Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75127.07173059873,
-      "required_pages_per_second_upper_estimate": 0.00651182416978579
+      "required_token_positions_per_second": 75399.57249007415,
+      "required_pages_per_second_upper_estimate": 0.006535443844970258
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9390.883966324842,
-      "required_pages_per_second_upper_estimate": 0.0037461640203944636
+      "required_token_positions_per_second": 9424.946561259268,
+      "required_pages_per_second_upper_estimate": 0.003759752098794985
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18781.767932649684,
-      "required_pages_per_second_upper_estimate": 0.0043448314633925
+      "required_token_positions_per_second": 18849.893122518537,
+      "required_pages_per_second_upper_estimate": 0.004360591027106267
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28172.651898974527,
-      "required_pages_per_second_upper_estimate": 0.0017155666901664307
+      "required_token_positions_per_second": 28274.839683777805,
+      "required_pages_per_second_upper_estimate": 0.0017217893901230784
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28172.651898974527,
-      "required_pages_per_second_upper_estimate": 0.0021688087300748985
+      "required_token_positions_per_second": 28274.839683777805,
+      "required_pages_per_second_upper_estimate": 0.002176675428622948
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28172.651898974527,
-      "required_pages_per_second_upper_estimate": 0.3498100487847142
+      "required_token_positions_per_second": 28274.839683777805,
+      "required_pages_per_second_upper_estimate": 0.3510788790714554
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11461328571851936,
-    "producer_p95_seconds": 0.24585257699800422,
-    "gpu_update_seconds": 0.9813772625716636,
-    "producer_batches_per_second": 8.72499199138149,
-    "gpu_batches_per_second": 1.0189761248182336,
-    "producer_headroom": 8.562508756461655,
+    "producer_batch_seconds": 0.06769961552336667,
+    "producer_p95_seconds": 0.24302303299919004,
+    "gpu_update_seconds": 0.977830477881102,
+    "producer_batches_per_second": 14.77113263154128,
+    "gpu_batches_per_second": 1.0226721529144172,
+    "producer_headroom": 14.443663679945148,
     "sustainable": true,
-    "buffer_seconds": 60.845390279443144,
+    "buffer_seconds": 60.62548962862832,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309650.50277249195,
-    "upload_bytes_per_second": 5932414.924067884,
+    "generation_bytes_per_second": 309563.81161967776,
+    "upload_bytes_per_second": 5994912.8714193,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.44378960019178,
+    "estimated_full_shards_per_hour": 66.4251876968646,
     "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": 0.8812644179997733,
-      "bytes": 1118965,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.006351252002787,
       "bytes": 5851026,
@@ -562,6 +556,12 @@
       "bytes": 2649444,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9738625030004187,
+      "bytes": 6071559,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -168,3 +168,4 @@
 {"step": 101800, "loss": 0.6054339203983545, "lr": 0.0003, "grad_norm": 0.14104400304823012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14896947164, "tokens_per_second": 148862.45410511692, "data_wait_s": 1.686799805611372e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 101900, "loss": 0.6645171958953142, "lr": 0.0003, "grad_norm": 0.13975844524589312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14911700228, "tokens_per_second": 149662.26764066803, "data_wait_s": 1.683399750618264e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 102000, "loss": 0.623924795538187, "lr": 0.0003, "grad_norm": 0.1287472829987993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14926361590, "tokens_per_second": 149002.16194981002, "data_wait_s": 1.3975997717352584e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 102100, "loss": 0.6249076779931784, "lr": 0.0003, "grad_norm": 0.1314781084798561, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14941062878, "tokens_per_second": 148819.64517445135, "data_wait_s": 1.5249999705702066e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:99eef3d5c49788a5adcc961826f3e37f7d70766db04d77688e5f9d35efe6bd59
-size 362574
+oid sha256:f3b1bdc0369777d1e3dbdf5241e39ef7c826ddf0e97f8af3247644136dfdbbd0
+size 364692