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": 76756.86364394701,
-      "required_pages_per_second_upper_estimate": 0.005310217927349363
+      "required_token_positions_per_second": 77067.7278844574,
+      "required_pages_per_second_upper_estimate": 0.005331724236812286
     },
     {
       "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": 9594.607955493377,
-      "required_pages_per_second_upper_estimate": 0.004017061992767476
+      "required_token_positions_per_second": 9633.465985557175,
+      "required_pages_per_second_upper_estimate": 0.004033331038507248
     },
     {
       "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": 19189.215910986753,
-      "required_pages_per_second_upper_estimate": 0.003996607620604334
+      "required_token_positions_per_second": 19266.93197111435,
+      "required_pages_per_second_upper_estimate": 0.0040127938264185835
     },
     {
       "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": 28783.823866480132,
-      "required_pages_per_second_upper_estimate": 0.0017255462904412478
+      "required_token_positions_per_second": 28900.397956671528,
+      "required_pages_per_second_upper_estimate": 0.0017325347291498928
     },
     {
       "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": 28783.823866480132,
-      "required_pages_per_second_upper_estimate": 0.0022378156989578462
+      "required_token_positions_per_second": 28900.397956671528,
+      "required_pages_per_second_upper_estimate": 0.002246878821714994
     },
     {
       "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": 28783.823866480132,
-      "required_pages_per_second_upper_estimate": 0.36920798689703993
+      "required_token_positions_per_second": 28900.397956671528,
+      "required_pages_per_second_upper_estimate": 0.3707032741585091
     }
   ],
   "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": 0.06076359718746005,
-    "producer_p95_seconds": 0.19642133199886302,
-    "gpu_update_seconds": 0.9605395074765295,
-    "producer_batches_per_second": 16.457221861222738,
-    "gpu_batches_per_second": 1.0410815923929446,
-    "producer_headroom": 15.807811781010862,
+    "producer_batch_seconds": 0.07642315499985841,
+    "producer_p95_seconds": 0.2048241690008581,
+    "gpu_update_seconds": 0.9566650272930786,
+    "producer_batches_per_second": 13.085039475298457,
+    "gpu_batches_per_second": 1.0452979585022977,
+    "producer_headroom": 12.517999646767409,
     "sustainable": true,
-    "buffer_seconds": 60.513988971021355,
+    "buffer_seconds": 60.26989671946395,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 359086.1264012401,
-    "upload_bytes_per_second": 5906921.105498973,
+    "generation_bytes_per_second": 443793.782803533,
+    "upload_bytes_per_second": 5898209.082619942,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.05152362850096,
+    "estimated_full_shards_per_hour": 95.2278147991132,
     "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": 0.9096160310000414,
-      "bytes": 6003921,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0415576699997473,
       "bytes": 6858425,
@@ -562,6 +556,12 @@
       "bytes": 7841640,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.0024716899988562,
+      "bytes": 5917383,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -90,3 +90,4 @@
 {"step": 494000, "loss": 0.5917806029319763, "lr": 0.0003, "grad_norm": 0.17410014528400036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72305737137, "tokens_per_second": 155062.05867281987, "data_wait_s": 1.4626999472966418e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 494100, "loss": 0.5667669773101807, "lr": 0.0003, "grad_norm": 0.1641166069705547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72320479632, "tokens_per_second": 153706.58341545094, "data_wait_s": 1.4867000572849065e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 494200, "loss": 0.5927258729934692, "lr": 0.0003, "grad_norm": 0.16062092644810683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72335214197, "tokens_per_second": 153688.4105405821, "data_wait_s": 1.3154998669051565e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 494300, "loss": 0.6222625970840454, "lr": 0.0003, "grad_norm": 0.16598529522696565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72349821711, "tokens_per_second": 152517.2418838022, "data_wait_s": 9.6160001703538e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0d47537011703d3f39388a1fa04d2f15831d0faa6a64b32e17ab2268b47ee95b
-size 195804
+oid sha256:fc01dff1596194208af031e076a528e8ee82c9fa1dcd0b602ab423f20ea2ca23
+size 197922