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": 77263.33364209901,
-      "required_pages_per_second_upper_estimate": 0.005345256696993822
+      "required_token_positions_per_second": 77270.2600867669,
+      "required_pages_per_second_upper_estimate": 0.005345735884507511
     },
     {
       "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": 9657.916705262376,
-      "required_pages_per_second_upper_estimate": 0.004043568044258727
+      "required_token_positions_per_second": 9658.782510845862,
+      "required_pages_per_second_upper_estimate": 0.004043930538976456
     },
     {
       "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": 19315.833410524752,
-      "required_pages_per_second_upper_estimate": 0.0040229787066300895
+      "required_token_positions_per_second": 19317.565021691724,
+      "required_pages_per_second_upper_estimate": 0.004023339355570512
     },
     {
       "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": 28973.75011578713,
-      "required_pages_per_second_upper_estimate": 0.0017369320790866112
+      "required_token_positions_per_second": 28976.347532537584,
+      "required_pages_per_second_upper_estimate": 0.0017370877902547753
     },
     {
       "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": 28973.75011578713,
-      "required_pages_per_second_upper_estimate": 0.002252581629444182
+      "required_token_positions_per_second": 28976.347532537584,
+      "required_pages_per_second_upper_estimate": 0.002252783567171701
     },
     {
       "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": 28973.75011578713,
-      "required_pages_per_second_upper_estimate": 0.3716441568962318
+      "required_token_positions_per_second": 28976.347532537584,
+      "required_pages_per_second_upper_estimate": 0.37167747376941784
     }
   ],
   "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": 0.12284430739842378,
-    "producer_p95_seconds": 0.2965203729982022,
-    "gpu_update_seconds": 0.9542430610297574,
-    "producer_batches_per_second": 8.140385347744905,
-    "gpu_batches_per_second": 1.0479510313869767,
-    "producer_headroom": 7.7679062321938845,
+    "producer_batch_seconds": 0.10998530078916247,
+    "producer_p95_seconds": 0.3066025849984726,
+    "gpu_update_seconds": 0.954157523440593,
+    "producer_batches_per_second": 9.09212406407799,
+    "gpu_batches_per_second": 1.0480449773053235,
+    "producer_headroom": 8.675318579795274,
     "sustainable": true,
-    "buffer_seconds": 60.11731284487472,
+    "buffer_seconds": 60.11192397675736,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412577.6043317872,
-    "upload_bytes_per_second": 4677588.592484768,
+    "generation_bytes_per_second": 411694.7088502274,
+    "upload_bytes_per_second": 4636235.518210677,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.52954957452023,
+    "estimated_full_shards_per_hour": 88.34010075693242,
     "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": 1.173846097997739,
-      "bytes": 6578162,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.273192676002509,
       "bytes": 7543138,
@@ -562,6 +556,12 @@
       "bytes": 5890041,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.0961665050017473,
+      "bytes": 2428850,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -317,3 +317,4 @@
 {"step": 874200, "loss": 0.5261496901512146, "lr": 3e-05, "grad_norm": 0.3134866176176676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127958325044, "tokens_per_second": 151313.49266729451, "data_wait_s": 1.093399987439625e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 874300, "loss": 0.5033455491065979, "lr": 3e-05, "grad_norm": 0.30888795372106576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127972816590, "tokens_per_second": 150596.93019013514, "data_wait_s": 7.747999916318804e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 874400, "loss": 0.5264037847518921, "lr": 3e-05, "grad_norm": 0.3107110612346418, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127987419329, "tokens_per_second": 151784.92511335204, "data_wait_s": 1.385100040351972e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 874500, "loss": 0.5676658153533936, "lr": 3e-05, "grad_norm": 0.31331448117095145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128002110219, "tokens_per_second": 152396.89604623933, "data_wait_s": 1.6763002349762246e-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 sha256:1715c1fc43242ad541ca01b642e4a0297f2e164cd3a914b5871623dc3f009dea
-size 680505
+oid sha256:6f4cecd8b7dcda90f647743407a256d579a9f8dd5b2e1bffbda05d8e357d6f77
+size 682623