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": 77634.8613087369,
-      "required_pages_per_second_upper_estimate": 0.005370959843034787
+      "required_token_positions_per_second": 77325.06236788606,
+      "required_pages_per_second_upper_estimate": 0.005349527233474146
     },
     {
       "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": 9704.357663592113,
-      "required_pages_per_second_upper_estimate": 0.004063011903713898
+      "required_token_positions_per_second": 9665.632795985757,
+      "required_pages_per_second_upper_estimate": 0.004046798610314309
     },
     {
       "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": 19408.715327184225,
-      "required_pages_per_second_upper_estimate": 0.004042323560409396
+      "required_token_positions_per_second": 19331.265591971514,
+      "required_pages_per_second_upper_estimate": 0.004026192823051457
     },
     {
       "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": 29113.07299077634,
-      "required_pages_per_second_upper_estimate": 0.001745284272708502
+      "required_token_positions_per_second": 28996.89838795727,
+      "required_pages_per_second_upper_estimate": 0.0017383197826578462
     },
     {
       "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": 29113.07299077634,
-      "required_pages_per_second_upper_estimate": 0.0022634133701580307
+      "required_token_positions_per_second": 28996.89838795727,
+      "required_pages_per_second_upper_estimate": 0.0022543813057869216
     },
     {
       "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": 29113.07299077634,
-      "required_pages_per_second_upper_estimate": 0.3734312411433453
+      "required_token_positions_per_second": 28996.89838795727,
+      "required_pages_per_second_upper_estimate": 0.3719410780769522
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10139773190623202,
-    "producer_p95_seconds": 0.18660580399955506,
-    "gpu_update_seconds": 0.9496764566474825,
-    "producer_batches_per_second": 9.862153533421775,
-    "gpu_batches_per_second": 1.052990197872408,
-    "producer_headroom": 9.36585502253344,
+    "producer_batch_seconds": 0.3946124600546739,
+    "producer_p95_seconds": 0.1712527739982761,
+    "gpu_update_seconds": 0.9534812872083767,
+    "producer_batches_per_second": 2.5341318413043754,
+    "gpu_batches_per_second": 1.0487882808144267,
+    "producer_headroom": 2.4162472900026297,
     "sustainable": true,
-    "buffer_seconds": 60.77929322543888,
+    "buffer_seconds": 60.06932109412773,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 336126.29702411336,
-    "upload_bytes_per_second": 4262305.280323286,
+    "generation_bytes_per_second": 301945.3812618076,
+    "upload_bytes_per_second": 4290878.952557519,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.12487872164297,
+    "estimated_full_shards_per_hour": 64.79044989004774,
     "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": [
@@ -466,6 +466,12 @@
       "bytes": 1201273,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.2199994860002334,
+      "bytes": 7355179,
+      "pages": 26,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 532500, "loss": 0.6018000245094299, "lr": 0.0003, "grad_norm": 0.16666585614226395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77940993934, "tokens_per_second": 154926.7151380044, "data_wait_s": 1.0138999641640112e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 532600, "loss": 0.6273928880691528, "lr": 0.0003, "grad_norm": 0.14651706984761576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77955790238, "tokens_per_second": 154786.6819423123, "data_wait_s": 1.268699998036027e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 532700, "loss": 0.6021421551704407, "lr": 0.0003, "grad_norm": 0.16543119273718426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77970485982, "tokens_per_second": 153435.46615082206, "data_wait_s": 8.90299997990951e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 532800, "loss": 0.8693204522132874, "lr": 0.0003, "grad_norm": 0.3004651123097426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77984952000, "tokens_per_second": 151148.54085587242, "data_wait_s": 1.012800021271687e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6af83ece9987f5e0a03b2379009b54f3859cba2aabb7c49ec6c7c1b4da79ac5c
-size 164034
+oid sha256:ced2a6af6a16538bd6b439c2e4290ca0bced0b8874106c132033ea48335abb76
+size 166152