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": 77078.18082333574,
-      "required_pages_per_second_upper_estimate": 0.0053324473953780495
+      "required_token_positions_per_second": 77083.85365956939,
+      "required_pages_per_second_upper_estimate": 0.0053328398553514865
     },
     {
       "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": 9634.772602916968,
-      "required_pages_per_second_upper_estimate": 0.00403387809191052
+      "required_token_positions_per_second": 9635.481707446173,
+      "required_pages_per_second_upper_estimate": 0.004034174979169112
     },
     {
       "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": 19269.545205833936,
-      "required_pages_per_second_upper_estimate": 0.004013338094295039
+      "required_token_positions_per_second": 19270.963414892347,
+      "required_pages_per_second_upper_estimate": 0.004013633469841218
     },
     {
       "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": 28904.317808750908,
-      "required_pages_per_second_upper_estimate": 0.001732769718296784
+      "required_token_positions_per_second": 28906.445122338522,
+      "required_pages_per_second_upper_estimate": 0.0017328972474981417
     },
     {
       "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": 28904.317808750908,
-      "required_pages_per_second_upper_estimate": 0.0022471835729725585
+      "required_token_positions_per_second": 28906.445122338522,
+      "required_pages_per_second_upper_estimate": 0.002247348962246932
     },
     {
       "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": 28904.317808750908,
-      "required_pages_per_second_upper_estimate": 0.3707535538121741
+      "required_token_positions_per_second": 28906.445122338522,
+      "required_pages_per_second_upper_estimate": 0.37078084070674466
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.29475893184391566,
-    "producer_p95_seconds": 0.24974660699808737,
-    "gpu_update_seconds": 0.9565352893964323,
-    "producer_batches_per_second": 3.3926028763380516,
-    "gpu_batches_per_second": 1.045439735559567,
-    "producer_headroom": 3.2451443741251866,
+    "producer_batch_seconds": 0.08838218873441406,
+    "producer_p95_seconds": 0.25317216600160464,
+    "gpu_update_seconds": 0.9564648950428701,
+    "producer_batches_per_second": 11.314496894900072,
+    "gpu_batches_per_second": 1.0455166783253227,
+    "producer_headroom": 10.821919085043477,
     "sustainable": true,
-    "buffer_seconds": 60.26172323197523,
+    "buffer_seconds": 60.25728838770082,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354414.40509992925,
-    "upload_bytes_per_second": 5393589.597304088,
+    "generation_bytes_per_second": 354826.06452910317,
+    "upload_bytes_per_second": 5349799.072933784,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.04908098934563,
+    "estimated_full_shards_per_hour": 76.1374135199053,
     "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": 1.138313076000486,
-      "bytes": 10599364,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.9668768979991,
       "bytes": 2353349,
@@ -562,6 +556,12 @@
       "bytes": 11594708,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.083468554999854,
+      "bytes": 6825645,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -231,3 +231,4 @@
 {"step": 580200, "loss": 0.596977949142456, "lr": 0.0003, "grad_norm": 0.22100880422977204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84924218816, "tokens_per_second": 153454.72095366375, "data_wait_s": 1.2499000149546191e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 580300, "loss": 0.5834414958953857, "lr": 0.0003, "grad_norm": 0.159024908062277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84938873882, "tokens_per_second": 142447.16291871038, "data_wait_s": 1.0414998541818932e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 580400, "loss": 0.6290363073348999, "lr": 0.0003, "grad_norm": 0.18678624152415044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84953539783, "tokens_per_second": 152079.353037956, "data_wait_s": 1.0726998880272731e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 580500, "loss": 0.6079675555229187, "lr": 0.0003, "grad_norm": 0.1833229334442007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84968185082, "tokens_per_second": 152581.97426825776, "data_wait_s": 1.1011998140020296e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5771344a9b5d0bde8b4c6461e6c0849c205709c09975a3d643229faf0e4a367f
-size 497574
+oid sha256:cc3056721120b0b39e4c35da3da9533e4062197806366b2136e8c6794847bf74
+size 499692