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": 76448.78444575127,
-      "required_pages_per_second_upper_estimate": 0.005288904293575956
+      "required_token_positions_per_second": 77010.94428116515,
+      "required_pages_per_second_upper_estimate": 0.005327795815380417
     },
     {
       "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": 9556.098055718909,
-      "required_pages_per_second_upper_estimate": 0.0040009387019100595
+      "required_token_positions_per_second": 9626.368035145644,
+      "required_pages_per_second_upper_estimate": 0.004030359274892
     },
     {
       "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": 19112.196111437817,
-      "required_pages_per_second_upper_estimate": 0.003980566427506993
+      "required_token_positions_per_second": 19252.73607029129,
+      "required_pages_per_second_upper_estimate": 0.004009837194648219
     },
     {
       "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": 28668.294167156728,
-      "required_pages_per_second_upper_estimate": 0.0017186204613704465
+      "required_token_positions_per_second": 28879.10410543693,
+      "required_pages_per_second_upper_estimate": 0.001731258195282207
     },
     {
       "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": 28668.294167156728,
-      "required_pages_per_second_upper_estimate": 0.0022288337729968943
+      "required_token_positions_per_second": 28879.10410543693,
+      "required_pages_per_second_upper_estimate": 0.002245223318443257
     },
     {
       "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": 28668.294167156728,
-      "required_pages_per_second_upper_estimate": 0.36772609596024586
+      "required_token_positions_per_second": 28879.10410543693,
+      "required_pages_per_second_upper_estimate": 0.3704301394984278
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.5394387050465923,
-    "producer_p95_seconds": 0.2438600889981899,
-    "gpu_update_seconds": 0.9644103635463038,
-    "producer_batches_per_second": 1.8537787345341268,
-    "gpu_batches_per_second": 1.0369030008375553,
-    "producer_headroom": 1.7878034233064644,
+    "producer_batch_seconds": 0.08798968057007528,
+    "producer_p95_seconds": 0.24350464799863403,
+    "gpu_update_seconds": 0.9573704190773301,
+    "producer_batches_per_second": 11.3649690909333,
+    "gpu_batches_per_second": 1.0445277815913243,
+    "producer_headroom": 10.880485221387717,
     "sustainable": true,
-    "buffer_seconds": 60.75785290341714,
+    "buffer_seconds": 60.3143364018718,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324861.7370081625,
-    "upload_bytes_per_second": 4934107.96965362,
+    "generation_bytes_per_second": 322681.66687869857,
+    "upload_bytes_per_second": 5009417.93345262,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.70776636775642,
+    "estimated_full_shards_per_hour": 69.23997406740874,
     "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": 0.9889250640007958,
-      "bytes": 952891,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.1961971269993228,
       "bytes": 12036916,
@@ -562,6 +556,12 @@
       "bytes": 10074712,
       "pages": 25,
       "shards": 3
+    },
+    {
+      "seconds": 2.2123866580004687,
+      "bytes": 13632293,
+      "pages": 49,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -243,3 +243,4 @@
 {"step": 784300, "loss": 0.4945064187049866, "lr": 3e-05, "grad_norm": 0.25559666003466647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114799206308, "tokens_per_second": 151253.90197177566, "data_wait_s": 1.1760999768739566e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 784400, "loss": 0.5669555068016052, "lr": 3e-05, "grad_norm": 0.2651055261684914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114813828782, "tokens_per_second": 121463.19791131541, "data_wait_s": 1.37330025609117e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 784500, "loss": 0.5469583868980408, "lr": 3e-05, "grad_norm": 0.25180282118908853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114828499502, "tokens_per_second": 153061.59005084969, "data_wait_s": 1.019100091070868e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 784600, "loss": 0.553794801235199, "lr": 3e-05, "grad_norm": 0.2710875305550052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114843234669, "tokens_per_second": 153154.4835286263, "data_wait_s": 7.327998901018873e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7421f6bee5f3635381fdca8ec1f082937a272072ee9eae2c57cef2fcf7e37fcd
-size 522207
+oid sha256:74609bd356ae143353f2cdd417d36ab2e9c5fb673fa8f71f36aa16614a731633
+size 524325