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": 76943.07036535782,
-      "required_pages_per_second_upper_estimate": 0.005323100140395683
+      "required_token_positions_per_second": 76713.09659255906,
+      "required_pages_per_second_upper_estimate": 0.005307190021180794
     },
     {
       "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": 9617.883795669728,
-      "required_pages_per_second_upper_estimate": 0.0040268071009945
+      "required_token_positions_per_second": 9589.137074069882,
+      "required_pages_per_second_upper_estimate": 0.004014771448960454
     },
     {
       "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": 19235.767591339456,
-      "required_pages_per_second_upper_estimate": 0.004006303107971435
+      "required_token_positions_per_second": 19178.274148139764,
+      "required_pages_per_second_upper_estimate": 0.003994328739956997
     },
     {
       "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": 28853.651387009188,
-      "required_pages_per_second_upper_estimate": 0.0017297323436765144
+      "required_token_positions_per_second": 28767.41122220965,
+      "required_pages_per_second_upper_estimate": 0.001724562377477889
     },
     {
       "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": 28853.651387009188,
-      "required_pages_per_second_upper_estimate": 0.002243244481540177
+      "required_token_positions_per_second": 28767.41122220965,
+      "required_pages_per_second_upper_estimate": 0.0022365396880574084
     },
     {
       "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": 28853.651387009188,
-      "required_pages_per_second_upper_estimate": 0.37010365935543654
+      "required_token_positions_per_second": 28767.41122220965,
+      "required_pages_per_second_upper_estimate": 0.3689974631188626
     }
   ],
   "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": 0.331411894492021,
-    "producer_p95_seconds": 0.35581488699972397,
-    "gpu_update_seconds": 0.9582149458022492,
-    "producer_batches_per_second": 3.017393209536949,
-    "gpu_batches_per_second": 1.0436071826898576,
-    "producer_headroom": 2.8913112707405224,
+    "producer_batch_seconds": 0.12801498380468956,
+    "producer_p95_seconds": 0.36095250400103396,
+    "gpu_update_seconds": 0.961087523185075,
+    "producer_batches_per_second": 7.811585568183832,
+    "gpu_batches_per_second": 1.0404879637662634,
+    "producer_headroom": 7.507617425874075,
     "sustainable": true,
-    "buffer_seconds": 60.3675415855417,
+    "buffer_seconds": 60.54851396065972,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430598.00361536926,
-    "upload_bytes_per_second": 3945085.371121359,
+    "generation_bytes_per_second": 428776.1893345945,
+    "upload_bytes_per_second": 3953374.7422536113,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.39630776735123,
+    "estimated_full_shards_per_hour": 92.00538883236291,
     "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": 1.8466758049980854,
-      "bytes": 10680095,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.7479712950007524,
       "bytes": 6009903,
@@ -562,6 +556,12 @@
       "bytes": 976930,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.5532393190005678,
+      "bytes": 10410458,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -374,3 +374,4 @@
 {"step": 837400, "loss": 0.4866166412830353, "lr": 3e-05, "grad_norm": 0.2833518862650266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122571372745, "tokens_per_second": 151532.1367779096, "data_wait_s": 1.2470001820474863e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 837500, "loss": 0.5189028382301331, "lr": 3e-05, "grad_norm": 0.28003621527505296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122586023309, "tokens_per_second": 151814.67107442184, "data_wait_s": 1.0933996236417443e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 837600, "loss": 0.5480495095252991, "lr": 3e-05, "grad_norm": 0.2912639937193443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122600658392, "tokens_per_second": 151211.46516290552, "data_wait_s": 1.3545002730097622e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 837700, "loss": 0.5169020295143127, "lr": 3e-05, "grad_norm": 0.28522732600895295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122615260985, "tokens_per_second": 150896.4012463579, "data_wait_s": 1.5616002201568335e-05, "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 sha256:890725b207213624b6b6fe41d3fb18525f7fb6de90164f4a78c77fe1c23f43f8
-size 802014
+oid sha256:d4abbd7d8711f4af485ac23db78c4e122544a8491f701ec12eaf2963ee4f2c7e
+size 804132