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": 76226.06105870237,
-      "required_pages_per_second_upper_estimate": 0.005273495772870456
+      "required_token_positions_per_second": 76942.0093304089,
+      "required_pages_per_second_upper_estimate": 0.005323026735535976
     },
     {
       "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": 9528.257632337796,
-      "required_pages_per_second_upper_estimate": 0.003989282498014539
+      "required_token_positions_per_second": 9617.751166301112,
+      "required_pages_per_second_upper_estimate": 0.004026751571847477
     },
     {
       "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": 19056.515264675592,
-      "required_pages_per_second_upper_estimate": 0.003968969575529097
+      "required_token_positions_per_second": 19235.502332602224,
+      "required_pages_per_second_upper_estimate": 0.004006247861571812
     },
     {
       "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": 28584.772897013387,
-      "required_pages_per_second_upper_estimate": 0.0017136134887549485
+      "required_token_positions_per_second": 28853.253498903337,
+      "required_pages_per_second_upper_estimate": 0.0017297084908921086
     },
     {
       "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": 28584.772897013387,
-      "required_pages_per_second_upper_estimate": 0.002222340361614492
+      "required_token_positions_per_second": 28853.253498903337,
+      "required_pages_per_second_upper_estimate": 0.0022432135474900715
     },
     {
       "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": 28584.772897013387,
-      "required_pages_per_second_upper_estimate": 0.366654774785
+      "required_token_positions_per_second": 28853.253498903337,
+      "required_pages_per_second_upper_estimate": 0.37009855567403366
     }
   ],
   "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": 0.3934599622811561,
-    "producer_p95_seconds": 0.35930893299519084,
-    "gpu_update_seconds": 0.9672282546939086,
-    "producer_batches_per_second": 2.5415546583248703,
-    "gpu_batches_per_second": 1.033882121564431,
-    "producer_headroom": 2.4582634763807376,
+    "producer_batch_seconds": 0.24915556036705766,
+    "producer_p95_seconds": 0.35227929699613014,
+    "gpu_update_seconds": 0.9582281596441404,
+    "producer_batches_per_second": 4.013556825811125,
+    "gpu_batches_per_second": 1.0435927914823255,
+    "producer_headroom": 3.845903170824172,
     "sustainable": true,
-    "buffer_seconds": 60.935380045716244,
+    "buffer_seconds": 60.368374057580844,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 568939.476889592,
+    "generation_bytes_per_second": 570697.6114910742,
     "upload_bytes_per_second": 3802259.5495382077,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.04058375366125,
+    "estimated_full_shards_per_hour": 61.22921113276086,
     "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": [

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -353,3 +353,4 @@
 {"step": 835300, "loss": 0.5298817157745361, "lr": 3e-05, "grad_norm": 0.29290207557855097, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122264067910, "tokens_per_second": 150097.1598721705, "data_wait_s": 1.545899431221187e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 835400, "loss": 0.5379299521446228, "lr": 3e-05, "grad_norm": 0.305677779022904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122278756085, "tokens_per_second": 152780.8836670833, "data_wait_s": 1.3860997569281608e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 835500, "loss": 0.48333069682121277, "lr": 3e-05, "grad_norm": 0.29217728596827114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122293481709, "tokens_per_second": 151981.18707182675, "data_wait_s": 1.6933001461438835e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 835600, "loss": 0.5196800827980042, "lr": 3e-05, "grad_norm": 0.2856584328634872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122308166701, "tokens_per_second": 152360.93729544472, "data_wait_s": 6.574999133590609e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b9f39ead00875dafa091f0b57defa56ed69d7184ccd1aeaf4aca95a58b2cde53
-size 757536
+oid sha256:bd8002bb132fab48dedc6c3d0b4bf51f0727df89e0fb0fd2bafe6b34f7cb02b8
+size 759654