0.35438324296253954
+      "required_token_positions_per_second": 28308.839226946846,
+      "required_pages_per_second_upper_estimate": 0.35150103960846374
     }
   ],
   "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": 0.07855621991412676,
-    "producer_p95_seconds": 0.2089661559994056,
-    "gpu_update_seconds": 0.9687129256636192,
-    "producer_batches_per_second": 12.729736755321778,
-    "gpu_batches_per_second": 1.0322975708360116,
-    "producer_headroom": 12.331460535175466,
+    "producer_batch_seconds": 0.44149790242970255,
+    "producer_p95_seconds": 0.20993863700005022,
+    "gpu_update_seconds": 0.976656081810737,
+    "producer_batches_per_second": 2.2650164236266668,
+    "gpu_batches_per_second": 1.023901881761677,
+    "producer_headroom": 2.212142065536189,
     "sustainable": true,
-    "buffer_seconds": 60.06020139114439,
+    "buffer_seconds": 60.552677072265695,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315994.84641221,
-    "upload_bytes_per_second": 4874278.025479407,
+    "generation_bytes_per_second": 302364.9274024608,
+    "upload_bytes_per_second": 5006641.475695034,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.80513805651402,
+    "estimated_full_shards_per_hour": 64.88047472529762,
     "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": [
@@ -388,6 +388,12 @@
       "bytes": 848800,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.479707024000163,
+      "bytes": 13318632,
+      "pages": 43,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -71,3 +71,4 @@
 {"step": 162100, "loss": 0.5705337896943092, "lr": 0.0003, "grad_norm": 0.14917666228697132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23724411562, "tokens_per_second": 150661.94945537805, "data_wait_s": 1.3358000614971388e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 162200, "loss": 0.5789471492171288, "lr": 0.0003, "grad_norm": 0.1280685178990801, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23739045515, "tokens_per_second": 134287.53340188722, "data_wait_s": 1.347699981124606e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 162300, "loss": 0.5677263271063566, "lr": 0.0003, "grad_norm": 0.15410590502460117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23753576021, "tokens_per_second": 148492.45933231778, "data_wait_s": 1.364699983241735e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 162400, "loss": 0.5808803290128708, "lr": 0.0003, "grad_norm": 0.13058085458923926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23768096827, "tokens_per_second": 148903.8294180968, "data_wait_s": 1.1887999789905734e-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 sha256:7e65bdbee1c1f552b4abd73f724bb6d85625be848445e8e0e72316e2c98aeff5
-size 155562
+oid sha256:43f66cd277969a974b70808df548f6f056b0dd7bed662474896b5ad6eb500b91
+size 157680