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": 77095.93980537626,
-      "required_pages_per_second_upper_estimate": 0.005333676002962128
+      "required_token_positions_per_second": 77105.51629499222,
+      "required_pages_per_second_upper_estimate": 0.005334338526734278
     },
     {
       "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": 9636.992475672032,
-      "required_pages_per_second_upper_estimate": 0.004034807506276851
+      "required_token_positions_per_second": 9638.189536874028,
+      "required_pages_per_second_upper_estimate": 0.004035308690804646
     },
     {
       "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": 19273.984951344064,
-      "required_pages_per_second_upper_estimate": 0.004014262776200836
+      "required_token_positions_per_second": 19276.379073748056,
+      "required_pages_per_second_upper_estimate": 0.004014761408760333
     },
     {
       "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": 28910.9774270161,
-      "required_pages_per_second_upper_estimate": 0.001733168952243134
+      "required_token_positions_per_second": 28914.568610622086,
+      "required_pages_per_second_upper_estimate": 0.001733384238216893
     },
     {
       "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": 28910.9774270161,
-      "required_pages_per_second_upper_estimate": 0.0022477013290001123
+      "required_token_positions_per_second": 28914.568610622086,
+      "required_pages_per_second_upper_estimate": 0.0022479805277295303
     },
     {
       "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": 28910.9774270161,
-      "required_pages_per_second_upper_estimate": 0.37083897624473905
+      "required_token_positions_per_second": 28914.568610622086,
+      "required_pages_per_second_upper_estimate": 0.3708850400921241
     }
   ],
   "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": 0.0769696649531113,
-    "producer_p95_seconds": 0.1849754730001223,
-    "gpu_update_seconds": 0.9563149523324003,
-    "producer_batches_per_second": 12.992131388504601,
-    "gpu_batches_per_second": 1.0456806071692744,
-    "producer_headroom": 12.42456950949406,
+    "producer_batch_seconds": 0.051265071757789826,
+    "producer_p95_seconds": 0.17385158500019315,
+    "gpu_update_seconds": 0.9561961782076598,
+    "producer_batches_per_second": 19.506458602548392,
+    "gpu_batches_per_second": 1.0458104966226158,
+    "producer_headroom": 18.6520011661227,
     "sustainable": true,
-    "buffer_seconds": 60.24784199694122,
+    "buffer_seconds": 60.240359227082564,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330658.769732098,
-    "upload_bytes_per_second": 5410771.937301664,
+    "generation_bytes_per_second": 337616.4693744961,
+    "upload_bytes_per_second": 5258899.775499597,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.95167464229779,
+    "estimated_full_shards_per_hour": 72.44463501859819,
     "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": [
@@ -364,6 +364,12 @@
       "bytes": 12213784,
       "pages": 48,
       "shards": 4
+    },
+    {
+      "seconds": 1.305295600000136,
+      "bytes": 1053111,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -72,3 +72,4 @@
 {"step": 767200, "loss": 0.47592201828956604, "lr": 3e-05, "grad_norm": 0.23324845216156045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112295468598, "tokens_per_second": 146341.6405724422, "data_wait_s": 1.0621000001265202e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 767300, "loss": 0.5454779863357544, "lr": 3e-05, "grad_norm": 0.25132323861385564, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112309975559, "tokens_per_second": 151041.8578830546, "data_wait_s": 1.0212000233877916e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 767400, "loss": 0.5483013391494751, "lr": 3e-05, "grad_norm": 0.24458661872707071, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112324634971, "tokens_per_second": 152686.65767280434, "data_wait_s": 8.879000233719125e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 767500, "loss": 0.529920220375061, "lr": 3e-05, "grad_norm": 0.24821814876417622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112339415179, "tokens_per_second": 153615.94682883014, "data_wait_s": 1.080899983207928e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3794fe06a499bd2af3539e191816e8c839007a6a8926ecc32cf14706e5c47b61
-size 157680
+oid sha256:6627e39fa04a75cea02ecbc0ff7487518a2bf3e0942bcd9aad8a157f7e16a8a2
+size 159798