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": 77123.97976209222,
-      "required_pages_per_second_upper_estimate": 0.005335615872229402
+      "required_token_positions_per_second": 77149.02252247771,
+      "required_pages_per_second_upper_estimate": 0.00533734838850008
     },
     {
       "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": 9640.497470261527,
-      "required_pages_per_second_upper_estimate": 0.004036274974318862
+      "required_token_positions_per_second": 9643.627815309714,
+      "required_pages_per_second_upper_estimate": 0.0040375855844215
     },
     {
       "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": 19280.994940523055,
-      "required_pages_per_second_upper_estimate": 0.004015722772080973
+      "required_token_positions_per_second": 19287.255630619427,
+      "required_pages_per_second_upper_estimate": 0.004017026708722547
     },
     {
       "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": 28921.492410784584,
-      "required_pages_per_second_upper_estimate": 0.001733799309464604
+      "required_token_positions_per_second": 28930.88344592914,
+      "required_pages_per_second_upper_estimate": 0.0017343622876822397
     },
     {
       "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": 28921.492410784584,
-      "required_pages_per_second_upper_estimate": 0.002248518822737586
+      "required_token_positions_per_second": 28930.88344592914,
+      "required_pages_per_second_upper_estimate": 0.0022492489344132774
     },
     {
       "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": 28921.492410784584,
-      "required_pages_per_second_upper_estimate": 0.3709738511664112
+      "required_token_positions_per_second": 28930.88344592914,
+      "required_pages_per_second_upper_estimate": 0.37109430928193765
     }
   ],
   "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": 0.13586633139840387,
-    "producer_p95_seconds": 0.3301817600004142,
-    "gpu_update_seconds": 0.9559672650118944,
-    "producer_batches_per_second": 7.3601751788504375,
-    "gpu_batches_per_second": 1.046060923422475,
-    "producer_headroom": 7.036086535734084,
+    "producer_batch_seconds": 0.24535711135138172,
+    "producer_p95_seconds": 0.32691912899463205,
+    "gpu_update_seconds": 0.9556569557121611,
+    "producer_batches_per_second": 4.075691935286426,
+    "gpu_batches_per_second": 1.0464005875987101,
+    "producer_headroom": 3.8949633472964322,
     "sustainable": true,
-    "buffer_seconds": 60.22593769574935,
+    "buffer_seconds": 60.206388209866155,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 389228.54675722675,
-    "upload_bytes_per_second": 5111397.04724788,
+    "generation_bytes_per_second": 409598.8519661937,
+    "upload_bytes_per_second": 5026975.921434505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.51938535726167,
+    "estimated_full_shards_per_hour": 87.89037865867003,
     "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": 1.0794645450005191,
-      "bytes": 10910956,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.1706882670005143,
       "bytes": 13271387,
@@ -562,6 +556,12 @@
       "bytes": 7458410,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.3072389689987176,
+      "bytes": 4796842,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -352,3 +352,4 @@
 {"step": 639600, "loss": 0.5480508208274841, "lr": 0.0001, "grad_norm": 0.17035600415695004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93621211222, "tokens_per_second": 151809.97375823016, "data_wait_s": 1.2152995623182505e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 639700, "loss": 0.5964689254760742, "lr": 0.0001, "grad_norm": 0.1727498107872469, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93635917622, "tokens_per_second": 152252.92209369232, "data_wait_s": 1.6600999515503645e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 639800, "loss": 0.5595235824584961, "lr": 0.0001, "grad_norm": 0.1741859342505281, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93650638788, "tokens_per_second": 152903.79357081215, "data_wait_s": 1.479500497225672e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 639900, "loss": 0.5547839403152466, "lr": 0.0001, "grad_norm": 0.1854655156901785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93665222126, "tokens_per_second": 150913.60097193083, "data_wait_s": 1.2490003427956253e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:957bfdd2a0ae72ed938611b29a62864d277dc63297d38902dfd5f98b83982fd5
-size 755418
+oid sha256:07b61be8dc566ae2a9f4a4720c1a65e1fb24f5bc2f85e752b3cc48cd1cc56b23
+size 757536