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": 76490.54748296627,
-      "required_pages_per_second_upper_estimate": 0.005291793557394087
+      "required_token_positions_per_second": 76346.09181011362,
+      "required_pages_per_second_upper_estimate": 0.0052817997787628
     },
     {
       "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": 9561.318435370784,
-      "required_pages_per_second_upper_estimate": 0.004003124365856376
+      "required_token_positions_per_second": 9543.261476264202,
+      "required_pages_per_second_upper_estimate": 0.003995564294150635
     },
     {
       "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": 19122.636870741568,
-      "required_pages_per_second_upper_estimate": 0.003982740962328626
+      "required_token_positions_per_second": 19086.522952528405,
+      "required_pages_per_second_upper_estimate": 0.003975219385552893
     },
     {
       "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": 28683.95530611235,
-      "required_pages_per_second_upper_estimate": 0.0017195593227376615
+      "required_token_positions_per_second": 28629.78442879261,
+      "required_pages_per_second_upper_estimate": 0.0017163118613564834
     },
     {
       "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": 28683.95530611235,
-      "required_pages_per_second_upper_estimate": 0.0022300513576646243
+      "required_token_positions_per_second": 28629.78442879261,
+      "required_pages_per_second_upper_estimate": 0.002225839810225522
     },
     {
       "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": 28683.95530611235,
-      "required_pages_per_second_upper_estimate": 0.36792698023514775
+      "required_token_positions_per_second": 28629.78442879261,
+      "required_pages_per_second_upper_estimate": 0.3672321343850465
     }
   ],
   "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": 0.08709717692207164,
-    "producer_p95_seconds": 0.2504955990007147,
-    "gpu_update_seconds": 0.9638838055959599,
-    "producer_batches_per_second": 11.481428392274172,
-    "gpu_batches_per_second": 1.03746944828242,
-    "producer_headroom": 11.06676289242273,
+    "producer_batch_seconds": 0.06978306571861026,
+    "producer_p95_seconds": 0.25507720400128164,
+    "gpu_update_seconds": 0.9657075857055619,
+    "producer_batches_per_second": 14.330124217132418,
+    "gpu_batches_per_second": 1.0355101428238067,
+    "producer_headroom": 13.838709660587755,
     "sustainable": true,
-    "buffer_seconds": 60.72467975254547,
+    "buffer_seconds": 60.839577899450404,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322629.92995218607,
-    "upload_bytes_per_second": 3509116.292167665,
+    "generation_bytes_per_second": 321844.14356477564,
+    "upload_bytes_per_second": 3512872.2481645616,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.22887252735316,
+    "estimated_full_shards_per_hour": 69.06026106078579,
     "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": 1.2942262079995999,
-      "bytes": 1101377,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.815276637998977,
       "bytes": 10307802,
@@ -562,6 +556,12 @@
       "bytes": 7708812,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.0786898769983964,
+      "bytes": 798109,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -188,3 +188,4 @@
 {"step": 861300, "loss": 0.5280181169509888, "lr": 3e-05, "grad_norm": 0.2920627687392126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126069751600, "tokens_per_second": 152517.52313176062, "data_wait_s": 1.7241000023204833e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 861400, "loss": 0.5860153436660767, "lr": 3e-05, "grad_norm": 0.307238635792655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126084499905, "tokens_per_second": 152993.05664551342, "data_wait_s": 7.493003067793325e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 861500, "loss": 0.532051146030426, "lr": 3e-05, "grad_norm": 0.2918758091793663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126099099552, "tokens_per_second": 151780.9125423944, "data_wait_s": 1.42680000863038e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 861600, "loss": 0.49249354004859924, "lr": 3e-05, "grad_norm": 0.2872346917661992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126113604782, "tokens_per_second": 150743.31168528905, "data_wait_s": 1.1613999959081411e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c10bf6116d5c52c540c89b88fecb16b283bf558b34489ee32e842f11b60731a2
-size 405717
+oid sha256:9d473300ddc45495a351ed5378d7c5bc2fb79d3da41461dfd0eb0d96b05a2905
+size 407835