Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77101.40166691653,
-      "required_pages_per_second_upper_estimate": 0.005334053867216749
+      "required_token_positions_per_second": 77346.73641265622,
+      "required_pages_per_second_upper_estimate": 0.005351026694182058
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9637.675208364566,
-      "required_pages_per_second_upper_estimate": 0.004035093352198135
+      "required_token_positions_per_second": 9668.342051582027,
+      "required_pages_per_second_upper_estimate": 0.0040479329190567775
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19275.350416729132,
-      "required_pages_per_second_upper_estimate": 0.004014547166630801
+      "required_token_positions_per_second": 19336.684103164054,
+      "required_pages_per_second_upper_estimate": 0.004027321356037114
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28913.025625093695,
-      "required_pages_per_second_upper_estimate": 0.0017332917385904684
+      "required_token_positions_per_second": 29005.02615474608,
+      "required_pages_per_second_upper_estimate": 0.001738807029866454
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28913.025625093695,
-      "required_pages_per_second_upper_estimate": 0.002247860567391574
+      "required_token_positions_per_second": 29005.02615474608,
+      "required_pages_per_second_upper_estimate": 0.0022550132039044837
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28913.025625093695,
-      "required_pages_per_second_upper_estimate": 0.37086524833049467
+      "required_token_positions_per_second": 29005.02615474608,
+      "required_pages_per_second_upper_estimate": 0.372045332342403
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.41070608171878575,
-    "producer_p95_seconds": 0.20685095000044385,
-    "gpu_update_seconds": 0.9562472070029303,
-    "producer_batches_per_second": 2.4348312443172175,
-    "gpu_batches_per_second": 1.0457546884076134,
-    "producer_headroom": 2.328300576901809,
+    "producer_batch_seconds": 0.05631322054686905,
+    "producer_p95_seconds": 0.18836101799934113,
+    "gpu_update_seconds": 0.9532141033934551,
+    "producer_batches_per_second": 17.757819394607488,
+    "gpu_batches_per_second": 1.049082253860897,
+    "producer_headroom": 16.927003892453683,
     "sustainable": true,
-    "buffer_seconds": 60.24357404118461,
+    "buffer_seconds": 60.05248851378767,
     "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": 320354.18532763387,
-    "upload_bytes_per_second": 3116815.615778697,
+    "generation_bytes_per_second": 319869.82043033105,
+    "upload_bytes_per_second": 3161427.9926424385,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.74055070754778,
+    "estimated_full_shards_per_hour": 68.63661727602432,
     "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": [
@@ -502,6 +502,12 @@
       "bytes": 8387056,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.502407841000604,
+      "bytes": 9864729,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -108,3 +108,4 @@
 {"step": 695800, "loss": 0.5793548822402954, "lr": 8e-05, "grad_norm": 0.21585714774418002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101846698572, "tokens_per_second": 153571.48304085416, "data_wait_s": 1.3446000593830831e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 695900, "loss": 0.5242658257484436, "lr": 8e-05, "grad_norm": 0.20053522714558836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101861235743, "tokens_per_second": 151289.95061557947, "data_wait_s": 1.521600097476039e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 696000, "loss": 0.5266731977462769, "lr": 8e-05, "grad_norm": 0.23273954291675364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101875729910, "tokens_per_second": 151681.54074112067, "data_wait_s": 1.4149998605716974e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 696100, "loss": 0.4949779808521271, "lr": 8e-05, "grad_norm": 0.2101428819671386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101890331384, "tokens_per_second": 151445.37746768864, "data_wait_s": 1.6972000594250858e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3ffdd8ed361b9794622b65488cda26e0de83272148ac72378fa67f3a7458e925
-size 234711
+oid sha256:405c571269ed535b813213bc25712db1d05f599ed26684fd62ef561356c39a80
+size 236829