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": 76742.70439241717,
-      "required_pages_per_second_upper_estimate": 0.0053092383574745395
+      "required_token_positions_per_second": 76670.67574556493,
+      "required_pages_per_second_upper_estimate": 0.005304255248555815
     },
     {
       "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": 9592.838049052147,
-      "required_pages_per_second_upper_estimate": 0.0040163209699003825
+      "required_token_positions_per_second": 9583.834468195617,
+      "required_pages_per_second_upper_estimate": 0.004012551358611902
     },
     {
       "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": 19185.676098104293,
-      "required_pages_per_second_upper_estimate": 0.0039958703709320606
+      "required_token_positions_per_second": 19167.668936391234,
+      "required_pages_per_second_upper_estimate": 0.003992119954028019
     },
     {
       "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": 28778.514147156442,
-      "required_pages_per_second_upper_estimate": 0.0017252279808752647
+      "required_token_positions_per_second": 28751.503404586852,
+      "required_pages_per_second_upper_estimate": 0.0017236087281012336
     },
     {
       "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": 28778.514147156442,
-      "required_pages_per_second_upper_estimate": 0.002237402891635416
+      "required_token_positions_per_second": 28751.503404586852,
+      "required_pages_per_second_upper_estimate": 0.002235302925208331
     },
     {
       "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": 28778.514147156442,
-      "required_pages_per_second_upper_estimate": 0.3691398795186881
+      "required_token_positions_per_second": 28751.503404586852,
+      "required_pages_per_second_upper_estimate": 0.3687934147148812
     }
   ],
   "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": 0.1648581179296542,
-    "producer_p95_seconds": 0.19809514199914702,
-    "gpu_update_seconds": 0.9607167298014186,
-    "producer_batches_per_second": 6.065822008393333,
-    "gpu_batches_per_second": 1.0408895452530542,
-    "producer_headroom": 5.8275366834611155,
+    "producer_batch_seconds": 0.4443940963594173,
+    "producer_p95_seconds": 0.2081990679998853,
+    "gpu_update_seconds": 0.9616192799013492,
+    "producer_batches_per_second": 2.2502549160581546,
+    "gpu_batches_per_second": 1.0399125942052536,
+    "producer_headroom": 2.1638885119743136,
     "sustainable": true,
-    "buffer_seconds": 60.525153977489374,
+    "buffer_seconds": 60.582014633785,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375296.30858700874,
-    "upload_bytes_per_second": 4130574.7776954207,
+    "generation_bytes_per_second": 331605.5813873148,
+    "upload_bytes_per_second": 4116138.398159721,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.52985137183853,
+    "estimated_full_shards_per_hour": 71.15483838286002,
     "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": 12131753,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.6632540899991,
+      "bytes": 5726975,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -98,3 +98,4 @@
 {"step": 809800, "loss": 0.5648759603500366, "lr": 3e-05, "grad_norm": 0.34168568017784234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118531461221, "tokens_per_second": 150707.84311820107, "data_wait_s": 7.571001333417371e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 809900, "loss": 0.5182192325592041, "lr": 3e-05, "grad_norm": 0.26849332831787315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118546013215, "tokens_per_second": 150804.6009138714, "data_wait_s": 1.5571999028907157e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 810000, "loss": 0.5237100124359131, "lr": 3e-05, "grad_norm": 0.28498257341339567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118560875242, "tokens_per_second": 133640.02870263596, "data_wait_s": 1.4471999747911468e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 810100, "loss": 0.5151100754737854, "lr": 3e-05, "grad_norm": 0.27117276915334454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118575583382, "tokens_per_second": 146496.49573108696, "data_wait_s": 2.8355998438200913e-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 sha256:379f086bab2499dc7befb2b5f26bfa21abd1c8db0c750eeed806438851a3f139
-size 213531
+oid sha256:459b3efd25ec26a0a1825e1c79ec8b4e93b365fd1614214694949b5eb92694d3
+size 215649